How to design a fast electric race motorcycle

General guidance and one project's case, kept apart on purpose. Only the first travels to your project unchanged.

This is a design guide, not a build manual and not a specification. It answers one question — in what order are you forced to make the decisions, and what does each one lock down? — rather than which parts to buy. Parts go out of date within a few years, while the order in which the decisions have to be made stays the same for much longer.

How the book is organised

Five parts, in the order the work has to happen. Part I is about deciding what to build: what “fast” means, how you know, and why the sequence of decisions is forced. Part II is the powertrain and the one system-level number, voltage, that most people learn about on track. Part III is the battery, and specifically heat, because heat is what ends a session. Part IV is the chassis, where the remaining lap time lives. Part V is making it work: settings, safety, and the method that keeps a three-year project's numbers trustworthy. The appendices close the book: a checklist of the order of work, the field-survey data behind chapter 01, an errata page with the open conflicts, and the list of references.

Two kinds of claim

The running text is general; it should hold for any electric race bike. Where a principle meets a concrete number from our own project, that is set apart so you can tell the two kinds of claim from each other:

Five block types, five colours: copper for our case, black for a rule, green for an explanation, red for a warning and grey for the limits of the evidence. A black-ruled block tagged Summary is the same type as a rule, used to close a long section in one paragraph. Beyond these there are charts, photographs and, in chapter 08, a strip of key figures; each carries its own caption.

How to read the numbers

Every figure carries a small reference key. You don't need to open the reference to follow the argument (the text is written to stand on its own), but the key is there so that when the figure matters to your own decision you can go and check it. The list of references at the back groups them by what kind of evidence they are: a measurement of our own bike, a configuration file as it actually ran, a manufacturer's statement and a thesis carry different weight, and you should be able to see which is which.

Numbers in this handbook use a decimal point, a thin space for thousands (1 080 cells), and lap times as minutes:seconds.hundredths (2:00.50). Where the project's own sources disagree with each other, the disagreement is recorded in the errata rather than resolved silently.

Every Mk4 lap time in this book is a prediction, not a result. Mk4 has not been built. Its lap times come from a lap model calibrated to one measured Mk3 lap, 2:00.50 at KymiRing, and two of them recur throughout: the baseline, 1:57.23, and the full potential, 1:53.03. Chapter 02 shows where both come from. The differences between Mk4 configurations are more reliable than any one of the absolute times.

The four bikes

Mk1, Mk2, Mk3 and Mk4 appear throughout this book as shorthand. They are four states of one machine rather than four separate builds. Students join and leave the project continuously rather than as a year group, and at each stage the part that gets redesigned is the one expected to give the most lap time, or the most safety, for the effort, while everything else is carried over. It is worth knowing which state is which before the examples start; the closing section describes how the project runs, and the J. Hyneman Center staff who have kept it going. TEAM

Mk1 (2019–2022) began at the invitation of Imatran Moottorikerho, the club behind Imatranajo, which wanted electric machinery at its event. The work started essentially from zero: nobody in the group had particular experience of motorcycles or of electric powertrains. That included the frame. Rather than fitting an electric powertrain into a donor motorcycle, the project designed its own frame at the J. Hyneman Center, so that the motor could be placed to leave the most room for cells and the frame's width matched to the pack; it was made from welded billet parts in 5083 aluminium by LUT's prototype-manufacturing unit, LUT Voima. The same frame carried Mk1, Mk2 and Mk3, and chapter 09 comes back to what that choice buys. TEAM It was raced at Imatranajo in July 2022 against Electric Superbike Twente, a University of Twente student team from the Netherlands, and lost. That team appears throughout this book as "the Dutch team"; its machine was the Delta XE in 2022 and the Vector-ST in 2024. The bike weighed 262 kg without a rider, and on a closed airfield it reached 240.4 km/h, the fastest any version of the bike has been, and a number chapter 06 comes back to.

Mk2 (2023) was a response to two specific complaints about Mk1: the inverter's heat and the mass. The pack was rebuilt far lighter and the inverter was replaced. It was tested at Motopark (a club circuit in Virtasalmi, eastern Finland, and the project's usual test track) in October 2023, and never raced: an error in the new inverter's settings prevented proper testing for most of the season.

Mk3 (2024) repackaged the battery again, changed the donor swingarm for one that allowed better suspension geometry and linkage, and lightened the rear subframe. Most of the measured data in this handbook comes from it, and it is the first version that did not have to be taken apart at the end of the season. It is also the one that reversed the 2022 result, at KymiRing in 2024; chapter 01 tells that comparison carefully.

Mk4 is in design as this is written. It is not a new bike so much as an optimisation of Mk3 for a ten-lap race with the technology now available: a 180S6P pack (180 cell groups in series, six cells in parallel in each; chapter 07 explains the notation), a higher-voltage inverter, and a liquid-cooled pack. CAD work has started and, like the rest of the design, proceeds in iterations. The ten-lap race is a design specification, not a booking. The bike has always been two things at once, a race machine and a demonstrator and teaching platform, and the two count equally, as the opening of this book says. The race specification is the hardest target the project has; the closing section explains what the platform is for.

Top speeds are the highest GPS reading from any logged circuit session, cross-checked against wheel speed, with GPS start-up artefacts removed: Mk1 at Imatranajo (1 July 2022), Mk2 at Motopark (8 October 2023), Mk3 at KymiRing on 9 August 2024, the session of the calibration lap HISTAiM. The three-generation figure in chapter 01 uses a different selection on purpose: the same circuit for Mk1 and Mk2 (Motopark), and for Mk3 the earlier KymiRing session of 13 July 2024, so its numbers are lower. Mk1's straight-line best, 240.4 km/h, was set on an airfield and is not a circuit figure.

The motor is the one thing that never changed: an EMRAX 268 MV has been in all three built so far, and Mk4 is designed around it too. That is not an accident of inertia (chapter 05 sets out why it survived a comparison with other candidates), but it does mean that everything this book says about batteries, voltage and cooling was learned around a fixed motor.

Three people have ridden the bike: Pauli Pekkanen on Mk1, Eemeli Lahti on Mk2 and Mk3, and Jere Honkanen as the reserve for straight-line testing. None of them rode it as a student of the university, and all three brought professional racing experience. Who they are, what they have won, and why an experienced rider and mechanic turned out to be the cheapest thing the project ever bought is told in the closing section. TEAM

Where this comes from, and what that limits

The handbook is written out of one project's material: three bikes built, one lap simulation calibrated against measured data, battery load tests, a dyno session, and the eighty-odd entries in the list of references, from PhD theses to manufacturer datasheets. The limits should be stated plainly. The lap simulation is calibrated on one circuit, KymiRing; the measured lap times and speeds come from several (Motopark, Imatra, KymiRing and an airfield), and each is labelled with where and when it was set, because they do not compare across circuits. All the battery experience is cylindrical cells. None of the comparisons are blind. It is one team's experience, not a literature review, and its weighting follows the mistakes: the case material is densest on heat, voltage and settings, because those three cost the most.

What transfers is wider than motorcycles. The numbers belong to one bike, but a good part of the method does not. The order of work in chapter 04, the requirement list, a reference against every number, an errata page and a written list of open questions (chapter 12), safety designed in rather than added (chapter 11), settings verified rather than assumed (chapter 10), and the paddock treated as part of the design all apply to any student engineering project that builds one machine over several years with people who come and go: a Formula Student car, a planetary rover for a university rover competition, a solar car or a rocket.

Keeping it current

The document is meant to grow with the project, by the four conventions set out in chapter 12, which are also those of the project's internal design handbook HB. The short version: every number carries a reference key, a number without one is marked as an estimate, and in the running text a number appears only in its current form. The value it replaced, and why, goes to the errata page and the decision log LOG.

“Fast” is three different design problems. On a circuit it means lap time, and lap time is not the same thing as power or top speed.

This chapter settles the metric. Everything after it is judged in seconds of lap time, so the first job is to be clear about what that measures and what it doesn't, and to see where an electric race bike actually stands against the petrol classes it shares a paddock with.

It comes in two halves. The first asks what “fast” is as a design problem; the second asks where the category actually stands, using the published field and this project's own three generations. Along the way it points forward a good deal: to the sensitivity list in chapter 03, the order of work in chapter 04, the motor in 05, voltage in 06 and mass and grip in 09. This chapter sets the questions, and the rest of the book answers them one at a time.

Three kinds of fast

You can design an electric motorcycle to be fast in three ways, and they give you three different machines:

The third is the hard one, because it is the only one where lasting is part of the metric. A drag bike is allowed to cook itself at the finish line. A race bike isn't, because it has to do the same thing nine more times. That single difference moves the centre of gravity of the whole design job from the powertrain to thermal management, and it is the main claim this handbook makes.

The circuit is a design constraint

Lap time isn't built the same way everywhere. The rough split:

• Grip-limited circuit. Most of the lap is spent in corners where the tyre, not the motor, sets the limit. Extra power has nowhere to go. Time comes from grip, mass and centre of gravity.
• Power-limited circuit. Long straights where the bike accelerates all the way to terminal velocity. Power, aero and gearing decide it.

Most modern circuits are the first kind. Which means adding power is usually the wrong answer — and it will still feel like the right one. That is exactly why the metric has to be measured rather than assumed.

What fast looks like in the field

The second half of the chapter. Everything from here on is measurement rather than definition: first other people's bikes, then our own.

It helps to know what you are aiming at, and the answer is that the field is very small. Purpose-built electric race motorcycles with published lap times are rarer than most people assume. As of 2026 there is essentially one active programme in public view, Lightfighter in MotoAmerica's Super Hooligan class. MotoE, which ran eighteen Ducati V21Ls for Dorna and Ducati Corse, was put on hiatus after the 2025 season MOTOE; its results remain the best-documented electric racing data there is, and this chapter uses them. Mugen's Isle of Man programme ended in 2019 when TT Zero stopped. Energica's MotoE era ended in 2022. Lightning is worth mentioning, but no comparable circuit lap times from recent years are available for it. Almost everything else is a university project. CMP

These bikes can't be ranked against each other, because they have never run on the same circuit. What can be done is to measure each one against the petrol classes at its own track on its own day, which is what the chart below does. That comparison turns out to be far more useful than a ranking, because the same pattern shows up every time.

Where that puts them on the petrol ladder

A gap to the premier class is abstract. It gets concrete when you ask which petrol class an electric bike would actually be racing. Most circuits run a ladder of three or four classes, and measuring against all of them rather than just the top one is what makes the position legible.

Read the last column, and read it across the three Ducati rows. The best-funded electric race motorcycle in the world and a Moto3 bike set nearly the same lap time: a percent or so faster at Assen and Mugello, on laps from the same meeting, and 2.6 % slower at Jerez, where the Ducati figure is a test lap set against Moto3's race record and is the weakest row in the table. Moto3 is the entry class of Grand Prix racing: a 250 cc single-cylinder bike weighing about 80 kg with 40 kW. The Ducati carries nearly three times the power and nearly three times the mass, and arrives at the same number. That comparison does more to locate electric powertrains than any of the others in the table.

The spread across the three circuits is itself informative, and it comes back below: the electric bike's deficit is smallest at the fastest, most flowing circuit and largest at the tightest one.

The same shape repeats. Mugen, which through the Isle of Man speed trap was quicker than a Supersport bike, finished 2.1 % off the Supertwin record. Ukkonen Mk3 is already faster than the Retro SBK pace at KymiRing and sits mid-field in Supersport. So the summary of where electric powertrains stand on a road circuit today is this: they place around the entry and lower-middle of the petrol ladder, not near the top of it, while carrying two to three times the power of the class they're matching.

The data behind this section is in appendix B: the reference times each percentage is calculated from, the official records and same-weekend laps at three circuits MotoE visits, how the surviving programmes reach their torque, and the pack architecture compared line by line with the MotoE bike. Its short version is two findings. Whichever baseline is used, MotoE sits seven to eleven percent off the front, so the choice of yardstick does not change the answer. And the size of the gap depends on the circuit in the way the mass argument predicts: smallest at fast, flowing Assen, largest at tight, stop-and-go Jerez. It tracks the number of hard accelerations per lap, which is the signature of a machine limited by acceleration rather than by grip or top-end power; the explainer later in this chapter has the algebra.

What's actually holding them back

More interesting than the size of the gap is where it is, and the answer differs from bike to bike, which is itself the finding.

• Mugen went through the Isle of Man speed trap at 176 mph (faster than the Supersport bikes) and still lost 7 % on lap time.
• Lightning is a maker of road motorcycles rather than a racing programme, but it has raced race-prepared versions of its bikes, and they are the sharpest version of the point. The LS-218 ran 351 km/h at Bonneville and won Pikes Peak outright in 2013, ahead of the combustion motorcycles LSWIKI. It is worth mentioning, but no comparable circuit lap times from recent years are available, so it has no place on the ladder above.
• Ukkonen Mk3 is the opposite case. At 232 km/h it is markedly slower than a Supersport 600 down the same straight (and that particular number is not an aerodynamic limit or a power limit but a voltage one, which chapter 06 takes apart), and it still lands mid-field in that class on lap time.

Seen from the road circuit, the first two are the category's stereotype: top speed in hand, lap time missing. The third case breaks it, and it is the most useful of the three, because a bike that gives away the straight and still runs the class pace has to be getting that time back somewhere. It is getting it back in the corners.

Our case Mk3 · where the lap actually goes AiMWB s0/s1CMP

Decomposed from the measured 2024 lap rather than assumed, Mk3 is three different bikes in one:

Phase	Measured	Verdict
Mid-corner, lateral	1.49…1.51 g	At or above class
Corner exit, longitudinal	0.62 g	Less than two-thirds of a 600
Straight	232 km/h	Slower than the class

So the bike corners well and accelerates badly. The lateral figure is measured in four corners, and 31 % of the lap is spent above 1.0 g; those are peaks, and the corner-by-corner 99th percentile is 1.38 g, which is what the simulation's 1.40 g is based on. The peaks are why the grip line in chapter 03 is an opportunity rather than a complaint: the tyre has already given more than the model assumes, briefly. A petrol 600 is at its wheelie limit around 1.0 g on the exit, where Mk3 manages 0.62. What it cannot do is use the tyre on the way out. Mk3 puts 579 Nm to the rear wheel, and the exit figure above is what that buys: two-thirds of a 600's. The same shortfall stated as a design target is in chapter 05's motor figure: Mk4 aims at 843 Nm and the tyre would take 1 127.

Which sharpens the diagnosis for the whole category. The deficit lies in acceleration rather than in cornering grip or in power as such, and mass is the biggest term in acceleration. MotoE's Ducati weighs 224.5 kg without a rider; a Moto3 bike weighs about 80 kg and has 40 kW, and with the whole MotoGP infrastructure behind it Ducati Corse beats that Moto3 bike by only a percent or so on a good day, against a machine with a third of its mass and a third of its power. When MotoE changed generations, the fix was 35 kg, not more power. CMPDUCATI

Lightfighter looks like an exception, and on closer reading supports the same conclusion. Its V3-RS is published at 181 kg ready to race, with 115 kW LFV3NEWATLAS, and its Super Hooligan version, the V3-RH, is the measured electric bike closest to the petrol classes: 1.5 % off Supersport at Barber, against a softer race-pace baseline. Take the mass problem away and most of the gap goes with it. It got there by packaging for its race length: pouch cells, a pack of about seven minutes at full throttle, and eight-lap races FARASISCMP. It is the balance between energy and mass that chapter 07 works through for Ukkonen's ten-lap specification, struck for a shorter race; appendix B has the detail.

The same argument, in one project's own history

Everything above is drawn from other people's bikes. A project's own generations are a better test, because the circuit, the logger and the processing can be held constant while the machine changes, and because nobody can accuse the result of being selected. This project has three generations of logged data and had never plotted them against each other.

The pattern is the one the rest of this chapter argues for, and it is stronger in the project's own data than in the published field. Three winters of work moved top speed by a tenth and cornering by almost a half. The generation that could not corner was also the generation that had the most power available on paper, because Mk1's pack was the largest of the three.

The other column: what the electric architecture gives back

This chapter has so far been an account of what an electric race bike gives away: the straight, the corner exit, the mass it carries to get there. Before it closes, the other column of the ledger: what the same architecture gives a race bike that a combustion one cannot have. Everything so far in this chapter is a deficit, and read on its own it makes a poor case for building the thing at all. That is not the whole picture, and the balance belongs here rather than in a chapter of its own — a book that lists advantages separately is advocating, and a reader is entitled to see both in the same place.

Some of what an electric powertrain gives back is worth lap time and some is not, so they are worth separating.

Read together, the two halves say something more useful than either does alone. The deficits are in the physical plant and the advantages are in the control of it. That is a real asymmetry, and it points the work: the seconds that are missing come back through mass, grip and heat, which are mechanical problems; what the architecture hands you for free is precision, which is only collected if somebody builds the controller to collect it. Neither half substitutes for the other, and a project that spends all its time on one of them is leaving the other on the table.

Put a number in seconds on every development item

The practical reason to do this is work order. A project always has more good ideas than hours. When every item carries a value in seconds, two competing ideas don't need an argument; the number settles it. Without one, priority gets decided by whatever is most fun or most obvious, and that is rarely what is fastest.

A simple point-mass model is enough, as long as it is calibrated against one measured lap. What the model produces is sensitivity, not an absolute lap time.

This chapter settles how you know. Every value in seconds in this book comes from a lap simulation, and the simulation is only as good as the one measured lap it was calibrated to. So the calibration point and the outside check on it are worth more than any amount of model sophistication.

What the model needs to be

A lap simulation doesn't have to be sophisticated to be useful. A point-mass model that knows mass, the torque curve, gearing, drag, a grip limit and the track geometry will give you sensitivities you can work with. Tyre models, suspension kinematics and a rider model sharpen the absolute number, but they rarely change which of two development items is bigger, and that is the only thing the model is used for. If you want a worked example of the full version, a validated electric race motorcycle model with a Pacejka tyre model, an entropic battery thermal term (the small reversible heat a cell produces or absorbs as it charges and discharges, separate from ohmic loss) and motor and inverter loss maps exists in the literature FLA, and it is also where usable starting values for drag and mass sensitivity come from.

Calibration is the whole value of the model

An uncalibrated simulation just tells you the consequences of your own assumptions. Calibrating means forcing the model to reproduce one real, measured lap, and solving the coefficient that cannot be set any other way from that: here the rider factor. The rest are set from measurements or stated as assumptions (grip from the logged lateral acceleration, drag from the energy balance on the long straight, drivetrain efficiency assumed). Only after that does a predicted change mean anything.

Two things follow. First, the calibration point is sacred: change it and every prediction moves with it. Second, only a measured clean lap will do, not an official result. Official times include traffic, flags and a standing start. The model has none of those.

Validate against somebody else's measurement

Your own calibration tells you the model reproduces your bike. It doesn't tell you whether the model's sensitivity is right: whether it predicts the size of a change. For that you need a case where somebody else made a change and measured the result.

What the model predicts for Mk4

Mk4 has not been built, so it has no lap time. What it has is a prediction, and the chain behind it is short. The logged Mk3 lap of 2:02.588, less the traffic, is the 2:00.50 calibration lap. The model fitted to that lap is given the Mk4 configuration in place of Mk3's. The simulated lap, multiplied by the same rider factor, is the Mk4 figure. Every Mk4 lap time in the rest of the book is that chain, run on a different configuration.

Put against the field of chapter 01, the prediction looks like this.

The two Mk4 rows carry everything that is uncertain about the calibration: one lap, one rider, one day, and a rider factor assumed to be constant. They say where the configuration would sit if the model's absolute level is right. The step from one row to the next is the safer reading.

Grip, then the braking limit, then mass. Power is worth something on a single lap, and the race, not the lap, decides how much of it you can use.

This chapter settles the priority order. Once you have the numbers in seconds, the ordering usually comes out the same on any grip-limited circuit: grip first, then the two ways of using the tyre harder in braking, then mass, then a long tail. The gap between the top and the bottom is two orders of magnitude.

One term needs fixing before the chart. The braking limit is the hardest deceleration the simulated bike is allowed to use. It is capped by the rear wheel lifting — the stoppie limit, which a lower centre of gravity raises (chapter 09 has the geometry) — and extended by letting the rear wheel do some of the braking through regeneration. That is why the two appear in the chart as one bar.

The baseline is Mk4 at 1:57.23, the prediction defined in chapter 02. Applied in order rather than one at a time, the top of the list is the development path the project is working to, and its end point is what this book calls Mk4's full potential:

Why the power cap stays at 135 kW

On a grip-limited circuit much of the lap is spent where the tyre is already at its limit: corner entry, mid-corner, corner exit. Extra power does nothing there, because the grip is spent. On Mk4's locked configuration, with an overall ratio of 2.50 and an rpm ceiling at 264 km/h, about 30 % of the lap is limited by the power cap, and there 25 kW more is worth 0.44 s on a single lap SIM. So for one lap, more power helps.

The reason not to take it is the race. The 135 kW cap is sized for ten laps: at that power the cell current and the heat stay within limits and the energy lasts to the flag (chapters 04, 07 and 08). More power is a qualifying setting, not a race setting. On an electric bike, what the pack can deliver for the whole race sets the power more than what power is worth on one lap.

Aero is worth little for a related reason. Cutting drag only buys time if you have rpm headroom left on the straight. If the bike is already against its speed ceiling before the braking marker, a better CdA just gets you to the same ceiling sooner. For reference, a coast-down measured CdA of 0.335 and a sensitivity of 0.22 s per CdA point on a comparable machine FLA are the right order to sanity-check your own figure against.

The list of things that don't matter is just as useful

Write that one down separately and keep it visible. Its job is to stop work drifting into things that feel productive. On this circuit it is three items long: aerodynamic drag 13 % lower, 0.11 s; moving the gearing anywhere between 2.4 and 2.7, 0.05 s at most; rolling resistance 30 % lower, 0.01 s. Together they are worth under two tenths of a second, less than a tenth of what grip alone is worth. Without the list, a team spends the winter on aerodynamics and gearing and collects a tenth or so for the trouble.

Some choices are the boundary conditions for others. Done in the wrong order, they get done twice.

This chapter settles the sequence. There is a chain in electric powertrain design where each link constrains the next. You can walk it in any order you like, but only one order avoids throwing work away. The classic expensive mistake is picking the cell and pack size first, which feels natural because the battery is the biggest single purchase, and only later discovering that the inverter's voltage ceiling, or the number of cells the BMS (the battery management system that monitors every cell) can handle, dictates the series count, at which point the entire pack geometry gets recalculated.

The chain above is the one you can draw. There is a second kind you can't.

The sequence in the diagram is a chain of specifications: each link hands the next one a number, and it can be drawn in advance because every quantity in it belongs to somebody. The couplings that actually hurt are the other kind: where a decision in one subsystem silently removes an option in another, through geometry rather than through a number. Nobody writes those down, because at the moment the first decision is made the second subsystem has no reason to object. They surface in the workshop.

There is no general method for finding them in advance. There is a partial one, and it is cheap: whenever a decision fixes which physical face of a component gets used for something, ask what else that face was for. Cooling, current, mounting and sealing all compete for the same surfaces, and a component has only so many.

The same order, turned into twenty questions with what each one locks, is the checklist in appendix A. It is the page to keep open while you work.

Mk4 as it stands: the requirements, and what is locked

Two tables close the chapter, because the rest of the book keeps returning to them. The first is the set of requirements the Mk4 design is sized against; each of them moves the pack, the cooling or both, so each carries where it came from and what changing it would cost. The second is the configuration as it is currently locked, gathered from the chapters that decided it. Locked means it is not reopened without new measured data; to confirm means the choice is made but waits for a check; open means the question is still live.

A single-speed electric bike is described by one product, not by peak power and peak torque separately.

This chapter settles what a motor is chosen on. On a combustion bike the gearbox decouples engine speed from road speed. An electric bike usually has no gearbox, so the motor's whole operating range has to cover the whole speed range on one ratio. What matters then is the torque you actually have at the rpm you are actually at. Integrated across the range, that collapses to the product of torque and usable rpm.

The four numbers that decide a motor

There's more than one way to reach the same product

Torque × rpm can be arrived at from either end, and the motor's design decides which end you get. A high-speed machine gets there with modest torque and a lot of rpm; a large-diameter, low-speed machine such as the EMRAX gets there with a lot of torque and a modest ceiling. Both ends exist within the same topology: every motor in this chapter's comparison is axial-flux, and they range from 5 500 to 14 000 rpm. Both can produce the same figure, but they hand you completely different downstream problems: gearing, driveline strength, bearing loads, and how much of the corner you can actually use.

A datasheet is not a measurement

A permanent-magnet machine's torque constant depends on magnet flux, and flux falls with temperature. The datasheet K_t is typically a cold figure. If you tell the inverter the wrong flux, it doesn't just report torque wrongly: it computes the torque command wrongly, so the error is in the control loop, not only in the telemetry.

The rpm ceiling: an assumption to own

A datasheet's speed limit multiplies straight into the torque × rpm product and therefore into lap time, so it is worth establishing rather than assuming. It is also worth understanding before you go near it, because this is the one place in the chapter where the engineering question and the responsibility question are not the same question.

Choosing between motors: the tyre decides before the motor does

Comparing candidate motors looks like a torque contest, and for the first part of the range it is. Then it stops being one. Tractive torque at the rear wheel has a ceiling that has nothing to do with the motor: the tyre and the wheelie limit. Past that point extra torque is not slow to arrive or expensive — it is unusable, and a comparison that keeps ranking motors by it is measuring something the bike cannot spend.

So a motor comparison has three questions in it, and they are worth separating because they have different answers:

1. Does it reach the tyre? Torque × gearing against the traction limit at your mass. Below the traction limit, more torque turns into lap time; once the tyre is at its limit, extra torque adds nothing.
2. How high does it rev? This is the one that survives saturation. Once every candidate saturates the tyre out of slow corners, the only remaining difference is how much of the speed range it covers before it runs out of revs, and that difference only shows up when you have power to spare.
3. What does the same lap time cost elsewhere? Mass and grip buy seconds too, usually at a lower price than a motor change, and they do not restart the inverter characterisation.

The numbers a datasheet does not print

The second lesson from the same exercise is about elimination. A motor is rarely separated from its rivals by its headline figures. It is separated by a number the datasheet does not print in a usable form: the torque constant at the current you will actually run, whether the speed limit is a hard ceiling or a life rating, what the insulation system is qualified for. The way to get that number is to ask. Write to the supplier with short, specific questions; manufacturers generally answer clearly and quickly, and they are the only source of reliable information about their own product.

Three habits make the answers worth more. Ask the disqualifying questions before the interesting ones, because one figure can settle a selection before a long letter is ever needed. File the answer in the project's decision log, not in somebody's inbox, and respect it if it was given in confidence: an answer written for your project is not yours to publish. And treat it as the manufacturer's statement, not a measurement. This chapter opens by saying that a datasheet is not a measurement, and an email is not one either; if a candidate is close enough to matter, bench it before the choice is final.

The right answer is a winding you specify — and most projects can't have it

Everything above treats the motor as something you pick off a shelf, but a motor can also be wound to order. A permanent-magnet machine's K_v is set by turns per coil, and the insulation system may be the same across a manufacturer's low-, medium- and high-voltage variants, with only the turn count and wire gauge changing. Whether that is so for a given motor is a question only its manufacturer can answer, and it cannot be assumed. In principle, then, the correct move is not to pick the winding closest to your bus but to have the winding wound for your bus, so the motor reaches its rpm ceiling exactly where your pack runs out of volts and no current is wasted on field weakening — the trick, explained in chapter 06, by which an inverter buys rpm above the pack’s natural limit at the cost of current that makes no torque.

In practice this is available to very few projects, and it is worth saying why:

• Somebody has to agree to wind it. That means a manufacturer with the appetite for a one-off, which in practice means a commercial or sponsorship relationship rather than a purchase order.
• You inherit the qualification question. A non-standard winding on a higher bus raises volts-per-turn, and turn-to-turn insulation is what that stresses. The failure mode is slow partial-discharge erosion, so it shows up mid-season rather than on a test bench, which means it cannot be dismissed by "it ran fine on the dyno".
• It restarts the inverter work. A stock motor that your inverter already has a parameter set for is worth real weeks.
• Lead time and one-off price are both unknown until you ask, and both land on the critical path.

So the practical order is: take the standard winding that best matches the bus you can build, and treat a custom winding as a question you ask early, because the answer, whichever way it goes, changes the pack.

Gearing: the optimum is flat

Optimising the ratio on a single-speed machine usually returns very little. Sweeping overall ratio across a sensible window moves lap time by hundredths, because shorter gearing buys acceleration and gives back top speed in almost equal measure. That is good news: you get to choose the ratio on mechanical grounds (chain tension, chain speed, sprocket diameter, what you can actually make) because the calculation isn't imposing a tight requirement. What the calculation does care about is efficiency: on a comparable bike two percent of driveline loss is worth about 0.4 s a lap, more than the whole gearbox weighs in lap time. WB s15

Pack voltage is less a battery parameter than the performance ceiling for the whole machine, and it is the one most people learn about on track.

This chapter settles the series count, which is the same thing as settling the voltage. Back-EMF in a permanent-magnet machine rises with speed. When it reaches the available pack voltage, the inverter can no longer push current into the windings normally. That point is the base speed:

Above base speed the inverter can keep going with field weakening: it injects negative d-axis current that opposes the magnet flux and lets the machine turn faster. It works, and the price is direct — d-axis current produces no torque. It heats the winding and the inverter, and it eats part of a current budget that would otherwise have been pulling you forward.

So the voltage is chosen for the motor, or the motor is wound for the voltage. What decides is where the motor's base speed lands against the rpm the gearing needs at top speed, and that is set by K_v and pack voltage together. A motor with a low K_v needs a high voltage, which is Mk4's case below; if the voltage is fixed by other constraints, the motor has to be wound to suit it (chapter 05, last section). The voltage class on its own buys little. An 800 V system can in principle be faster, because the same power at half the current lets cables, busbars, connectors and fuses be lighter, but against a well-matched 400 V system the project's estimate is that the difference is small TEAM. The project's own simulation points the same way without isolating it: two Magelec variants, one per voltage class, on the same cells rewired from 180S6P to 108S10P, differed by 0.02 s (chapter 05), a comparison that changes the motor as well as the voltage. Lightfighter's current V3, one of the fastest electric race motorcycles in the field of chapter 01, runs a 400 V-class pack: 383 V nominal, 440 V peak RIDEAPART.

Three ceilings decide the series count

Series count is set by the lowest of three independent ceilings:

1. The inverter's voltage window. The upper bound is usually the trip voltage rather than the nominal maximum, and a fully charged pack sits well above nominal. You need the lower bound too: an empty pack under load must not drop out of the window.
2. The BMS cell count. Surprisingly often this is the binding one, and it is completely hard — a BMS has a fixed maximum, and frequently a connector-group structure that favours certain multiples.
3. Insulation. Motor winding class, connectors, cables and measurement circuits.

Check rather than assume on the third one. Motor insulation may be specified the same way across a manufacturer's range regardless of winding voltage, in which case it isn't the constraint at all; ask the manufacturer, because it varies.

Two of those three ceilings are only ceilings because you bought them

Worth saying plainly, because the table above reads as physics and only one row of it is. The inverter's voltage window and the BMS's cell count are product limits, not natural ones. Design your own battery management electronics and the 180-cell maximum disappears, along with the constraint it puts on module size and series count — exactly as a motor wound to your own specification removes the K_v constraint in chapter 05. Both routes are available, and both have a cost.

What they cost is not money, it is what the project becomes. A team that designs its own BMS is a team doing BMS development, with its own validation, its own failure modes and its own share of the season; the same is true of winding a motor. That may be the right choice; for a research group it can be the whole point. But it has to be chosen rather than drifted into, and the question to ask out loud is the one this book keeps returning to: does this work produce lap time, and is it the cheapest second available? For this project the answer was no on both counts, so the ceilings stayed where the catalogue put them.

One consequence is worth keeping even if you never build your own: which supplier's limit binds is a purchasing decision and can be revisited. Ours happens to be an Orion, and if a better or more workable unit appears the ceiling moves with it. A ceiling with a part number in it is negotiable in a way that K_v × U is not.

One consequence of the series count: the 12 V supply

The series count also decides the low-voltage side. The 12 V supply looks like a catalogue item and usually isn't: a traction pack swings widely between empty and full, and a fixed-ratio converter alone can't hold a regulated low-voltage output across that range. The usual answer is two stages (an isolating fixed-ratio stage from HV down to a safe level, then a regulating stage), and the selection criterion on a race bike is mass, where the difference between module types is not marginal.

The same ceiling, three bikes and two inverters apart

A diagnosis is much stronger when the same mechanism shows up on hardware that shares almost nothing. This one does: only the motor is the same, and its K_v is the point. It was in the project's logs from the first season.

The series/parallel split doesn't change per-cell current for a given cell count. That is the most common misconception in pack sizing.

This chapter settles the cell count. Once the series count is locked, sizing the pack is the intersection of three requirements: energy (how far), current (how hard) and heat (how long). On a race bike the third is usually the binding one, even though the first two are the ones you calculate first.

Energy budget

Energy comes from per-lap consumption: kWh per lap × number of laps, minus whatever regeneration gives back. You can get per-lap consumption from the lap simulation or straight out of measured ride data, and the second is better.

Leave margin at the bottom. A pack that is emptying sags under load, and if it sags below the inverter's lower limit or the BMS undervoltage cut, the session ends before the energy does.

Per-cell current doesn't depend on the S/P split

This is worth deriving once yourself, because the result is counter-intuitive. For a given power and a given number of cells, per-cell current is the same regardless of how you split them into series and parallel. A higher series count raises voltage and lowers total current, but it removes parallel branches in exactly the same proportion. Net result: per-cell load, and therefore I²R loss per cell, is set by the cell count — not by the split.

Two things follow. A freedom: choose the S/P split on voltage and BMS grounds, because it doesn't cost you heat. And a hard constraint: if the heat doesn't fit the budget, the only fix on the battery side is more cells. Rearranging them won't do it.

Cylindrical or pouch

Several of the machines this book compares itself against run pouch cells (the Dutch team's 2022 Delta-XE MOTORRAD and Lightfighter FARASIS among them), while Ducati's MotoE bike, like Ukkonen, uses cylindrical 21700s; Ukkonen has run cylindrical 18650 and then 21700 cells through its three built generations, and Mk4 is designed on 21700s. The reasons were practical, not electrochemical. Availability: automotive-grade pouch cells were hard for a small team to buy at all, and hard to buy in small numbers; 21700 cells of known type were on a web shop with a delivery time of days. Fit: the standard pouch formats on offer did not fit the frame's dimensions, whereas a pack built from cylindrical cells can be shaped, in steps of one cell, around whatever volume the chassis leaves. Neither reason is a verdict on the chemistry. They are a verdict on the project's purchasing position, and a team with an automotive supplier should re-run the comparison.

The trade itself, in outline. Pouch: higher energy and power density at the cell level, a large flat face that suits plate cooling, and tabs that carry high current without welding many small joints; against that, the need for compression fixtures and a rigid enclosure, swelling over life, a vent that is not a defined feature, and a format that dictates the pack's shape. Cylindrical: a steel can that is its own structure and its own vent, a format that can be shaped around the bike and bought in any quantity, and a failure that tends to stay in one can; against that, a lower packing density, thousands of welded joints (the interconnect problem later in this chapter), and a heat path that has to leave through the can's ends or its curved side rather than a flat face. On this project the joint count and the heat path are the two costs that have been paid for choosing cylindrical, and both are visible in this chapter and the next.

Power cell or energy cell

Before the cell choice, one framing from outside the project, because it is easy to get backwards. Capping power costs very little lap time on a grip-limited circuit, and people who have raced these machines for a long time treat that as the design margin rather than as a compromise. The outside reviewer of chapter 12 made the same point: it is better to have more power on tap than you need and turn it down than to chase every last bit out of a powertrain that has just enough; and, concretely, some Isle of Man machines ran a peak of around 80 kW purely so the riders would reach the end REVIEW22. Size the powertrain for headroom and the battery for the distance; the cap is a setting.

Cylindrical cells split roughly into power and energy types. A power cell gives a higher continuous C-rate and lower internal resistance but fewer watt-hours; an energy cell the reverse. On a race bike the choice usually falls out of the energy budget: if ten laps don't fit, a power cell doesn't help, because the pack is empty before its current capability ever becomes the limit.

Three numbers are worth measuring rather than reading off a datasheet: internal resistance at operating temperature rather than room temperature, capacity at the real discharge current, and the cell's actual outside diameter. The last one sounds trivial. It is the start of the dimension chain.

Joints: nickel for weldability, copper for conductivity

Cells are normally joined by spot-welding nickel strip to the terminal. Nickel welds well and conducts badly: hot, its resistivity is roughly four times copper's. In a high-current pack, nickel alone won't do as a collector or as a series link.

The division of labour that works is nickel from cell terminal to collector (low current, weldable, and the strip's cross-section can be sized so it also acts as the per-cell fuse) and copper as the collector and series link, where the current actually is. The nickel strip is soldered onto the copper; it adds a few percent to conductivity, so it isn't a conductor, it is a joining surface.

Which welding process to use is a separate question with a good published answer: ultrasonic, laser, micro-TIG, resistance and mechanical joints have been compared with manufacturing-readiness ratings and separate decision matrices for cylindrical, pouch and prismatic cells DAS. Worked pack designs from student and record projects are worth reading alongside it, less for their numbers than for the checklist of what a pack design has to demonstrate PACKS.

Why not aluminium

On the properties you can look up, aluminium is the obvious answer and it isn't close. Per kilogram it conducts about twice as well as copper and seven times as well as nickel, and mass is the thing this whole book is trying to remove. A selection table built from resistivity and density will choose aluminium every time, and the table will be arithmetically correct.

It is still the wrong answer, because the decisive property isn't in the table. A cylindrical cell's can is typically nickel-plated steel, so a busbar-to-can weld is not an aluminium joint: it is a dissimilar joint involving aluminium, nickel and steel. Aluminium to steel is the hard pair. A fusion weld between them forms Al–Fe intermetallic compounds at the interface, among them Fe₂Al₅ and Fe₄Al₁₃ (also written FeAl₃). These aluminium-rich phases are hard and brittle, and a thicker layer generally means a weaker joint ASIRVBEYGI. Fe₂Al₅ is reported as the thermodynamically stable phase; which phase forms first depends on the thermal and compositional conditions, and the literature reports both orders ZHANGBEYGI. The layer limit usually quoted is about ten micrometres, and it is a rule of thumb rather than a universal limit: the optimised busbar welds in one study held 5–20 µm, and 5–10 µm over part of the interface ASIRVZHANG.

Copper to steel avoids that reaction system: copper and iron form no harmful intermetallics ASIRV, although the pair has difficulties of its own: copper and iron have limited mutual solubility, and laser-welded steel–copper joints segregate and are prone to solidification cracking CUFE. Nickel is the compatible metal: it is soluble in both iron and copper, which is why it is the usual intermediate layer when copper is welded to steel TMAT.

Nickel-plated aluminium busbars are used in battery interconnects, and the plating changes how they weld. One study laser-welded nickel-plated 1 mm aluminium sheet onto unplated aluminium sheet, an aluminium-to-aluminium joint rather than one to a steel can. The lap-shear strength was 840 N without plating, about 1.1–1.2 kN with the thinnest plating (3–4 µm) and about 1.6 kN with 15 µm, while thicker plating generally raised weld porosity. For that sheet the authors recommend sulfamate nickel under about 4 µm as the compromise between strength, porosity and conductivity NIPLATE. The study does not test a joint to a cell can, so it says nothing directly about the aluminium-to-steel problem above. What it does show is that where aluminium busbars are used, the nickel has not gone away: it has moved to a plating line.

On a race bike you rarely run out of energy. You run out of heat capacity, and you run out of it mid-session. It is also where Ukkonen lost the most.

This chapter settles the cooling requirement and how to meet it. A typical electric race session ends with a power limit rather than an empty battery: the BMS or the inverter sees an over-temperature and pulls the current limit down, and the bike slows progressively. To the rider it feels like the bike dying. In reality there is plenty of energy left.

The project's first race tells the same story at a higher price, and it is told later in this chapter, after the method, where it is easier to read, because its whole point is a temperature limit that nobody had decided.

The requirement is derived from a temperature limit, not chosen

The cooling requirement falls straight out of three numbers: cell dissipation at race load, the cell's maximum allowable temperature, and the coolant temperature. The rest is division.

Dissipation comes from per-cell current and measured internal resistance: P = I²R. Do it with the RMS current across the whole lap, not peak current: the peak lasts seconds, and thermal equilibrium settles in minutes.

One term is deliberately left out of that number, and it is worth saying so rather than letting a reader wonder. Regeneration adds heat — it can never subtract it, because ohmic loss does not care which way the current flows. On this pack it comes to one or two percent of the lap's cell heat at a 15 % recovery, and it is very nearly cancelled by the fact that recovering energy keeps the pack at a higher voltage for the whole race, which lowers the current everywhere else. Chapter 10 does that arithmetic. The conclusion for cooling design is that the requirement above does not need a regen allowance, but the charge current limit does need checking, because that is what really caps recovery.

Draw the thermal chain and find out where the resistance actually is

It is easy to spend most of your cooling effort on the link that is easiest to buy — the thermal interface material. Draw the chain out and put a K/W on every link, and the weighting usually turns out to be somewhere else entirely.

Read the chain against the requirement. The chain adds up to 2.64 K/W in the cooling document (2.61 for the four links in the figure), which is 0.38 W/K: about twice the 0.187 W/K the requirement asks for. At 5.62 W the core of an average cell then sits 14.8 K above the coolant, near 45 °C against a 60 °C limit. That margin is not spare, for two reasons. The coolant warms by about 8 K along the pack and cells at the edge run differently from those in the middle, so the hottest cell sits 8–12 K above the average, and the 60 °C limit applies to the hottest one CC §02; counted that way the requirement for the average cell is closer to 0.26–0.31 W/K. And the chain was computed for a contact area the channel does not have: drawn with its bond flange, manifolds and run-out, the channel touches 58 mm of the cell's 70 and the contact area is 17 % smaller, which moves the estimate about 15 % down CC §07. The design's range then becomes roughly 0.21–0.44 W/K around a central 0.32. The middle just meets the hottest-cell requirement and the bad end clearly does not. The cooling document calls this the largest single open item in the design; it is why the first module is measured before the others are built, and why chapter 12 uses this range as its example of a number that is not ready to lock.

Opening the chain up moves the design effort in two directions. The coolant boundary layer depends on channel height: in laminar flow, heat transfer goes inversely with channel height, so halving the channel doubles the heat transfer. And because pressure drop is typically a small fraction of what the pump can deliver, thinning it is close to free. The cell's shrink sleeve, on the other hand, is a given: you can't design it, but its thickness varies between cell types and it is worth actually measuring.

The third link, the cell's own radial conductivity, is the least certain number in the chain. Published measurements on cylindrical cells range from 0.15 to 2.6 W/mK depending on method, with careful measurements on 18650s clustering around 0.5–0.7; the cooling design uses 0.83 as an upper-end assumption KOLLER. That spread alone moves the core-to-surface link by a factor of two, which is one of the three reasons the first module has to be measured before the rest are built.

Parallel paths don't share flow evenly

This is the most common and most expensively learned mistake in cooling design, and it applies equally to air and liquid. When one manifold feeds many parallel channels, flow distributes according to pressure drop. If an individual channel's own pressure drop is small compared to the manifold's, the manifold decides the split — the nearest channels take everything and the far ones run dry.

The fix is counter-intuitive: deliberately add resistance. A restrictor at each channel inlet, with a pressure drop clearly larger than the channel's own, makes the channels look identical from the manifold's point of view, and the flow splits evenly.

Why not just run it hotter

The question always comes up: raising the temperature limit would relax the cooling requirement directly and save kilograms. The answer has three parts, and only one of them is about performance.

• Ageing. The Arrhenius relationship means a 40-degree increase multiplies the ageing rate, typically by six to ten.
• The feedback is positive. An aged cell has higher internal resistance, which produces more heat, which ages it faster.
• Safety margin. This one is a different category from the other two. Ageing is money; margin to the onset of an exothermic reaction is the rider. A thousand cells a millimetre apart means that if one goes, it propagates.

And the cell core is always hotter than its surface. If you measure at the surface and the limit is in the core, that difference is part of the sizing; it isn't something you can correct afterwards.

What happens when nobody decides the limit

Everything above assumes that somebody has chosen the temperature limit and written it down. The project's first race is what it looks like when nobody has. None of it is about a component failing. It is about two numbers, a BMS cut-off and an inverter temperature, that existed in the machine and in the paddock notes and were never compared to anything.

Our case Mk1 · fastest in qualifying, a minute and a half down in the race RACE22DIARYHISTLUT

The project's first race, Imatranajo 2022, is the cleanest statement of this chapter's argument that its own material contains. It is a result sheet, not a calculation.

	Qualifying best	Race best lap	Three laps	Spread
Winner · Delta XE	2:26.923	2:28.783	7:30.201	1.3 s
Ukkonen Mk1	2:20.768	2:32.299	9:04.667	29 s
Gap	−4.2 %	+2.4 %	+21 %

Mk1 qualified as the faster machine, and then lost the race by a minute and a half. The comparison worth relying on is the one that stays inside a single machine, because it assumes nothing about anybody else's day: Mk1's own qualifying best was 2:20.768 and its own best race lap was 2:32.299, 11.5 s slower, same bike, same rider, same circuit, a few hours apart. The winner's equivalent figure was 1.9 s, and its three race laps were within 1.3 s of each other where Mk1's varied by 29. The cross-team qualifying gap is weaker than it looks (the winner completed one timed qualifying lap in the whole weekend, so 2:26.923 says little about that machine's pace), and the chapter does not lean on it. The lap and race times are official timing; the competitor's power, mass and voltage figures used elsewhere in this book are that team's own published claims, and are treated as claims.

What the log says. Over the race the motor climbed from 48 to 93 °C. The inverter's reported temperature did not climb at all: it swung between about 50 and 109 °C on every straight, up under load and back within seconds — the "40 degrees up and down within one second" the team saw in the paddock tool, and the same 102–110 °C the inverter had reached at Motopark in June and would reach again at the airfield in September. In the inverter's own terms, its current target fell from near 300 A to 150 A within six to nine seconds of full throttle, so every straight began at full current and ended at half; the team's estimate that at most half the bike's power was ever in use is consistent with the pack side, where peak power fell from 87 kW in the first minute to 73–79 kW in the last. The pack's hottest cell went from 38 °C on the grid to 50. The ambient was over 30 °C, exceptional for a Finnish July, and every temperature in the log starts from it. HISTDIARY

And then the pack reached 50 °C. At 430 s into the race log, accelerating through 130 km/h with 209 A flowing, the pack current went to zero within a second and the inverter's status word changed, at the same moment the hottest-cell channel ticked from 49 to 50. The BMS's upper temperature limit was 50 °C, and it opened the circuit with the throttle wide. The bike coasted to a halt. About a minute later the channel read 49, the system was reset, the bike ran another 39 seconds and reached 185 km/h — and cut out again at 50. It then crawled to the flag at under 75 km/h. That is where most of the minute and a half went: not in the power fade, which cost seconds a lap, but in one integer nobody had been asked to decide, on a day when the pack started at 38. At the time the team attributed the stop to a voltage drop at the inverter, or possibly a fuse; the log, read four years later, says otherwise. Chapter 12 is about why that happens.

Two lessons, and the second is the one worth carrying. A qualifying figure tells you almost nothing about a race — here it told the opposite of the truth. And a thermal problem does not announce itself as a fault: nothing broke, no light came on, the bike simply got slower every lap while the rider kept riding it the same way — and when it finally did stop, the team blamed a fuse.

Evaluate the alternative properly, then keep the negative result

Before committing to a liquid circuit it is worth putting the passive options through the same analysis, because they are lighter and simpler if they work. A published study of phase-change material and heat pipes on a generic pack at a 3C discharge (not this pack, and not commissioned by this project) found that performance improves as you add thickness but the target isn't reached LI. That is a negative result, and it is worth as much as a positive one: it is what makes the liquid-circuit decision defensible rather than merely conventional, and it stops the question being reopened every six months.

The time constant tells you when equilibrium arrives

A cell's thermal time constant is mass × specific heat / heat transfer coefficient. If it is roughly one lap long, the pack reaches equilibrium early in the race and your calculated temperatures are running temperatures rather than end-of-race peaks. If it is hours, the pack heats throughout and never equilibrates — and then you have to size by integrating rather than from a steady state. For a 70 g cell at 0.38 W/K the constant is about three minutes, so a KymiRing pack settles within a lap or two.

What actually sizes a cold plate

Once the required heat transfer is known, the plate itself is sized by three things, and only the third is thermal.

1. Deflection, not stress. A pressed plate under internal pressure is a plate in bending, and the allowable deflection is reached before the allowable stress is. Size it on deflection and the stress case will pass; do it the other way round and the plate bulges away from the cells it is supposed to be touching.
2. Off-the-shelf sheet thickness. A calculated optimum of 0.78 mm is not a thickness you can buy. Round to stock and re-check, rather than the reverse.
3. Coolant layer thickness. Thicker is not better — past a few millimetres the extra fluid slows down and stops carrying heat away. Published optimisation studies for cold plates put the useful range at 2–3 mm COLDPLATE.

An idea that was right, arrived late, and failed on the joint

One of this project's better ideas never got tested, and the way it died is more instructive than the idea itself. A team member pointed out that the frame is thirteen kilograms of aluminium that is already on the bike, already in the airstream, and doing nothing thermally. Used as a buffer in the cooling system it would absorb heat the pack cannot hold.

Pick the adhesive by service temperature, not by strength

A detail that bites late. Structural adhesives are usually compared on room-temperature strength, but a bonded joint inside a battery pack sits continuously at elevated temperature, and what matters is the glass transition. An adhesive with a Tg below the joint's service temperature is a soft joint in service regardless of what the headline figure says.

Sensor coverage is the cheapest performance you can buy

If the current limit is set by the hottest measured temperature but sensors are sparse, the system has to assume the worst. Sensor count and placement are therefore a direct performance question — and what settles it is the same measurement the thermal model needs: how much hotter is the hottest cell than the sensor reads? Nobody knows that number until it is measured, and it sets the cut-off threshold. It is also the 8–12 K spread earlier in this chapter seen from the BMS's side, which is one more reason to instrument the first module well.

Two of the biggest lap-time items live here. The thing most easily got wrong about mass is what kind of quantity it is. And some of it spins, which makes it two quantities at once.

This chapter settles how to think about mass, and where the rest of the time is. Mass sensitivity on a grip-limited circuit is around 0.02–0.025 s/kg in the project's own model and in Flanagan's validated model, and the MotoE generation change agrees in order of magnitude, not in the second digit (chapter 02) FLAWB s15. So thirty kilos is a little under a second. Since the battery on an electric bike is a hundred kilos, that is also where the kilos are.

Mass is an output, not a target

It is tempting to read a project's mass history as drift: every generation a little heavier than the last, discipline slowly failing. That is rarely what is happening, and reading it that way leads to the wrong fix, because the fix it suggests is “be stricter next time”, when the actual problem is usually that the number was chosen instead of derived.

So the chain runs the other way round from the way it is usually discussed:

Chapters 07 and 08 are that chain. Cell count is the largest single line in an electric bike's mass budget, and it is not a free variable; it is whatever the thermal requirement says it is. Which means the right way to state a mass figure is as the outcome of a mission spec, never as a number picked in advance and defended afterwards.

And then fight for every gram

None of that is an argument for slack. The opposite: once the cell count is fixed, everything else is fair game, and the discipline is to size each part for what it actually carries rather than adding margin out of unease. Mk1 is the cautionary case in the other direction: redundant fusing layers added to be safe, never traced back to a requirement, and paid for in kilograms.

The practical test is whether a dimension can be traced to a number. If a part is 2 mm thick because someone worked out that it needs to be, that is engineering; if it is 2 mm because 1 mm felt thin, that is an unpriced kilogram. The busbar in chapter 07 is the live example, and so is the driveline: on a comparable bike two percent of gearbox efficiency is worth about 0.4 s a lap, which is more than the whole gearbox's mass is worth, so a lighter case is not the first place to look there either WB s15.

The mass budget is a living document

Start one on day one and keep it at line-item level: component, estimate, weighed value, target. Three things make it work:

• Estimate and weighed value are separate columns. Otherwise you have no idea how much to trust the total.
• Every new component gets a line the moment it is decided, not when it arrives. Otherwise the budget looks healthy because half the bike is missing from it.
• Decisions that add mass get logged where they are made. Going to liquid cooling, thicker busbars, a bigger enclosure: all of these originate somewhere other than the mass budget, and that is exactly why they go missing from it.

And then weigh the bike, because that is the only number that settles it. Mk3's line-item budget totalled 190.6 kg and the scale said 194 kg, 1.8 % light. That is a good budget, and knowing it is good is worth as much as the figure itself: it tells you how much to trust the next one before anything exists to put on a scale. A budget that has never been checked against a weighing is an assumption with a decimal point.

Centre of gravity

CG height sets the wheelie and stoppie limits, which is to say how much acceleration and braking force you can use before the bike tips over lengthwise. On an electric bike CG is designable, because the battery is massive and you have freedom in where you put it. But that also means battery placement is the first thing that constrains CAD, before anything else can be locked.

CG also wants measuring rather than guessing: axle weights level and tilted will do it, and it is an hour's work. The geometry is one line: the stoppie limit is g × (horizontal distance from the front axle to the CG) / (CG height). On Ukkonen, with 0.70 m and a CG estimated at 0.60 m, that is 1.17 g; 50 mm lower gives 1.27 g, which is about 0.4 s of the braking-limit bar in chapter 03; the rest of that bar is regeneration. The project's own workbook carries the CG at 0.65 m in one place and 0.60 m in another, and at 0.65 m the same formula gives 1.08 g. The 0.60 m behind the 1.17 g, and the 0.4 s, come from the workbook's older sensitivity sheet, computed for the previous 104S pack WB s4; the locked-choices sheet repeats them, while its own open item carries 0.65 m WB s0. At 0.65 m the stoppie limit, 1.08 g, is below the 1.10 g braking limit the baseline simulation uses, so if the higher estimate is right the baseline is slightly optimistic. Until the bike is on the scales the 0.4 s is an order of magnitude, and that disagreement is the argument for the hour (open item O17). The 50 mm itself is a target, locked in the workbook.

Grip is the biggest single item — and you can't buy it as a part

Grip tops the sensitivity list, and it is a collection of decisions rather than one: tyres and their temperature, pressures, geometry and setup, how the suspension actually works, and rider time. The last is the one most easily overlooked. If the bike only ever gets ridden in single-lap bursts, the rider never reaches the tyre's real limit and the setup can't be developed.

Suspension, linkage and geometry follow the same rules on an electric motorcycle as on a combustion one, applied to the electric bike's own mass and centre of gravity. This book does not go into them, because the literature on them is extensive; Cossalter's Motorcycle Dynamics is a standard reference COSSALTER.

An electric bike also has a software route to grip: traction control and a corner-dependent torque limit. Those need a supervisory controller and a redundant throttle sensor, so they belong to electrical system design, but the benefit shows up on the grip line. If your simulation uses a fixed grip figure, a slip-ratio study for your own bike is what tells you how much of the gap that software could realistically close SAL.

Grip and braking are bought with track time

Together, grip and the braking limit are about 3.4 s of chapter 03's list: more than every other item on it put together. Neither is a part you can order, and on this bike that includes the brakes. The brakes are already from Brembo's racing range TEAM, so the hardware is not what limits braking. Mk3's hardest measured deceleration was 1.04 g, where the simulation assumes 1.10 g and a Superbike brakes at 1.3–1.4 g WB s4/s6. What raises the braking limit is the rest of the machine: the centre of gravity, through the stoppie limit above; the suspension setup, which decides how the load transfer arrives at the front tyre; engine braking from regeneration, which Mk3 barely had, so its rider was making up for it with the rear brake and was given a hand lever for it (chapter 12) WB s4; tyre pressures and temperatures; and the rider learning how this particular bike behaves on the brakes.

Grip is the same list with the tyre at the head of it: compound and size (the 180/60-17 in use now, a 200/60-17 under study, which needs a wider rim, room in the swingarm and a new chain line), temperature and pressure, geometry, and suspension that works in the range where the tyre is actually loaded. None of it can be found on a dyno or in a simulation. It is found by traditional setup work at a circuit: change one thing, ride a stint, read the tyres and the data, change the next, and do it again at the next circuit, because a setup belongs to a track.

That rule has a practical form. The donor bike's own race-kit manual is the original source document for anything inherited (part numbers, suspension and geometry settings) and worth keeping rather than re-deriving. Against it, published stiffness targets give you something to design to: lateral stiffness of 0.8–1.6 kN/mm and torsional 1–2 kNm/° for a swingarm, and the separate finding that a frame's lateral stiffness can be reduced without giving up longitudinal stiffness CHASSIS. Ukkonen's inherited swingarm weighs 7.65 kg; a purpose-designed one in 7075-T6 within those stiffness ranges came out at 4.86 kg, which is the size of the prize.

Packaging serves the geometry, not the other way round

The largest components on an electric bike, the pack and the motor, are also the ones that arrive with their dimensions fixed, and there is a quiet way for them to take over: the pack is sized in chapter 07, the cells are laid out to fit, the motor goes where the chain line puts it, and the wheelbase, the lean clearance and the centre of gravity turn out to be whatever was left. Do not let the battery and motor packaging define the motorcycle's geometry by accident. A cell count that is optimal on the energy and heat arithmetic is not optimal if it costs a compromise on geometry that the rider will feel in every corner; 1 080 cells is this project's answer to a thermal question, and it stays the answer only as long as it fits.

Three practical consequences. Start from a known geometry (a donor chassis, or a set of published numbers, even when the frame itself is your own), because a known starting point is a risk removed, and this chapter has already said what it costs in mass. Package inside that geometry: the pack's envelope is an input to the module layout, not an output of it, which is the discipline chapter 04's 100S8P case learned by having to redo the layout. And check lean clearance in CAD, with the suspension at full compression and the tyre at its real profile, before anything is machined. Mk1's first test-day note was that the fairings touched the ground in a corner and the frame "had not been far from it" DIARY. That is a lean-angle limit set by a bodywork draughtsman, and the rider found it before anyone measured it.

Rotating mass counts twice, and the two counts scale differently

Last in this chapter because it is the smallest item in it: about a tenth of a second against the lighter rotor of a radial motor WB s15, 0.14 s against no rotor at all, and an effect on handling that is real but has no number in seconds. It is here because it is a design choice that costs nothing if it is made early.

Everything above treats mass as a single number. Rotating mass is not one: it appears in two places, with two different laws, and a design decision that is bad in one can be good in the other. Both laws are about the same quantity, the moment of inertia I of whatever spins, and about the gear ratio G between it and the wheel.

During acceleration, inertia is reflected as I · G². The motor has to accelerate itself as well as the bike, and because it spins G times faster than the wheel, its inertia is felt at the wheel multiplied by G squared. Divide by the wheel radius squared and you have an equivalent mass in kilograms, which goes straight into the mass sensitivity from chapter 03.

In direction changes, angular momentum is reflected as I · G. Angular momentum is I·ω, and ω scales with G to the first power, so the gyroscopic term grows more slowly with gearing than the inertia term does. And unlike inertia, angular momentum has a sign: something spinning the other way subtracts from the total instead of adding to it.

The second one is why MotoGP engines rotate backwards: angular momentum couples roll into yaw, so a machine with a lot of it resists being placed, and turning part of the total around reduces the coupling SPALD. The same option exists on an electric bike, and it is very much cheaper there: a crankshaft has to be reversed with an idler gear and a redesign, while a motor's direction is a phase order and a gear pair.

A setting can cost more than a component. It is the cheapest place to find time and the hardest place to notice it's missing.

This chapter settles what has to be verified rather than assumed. Inverter and BMS parameters are design choices as much as anything mechanical, but they don't appear on any drawing. A wrong parameter doesn't break anything either: the bike works, it just works worse than you think, and you only see the difference in the data.

Two limiters, and which one bites

Current limiting typically comes from two places: the BMS requests a limit and the inverter enforces its own. You need to know which one bites when, and in particular whether the BMS request is connected to the inverter at all. An integration implemented over a bus looks identical from the outside whether or not it works.

Regeneration isn't just a setting

Getting regen to work needs three things at once, and if one is missing it doesn't work at all: the inverter's regen limit set high enough, the BMS charge current limit raised to what the cell will actually take, and a starting state of charge below full so there is room to put energy back on lap one. On a fully charged pack regen is off until SOC drops — which is exactly the laps you are riding hardest.

It is worth being clear about why it is being tuned at all, because regen is three different things at the same time and they do not have the same answer.

1. It is energy. On a pack sized for the race distance it can be the difference between finishing with margin and finishing on fumes.
2. It is how the bike behaves on corner entry. A rider arrives at a corner expecting the machine to slow when the throttle closes. Without it the bike free-wheels in, the rear stays light, and the chassis is harder to settle. This is the one that sets a floor. Some retardation on a closed throttle is a handling requirement whatever the energy budget says, so the question is never whether to have regen, only how much, and shaped how.
3. It is heat, and it can only ever add. Ohmic loss is I²R, which does not care about the sign of the current: ten amps in heats the cell exactly as much as ten amps out.

What it actually costs in heat

The third item sounds worse than it is, and the arithmetic is worth doing once because the result changes how you tune it. The drive side is unchanged (the bike needs the same traction energy either way), so the extra heat is only the ohmic loss of the charging events themselves:

Q_extra ≈ E_regen × ( I_regen · R_cell / U_cell )

That is the recovered energy multiplied by the cell's ohmic loss fraction at the regen current, and because regen current is a fraction of drive current, so is the loss fraction. It also gives the ceiling directly: even if regen were taken at the full drive current, the penalty could not exceed the drive-side loss fraction times the recovered energy.

Which is the real conclusion: heat is not what limits regeneration. The cell's charge current limit is, and it is usually far below its discharge limit; after that comes the BMS's own limit and its derate near full charge, and then the inverter's regen torque setting. All three are settings or datasheet numbers rather than thermal physics.

One more term exists and is worth knowing about without being designed around. Reversible, entropic heat changes sign with current direction: a cell that produces a little extra heat on discharge absorbs a little on charge. At the regen currents above it is of the same order as the ohmic term, so it works in regen's favour — but its sign varies with state of charge, and this project's thermal model is ohmic only, so it is a reason the estimates above are conservative rather than a number to spend.

What you have to build to collect the electric advantages

Chapter 01 tabulates what the architecture gives back. None of it arrives by itself; each item is collected only if something is built to collect it.

• Commanded torque needs an accurate K_t. The inverter knows the torque it just applied only as well as it knows the torque constant, which is why chapter 05 asks for it to be measured. Any loop closed around wheel slip depends on it.
• Fine torque control needs a supervisor. The current loop can move torque in milliseconds and in small steps, but only a vehicle controller in CAN control mode, with a redundant throttle sensor, can use that for traction control, launch control or a corner-dependent torque limit. Mk3 had an analogue throttle and none of the three.
• Engine braking has to be mapped. Drive and regeneration are one control law, so retardation on a closed throttle is a map by speed and by position on the circuit, not a property of the engine. Someone has to write the map, and the regeneration sections above are what it has to respect.
• The mass does not change during the race. A combustion bike finishes twenty-odd kilos lighter than it starts; setup has to be right for the whole distance at the start weight.

Check the resolver's electrical alignment, not just the mechanical one

The resolver is the rotor-position sensor: it tells the inverter where the magnets are at every instant, so that the current can be placed correctly against them. It gets three sections here, because this project got the same component wrong three separate times and each failure had a different shape: a wrong offset, a wrong timing and a wrong direction, in the order of the three sections below. In each case the symptom pointed somewhere other than the sensor.

A subtle one worth knowing about in advance. If the position sensor and the motor have different pole-pair counts, mechanical and electrical angle are not the same thing, and a small mechanical misalignment becomes a large electrical offset. Changing the sensor without recalculating the offset leaves a value behind that looks plausible and quietly costs torque.

The offset matters because it is the angle at which the current vector is placed relative to the magnets, so it sets torque per amp and efficiency directly. The inverter manual asks for it within ±0.7 electrical degrees, and the procedure is a coast-down with the inverter disabled, reading the angle between back-EMF and resolver CASC §4.1.3. It is half an hour on a bench.

And it follows from this that the sensor's physical position is a calibration parameter. Taking the drive apart and reassembling it in a different clocking invalidates the number, without anything looking wrong afterwards. Any procedure that separates motor and sensor has to end with "recalibrate the offset" written in it.

An error that grows with speed is a delay, not a fault

The offset above is a constant: get it right once and it is right everywhere. The second failure is the one that survives a correct calibration, because it is zero at the speed you calibrate at and grows from there.

The diagnostic move is general and worth having: plot the error against the thing it varies with, and the shape names the mechanism. An error that is constant is an offset. An error that scales with speed is a time — something measured at one instant and acted on at a later one. An error that scales with current is usually saturation. You do not need to know the cause to read the shape.

It was a battery problem too. The 129 A of unrequested charge current was more than a control nuisance: it goes into the pack, past whatever the BMS charge limit was set to expect, in the part of the lap where the cells are already warm. A torque-control fault and a battery event can be the same event, which is an argument for logging pack current at the same rate as torque, and for reading them together rather than one at a time.

A sign convention can be a feature switch

The third failure belongs to the same commissioning step, and it was the most expensive of the three. Every inverter has a notion of which way is forwards, and it is set by two independent things: how the resolver is wired, and the order of the motor phases. Get either backwards and the machine still runs perfectly well; it just reports a negative speed. That can look cosmetic, although it may not be.

The manual will usually tell you how to check the direction. What it is much less likely to tell you is which functions stop working if the sign is wrong, because from the manufacturer's point of view the direction is simply part of a correct installation and not a setting with consequences. So the check reads as informational, and gets treated as informational.

And it is not a beginner's fault. When the team described the crossed resolver wiring to the people at the Lightning Motorcycle factory in January 2024, one of them finished the story before it was told: field weakening cannot have worked. Nobody had told them that part; the symptom was recognised from three sentences LIGHT. That is what a mature field looks like from the outside: the knowledge exists, it is written down nowhere, and it passes from team to team when somebody recognises your symptom.

It is worth noticing how early this class of fault first appeared, and how harmless it looked then. On Mk1's very first test day, in June 2022, an adjustment to the angle offset was deliberately made large — "not just by two units but by twenty", to see what it would do, and the rear wheel turned the wrong way in the paddock DIARY. Somebody laughed, the value was reduced, the bike went back out. That was the same relationship between believed angle and actual direction, in its visible, instantly diagnosed form. A year later the same relationship appeared in its invisible form and cost a test day.

Four things come out of that, and none of them are about resolvers.

• An unexplained sign is an unexplained fault. Resolve it before the test day, not after it. Anything a system reports that you cannot account for is a finding, however cosmetic it looks.
• Commission against the envelope you will use, not the one you are in. A workshop run never reaches the base speed, so an entire region of the control law is untested by construction. Write the commissioning plan around the features you will need later (high speed, field weakening, regen, current limits) and find a way to exercise each one deliberately.
• What the previous component tolerated is a debt, not an inheritance. The old inverter did not care about direction, so the check had quietly left the list. When you replace a component, the dangerous items are the ones you stopped checking because the last one forgave them — and there is no list of those anywhere.
• A procedure tells you how to check; it rarely tells you what breaks. When a manual asks you to verify something and does not say why, that is the point to find out, because the answer is the difference between a five-minute check and a lost day.

Two words, two documents, one shaft. If you reverse the motor's rotation on purpose, as chapter 09 describes, then “backwards” mechanically has to be “forwards” electrically, and those two words come from different documents written by different people. The condition is not one thing but two, and they have to be recorded together: the mechanical direction relative to the wheels, and the requirement that the inverter sees that direction as positive. If the motor mounting, the gear layout or the phase order changes, the same fault comes back — and again it will only show above the base speed.

A settings mistake can cost the season, not the lap

Everything above is about settings that cost performance. There is a second category, and it is worse: settings that cost availability. A machine that is slower can still be developed. A machine that will not turn on cannot.

Three properties make configuration work unusually dangerous, and they compound.

• Scaled parameters. Many inverter and BMS parameters are stored as integers with an implied scale, and the scale is in the parameter's name: ..._x_10 for tenths, ..._x_100 for hundredths. A factor of ten in a limit does not produce an obvious error; it produces a machine that refuses to operate for a reason that points somewhere else.
• Some of the configuration is unique to your unit. Alongside the settings you choose sit factory calibration constants: the gains and offsets of that individual unit's voltage and current sensing, measured on that serial number at the end of the production line. They are not in any manual and not in a generic defaults file.
• Recovery may depend on the supplier's support, so plan risky configuration work for when it can be reached, and never against a booked dyno slot.

One smaller habit falls out of the same episode, and it is free: check which variant of the firmware your hardware wants: this inverter ships two builds selected by motor type number, a distinction that is documented and was missed, and which muddied a week of diagnosis for no reason.

Some requirements can't be added afterwards and belong in the CAD phase rather than in race week.

This chapter settles what has to be in the design from the start. Above 60 V DC the FIM's technical guidelines for electric motorcycles put a system in voltage class B, the high-voltage class FIM23; other codes draw the line in their own place, so check the one that applies to you. There is no way around that, and no reason to want one: most of the requirements are things a well-designed machine would meet anyway.

Breaking capacity is the component number that gets forgotten

A fuse or contactor's rated voltage is not the same thing as its breaking capacity. A DC circuit has no zero crossing to extinguish an arc, so breaking capacity has to be designed into the component. An under-rated fuse doesn't extinguish the arc, it sustains it — it is the single component that turns a fault into a fire.

Insulation: two independent barriers

Insulation requirements are usually expressed in ohms per volt and multiplied by system voltage. The level varies between sources, so confirm it against the code that actually applies to you; the difference can be five-fold, and it sizes the whole insulation structure.

Structurally the principle is the same whatever the level: two independent barriers between live parts and the chassis, each sufficient on its own. And a third requirement that is easy to skip past: insulation monitoring has to open the contactors, not just log a fault code.

The high-voltage system, drawn before it is bought

The components above only make a safe system if they are connected in the right order and switched in the right sequence, and both can be drawn on one page before anything is ordered. The architecture is the same on almost every electric race machine; what differs is the part numbers.

The switching sequence is part of the design, and it is worth writing down as a short list of states, because every fault the system can have is a question about which state it is in:

1. Off. Both contactors open, the bus discharged. The 12 V system runs from its own battery, the BMS checks the cells, and the insulation monitor checks the pack against the chassis before anything else is allowed.
2. Precharge. The negative contactor and the precharge relay close, and the inverter's capacitors charge through the resistor. Closing the main contactor straight onto an empty capacitor bank welds contacts; the precharge exists to prevent that, and the controller waits until the bus voltage is close to the pack's before moving on. A precharge that never completes is a fault, not a delay.
3. Ready. The main positive contactor closes and the precharge relay opens. The bus is live, drive is not enabled, and the lights say so.
4. Drive. Drive is enabled by a deliberate action, and the lights change to a signal that is different from both "ready" and "fault".
5. Fault or stop. Any break in the safety loop, an emergency stop or the lanyard pulling out, opens the contactors, and the bus is actively discharged. The FIM guideline asks for the bus to fall below 60 V within five seconds of the general circuit breaker opening, for both poles to be isolated, and for a rider cut-out on a lanyard that is separate from the emergency stop FIM23.

Two items in the figure are worth a word because they are easy to leave out. The interlock loop (HVIL) is a low-voltage circuit threaded through every high-voltage connector, so that unplugging one drops the contactors before the connector can be opened live. The FIM guideline does not require it; production electric vehicles use it as standard, and on a machine that is taken apart between sessions by students it is cheap insurance. And the status lights are for the marshal, the mechanic and the visitor rather than the rider, which is why the guideline wants them on both sides of the bike and visible from ten metres FIM23.

Design it to be testable

A self-built pack typically needs test evidence for overcharge, overvoltage, over-temperature, spray and drop. These are far easier to pass if the structure was designed for them from the start than if the pack was first made to work and then made testable. The same goes for the vent path: when a cell vents, the gas has to be routed out somewhere that doesn't point at the rider, the rest of the pack or the inverter, and that is a volume allocation, which has to happen before the module height is locked.

Who is allowed to ride it is a decision, and it is made in advance

A university prototype that does 240 km/h eventually needs somebody to sit on it, and the question of who is not an administrative afterthought. It belongs with breaking capacity and vent paths: cheap to settle months early, and impossible to settle at the moment it matters — which is a booked airfield, a loaded van and eight people already standing on the runway.

Two things want separating, because they fail differently. One is competence: a current competition licence, recent time at speed, and machinery of comparable performance in the rider's own history. The other is the capacity in which the person is riding: an institution's own student or employee riding institutional equipment is a different arrangement from a licensed racer riding on their own competition licence, and it is worth settling which applies before the day rather than on it.

Race-day operations are part of the design

Four things decide whether a ride day is one session or a full day, and none of them shows up in lap time: where the bike charges, what is in the paddock, what sets the schedule, and where the heat goes between sessions.

The first is charging where you actually ride. The bike can be finished and still be a one-session machine if there is nothing to charge it from. Work out the numbers early (energy per session, the voltage the pack charges at, and the power you need to turn it around between sessions), because they decide whether you need a portable unit, a fixed installation, or an arrangement with someone who builds them. A street circuit has no infrastructure to fall back on, which makes portability a design requirement rather than a convenience KEMP. For scale: the MotoE bike charges through a 20 kW socket from an external charger and takes about 45 minutes to 80 % DUCATI, which is roughly the turnaround a race weekend schedule gives you.

The second is the paddock itself. The PPE and equipment list for a high-voltage bike is short, cheap and much easier to buy in advance than during a race weekend: class 0 insulating gloves, insulating mat, a rescue hook, a voltage tester, a CO₂ extinguisher for the HV system and a lithium-rated extinguisher for the pack FIM24. Event organisers' guidance covers paddock operations in more practical detail than the technical codes do, and it is worth reading even from the wrong discipline MSUK. For the risk assessment itself (hazard mechanisms, firefighting and what the authorities expect), a national safety-authority report is a better starting point than a sporting code GAIA.

The third is the schedule, and what actually sets it was learned at a first race weekend. Tyre warmers set it, not the charger. Getting slicks to temperature takes close to an hour, which on a test day is the real interval between runs and on a race day is what every other preparation has to fit around; an electric bike's turnaround is rarely the thing that binds. And the connection between your charger and the local supply is a part you should own, in advance, and have plugged in once. The project's first race weekend included an emergency drive to a hardware shop for a three-phase cable at nearly €300 (which turned out not to be needed, and was promptly lent to the race office, which did not have one either) and then a Saturday morning on which the bike simply would not begin charging. That is a trivial fault on a bench and an expensive one at eight o'clock before qualifying. DIARY

The fourth is where the heat goes between sessions, and it is larger than it looks, because paddock time is five things at once: charging time, cooling time, maintenance time, data-download time and preparation time for the next session. And a hot pack is not necessarily ready to accept charge the moment the bike stops, so the first two are in series before they are in parallel. If the pack's passive time constant is hours, it starts the next session warm unless something cools it in the paddock. On Mk2 and Mk3 that something was a cordless leaf blower with a home-made nozzle, blowing air straight through the pack, which is the kind of improvisation that works and never gets written down. Mk4's liquid circuit makes the proper version possible: cold water through the pack while it charges and while it waits.

Over three years the biggest single risk is losing track of where a number came from, more than getting a calculation wrong.

This chapter settles how the numbers stay trustworthy. A technical project generates more numbers than one person can hold. When calculations get updated at different times in different files, you get sediment: the same quantity exists as three different numbers and nobody knows which is current. That is the normal consequence of design progressing unevenly rather than carelessness, and the errata page at the back of this book is a worked example of what it looks like when someone goes back and checks.

The chapter runs from the discipline of writing things down, through what to measure and how to judge a result, to the two sources of evidence that are not instruments (the rider and the outside expert), and ends with when to stop deciding.

Four rules that stop the sediment

1. Every number has a source. A workbook sheet, a datasheet, a test record or a calculation. If there is no source, mark it an estimate. This is the only mechanism by which somebody three years from now can trust the page.
2. One place per number. When a decision changes, rewrite the page and move the old version into the decision log as a line. Pages don't accumulate layers.
3. Calculation and readable form are different documents, and one of them wins. The workbook calculates, the handbook explains. In a conflict the workbook is right and the handbook has fallen behind, but that has to be agreed in advance.
4. Conflicts get written down, not quietly fixed. When the material contains two different numbers for the same quantity, record it as a finding with both sources. A silent correction destroys the information that the other number is in use somewhere.

Three registers

In practice this means three lists, kept separate from the subsystem descriptions:

The answer is often already in the data

The rules above are about keeping numbers findable. There is a second reason to keep them, and it is the one that pays: a question you ask in year four is often answered by a log from year one. Logs are cheap to keep and impossible to recreate, and the value of an archive is not what you thought you were recording; it is what somebody later thinks to ask of it.

Three ways a measurement can be worthless, all of them cheap to prevent

The archive also showed how data goes bad, and the failures are worth naming because each is a two-minute check that nobody did.

Measure what you can't reconstruct later

Bench and track time is always short, so measurements need prioritising. A good test is simple: what can't be reconstructed afterwards by calculation? Thermal time constants, the temperature dependence of resistance, and how a limiter actually behaves all disappear unless they are logged under load. Efficiency maps and driveline losses can be estimated, less accurately, on paper.

One logging requirement applies to everything: all sources on one timebase, from one common trigger, at 10 Hz or better: the inverter's currents, voltages and temperatures, every cell voltage and thermistor from the BMS, and torque and speed from the bench DYNO. If those don't share timestamps you can't compute resistance or efficiency afterwards and you can't repeat the run comparably. It is the one place where carelessness invalidates the whole session.

And one requirement applies to every lap time this project ever records, because chapter 01 showed how little a bare figure means. Log the context of the lap, not only the lap. At minimum, chapter 01's five variables with the map in use added and the lap's place in the session made explicit: state of charge at the start of the lap; pack, motor and inverter temperatures at the start; the power or torque map in use; the tyre type and its history (new, and how many laps if not); the lap's position in the session; and whether the session was a single flying lap, a race-distance simulation or a settings test. Six fields, written on the same line as the time. Without them a lap time is not comparable with another lap on the same bike a week later, and the project has three years of times that cannot now be told apart on any of the six.

A requirement list is what makes "did it work" answerable

The registers above record what was decided. A requirement list records what was asked for, and it is the only document that lets anyone judge the result without re-arguing the design. Two details make it work: each line is marked as a requirement or a wish, so that missing a wish is not a failure; and each line carries the date it was added, which is how a later reader tells a stable target from a moving one. At the end you walk the list and name only the lines that did not come true. That is the whole evaluation.

Two habits sit next to it, and both come from the same failure mode: the document says one thing and the workshop did another.

• Record substitutions where the choice was made. The next reader will assume the specified part is the part that exists.
• Treat generated files as their own artefact. A drawing that is correct can still produce a toolpath that is not. Anything derived from a model (DXF, toolpath, weld path, BOM export) needs a check of its own, because it is what the machine actually obeys.

The rider is an instrument, and the readings need writing down

A race bike carries one sensor that is on no wiring loom, samples continuously, and reports in sentences. It is also the only one that can say what the machine is like to use. The difficulty is that its output arrives as opinion, in a paddock, between sessions, and almost never reaches the place where the telemetry is kept.

Two habits make it usable. The first is to write the rider's words down verbatim, and separately from your reading of them — the same register discipline as the rest of this chapter. A rider reports symptoms accurately and attributes causes no better than anyone else: the sentence is evidence, the cause is yours to find. The second is to ask a question that has an answer. "How did it feel?" produces politeness. A question with a number in it produces data.

Getting an expert to look, and what happens to what they say

A student project can get expert attention more easily than it expects. People who have spent careers on a problem are often willing to look at somebody else's version of it, and one introduction from someone they trust is usually all it takes. That part is cheap.

What is not cheap is doing something with the answer, and this is where a review goes to waste. Expert feedback arrives as a flat list: eight or ten observations in the order they occurred to the writer, none marked as more important than the others, most of them small. The reader triages by apparent effort (the easy ones get done), and the item that would have changed everything looks the same size as the rest.

The fix is a different triage question. Not which of these can I do this week, but: which of these, if true, would mean a measurement I already trust is wrong? That question sorts the list in a completely different order, and it is the only order that matters.

Lock late, but lock

Locked choices are necessary, because CAD can't start from moving inputs. But locking before the first measurement is a mistake whose price is a whole production batch. The practical compromise: lock the architecture early and the dimensions only after the first measured part.

How the project actually runs, what this document is for, the problem it is trying to solve (which is not a technical one), the people who kept it going, and a last word.

Work continues at the J. Hyneman Center with Mk4, as one platform project among several. That phrasing is deliberate and it sets the pace: this is not a motorcycle programme, it is a teaching workshop that happens to build a motorcycle, and the bike advances in the gaps between course assignments, theses and everything else the centre does. It will not move as fast as it would somewhere that builds motorcycles and nothing else, and it is not meant to.

Nor does it run the way a student project is usually pictured. There is no year group that arrives in September, builds a bike and hands it over in spring. Students join when a course, a thesis or an assistant's post brings them in and leave when it is done, so at any moment the team is a mix of people in their first week and people in their last. The bike changes the same way. It is never rebuilt for the sake of a new group: what gets redesigned is the part expected to give the most lap time for the effort, or to make the machine safer, and everything else is carried over. Mk1 to Mk4 are the states of one machine at the points where enough had changed to give it a new name. TEAM

What the platform is supposed to do is produce topics. Each subsystem in this book is a thesis or a course project waiting for someone: the cold plate was one, the air-cooling review was another, the interconnect study a third, and chapters 05 through 11 are full of open items that are the right size for a student and that have a real machine at the end of them. That is the point of the platform, and it is why a bike that is never finished is a feature and not a weakness.

Around the theses there has also been a steady stream of smaller exercises, many of them from the machine design course: one version of the rear subframe was designed that way, and so was the steering-head bearing arrangement. Some of that work went onto the bike as it was, some after modification, and some was set aside. That is ordinary design work, and it would be no different with a team of full-time professionals: some of the ideas and designs always end up rejected. Nor has the work come only from mechanical engineering: visual artists, business school students, electrical engineering students and even a practical nurse have taken part, to name a few. TEAM

The problem this document exists to solve

It is also, though, exactly where student projects fail — and the failure is almost never technical. A student finishes, and what they knew leaves with them. In a project where people come and go all year there is not even a handover day: nobody is ever formally handed the project, so nothing is ever formally handed over. Whoever joins next inherits hardware they did not design, settings nobody can explain, and a folder of files with no index, and spends their first months rediscovering things that were known two years earlier. This project has done that more than once, and this book is a record of it: a voltage fix that was known from a 2022 log and waited four years for the money to make it; an expert review, two of whose findings were later met again the hard way; a published power figure that was wrong for four years because it travelled next to a speed figure that was right (chapter 12).

That is what this handbook is for. Not to be right: the errata page is where its mistakes will be recorded, and there will be some. To be the one place where the project's numbers, decisions and mistakes are written down together, with a source against each, so that whoever joins next starts from where the last person finished rather than from where the project began. A handbook that is updated and occasionally corrected is worth more than a report that was accurate once.

Every convention in chapter 12 exists for that reason, and they are worth restating as the thing to take away if nothing else is: every number carries a source; a decision that is reversed is logged with its reason; a question that is open is written down as open; and a correction is published rather than quietly applied. None of it requires software, a budget or anyone's permission. All of it requires somebody to decide that the next person to join matters as much as the people already here.

The people who kept it going

First the riders, because every measured number in this book exists because one of them went and got it.

If students come and go, the continuity has to come from somewhere, and in this project it has come from the J. Hyneman Center's staff. That the project has had funding, has survived the gaps between one set of students and the next, and has actually reached a track is largely their work, together with a close partner elsewhere at LUT, and it is right to end with them. TEAM

• Terhi Virkki-Hatakka carried much of the project's documentation and communication (including most of the project diary), along with photography and the general arrangements around events.
• Markku Ikävalko has acted as the project's general manager: relations with race organisers and circuits, the transport to events, and hands-on work in the workshop, the fairings among it.
• Petri Tuhkanen did the electrical design and installation, a good deal of the mechanical installation work besides, and has had an important role at the track.
• Jamie Hyneman, Professor of Practice at LUT, has been there whenever the project needed new ideas or had materials to choose. Through him the project reached expert reviewers, and he made possible the team's visits to the Lightning and Zero Motorcycles factories.
• Marko Kasurinen, who compiled this handbook, has been lead designer and project manager.
• The centre's student assistants of each year have each added their own share, on the bike and around it.
• LUT Voima, the prototype workshop and laboratory of LUT School of Energy Systems, is where the project turns whenever a part is too difficult to make on the centre's own machines. It has made numerous parts that would otherwise have gone to subcontractors; subcontracting works, but it slows every round of iteration and development LUTVOIMA.

Documentation, photography and presenting the bike have been everybody's work, and the last of these matters more than a design handbook would usually admit. Ukkonen has been the J. Hyneman Center's most visible demonstration of what the whole workshop is for: a machine students designed and built, shown to the public several times at the MP Motorcycle Show in Helsinki, Finland's largest motorcycle show, and at many other large events, and sat on by the President of Finland when he visited LUT University in January 2025 PRES. For a project that lives on funding and goodwill, that visibility is part of the result, for LUT and for the centre alike. Count the time spent presenting the bike as project work, not as a distraction from it: nobody can tell in advance which stand conversation turns into a partner or a sponsor.

And then the point of all of it: the students. They have had real topics with a real machine at the end of them, and a place in the design office, in the workshop and in the pit crew at a race meeting. Much of what this book records, they learned first.

A last word

This book is partly a handbook and partly the story of how one team learned its trade. We are not claiming to be at the front of this field, and nothing here should be read that way. It is what we have been through and what we think we know today, written down so that the next people do not have to start from zero.

It is also worth saying where the calculations stop. A lap simulation, a pack sized on paper and a locked list of components are a long way from a motorcycle that works. Between the two lie CAD, electrical design, manufacturing methods, procurement, assembly and testing, and each of those holds more work than the arithmetic that set it up. As this book shows more than once, the first version of anything is the first prototype and a starting point, not an answer. Getting to a good result takes more iterations than anyone plans for, across hundreds if not thousands of details.

We are still on that road. The book will be updated as Mk4 produces results, and those results may show that we were heading in the right direction, or prove us spectacularly wrong. Either way it will be written down here. But the work is what pays you back. There are not many feelings better than watching something you built go past the others on a race track for the first time.

Nothing new here — just a place to look up something the book has already defined, without hunting for the chapter that defined it.

The book in twenty questions, in roughly the order the answers are needed. Each one locks something that later questions depend on; chapter 04 explains why the order is forced where it is.

The evidence behind the second half of chapter 01, kept here so that the chapter can carry the argument and this appendix the arithmetic. The percentages for Ukkonen and for the MotoE rows can be recomputed from these tables; those for the other programmes come from the project's benchmark CMP.

B.1 Reference times

The baselines the percentages in chapter 01 are calculated from:

B.2 Does the choice of baseline change the answer?

A fair objection to all of the above: MotoGP is not an obvious yardstick for an electric bike of about 225 kg. Superbike is the closer class on mass and power, and the comparison is testable rather than arguable, because MotoE and WorldSBK share several circuits. Better still, MotoE runs on the same weekend as MotoGP, Moto2 and Moto3, which allows a properly controlled comparison: one track, one weekend, one set of conditions, four classes.

Start with the official race lap records at three circuits MotoE visits.

Two caveats belong with that table rather than under it. Mugello has no usable Superbike baseline at all: the WorldSBK record there was set in 1994 and the series no longer visits, so the cell is empty rather than unknown. And the Jerez MotoE record still belongs to the Energica, which means that row compares a 2021 machine against 2025–26 records for everyone else. It is the weakest line in the table, and it is the one the project's own benchmark was built on.

The same-weekend figures avoid all of that. These are the best race laps set by each class at one meeting.

And a Superbike anchor for the same circuit and season: at Assen in April 2023 Bautista's fastest lap of the Superpole race was 1:33.780, which puts the V21L +6.9 %, the smallest MotoE gap anywhere in this book.

Two things fall out of all this, and the second is the one worth having.

The ordering never moves. Whichever baseline, whichever circuit, MotoE sits seven to eleven percent off the front. Choosing Superbike instead of MotoGP as the yardstick moves the number by roughly half a percentage point, because WorldSBK itself is within half a second of MotoGP at these circuits; a prototype's aero and power advantage counts for much less than intuition suggests. The conclusion does not move at all.

But the size of the gap depends on the circuit, and it depends on it in exactly the way the mass argument predicts. The deficit is smallest at Assen, which is fast and flowing; larger at Mugello, which has a 1.1 km straight but also heavy braking into slow corners; largest at Jerez, which is tight and stop-and-go. It tracks the number of hard accelerations per lap, not the length of the straights. That is the signature of a machine limited by acceleration (the drive force it can put down, divided by its mass) rather than by grip or by top-end power, and the explainer in chapter 01 has the algebra for why.

The Moto3 row deserves a second look, because it overturns something. A figure that is often repeated has MotoE about 2 % slower than Moto3, and the ladder table in chapter 01 shows 2.6 % for the Ducati at Jerez. On same-weekend data the Ducati is faster than Moto3 at both Assen and Mugello. Those figures compare laps of different kinds: a 2021 Energica lap, or, in the Jerez row, a Ducati test lap from March 2023, each against Moto3 records from other years and conditions. A test lap is like-for-like against qualifying records at best, and against race pace not at all. Neither survives a controlled comparison.

B.3 The choices that separate them

Within that shared constraint, the surviving programmes have made noticeably different bets. The clearest one is how they reach their torque × rpm product, the number chapter 05 is about.

Ukkonen is the outlier here, and this was a consequence of the motor choice rather than a strategy chosen on its merits. It isn't obviously wrong, but it does mean the driveline strength requirement is an open item on this project and not on the others; it is recorded as O20.

The second bet is the pack, and here the convergence is more striking than the divergence:

None of that was copied. It came out of cell testing, lap data and a thermal model, and it landed within about eight percent of a works MotoGP-adjacent prototype on every line but mass, where the analysed figure is 13 % heavier and the target 2 % lighter. That is worth knowing when you are deciding whether your own numbers are plausible: if an independent route lands you somewhere very close to where the best-funded team in the category ended up, that is evidence the constraints are real and not a coincidence of your assumptions. DUCATIHB 30

One caveat on that table, in the spirit of chapter 12: the project's own benchmark document quotes 233 kg for Mk4 where the mass budget says 219.6 kg. Both are stated as being without the rider, so one of them is wrong or they are counting different things — recorded in the errata rather than quietly resolved. MBCMP

The third bet is generational: when MotoE replaced Energica with Ducati, torque fell 36 %, mass fell 35 kg, power stayed the same — and the bike got faster. Chapter 02 uses that case for a different job, checking that a simulation's mass sensitivity is the right size.

B.4 And if you're starting from nothing

The bikes above are not the right reference for a first project. The realistic entry point is MotoStudent Electric, which runs around eighty universities from nineteen countries, and where a typical bike is about 10 kWh at 100 V rather than 19 kWh at 650 V. That is a different design problem (low voltage changes the inverter, the safety case and the whole of chapter 06), but the order of work in chapter 04 is the same, and so is the finding that mass and grip come before power. CMP

B.5 Lightning and Lightfighter, in full

Lightning builds road motorcycles rather than running a racing programme, but its race-prepared machines are the sharpest version of the straight-line point, and the one that most deserves to be stated carefully. The LS-218 is named for the 218 mph (351 km/h) it ran at Bonneville, it has been listed as the fastest production motorcycle of any kind, and in 2013 a race-prepared one won the Pikes Peak International Hill Climb in open competition, ahead of the combustion motorcycles LSWIKI. It is worth mentioning, but no comparable circuit lap times from recent years are available, so Lightning is not part of the lap-time comparison.

Lightfighter's V3-RS is published at 181 kg ready to race, about the weight of a Ducati Panigale V2 by its makers' own comparison, with 115 kW LFV3NEWATLAS. The lap time in the comparison is from the V3-RH, the Super Hooligan version of the same bike: 1.5 % off Supersport at Barber, against 3.7–7.9 % for the heavier machines against their middle class. Barber's baseline is race pace, the softest of the three, so that flatters it somewhat; the direction is still clear. What is left against Superbike is power, since the Super Hooligan version races at 100 kW. The pack is not small: 12.5 kWh in one published account and 16 kWh on the team's own specification page, against 18 kWh in the MotoE bike DUCATI. It uses pouch cells, which the supplier says take about a fifth less space than cylindrical ones FARASIS; one published account puts it at roughly seven minutes of full throttle, and the Super Hooligan races in the comparison are eight laps long CMP. The pack is air-cooled with external fans between races, and the team lists pre-cooling among the conditions a lap time depends on FARASISLFFAST. None of the published material says whether heat or energy runs out first.

Corrections to figures the project published before this revision, and the conflicts in the project material that are still unresolved. Written in the spirit of chapter 12: recorded, not patched.

Corrections

Each line says what was stated, what the book states now, and what the correction rests on. Corrections to this revision will be added here in the same form. The numbers are identifiers, so that an entry can be referred to from elsewhere without ambiguity.

Checked and confirmed

Not everything that was checked needed changing. These held up against the source data and are worth listing so the reader knows they were checked rather than inherited.

• Per-cell current on Mk3. The logged lap gives 24 A RMS and 39 A peak per cell on 624 cells, consistent with the 40 A peak used throughout and with the 21 A RMS predicted for Mk4's larger pack. AiM
• Bench versus track. The 24-cell test modules at 21 A per cell reached the 52 °C cut-off in five to six minutes uncooled and in about eight at best with the bench cooling; the track session at 24 A RMS lasted nine. Same failure, same scale. LOAD
• The thermal chain arithmetic. 0.72 + 0.41 + 0.28 + 1.20 K/W ≈ 2.6 K/W, times 5.62 W ≈ 14.7 K, against the 14.8 K stated. The cell-core link at 0.72 K/W is 28 % of the chain, which the figure had inferred and the check confirms from the same document. CC §02/§09
• The 85 °C arithmetic. Arrhenius with 50 kJ/mol from 45 to 85 °C gives a factor of 8.3; the busbar resistances, powers and sags reproduce from resistivity and geometry; the adiabatic 30-second temperature rise on 1 mm copper comes out at 14–16 K.
• Ducati V21L and Molicel P50B headline figures against the manufacturers' own publications. DUCATIMOLI
• The nickel-versus-copper division of labour. Independent of the project's own calculation, the published literature gives the same answer from the other direction: nickel is the compatible metal against steel, copper-to-steel avoids the Al–Fe intermetallics but has cracking problems of its own, and aluminium-to-steel needs its brittle intermetallic layer kept thin, with about ten micrometres the usual rule of thumb. The project arrived at the industry's answer by arithmetic; the metallurgy agrees. ASIRVNIPLATECUFE
• WorldSBK is close to MotoGP. Checked at Assen as well as Jerez: 1:32.357 against 1:31.866, about half a percent. The observation holds at a second circuit. LAPREC
• Radial conductivity assumption. The 0.83 W/mK the cooling design uses sits inside the published range for cylindrical cells and above the careful 18650 measurements, so it is an upper-end assumption as the design document says. KOLLER

Open in the project material

These are conflicts or gaps in the source documents that this book found and does not resolve. Each belongs in the project's own conflict list.

Every reference key in the text points to an entry in this list. Paths are relative to the project folder 03 Ukkonen.

The list is grouped by what kind of evidence each item is, because that matters more than alphabetical order: a measurement of our own bike, a configuration file as it actually ran, a manufacturer's statement and a thesis carry different weight, and a reader should be able to see which is which.

What carries no reference

Four kinds of statement in this handbook don't carry a citation, and it's worth being explicit about which:

• General engineering relationships: the power equation, Arrhenius behaviour, laminar heat transfer scaling with channel height. Textbook material, not project findings.
• The order of work in chapter 4 and the method in chapter 12. These are the project's own working conclusions rather than anything measured. They're the part most likely to be wrong for a differently shaped team.
• The generalisation in each Rule block. Each one grew out of the example beside it, but the step from one case to a general rule is a judgement and not evidence.
• The explainer boxes, headed “In plain terms”. Standard definitions, written out rather than cited. Any textbook on electrical machines or battery engineering covers them properly, and should be preferred over these summaries for anything you're going to calculate with.