Test day

Rather than rewrite all the results again for this post, I've decided to just use a copy of the interim report from the first aero testing of the 2024 side-floors. I was originally going to do a separate post on preparing equipment and test plans for this test day, but it seemed unnecessary in the end. The report may contain some terminology I haven't used on this site, and may assume some context that is not given here, but in general should be followable. For a report that's meant to summarise results, I'll admit it's quite wordy, but I wanted to get all my thoughts to paper given there's no aero team to put the results to immediate use.

I'd been gone from the team for over a year by the time testing finally happened as it was a skeleton crew operating in 2025 and they barely managed to get a car running in time for competition, so aero testing wasn't a priority. I went and helped out at the test day when it finally happened in March 2026, and it was a good feeling to round out my year of work, two years later. The other aero team members have also since left the team, so I'm hoping I've left enough in the way of guides and data so that someone can come along and pick up for year two of the five year plan (see Defining the scope). I never said the years had to be consecutive...

A videos was created using onboard footage with data overlays to examine performance sensitivity in real time. The purpose of each overlay is briefly explained in the video description. Static taps are installed on the left side-floor.

The loose mounting cable seen on the left side at low speed was due to a chassis shape incompatibility between the EV24-designed floor and EV25 being used to test it. Downforce pulled the floor into the correct position and applied tension to the cable above 11m/s. All steady-state tests took place above 11m/s and so were not affected, but data in the video should be ignored below 11m/s. Also, I should mention that the yaw overlay is quite accurate, and at worst is only expected to be about 1 degree in error. The apparent independence of the yaw reading to what the car is doing visually is due to a gusty light/moderate wind. This is discussed in more detail in the report. 

1. Introduction

Testing was undertaken on 27th March 2026 at Orielton for the aerodynamic side-floors on EV25, designed and manufactured in 2024 for EV24. The primary goals of this test session were to validate performance predictions from the primary CFD design cases, to highlight areas for improvement in correlation, and to identify any performance sensitivities or otherwise unexpected behaviour that indicates a need for changes to or additional design cases.

A dedicated “skidpan” 9m radius test case, and installation of a total-pressure rake in the diffuser region were removed from the original test plan due to time constraints.

Tests were undertaken on EV25, which has some notable changes in chassis-shape over EV24, on which all previous CFD studies have been run, so not all results are directly comparable at this time. Due to abruptly losing the previous CFD license, there is very limited data from the 2024 CFD results for comparison to experimental results. Where affected, discussions will be minimal, with no conclusions at this time.  

A final decision on the preferred next steps, both for potential CFD setup alterations and surface design, will be deferred until new CFD benchmark cases have been created for the as-tested conditions in ANSYS (previous CFD results were obtained in STAR-CCM+), but some preliminary and conditional recommendations are made.  

Methods, processed results, and a brief discussion is included in each sub-section. Raw results are stored as log and mat files on SharePoint in /General/2025EV/Track data/27-APR-2026 Aero Floors Validation. MATLAB code used to produce key plots is included in R drive under continuing documents/design code/aerodynamics.

2. CFD validation tests

2.1 Static pressure on floor suction-side

Primary test method for performance validation. Static taps installed on the suction-side of the right-side floor (see Figure 1; total of 17 for Straight and Braking cases, and 14 for Cornering due to addition of pitot-static and yaw probes mounted on the nose, in-line with the front axle).

The taps were installed for steady-state cases for direct comparison to CFD results, and for full-lap autocross-style driving for analysis of typical responses to various driving scenarios and any particular sensitivities. The installed taps and pneumatic tubing are shown in Figure A1 (Appendix).

2.1.1 Straight 18m/s

Method – 10 runs at a constant 18m/s on straight track for 70m, avoiding wind where possible. Data extracted at 18m/s with a +/- 3% dynamic pressure window, and averaged.

The majority of points lie close to if not within the uncertainty window (Figure 3), so overall these results show acceptable accuracy, especially when considering the combined value, with CFD over-predicting local downforce by 7.5% (Figure 2). The largest individual errors occurred primarily at sensors 1, 4, 5, 6, and 7, all with lower-than-predicted magnitudes. Sensors 15, 16, and 18 had a notable increase in magnitude over CFD. This suggests the pitch of the car (or the floor) may have been slightly nose-up as a result of setup error, but equally the somewhat contained error magnitudes could just as easily be explained by the differences in chassis shape relative to the simulated EV24.

A particularly important and pleasing result is in the accuracy of sensors 12 and 13, suggesting the predicted vortex in this region has good correlation in both strength and approximate location.

Pressure distribution shows strong shape resemblance, and that the largest errors were typically located on high pressure gradients (Figure 4).

2.1.2 Braking 18m/s

Method – 10 runs accelerating to 20m/s then braking at maximum deceleration to below 18m/s, repeated on straight track for 70m, avoiding wind where possible. Data extracted at 18m/s with a +/- 3% dynamic pressure window, and averaged.

Results are generally similar in accuracy trends to the straight-airflow case, but average errors are larger, and experimental pressure are more typically of lower magnitude than CFD (Figures 5, 6). Combined error shows CFD over-predicting local downforce by 16.5%, which is still within a generally acceptable range of <20%, however the accuracy of the straight-airflow case suggests better correlation is possible.

Suspension changes during design and manufacture of EV24 (carried through to EV25) reduced dive and pitch over initial estimates. The EV24 CFD setup used the initial estimates, which resulted in a lower average ground clearance and larger diffuser expansion ratio, which could explain the larger errors. No CFD images for comparison of pressure distribution are available for this case.

2.1.3 Median corner radius

Method – 12 runs at a constant velocity through a 13.2m radius corner in alternative directions over a 90o arc, avoiding wind where possible. Data extracted at steady velocity and lateral acceleration, separated by corner direction, and averaged.

No numerical data exists to compare this case to, as none was extracted from CFD prior to the license removal. For completeness, the experimental data is still included here.

Sensors 0, 1, and 4 were removed for this case onwards, with sensor 19 being moved to replace sensor 4.

2.1.4 Pressure distribution comparison

Figures 9-12 show the average experimental static pressure distributions with lighter blues indicating larger negative magnitudes. The magnitudes have been normalised by freestream total pressure to allow direct comparison, despite differing velocities.

The braking case shows a forward-shift of the distribution and overall decrease in magnitudes, suggesting forward pitch, as expected. The inboard corner case also shows a decrease in magnitudes over the straight case, reflective of the increased ground clearance due to roll. Distribution remains similar. The outboard corner case shows the suction peak is both stronger, and positioned further outboard in the main diffuser tunnel, with a stronger vortex in the outboard tunnel. Suction magnitude is decreased alongside the vortex, and at the rearmost inboard sensor.

The outboard distribution image was calculated excluding instances where sensor 13 indicated positive pressures. These were initially assumed to be due to flow separation, but further analysis suggests this is unlikely. The difference in magnitude between sensors 12 and 13 therefore does not match that in Figure 8 (which used all relevant datapoints and is possibly a misrepresentation). This behaviour will be discussed further in section 3.2.3.

No CFD images for comparison of pressure distribution are available for this case.

2.2 Total pressures aft of brow wing

Some sensitivities were noted in CFD where the inboard vortex created by the brow wing combined with wishbone, clevis, and rocker “wake” and ultimately collapsed and shifted outboard, spreading loss across the top of the floor. The purpose of this rake was to locate the inboard vortex and determine if these predicted sensitivities carry over to reality under any conditions. Sensors 5-19 were attached (excluding 17 due to it not working, which along with sensor 3 is a permanent and previously known) from inboard-to-outboard, top-to-bottom, excluding the two off-grid probes and the bottom-outboard probe due to not having enough sensors. The rake is shown in Figure A3.

2.2.1 Straight and braking 18m/s

Method: as per 2.1.1 and 2.1.2, with rake installed in brow wing position.

Straight airflow results are shown in Figure 13. The orange cell indicates a predicted local velocity vector exceeding the maximum incident angle of the Kiel probes (30o), however for these tests this probe was not connected. Errors in magnitude should not be considered in isolation, both due to the rake being mounted 5mm higher during testing than in the original CFD setup, and because vortex size and position is more influential to the floor’s pressure-side performance than strength, to an extent.

A comparison of pressure distributions can be made from the super positioning of the measured data and predicted pressure field in Figure 14 (using the as-tested rake mounting position).

The experimental results do indicate the presence of a vortex; however, it is significantly further outboard, and larger and/or weaker than predicted. The smaller outboard vortex is not picked up at all, and the central area of higher total pressure is much weaker. These results more closely resemble the predicted field for the outboard floor during cornering, where the aforementioned sensitivity was noted in CFD (Figure 18). Mean and peak yaw angles were both non-zero during this test, however magnitudes were low (+0.5 degree mean, +4 degrees maximum). A possible explanation is the sensitivity observed in CFD is reflected in reality, and to a more extreme extent such that it occurs under even straight driving conditions. The discrepancy could be as simple as poor CFD accuracy, but could also be explained by the manufactured floor (particularly the brow wing) differing slightly to the CFD model, and/or by the EV25 chassis shape differing from EV24. Reduced front toe-out was used and would result in reduced in-wash over the EV24 CFD setup, but this is not expected to be significant enough in isolation to cause this.

Braking returned almost identical results (Figure 15), providing confidence that the pressures measured in both cases are an accurate reflection of typical behaviour.

2.2.2 Median corner radius

Method: as per 2.1.3, with rake installed in brow wing position.

No accurate pressure magnitudes from CFD are currently available for these cases, but the experimental data above is included for completeness. A comparison of outboard distribution is shown in Figure 18.

In this case, the measured data again suggests a larger area of low total pressure than was predicted, encompassing the entire rake area.

2.3 Flow-vis paint

Method: flow vis paint applied to brow wing suction side, floor pressure-side aft of brow wing, skirt lifted-inlet pressure and suction sides, skirt rear wall pressure-side, flap suction-side (see Figure A4). 5 fast and smooth laps of the track, ignoring cones.

This test was very experimental as it was the team’s first time using flow-vis paint. Uneven ground meant the scales used to measure the mixture ratio returned unusable values, so the ratio of kerosine, diesel, and paint pigment was approximated visually. The resulting mixture was powder-heavy which caused some issues with pooling/buildup, disrupting local airflow and thus the paint did not propagate.

The largest issues encountered were the surface finish of the floor, and the low average vehicle velocities. Porous or otherwise non-smooth surfaces quickly absorbed the paint and prevented propagation downstream. Low vehicle velocity meant the influence of gravity and lateral acceleration were significant compared to airflow effects (Figure 19).

All issues can be rectified for future testing. The largely unusable results mean there is a desire for further flow-vis testing. This is outlined in section 4.

Usable results were achieved on the brow wing, and forwards sections of the floor and lifted skirt inlet (Figures 20-21).

The purpose of the flow-vis in this area was to identify any unexpected flow separation on the brow wing, floor surface, or lifted skirt inlet, and any significant inboard or outboard shift of the brow wing vortex locations compared to CFD predictions.

There is no current surface shear imagery from CFD for direct comparison, but alterative comparisons are possible. Figure 22 shows the static pressure on the brow wing in CFD in the straight case (left) and the outboard median corner case (right).

The inboard-most edge (right side) is slightly obstructed by bodywork in both Figure 22 subfigures, but the left subfigure tends to suggest a relatively symmetric distribution, while the right subfigure (where the inboard vortex combined with wake from upstream suspension components and “blew up” while migrating outboard) is clearly asymmetric.

The flow vis shows highly asymmetric results (Figure 20), with long sparse streaks on the outboard side indicating high-speed flow as occurs in a strong vortex. The inboard edge, while showing some “S” patterns suggesting the presence of a vortex, is otherwise largely covered by denser and shorter streaks, indicative of low-speed or varying flow.

The outboard migration of this feature indicates a significant span-wise pressure gradient (such as that seen in the corner case in Figure 22). Such a gradient is caused by the absence of a strong vortex on the inboard side, which is further evidence of upstream wake causing vortex blowup or even failure to form. Attempts to replicate and subsequently resolve this issue in CFD should be made. 

Flow separation on the brow wing is minimal, occurring forward of the trailing edge only in a short and narrow central region which is not unexpected.

The flow vis on the floor surface and the lifted skirt inlet section shows attached flow, as predicted, which is important for the floors’ pressure-side downforce. 

3. Typical conditions analysis

3.1 Yaw angle (yaw probe)

Method: record yaw angles during autocross-style laps.

The purpose of recording yaw values was to generate a distribution of expected incident airflow angles during autocross-style driving.

Prior to testing, the yaw probe had only been calibrated to +/- 7.5° due to use of a temporary mount design in the wind tunnel that could not withstand the forces created by higher yaw angles.

The yaw distributions (Figures 23-25) are split into three categories based on instantaneous corner radius (from the vehicle’s IMU). Positive yaw values indicate local airflow moving from the centre of the corner outward towards the car, and negative values indicate the opposite (typical of slip/oversteer if not considering wind). Note that in the rest of the report, +/- yaw value refer to specific left/right directions.

Consistent wind including gusts up to 8m/s were present during testing. These instances were not filtered out as mild wind speeds are expected during typical test sessions and at the competition. B.O.M. data shows winds up to 10m/s are a regular occurrence in Adelaide during December (location of FSAE competition in 2026 onwards).

It is clear that for corner radii less than 9m (Figure 23), typical yaw angles are near or beyond the current 7.5° yaw probe limit. This is expected from a purely geometric aspect, with minimum yaw expected to be 9° for a 9m radius. The majority of yaw instances in the range shown in Figure 23 are therefore assumed to have been affected by wind.

In medium-speed corners (Figure 24), typical yaw angles were biased towards positive, but there were sufficient slip, oversteer, and/or wind instances to produce regular low and negative yaw instances. Values as low as -7.5° and below are expected to be dominated by wind.

In high-speed corners (Figure 25), the distribution is flatter and left-shifted compared to the medium speed corners. Despite the higher vehicle velocities and larger corner radii, it is still apparent that yaw exceeded both positive and negative bounds as a result of oversteer and/or wind.

Since even mild wind speed can easily match or exceed the vehicle velocity, the strong influence on yaw even in high-speed corners is not unexpected. The primary lesson is that optimising designs for a specific yaw angle is a waste of time; low sensitivity should be the main goal.

With the yaw probe angle bounds being a limiting factor for all three distributions, there is a pressing need to gather more yaw probe data and to extend the yaw probe calibration range. Based on an 8m radius (minimum radius before aerodynamic downforce is considered negligible, even with a “full aero” package) plus a wind buffer, calibration up to 20° is desirable if the probe is mounted at the front axle. By mounting the yaw probe as far rearwards on the nose as possible, the desired calibration range can be reduced to 17°.

While the limited yaw range of the probe meant distribution data was inconclusive, as a first test of the probe’s capability the results were excellent. Both the yaw and accompanying pitot-static probe trace data are shown for a two-lap run (Figure 26), plotted against lateral acceleration and wheel speed respectively. After applying 5-point moving average smoothing, the yaw data appears both sufficiently sensitive and unbiased (where wind appears negligible).

The effect of wind on the yaw readings is clear, particularly during the second lap. A period of minimal wind influence can be observed from 19s to 35s, during which the vehicle is going through a “slalom” section followed by a tight right and left corner. The same section is traversed on lap 2 from 57s to 73s.

3.2 Floor performance sensitivity

3.2.1 Suction-side total downforce

Method: record static pressured during autocross-style laps.

References to “local downforce” in this section are referring to the sum of all static tap pressures converted to pressure coefficients using freestream total pressure from vehicle airspeed measurements.

By examining local downforce, a metric for the performance sensitivity of the floor can be obtained through the standard deviation of the distribution during autocross-style driving. This will allow direct numerical comparison of sensitivity to future design iterations of aerodynamic floors.

Only instances with an airspeed greater than 11m/s (40kph) were included. Mounting incompatibilities between EV24 and EV25 meant the floor sat in an elevated position (approximately 10mm) until there was sufficient downforce to place tension on the mount cables, which occurred at approximately 11m/s (slower than all steady-state tests took place at). To allow comparison of future results, this velocity cutoff should be used as standard.

Standard deviation was found to be 1.85 for a mean of 13.72. The distribution is shown in Figure 27 with a two-SD window shown.

Two methods were used to assess conditional sensitivities:

• Plot of local downforce against yaw and corner direction (Figure 28)

• Onboard video recording with local downforce, local aero balance (see section 3.2.2), yaw angle, and instantaneous pressure distribution overlays

When yaw is positive (airflow from the right side), local downforce is typically greater during right turns, when the static taps are on the outboard side. This is expected due to the lower ground clearance. When yaw is from the left, there is no clear difference in downforce regardless of turn direction. Overall local downforce is typically lower during negative-yaw instances which is also expected due to a reduced “sealing” effect of the floor tunnel and possible increased ingestion of front wheel wake. Importantly, there are no significant low-magnitude outliers. Overall, this plot suggests predictable behaviour from the floor, but the range/standard deviation can not be commented on without data from alternate designs for comparison.

The onboard video with overlays confirms that local downforce is primarily influenced by high negative yaw magnitudes and vehicle roll, with instances such as combined corner/braking or oversteer showing no unexpected sensitivities.

3.2.2 Suction-side front/rear pressure balance

Method: record static pressured during autocross-style laps.

References to “local aero balance” in this section are referring to a zero-moment forward/aft location calculated using the sum of moments of all static tap data.

Under the same conditions as the data from 3.2.1, a distribution for local aero balance was produced (Figure 29). Mean is 48.6% with a standard deviation of 1.2%.

Local aero balance showed no clear sensitivity to yaw angle or vehicle roll, with the only significant correlation being vehicle pitch from longitudinal acceleration (Figure 30). The correlation is expected, but as with local downforce the range/standard deviation requires context from alternate designs to assess the sensitivity.

The outlying data between 52-54% occurred during a moment of strong negative yaw during a left turn, where local downforce was near the minimum recorded value (see Figure 28).

3.2.3 Diffuser and flap flow attachment

Method: install and record wool tufts on the right-side diffuser and flap during autocross-style laps.

Tufts were manually analysed from playback of the onboard video, considering only instances where vehicle velocity was sufficient to prevent gravity from dominating tuft behaviour. Four key results were identified:

• Consistent flow attachment was observed throughout the video on the flap suction-side. This is a good result and suggested that despite the unexpected low-energy region near/behind the brow wing, there is enough energy making it to and through the slot-gap to keep flow attached to the flap under all conditions. It should be noted, however, that compared to the EV24 CFD setup there is a significantly larger gap between the flap and the bodywork, allowing a vortex to form and potentially aid flow attachment, although this effect would diminish towards the flap outboard edge. 

• Consistent flow attachment was observed in the outboard portion of the main diffuser tunnel, as predicted. 

• Flow separation in the smaller outboard diffuser tunnel occurs during high roll scenarios where ground clearance is at a minimum (Figure 31). The most significant cases occur in the long sweeping left corner, which is both banked and concave. This allows atypical lateral acceleration and suspension compression and is not representative of any corner at the competition track. Some minor flow separation under high roll was predicted in CFD with minimal effect on downforce, so experimental reflection of this is a good result.

• Flow separation on the inboard portion of the main tunnel occurs during high negative yaw scenarios (airflow from the left side), shown in Figure 32. This is seen to occur regardless of vehicle roll. While the separation appears significant, it did not reduce upstream pressure magnitudes, with no relevant correlation between yaw angle and the closest upstream static tap data. No such separation was predicted by CFD, however there are possible explanations for the disparity.

None of the original CFD cases had significant yaw at the mid-car location, so the separation may be accurately predicted if such a scenario were set up. Another EV24/EV25 compatibility issue was the front suspension cover, which did not seal perfectly to the bodywork of EV25, potentially causing increased flow disturbance that was pushed into the right-side floor when yaw was negative (airflow from the left).

Recreating the separation in CFD should be attempted to help identify the cause of the separation and address it accordingly.  

Finally, a note on the positive-pressure spikes mentioned during median corner static-pressure tests in section 2.1.4. Initially thought to be instances of flow separation and//or vortex blow-up, these spikes did not occur at any point during the autocross-style driving, despite going through some of the same sections of track. There was also no clear yaw influence. Further, the downstream static pressure showed no change during these spikes, so it seems most likely that the pneumatic tube for the offending tap was either loose or kinked. The tubes were re-routed before the autocross-style driving to clear space around the steering wheel, during which such an issue may have been resolved without anyone noticing.

4. Conclusions and further work

Pressure correlation:

• Static pressure magnitudes and distribution showed generally good correlation to CFD during the straight airflow case, with errors reasonably explainable by geometric differences between the simulate EV24 and tested EV25. The braking case showed lower correlation but is also reasonably explainable with the additional pitch and ground clearance differences to CFD. Corner cases returned no unexpected results but are unable to be compared to previous CFD data. 

• Brow wing rake data combined with flow-vis analysis strongly suggests that an upstream disturbance is combining with or preventing formation of the inboard brow wing vortex, causing a large region of loss extending downstream. This was predicted in CFD for the outboard floor during cornering but was found to occur under all experimental scenarios and to a greater extent. The cause needs to be investigated to determine if it is EV24-EV25 chassis differences, floor manufacture differing from CAD design, or the presence of additional upstream disturbances such as the brake and wheel speed sensor lines that were not considered in CFD. 

Yaw probe:

• The yaw probe calibration range needs extending to ~17°, and further autocross-style testing run to produce expected yaw angle distributions.

Sensitivities:

• Metrics for local downforce and aero balance of the floor suction-side gave unsurprising results, but additional data for alternative (future) designs is needed to assess overall sensitivity. 

• Local aero balance is insensitive to yaw and roll.

• No unexpected conditional sensitivities were noted based on onboard footage with downforce, balance, yaw, and pressure distribution overlays during autocross-style driving. Strong airflow from the outboard edge of a floor towards the car centreline (i.e. airflow from left-to-right for the left floor and vice-versa) caused a moderate reduction in downforce on the suction-side of that floor, but this is not unexpected.

• No flow separation occurred on the flap suction-side under any conditions. 

• Flow separation in the small outboard diffuser tunnel occurs under high roll (when the affected floor had reduced ground clearance). This was predicted in CFD. 

• Flow separation on the inboard portion of the main diffuser tunnel occurs with airflow from the car centreline towards the outboard edge of the floor (i.e. airflow from left-to-right for the right floor and vice-versa). This was not predicted by CFD. It is expected that the reason for not observing the separation in CFD is one or both of not simulating high-yaw at the mid-car area in CFD, and a poor-fitting suspension cover producing upstream disturbances. Additional experimental testing with denser tufts (and a more suitable, thinner tape) may help diagnosis. Flow vis paint is not expected to be helpful as the separation is highly condition (yaw) sensitive.

CFD:

• A new CFD setup in ANSYS needs to be established for EV25, and validation cases set up to match the conditions of the experimental data (ride height, pitch, roll, steering angle, yaw). The results from testing should then be used to validate the new setup, and subsequently the new setup should be used to identify and resolve the issues identified (point 2 under pressure correlation, point 6 under sensitivities)

• Having observed the performance sensitivities are more closely tied to yaw than to corner radius or vehicle roll, it is recommended to remove the skidpan CFD design case and keep only the median corner case. The higher velocity of the median corner case means reduced uncertainty in experimental pressure readings, thus more effective validation. The skidpan case should be replaced with a high-yaw case. The ideal yaw angle and whether to use a straight or median corner scenario will depend on improved yaw distributions from additional yaw probe testing. 

5. Appended figures