WR-217e

Powertrain Architecture, Simulation, and Controls

WR-217e Architecture Design

The first step of architecture design, especially when given a more-or-less blank slate, is to create an
architecture development model. It is better to make one architecture development model that only
uses high level car parameters (e.g. a lap time simulator) and then make a few more specific models
(e.g. step steer torque vectoring response, quarter-car suspension model, etc), rather than to try to
incorporate all of the models together. When starting to make a model, ask yourself: “What am I trying
to get out of this model?” or “What will I be using this model to optimize?” From there, start simple and
add complexity to the model as needed in order to achieve your goals.

These are the goals for the architecture development model that will be outlined in this document:

Goals

• Determine the optimal power source characteristics required to achieve high level performance
targets* (e.g. peak torque, peak power, torque curve shapes)
• Determine optimal transmission ratio(s) to achieve high level performance targets
• Determine required onboard energy to achieve the desired vehicle range
• Determine aerodynamic properties that result in the optimal tradeoff between performance
and efficiency
• Determine optimal high level chassis parameters (e.g. vehicle CG location)

*note: High level performance targets can include metrics such as peak acceleration and top speed.
These are chosen at the architecture engineer’s discretion, and may need to be revised later as physical
constraints are realized (e.g. packaging, vehicle mass, etc).

Now, we can identify the vehicle parameters we will need to include in our model:

Propulsion

• Torque-speed curve for the vehicle power source

• Number of gears and corresponding transmission ratios (𝑁1 , 𝑁2 … 𝑁𝑛 )
• Torque-speed efficiency map for power source (optional)

Chassis

• Total vehicle mass w/ driver (𝑚)

• Height of the center of gravity (𝐶𝐺𝑧 )
• Wheelbase and track width(s) (𝑤𝑏, 𝑡𝑤𝑓 , 𝑡𝑤𝑟 )
• Static weight distribution (𝑠𝑤𝑑) (from 0 to 1, higher is
more rearward)

Tires

• Estimate of tire rolling radius (𝑟𝑡𝑖𝑟𝑒 )

• Estimate of maximum longitudinal coefficient of friction (𝜇𝑙𝑜𝑛𝑔 )
• Estimate of maximum lateral coefficient of friction (𝜇𝑙𝑎𝑡 )
• Sensitivity of coefficient of friction to normal load (optional)
Aerodynamics

• Coefficient of downforce (𝐶𝑑𝑓 )

• Coefficient of drag (𝐶𝑑 )
• Frontal area (𝐴𝑓 ) (area of the vehicle from a front view)
• Center of pressure (𝑐𝑝) (from 0 to 1, higher is more rearward)
• Air density (𝜌)

Notice what is missing:

• Suspension geometry
• Spring rates, damping rates
• 26 DOF Pacejka tire model
• KNC compliance data

This is on purpose! We don’t need any of that information to achieve our goals. Notice that all of the
parameters of the car that we are trying to optimize are included as input parameters for our model.
Because of this, we will need to make some educated guesses for our initial inputs. These guesses will be
driven by a combination of data from older vehicles and first principles.

The vehicle coordinate system

Steps to build your model:

Step 1: Point-mass acceleration event

Step 2: Bicycle model acceleration event

Step 3: Two-track model autocross event

Step 1: Part A – The Point Mass

To build your model, start with a point mass that accelerates in one dimension. We will call this type of
model a 1-DOF traction limit acceleration model, because we will calculate the “traction limit” based on
the given longitudinal coefficient of friction between the point mass and the ground. In this model, it is
assumed some propulsion system is capable of fully utilizing the available friction force at the contact
patch to accelerate the vehicle in all conditions.

Inputs:

• Mass of the point [kg]

• Coefficient of friction [-]

Equations:

1) Kinematic relationships using the symplectic (semi-implicit) Euler approximation:

• 𝑣𝑖 = 𝑣𝑖−1 + 𝑎𝑖−1 𝑑𝑡
• 𝑥𝑖 = 𝑥𝑖−1 + 𝑣𝑖 𝑑𝑡
2) 𝐹𝑛𝑒𝑡 = 𝑚𝑎
3) 𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 = 𝜇𝐹𝑁

Calculations:

𝐹𝑛𝑒𝑡 = 𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛

𝑚𝑎 = 𝜇𝑙𝑜𝑛𝑔 𝐹𝑁

𝐹𝑁 = 𝑚𝑔
𝑎 = 𝜇𝑙𝑜𝑛𝑔 𝑔
Mass cancels out in this first case, but is included in the model as we will need it later on. When you run
your simulation, you should be able to plot acceleration, velocity, and position vs. time. You should be
able to see a linearly increasing velocity and a quadratically increasing position.

For this first model and as we move forward, I will walk you through how to calculate the accelerations
of the vehicle. It is up to you to create a numerical (discrete time-step) model to calculate vehicle
velocity and position.

Q: How do I decide on what increment to make 𝑑𝑡?

A: It entirely depends on the time constants of different phenomena in your dynamic system. For this
application, I recommend 𝑑𝑡 = 0.001. As 𝑑𝑡 → 0, the behavior of your numerical model will
asymptotically approach the exact solution, as shown in the plot of an arbitrary function below.
Step 1: Part B – Aerodynamic Effects

We will now add some aerodynamic parameters to our point-mass acceleration event.

Additional Inputs:

• Coefficient of downforce [-]

• Coefficient of drag [-]
• Air density [kg/m3]

Additional Equations:
1
1) 𝐹𝑑𝑜𝑤𝑛𝑓𝑜𝑟𝑐𝑒 = 2 𝜌𝐴𝑓 𝐶𝑑𝑓 𝑣 2
1
2) 𝐹𝑑𝑟𝑎𝑔 = 𝜌𝐴𝑓 𝐶𝑑 𝑣 2
2

Calculations

𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 − 𝐹𝑑𝑟𝑎𝑔 = 𝐹𝑛𝑒𝑡

𝐹𝑁 = 𝑚𝑔 + 𝐹𝑑𝑜𝑤𝑛𝑓𝑜𝑟𝑐𝑒
1 2 1 2
𝜇𝑙𝑜𝑛𝑔 (𝑚𝑔 + 2 𝜌𝐴𝑓 𝐶𝑑𝑓 𝑣𝑖−1 ) − (2 𝜌𝐴𝑓 𝐶𝑑 𝑣𝑖−1 ) = 𝑚𝑎

1 2
𝜌𝐴𝑓 (𝜇𝑙𝑜𝑛𝑔 𝐶𝑑𝑓 − 𝐶𝑑 )𝑣𝑖−1
𝑎 = 𝜇𝑙𝑜𝑛𝑔 𝑔 + 2
𝑚
Now, mass no longer cancels out. Acceleration still has the 𝜇𝑙𝑜𝑛𝑔 𝑔 component from Part A, but now also
has an additional component due to combined aerodynamic effects. Even from this simple model, it can
be seen that the aerodynamic effects will impact vehicle acceleration more as the vehicle mass is
decreased. We also find that when 𝜇𝑙𝑜𝑛𝑔 𝐶𝑑𝑓 > 𝐶𝑑 , the net effect of aerodynamics will increase
acceleration. This is not necessarily true once a more complex model is developed, but still interesting to
note here. This is because in reality vehicles usually do not ride the traction limit all the way to top speed
due to the incredible power required to do so. When the vehicle become power limited (no longer
traction limited), downforce is no longer increasing the vehicle’s ability to accelerate, but drag is still
decreasing the vehicle’s ability to accelerate. From this, we can conclude that given the above condition
is satisfied, the net effect of aerodynamics is positive when the vehicle is traction limited, and negative
when the vehicle is power limited.

Note: since acceleration is now velocity-dependent, we use the velocity from the previous time step as an
approximation for the velocity at the current time step.
Step 1: Part C – The Propulsion System

In this part, we will add a defined propulsion system to our model. For example purposes, we will keep
our propulsion system simple: an electric motor that can be characterized by a peak torque, peak power
and a fixed gear ratio. Now, instead of our point-mass vehicle accelerating at the limit of traction, we
will compare the capability of the motor to the traction limit and use the smaller of these forces to
accelerate the point mass. This is, in effect, a rudimentary “traction control” system.

Additional Inputs:

• Motor peak torque (N-m) (𝑇𝑚𝑎𝑥 )

• Motor peak power (W) (𝑃𝑚𝑎𝑥 )
• Transmission ratio [-] (𝑁)

Additional Equations:
𝑇max∗𝑁 𝑃𝑚𝑎𝑥
1) 𝐹𝑚𝑜𝑡𝑜𝑟 = min ( 𝑟𝑡𝑖𝑟𝑒
,𝑣 )
𝑖−1
2) 𝐹𝑐𝑝 = min(𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 , 𝐹𝑚𝑜𝑡𝑜𝑟 )

(force at the contact patch of the tire)

Calculations:

𝐹𝑐𝑝 − 𝐹𝑑𝑟𝑎𝑔 = 𝐹𝑛𝑒𝑡

𝐹𝑁 = 𝑚𝑔 + 𝐹𝑑𝑜𝑤𝑛𝑓𝑜𝑟𝑐𝑒

𝐹𝑐𝑝 − 𝐹𝑑𝑟𝑎𝑔 = 𝑚𝑎
𝐹𝑐𝑝 − 𝐹𝑑𝑟𝑎𝑔
𝑎=
𝑚
Step 2: The Bicycle Model

In this part, we turn our point mass into a two-wheeled bicycle. We will need to make a couple
adjustments to our equations to accommodate this, mainly by introducing the concept of weight
transfer (or, more correctly and specifically, longitudinal load transfer).

When a force is applied at the bottom of an object with a center of gravity some distance above the
ground, a moment is applied about the CG. The result of this moment on an object like a bicycle (with
two points in contact with the ground and therefore two separate normal forces) is that some of the
normal load from the front tire is transferred to the rear tire during acceleration (and vice versa during
braking). By doing a moment balance about one of the tires, this load transfer can be described:
𝐶𝐺𝑍
𝑊𝑇𝑙𝑜𝑛𝑔 = 𝑚𝑎𝑙𝑜𝑛𝑔 ∗
(𝑤𝑏)
The magnitude of this load transfer will be added or removed from each tire depending on the direction
of acceleration. We will now begin distinguishing accelerations and weight transfers as longitudinal (in
the direction of the bicycle’s travel), and eventually when we get to the two track model as lateral (in
the direction pointing from the vehicle to the center of a turn). We will now also have separate
equations for the front tire and the rear tire. For this example, we will assume there is a motor for each
tire as well.

Additional Inputs:

• Static weight distribution (𝑠𝑤𝑑)

• Wheelbase of the bicycle (𝑤𝑏)
• Height of the center of gravity of the bicycle (𝐶𝐺𝑧 )
• Center of pressure (𝑐𝑝)

Additional Equations:
𝑍 𝐶𝐺
1) 𝑊𝑇𝑙𝑜𝑛𝑔 = 𝑚𝑎𝑙𝑜𝑛𝑔 ∗ (𝑤𝑏)
2) 𝐹𝑁,𝑓𝑟𝑜𝑛𝑡 = 𝑚𝑔 ∗ (1 − 𝑠𝑤𝑑) + 𝐹𝑑𝑜𝑤𝑛𝑓𝑜𝑟𝑐𝑒 ∗ (1 − 𝑐𝑝) − 𝑊𝑇𝑙𝑜𝑛𝑔
3) 𝐹𝑁,𝑟𝑒𝑎𝑟 = 𝑚𝑔 ∗ (𝑠𝑤𝑑) + 𝐹𝑑𝑜𝑤𝑛𝑓𝑜𝑟𝑐𝑒 ∗ (𝑐𝑝) + 𝑊𝑇𝑙𝑜𝑛𝑔

Calculations

𝐹𝑐𝑝,𝑓𝑟𝑜𝑛𝑡 + 𝐹𝑐𝑝,𝑟𝑒𝑎𝑟 − 𝐹𝑑𝑟𝑎𝑔 = 𝐹𝑛𝑒𝑡

For the sake of space, we solve more generally:

𝐹𝑐𝑝,𝑓𝑟𝑜𝑛𝑡 + 𝐹𝑐𝑝,𝑟𝑒𝑎𝑟 − 𝐹𝑑𝑟𝑎𝑔 = 𝑚𝑎

𝐹𝑐𝑝,𝑓𝑟𝑜𝑛𝑡 + 𝐹𝑐𝑝,𝑟𝑒𝑎𝑟 − 𝐹𝑑𝑟𝑎𝑔
𝑎=
𝑚
If the front and rear wheels are both traction limited, acceleration actually reduces to the same formula
in Step 1 – Part B. However, when a wheel is power limited (i.e. the motor cannot fully utilize the
available grip), non-linearities are introduced into our model. Now, we have longitudinal acceleration
that is self-affecting.

𝑎 = 𝑓(𝐹𝑛𝑒𝑡 ) → 𝐹𝑛𝑒𝑡 = 𝑓(𝑊𝑇𝑙𝑜𝑛𝑔 ) → 𝑊𝑇𝑙𝑜𝑛𝑔 = 𝑓(𝑎)

For example, let us say that we start accelerating from a standstill. Initially, we assume there is no
longitudinal weight transfer, so the dynamic normal load distribution is such that both wheels are
traction-limited. Now, we have the resulting acceleration that we got in Step 1 – Part B. However, now
that the bicycle is accelerating, there is longitudinal weight transfer. If we go back and recalculate this,
we find that the new dynamic normal load distribution is such that now the front wheel is traction-
limited and the rear wheel is power-limited due to the additional normal load on the rear wheel. As such,
if we subsequently recalculate the contact patch forces, the net force on the bicycle will be different than
our first calculation. So what is actually going on?

This is a good time to bring up that the model we have made is a quasi-steady state model. This is
because the weight transfer equations that we are using ignore the transient effects of the
spring/dampers on the vehicle. As such, in reality weight transfer has a transient component from the
suspension and does not reach a steady state value instantaneously. Since we are neglecting this, we
must implement an iterative solver to determine the steady-state weight transfer for each time step. For
more information on how to implement a simple iterative solver, see Appendix X.

Using the steady state value of weight transfer at each time step is perfectly fine given our goals for our
architecture design simulation. While the transient effects of load transfer through a suspension system
have a massive and very important impact on the real-life handling (and subsequently performance) of a
vehicle, they do not significantly affect the optimization of the high level parameters we have set out to
investigate. For a more accurate model of our suspension, we could make a smaller and more specific
quarter-car model.
Step 3: The Two-Track Model

Congratulations! We’ve arrived at the part where we begin modeling a four-wheeled vehicle. In this
section, we will introduce a couple new ideas and turn our acceleration event simulator into an
autocross event simulator. I recommend saving a copy of the acceleration event simulator at this point
because it will come in handy later on.

Lateral weight transfer occurs when a vehicle experiences a centripetal acceleration towards the center
of a turn. This type of weight transfer can only occur for a vehicle that has two tracks (i.e. not a bicycle).
This transfer of normal load is also a result of a force being applied to an object at the ground, below its
CG. By doing a moment balance about one side of the vehicle, this load transfer can be described:
𝐶𝐺𝑍
𝑊𝑇𝑙𝑎𝑡 = 𝑚𝑎𝑙𝑎𝑡 ∗
(𝑡𝑤)
It should be noted that vehicles can have different track widths for the front and rear axles. In that case,
later weight transfer must be calculated separately for each axle. In our example, we will assume equal
track widths: 𝑡𝑤 = 𝑡𝑤𝑓 = 𝑡𝑤𝑟

We will use a simple law of circular motion to calculate lateral acceleration:

𝑣2
𝐹𝑐𝑒𝑛𝑡𝑟𝑖𝑝𝑒𝑡𝑎𝑙 = 𝑚𝑎𝑐𝑒𝑛𝑡𝑟𝑖𝑝𝑒𝑡𝑎𝑙 = 𝑚 ∗
𝑅𝑐
𝑣2
𝑎𝑙𝑎𝑡 = 𝑎𝑐 =
𝑅𝑐
Tires have what we call a “traction ellipse”. A tire cannot produce its maximum longitudinal and lateral
forces simultaneously (otherwise it would be referred to as a traction rectangle). Therefore, for a given
lateral tire force, there is a finite amount of grip remaining for longitudinal force. Race tires usually have
higher peak lateral capability than peak longitudinal capability. Therefore, we need to be able to
calculate the available longitudinal force for a given lateral force. In our ellipse calculations, we will refer
to longitudinal force as 𝐹𝑥 and lateral force as 𝐹𝑦 .

𝐹𝑥2 𝐹𝑦2
2 + =1
𝐹𝑥,max 𝐹𝑦,2max

Additional inputs:

• Track width (𝑡𝑤)

• Estimate of maximum lateral coefficient of friction (𝜇𝑙𝑎𝑡 )
• Corner radius: (𝑅𝑐 )

[lap simulator introductory guide is unfinished and will

be expanded upon. To be continued…]
 Example Case: WR-217e Architecture

For the Formula SAE Electric competition, vehicles are restricted to drawing a maximum of 80 kW from
the tractive battery. This rule is actually very useful in helping constrain our architecture design. First, let
us run a two-track (four-wheeled) acceleration event simulator given the following constraints:

1. The FSAE Electric Acceleration Event is 75m long

2. The maximum usable power is 80 kW
3. The vehicle cannot propel itself beyond the capabilities of its tires.

In addition to these three constraints, we will need to determine some initial values for our vehicle
model. Luckily for us, we have data from our combustion car (cCar) to use as a jumping off point.

Parameter Variable Value [units] Reasoning

Total mass of vehicle w/ driver 𝑚 320 [kg] (cCar mass)*1.3 + heavy driver, conservative
Center of gravity height 𝐶𝐺𝑍 0.25 [m] (cCar CG height)*0.85
Vehicle wheelbase 𝑤𝑏 1.6 [m] Same as cCar
Static weight distribution 𝑠𝑤𝑑 0.55 [-] 2% more rearward than cCar
Estimate of tire rolling radius 𝑟𝑡𝑖𝑟𝑒 0.22 [m] Same as cCar (at std. pressure & load)
Estimate of maximum longitudinal 𝜇𝑙𝑜𝑛𝑔 1.4 [-] Correlated to cCar track data
coefficient of friction
Estimate of maximum lateral 𝜇𝑙𝑎𝑡 1.7 [-] Correlated to cCar track data
coefficient of friction
Coefficient of downforce 𝐶𝑑𝑓 3 [-] Same as cCar
Coefficient of drag 𝐶𝑑 1.5 [-] Same as cCar
Frontal area 𝐴𝑓 1.21 [m2] Same as cCar
Center of pressure 𝑐𝑝 0.55 [-] Same as cCar
Air density 𝜌 1.15 [kg/m3] In Lincoln, NE (where comp. is held)

We now have enough information to get this car model rolling! But first, some notes:

1. We will be using the metric of 0 – 75m time to gauge the vehicle’s performance.
2. We will keep the aerodynamics package at the standard angle of attack: this is more
conservative than trimming out wings since our goal is properly sizing our propulsion system.
This is because the additional downforce will result in more available grip that the propulsion
system will have to utilize.
3. When the amount of total available tractive power (∑ 𝐹𝑥,𝑡𝑖𝑟𝑒𝑠 ∗ 𝑣𝑣𝑒ℎ𝑖𝑐𝑙𝑒 ) exceeds 80 kW, force
will be removed from each contact patch proportional to where the grip is available. At the
power limit, force can be removed from the contact patches in any combination- doing so in this
particular way will ensure that neither the front nor the rear propulsion systems will have to
work disproportionately hard. This mechanical 80 kW power limit will be an approximation of
the rules-imposed 80 kW electrical power limit (since we have no way of quantifying electrical
system losses yet). In the real world, these electrical losses would result in some percentage (80-
95%) of the electrical power to be converted to mechanical power at the wheels of the vehicle.
Accel @ 80 kW Plots

Plot 1: Acceleration

This plot is useful mostly for sanity checking the simulation. Initial acceleration appears to be around
1.3G, which is to be expected with 𝜇 = 1.4 and some tire load sensitivity built into the model.
Acceleration increases quadratically in the first second or so, which is to be expected due to increasing
tire capability (as a function of downforce, which itself is a function of 𝑣 2 ). At around 𝑡 = 1.35 𝑠𝑒𝑐, the
mechanical power limit kicks in and acceleration starts to fall off.
Plot 2: Total Mechanical Power

This plot is also mostly useful for sanity checking the simulation. Power starts at zero (which makes
sense given 𝑣 = 0), and increases slightly more than linearly (which makes sense given slightly
increasing available traction due to downforce), and saturates at 𝑃𝑚𝑒𝑐ℎ = 80 [𝑘𝑊] as expected.
Plot 3: Individual Tire Powers

This plot gives us a good idea of the peak motor power our propulsion system should be capable of
producing for both the front and rear wheels. From the plots, we can generally say:

𝑃𝑝𝑒𝑎𝑘,𝐹 = 18 [𝑘𝑊]

𝑃𝑝𝑒𝑎𝑘,𝑅 = 28 [𝑘𝑊]

This is true if each wheel is individually driven by a motor. If, for example, one wanted to implement a
single inboard rear motor, the peak power requirement would be 2 ∗ 𝑃𝑝𝑒𝑎𝑘,𝑅 .
Plot 4: Individual Tire Torques

This plot gives us a good idea of the peak tire torque (post-transmission) our propulsion system should
be capable of achieving for both the front and rear wheels. From the plots, we can generally say:

𝑇𝑝𝑒𝑎𝑘,𝐹 = 150 [𝑁𝑚]

𝑇𝑝𝑒𝑎𝑘,𝑅 = 360 [𝑁𝑚]

This is true if each wheel is individually driven by a motor. If, for example, one wanted to implement a
single inboard rear motor, the peak torque requirement would be 2 ∗ 𝑇𝑝𝑒𝑎𝑘,𝑅 .
Plot 5: Torque-Speed Curves

This plot gives us an idea of the torque-speed relationship we want for our electric motors.
Justifying All-Wheel Drive Architecture

The heaviest component in an electric vehicle is the high voltage tractive battery. As such, the electric
motors and inverters account for a significantly smaller portion of the vehicle mass. Upon inspecting the
above plots, it is clear that having tractive force at the front tires (contact patches) is preferable from an
architecture standpoint, as it allows for ~30% higher longitudinal acceleration capability for a small
increase in mass. We can ground this idea in first principles:

𝐹 = 𝑚𝑎
In a simplified sense, as long as the addition of a front wheel propulsion system increases the
accelerating force on the vehicle 𝐹 more than it increases 𝑚, the net effect will be positive on
longitudinal vehicle performance.

Electric Motor Selection

In addition to hitting our peak torque and power specifications, the shape of the torque-speed curve is
important in the motor selection process. The general shape of the desired torque curves for our motors
include a relatively flat peak torque region followed by a downward sloping relatively constant power
region (due to the nature of the overall power limit on the car). Motors that have this characteristic
have what is called flux-weakening capability. Ideally, we would like motors with this capability.

That being said, there are many other factors that can influence the selection of a propulsion system,
chiefly packaging constraints and other electrical rules (such as maximum tractive voltage). During our
selection process we must meet a series of requirements before even considering performance targets.

1) Motors and motor controllers must obey all FSAE rules including tractive voltage limit.
2) Motors, motor controllers, and required transmissions must be able to package inside the
geometric constraints of the vehicle.
3) Motors and motor controllers must have a lead time that is compatible with the project
timeline.
4) Motors and motor controllers must be affordable by the team.

If a propulsion system meets all of these requirements, we can move onto achieving performance
targets, such as peak power, peak torque, and torque curve shape.
Case Study: WR-217e Motor Selection

Plettenberg Nova 15 Specifications (Front in-hub motors):

• Peak torque: 28 [Nm]

• Peak power: 20 [kW]
• Linear peak power curve, constant peak torque curve (no flux-weakening)
• Required gear reduction to achieve target tire torque: 5.3:1
• Mass: 3 [kg]
• Water jacket

Plettenberg Nova 30 Specifications (Rear inboard motors):

• Peak torque: 61 [Nm]

• Peak power: 30 [kW]
• Linear peak power curve, constant peak torque curve (no flux-weakening)
• Required gear reduction to achieve target tire torque: 5.9:1
• Mass: 6 [kg]
• Water jacket

Parameter Optimization

No matter what range of values are swept for vehicle mass, acceleration event time will be minimized as
mass gets smaller. On the other hand, the gear reduction ratio from a motor shaft to the tire will have a
global minimum in acceleration time. This is because as gear reduction increases, the top speed of the
vehicle decreases, and as gear reduction decreases, peak acceleration decreases. Because high
acceleration and high top speed are both desirable, there is a tradeoff involved in determining the
optimal gear ratio. Vehicle mass and gear reduction are examples of two kinds of parameters-
parameters that produce a global extrema of the output metric at some value within their defined
domain, and parameters that do so at a boundary of the domain. It is useful to sweep both types of
parameters around a nominal value in order to determine the sensitivity of acceleration event time to
that parameter. For example, by sweeping vehicle mass -20 kg to + 20 kg of the nominal value, we can
𝑑𝑡
determine the change in acceleration event time per unit mass (𝑑𝑚). Besides sensitivity analysis,
parameters with corner solutions are less interesting from an architecture optimization standpoint.

Gear Ratio Optimization

Now that we have selected the electric motors for our case study, we must determine the optimal gear
ratios to minimize acceleration event time. Previous lap time simulations have shown that optimizing
gear ratios for an acceleration event are sufficient for optimal performance in autocross events as well.
We will use an acceleration event due to less computational overhead.

It should be noted that the optimal gear ratios are not necessarily just the ratio required to achieve the
target peak tire torque. This is because that particular ratio may reduce the top speed of the car in such
a way that the overall acceleration event time goes up as a result.
First, we will sweep the rear gear ratio. It is imperative that we perform this analysis for varying
coefficients of friction- it is obvious from the result that the coefficient of friction changes the optimal
ratio drastically. This is because at low coefficients of friction, the car’s ability to accelerate is limited by
traction and the fastest acceleration event time will be achieved by increasing top speed (lower gear
ratio). On the other hand, at higher coefficients of friction the car has more grip it can utilize so the
optimal gear ratio will trend higher. As the coefficient of friction increases, the optimal ratio shifts from
3.5:1 to 4.5:1. Since higher coefficients of friction better represent the track surface at competition in
Lincoln, NE, a rear gear ratio of 4.5:1 was selected.
Next, we sweep the front gear ratio. This plot looks different from the previous plot due to the fact that
the front motors are significantly traction limited. Therefore, in the range of 6:1 to 9.5:1, lap time
simulation shows that there is no significant difference in performance. For a given output torque, the
motor itself will be more efficient with a higher gear ratio (since more torque is a result of the speed
reduction instead of phase current). That being said, packaging constraints limit our gear reduction to a
maximum of 6:1 in a single stage planetary gear set. Since packaging constraints must take priority (and
a compound gear reduction is a challenge to be tackled in a future year), a 6:1 reduction was selected.
Accumulator Energy Determination

Next, a full endurance lap simulator is utilized in order to determine the required accumulator energy to
finish our 22 km race. Using some relatively basic loss models for the electric motors/motor controllers
and the battery, we can include the effect of electrical losses in our analysis.

The electric motors have losses that are loosely categorized as torque-based losses (joule losses, I2R) and
speed-based losses (hysteresis losses, eddy current losses). Torque-based losses are easily accounted
for, and a linear speed-based loss model was developed for the motors in absence of dyno data.
2
𝐿𝑜𝑠𝑠𝑚𝑜𝑡𝑜𝑟 = 𝐼𝑝ℎ𝑎𝑠𝑒 ∗ 𝑅𝑝ℎ𝑎𝑠𝑒 + 𝑉𝑏𝑎𝑐𝑘𝐸𝑀𝐹 ∗ 𝐼𝑜 (𝑤ℎ𝑒𝑟𝑒 𝐼𝑜 = 𝑛𝑜 𝑙𝑜𝑎𝑑 𝑐𝑢𝑟𝑟𝑒𝑛𝑡)

The battery losses are tracked based on the battery’s internal resistance. The losses from the battery are
both electrical and chemical in nature, but can be roughly modeled as joule losses based on
2
𝐿𝑜𝑠𝑠𝑎𝑐𝑐 [𝑊] = 𝐼𝑎𝑐𝑐,𝑖𝑛𝑠𝑡𝑎𝑛𝑡𝑎𝑛𝑒𝑜𝑢𝑠 ∗ 𝑅𝑎𝑐𝑐,𝑛𝑜𝑚𝑖𝑛𝑎𝑙

The tricky part of this is that the accumulator losses depend on the resistance of the accumulator, which
depends on the number of cells in the accumulator. Since we are trying to determine the number of
cells to implement, an educated guess for accumulator resistance must be made at the beginning of
optimization.

One tool we will be implementing in this analysis is the idea of a software power limit. By rules, the
vehicle may not draw more than 80 [kW] from the accumulator. It is, however, within the ability of the
team to voluntarily lower the power limit on the vehicle in the control software. This has the potential
to significantly decrease energy consumption due to the fact that losses go up quadratically (or more!)
at higher power draws. A sweep of software power limit vs. endurance lap time and energy
consumption was performed.
It can be seen from this sweep that as the power limit is reduced from 80 [kW] to 40 [kW], lap time
increases from 80 [sec] to 82 [sec], or a 2.5% increase. At the same time, energy consumption goes from
8.1 [kWh] to 6.5 [kWh], or a 20% decrease! For the endurance event where the main goal is to finish at
all, this tradeoff between performance and probability of event completion is well worth it. From the
plot above, it can be seen that the lap time and energy consumption curves begin to change slope
significantly below 40 [kW], and so the endurance software power limit we selected was 40 [kW]. At this
power level, the required energy is 6.5 [kWh]. Our next question becomes: why is there such a large
drop in energy consumption between 80 [kW] and 40 [kW]?

For this, we will have to dig a bit deeper into the simulation to determine where the energy is going.
Simulation shows that the majority of accumulator losses occur in straightaways at peak accumulator
power draw.
The above plot was generated from an acceleration event. From t = 0 to t = 1.6, power increases more
or less linearly as the car accelerates at a relatively constant rate. From t = 1.6 to t = 2.1, the power
limiter saturates the usable electrical power at 80 [kW]. After t = 2.1, the car approaches top speed and
power decreases once more. The purple line represents the amount of power produced by the cells in
the accumulator, which is a combination of the usable electrical power measured by the energy meter
and the power lost to heat. It can be seen that the purple line diverges quickly from the blue (usable
electrical power at the battery terminals) line when the usable electrical power exceeds 40 [kW]. The
delta between the purple and blue lines is the amount of power lost to heat in the battery, and at peak
power the battery is ~80% efficient with losses close to 20 [kW]. The delta between the blue and yellow
lines is the power lost to electrical losses in the motors/motor controllers. The yellow line is the
mechanical power produced by the vehicle. The delta between the yellow and red lines is the power lost
to drag. The takeaways from this plot are a) the battery is very inefficient at high powers (in upcoming
years it would be worth developing a lower internal resistance battery), and b) the aggressive
aerodynamics package requires a significant amount of power to overcome drag at high speeds, in
excess of 10 [kW], indicating it would be worth designing an active aerodynamic drag reduction system
in future years. This is especially true given that the energy density of a lithium battery is an order of
magnitude worse than that of a chemical racing fuel, and as a result the weight savings achievable by
implementing a smaller battery due to the reduction in energy consumption could be substantial.
Optimal Split: Maximizing Efficiency

You may have noticed the delta between the blue and yellow lines decreasing as time progresses at the
80 [kW] power limit. This is because the simulation employs an optimal splitting strategy once the
electrical power limit is engaged in order to maximize the amount of those 80 [kW] being converted into
mechanical power (or, alternatively, minimize the amount of electrical losses). This is accomplished by
minimizing the torque-based joule losses in the system. Because the front motors are different than the
rears, the [W] of loss per [N] at the contact patch is different for each motor.

𝑃𝑒𝑙𝑒𝑐,𝑡𝑜𝑡𝑎𝑙 = ∑ 𝐹𝑐𝑝 ∗ 𝑣 + ∑(𝑗𝑜𝑢𝑙𝑒 𝑙𝑜𝑠𝑠𝑒𝑠) + ∑(𝑠𝑝𝑒𝑒𝑑 𝑙𝑜𝑠𝑠𝑒𝑠)

At the electrical power limit, the total electrical power is a constant, and the control system has no way
of directly controlling the speed-based losses. Therefore, we focus on the joule losses. We want to find a
way to split the tractive force between the front motors and rear motors to minimize these losses.

𝐽𝑜𝑢𝑙𝑒 𝑙𝑜𝑠𝑠𝑒𝑠 = 2(𝐼 2 𝑅)𝑓𝑟𝑜𝑛𝑡 + 2(𝐼 2 𝑅)𝑟𝑒𝑎𝑟

We will now rewrite the joule losses equation in terms of contact patch forces.
2 2
𝑟𝑡𝑖𝑟𝑒 ∗ 𝐹𝑐𝑝,𝑓𝑟𝑜𝑛𝑡 𝑟𝑡𝑖𝑟𝑒 ∗ 𝐹𝑐𝑝,𝑟𝑒𝑎𝑟
𝐽𝑜𝑢𝑙𝑒 𝑙𝑜𝑠𝑠𝑒𝑠 = 2 ( ) ∗ Ω𝑓𝑟𝑜𝑛𝑡 + 2 ( ) ∗ Ω𝑟𝑒𝑎𝑟
𝑁𝑓𝑟𝑜𝑛𝑡 ∗ 𝐾𝑡,𝑓𝑟𝑜𝑛𝑡 𝑁𝑟𝑒𝑎𝑟 ∗ 𝐾𝑡,𝑟𝑒𝑎𝑟
𝑁𝑚
where 𝐾𝑡 𝑖𝑠 𝑡ℎ𝑒 𝑚𝑜𝑡𝑜𝑟 𝑡𝑜𝑟𝑞𝑢𝑒 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑖𝑛
𝐴

Note that the dimension of the product inside the parentheses is equal to amps. To find where losses
are minimized, we want to take the partial derivative of losses with respect to front contact patch force
and rear contact patch force, and set both of them equal to zero. Since both are set equal to zero, we
can set the partial derivatives equal to each other and solve for a force ratio.
2
1
𝐹𝑐𝑝,𝑓𝑟𝑜𝑛𝑡 (𝑁 ) ∗ Ω𝑟𝑒𝑎𝑟
𝑟𝑒𝑎𝑟 ∗ 𝐾𝑡,𝑟𝑒𝑎𝑟
= 2
𝐹𝑐𝑝,𝑟𝑒𝑎𝑟 1
(𝑁 ) ∗ Ω𝑓𝑟𝑜𝑛𝑡
𝑓𝑟𝑜𝑛𝑡 ∗ 𝐾𝑡,𝑓𝑟𝑜𝑛𝑡

A quick sanity check is to think about what would happen if the front motors and rear motors were the
same. If this were the case, the optimal force ratio would be 1, or 50% of the force in the front and 50%
of the force in the rear. This makes sense because it would be less efficient to have either the fronts or
the rears do the majority of the work. In the case of our electric racecar, optimal split is around 25%
front. This also intuitively makes sense- a smaller motor will be less efficient at producing a force at the
contact patch than a larger one.
The above plot is one way to visualize optimal split. The contours are constant loss lines. It can be seen
that for constant losses, one can achieve higher total powertrain force by using a split of around 25%
front. This is why all of the contours have a peak around that percentage. The peak in the middle is a
result of the diminishing number of ways you can split tractive force as it increases. At the extreme,
around 3600 [N], there is only one way to split the force: around 45% front. There is only one solution
because this is the point where all motors are producing peak torque simultaneously. The lower the
total powertrain force is, the more flexibility the control system has in terms of splitting it up.
This is a different way to visualize optimal split. For a given amount of tractive force requested by the
driver, there is a finite power savings between the most and least optimal split percentages. Again, as
force approaches 3600 [N], the number of split possibilities approaches zero and therefore the power
savings between the best and worst split percentage also approaches zero. Around 1500 [N], the
maximum power savings are possible, with a delta of over 5.5 [kW]!

Note: the jagged line is just an artifact of numerical calculation error. The true curve is smooth except
for the peak.
Sensitivity Report

Variable Nominal Value Delta Delta Accel Time (ms) Sensitivity Sensitivity Units
𝑚 320 [kg] -20 [kg] -45.4 2.3 [ms/kg]
𝜇𝑙𝑜𝑛𝑔 1.4 [-] +0.1 [-] -18.7 -186.5 [ms]
𝐶𝐺𝑧 25 [cm] -1 [cm] -6.8 6.8 [ms/cm]
𝑤𝑏 160 [cm] +10 [cm] -9.9 -9.9 [ms/cm]
𝑠𝑤𝑑 55 [%] -5 [%] -41.0 8.2 [ms/%]
𝑟𝑡𝑖𝑟𝑒 22 [cm] +1 [cm] -35.2 -35.2 [ms/cm]
𝐶𝑑𝑓 3 [-] +0.1 [-] -0.5 -5.5 [ms]
𝐶𝑑 1.5 [-] -0.1 [-] -2.7 27.0 [ms]
𝐴𝑓 1.21 [m2] -0.1 [m2] -1.9 19.2 [ms/m2]
𝑐𝑝 55 [%] -5 [%] -2.1 4.2 [ms/%]
Ω𝑐𝑒𝑙𝑙 23 [mOhm] -5 [mOhm] -35.8 7.2 [ms/mOhm]
𝑁𝑐𝑒𝑙𝑙𝑠,𝑠𝑒𝑟𝑖𝑒𝑠 30 [cells] +1 [cell] -50.2 -50.2 [ms/cell]
𝑁𝑐𝑒𝑙𝑙𝑠,𝑝𝑎𝑟𝑎𝑙𝑙𝑒𝑙 24 [cells] +1 [cell] -6.8 -6.8 [ms/cell]
𝐾𝑡,𝑓𝑟𝑜𝑛𝑡 0.1 [N-m/A] -0.01 [N-m/A] -6.6 662.9 [ms/(N-m/A)]
𝐾𝑡,𝑟𝑒𝑎𝑟 0.217 [N-m/A] -0.02 [N-m/A] -56.8 2840.8 [ms/(N-m/A)]
𝑇𝑚𝑎𝑥,𝑓𝑟𝑜𝑛𝑡 28 [N-m] +1 [N-m] 4.6 4.6 [ms/N-m]
𝑇𝑚𝑎𝑥,𝑟𝑒𝑎𝑟 61 [N-m] +2 [N-m] -20.7 -2.1 [ms/N-m]
𝑁𝑓𝑟𝑜𝑛𝑡 6 [-] +0.5 [-] 10.6 21.2 [ms]
𝑁𝑟𝑒𝑎𝑟 4.5 [-] -0.5 [-] -61.9 123.8 [ms]

*highlighted in green are the areas in which the architecture has significant room for improvement. The
car could and should certainly be lighter. The static weight distribution should be closer to 50/50 in order
to better utilize the front powertrain. The rolling radius of the tire should be carefully identified in order
to ensure it is correct due to the high sensitivity of the accel event to it. Lower internal resistance cells
should be investigated in order to reduce losses at the power limit (and maximize mechanical power). A
higher voltage should be investigated, in this case in order to increase the top speed of the car. The rear
motors should either have a slightly lower gear ratio or lower Kt, as the rear motors are the limiting
factor for the top speed of the car, and a slightly higher top speed would further reduce accel times.
Overall Architecture Comparison

To wrap up, we should evaluate the architecture we settled on. We will benchmark our architecture in
an acceleration event against: a) riding the traction limit to 75 m, b) riding the traction limit and then
limiting to 80 kW mechanical, c) one of our competitor’s architecture, and d) Wisconsin Racing’s own
combustion car.

For these simulations, our architecture will be benchmarked against a vehicle of the same mass for a)
and b), and will use real car parameters for c) and d).
Takeaways:

1) Delta between yellow and red is our room to improve in terms of powertrain architecture alone
(chassis mass would shift blue, red, and yellow leftwards).
2) 217e succeeds at besting competitor UPenn Electric, whose car is significantly lighter but RWD
only.
3) Wisconsin Racing’s first ever electric car should be the quickest vehicle ever made by the
university!
Model Validation: Reality vs. Expectation and Model Correction

After dyno testing electric motors and motor controllers, we learned:

• Motor controller maximum continuous phase current: 200 A reality vs. 280 A expectation
• Rear motor torque constant: 0.175 Nm/A reality vs. 0.217 Nm/A expectation
• Transmission losses greater than expected: 15% reality vs. 5% expectation
• No regeneration capability from powertrain

Effects on the vehicle:

• Accel times increased by 40%

• Peak tire torque decreased by 50%
• Autocross times increased by 5%
• Peak electrical power draw from battery decreased by 20% during accel and autocross (80 kW ->
65 kW)
• Endurance energy consumption had negligible change
▪ 40 kW software power limit was selected for endurance to balance tradeoff of energy
consumption and laptime. This limiter remains the dominating factor in endurance energy
consumption- even with the loss in performance from expectation vs. reality, average
accumulator power draw during endurance was mostly unaffected
▪ Accumulator energy was designed for the contingency that regen would be nonfunctional.
As such, the car still would have been able to finish an endurance.

Processing track data: Twenty laps processed, dashed line is the mean lap velocity
Statistical analysis:

• Orange line is mean lap velocity from track data

• Blue region is the result of statistical analysis
• Dashed line is the LapSim velocity after tuning the model to track data
Lost Torque from Original Design Intent:

Design Intent vs Result Wheel: Torque Walk

300

250
Wheel Torque (Nm)

200

150

100

50

0
Original Torque Constant Current Limited Driveline Losses
Expectation Lower Than Back by Inverter
EMF Waveform
Rear Wheel Front Wheel
Takeaways:

1) Be very skeptical of “peak” specifications. Peak can mean anything from “maximum continuous
capability” to “highest possible value for 1 ms before damage”. Be fastidious and insistent- press
the manufacturer to be as clear as possible about their peak specifications. You should always
be able to make a plot of “pulse performance” vs. “pulse time”
a. For example, a battery or inverter could supply 250 A for 1 sec, 200 A for 20 sec, 150 A
for 20 min, and 100 A continuously, etc
2) Always ensure the correct safety measures are implemented- either by yourself or by the
manufacturer. Lack of standard safety mechanisms such as overcurrent protection, overvoltage
protection, reverse voltage protection, etc should be a red flag that the product has corners
being cut. This is important to ensure the safety of your team members, but also important to
ensure that in a fault condition your hardware does not become damaged.
3) If no data is available, be conservative in modeling loss mechanisms such that powertrain
performance has a reasonable factor of safety (ie motor losses, inverter losses, transmission
losses).
4) Take the “make vs. buy” decisions very seriously. We’re here to learn, and sometimes the best
way to do that is to make things yourself. Get out there and push yourself to design new things
and learn new concepts!

Max Liben

Wisconsin Racing Formula SAE Electric

Team President ’16-‘17

Powertrain Architecture, Simulation & Controls

(917) 608-1164 | liben@go.uwracing.com