Electrified vehicles present a whole new set of design challenges, and autonomous driving piles on an entire new level of complexity. For example, different levels of autonomous driving affect the power consumption, thermal management, and heat loads in an electrified vehicle and therefore the component sizing. A high-fidelity computational model can help automobile design engineers and manufacturers understand the complex interactions of the different vehicle systems to quantify the impact of upcoming developments in the field.
|Simulation tools can help to determine the extra load when an EV carries autonimous tools. (Image source: Siemens)rGa|
This article describes an approach that uses co-simulation between two system-design software tools (one at a vehicle level and another at the thermal system level). With a simulation model that is correlated with real test data, this method quantifies the potential change in heat generated from the battery, motors, and inverters for electric vehicles that result from additional load from the self-driving computer. We illustrate the effect using VW eGolf from PowertrainLive as the base reference vehicle.
There were three steps in quantifying the effect. Firstly, we quantified the baseline power consumption and head load for the battery and motors for a reference drive cycle. We picked a city drive cycle first as that is likely to be the immediate application for for autonomous vehicles in the form of robotaxis. Secondly, we added electric power consumptions and the driving behavior change of the autonomous vehicle using simulation. Lastly, we calculated the effect of the autonomous driver across many real-world drive cycles with different characteristics.
To model all of this information, we applied computational analysis using two tools: the PowertrainLive high-fidelity, vehicle-simulation model from CSEG and the 1D computational fluid dynamics (CFD) software Simcenter Flomaster from Siemens to analyze internal flow and model the thermal transients in the system. The vehicle model cumulated fuel economy, battery range, and the thermal and performance behavior of various components. The CFD simulation software allowed us to model the cooling system and to understand the effects on airflow.
We modeled the battery using an RC representation to speed up the calculations but still capture the transient operation of the battery with acceptable accuracy. The voltage and the resistances and capacitances are a function of state of charge of the battery and the temperature of the battery. We developed the battery performance characteristics for the complete operating range, from fully charged to fully depleted, and ambient temperatures from -7 to 45 °C. This enables an accurate representation of battery performance under all five EPA drive cycles.
Figure 1 shows some of the battery performance maps. The open circuit voltage is a function of battery temperature and state of charge. The voltage drops slightly with battery and state of charge. (Image source: Mentor)
|Figure 1: Battery performance maps.|
Figure 2 shows how this battery model predicted performance compared to the test data in the vehicle model. The correlation below shows the current demand for the USO6 drive cycle, an aggressive and highly transient drive cycle used by the EPA. The vehicle model was able to predict the peaks and transience of the current well enough to capture the sometimes subtle effects of driver behavior change.
|Figure 2: Overall vehicle model correlation with measurement data. (Image source: Mentor)|
Once an acceptable level of accuracy was achieved, we looked at the influence of autonomous driving features on the vehicle, namely the increased power consumption, increased weight, and less aggressive driving profile as a result of increased awareness of the surrounding.
The power consumption of autonomous vehicles (AV) varies significantly depending on the type of sensors used and the maturity of the technology. It can vary anywhere between an estimated 500 W for a Tesla to a 2.5 kW for an experimental autonomous vehicle with LIDAR and computers in the trunk. In our study, we assumed a power of 900 W for the sensors and computer based on interviews with few of AV vehicle manufacturers.
We assumed an increased weight of about 50 kg for the computer and the sensors. This is something that is widely reported as the added weight for the autonomous sensors and computers.
Finally, we looked at the effects of the driving profile change. The autonomous driving profile was slightly less aggressive and was represented with a smoother driving profile and with limited acceleration limits, which are usually in place for safety. The driver model is illustrated in Figure 3. The updated driving profile was calculated by the driver model with the changes in aggressiveness factor and limits on acceleration. The model used the drive cycle, the road conditions and the grade as the input to the autonomous driving controller to determine the accelerator pedal position and the new driving profile.
|Figure 3: Vehicle architecture of EV from PowertrainLive. (Image source: Mentor)|
With the computational models ready, we reviewed the base vehicle energy consumption reduction. The first drive cycle we considered was a city drive cycle, a typical city taxi ride (shown in figure 3b as orange line). The base energy consumption for the vehicle in this scenario is 0.47 kW/hr for an 8-minute and 2.2-mile drive. This rate of power consumption and driving profile gave us a baseline battery range of 99 miles. With the added power consumption of the AV package of 900 W, and a weight penalty of 50 kg, the battery range dropped from 99 to 62 miles. The smoother driving from the autonomous controller reduced the spikes in power consumption and thereby increased the battery range back up to 81 miles.
On thermal management, we observed some interesting trends for the battery and motor heat loads. Both peak and cycle average heat loads for the battery were down nearly 50%. Figure 4 illustrates this in the transient profile plots. Why such a drastic reduction in battery heat load? The battery heat is caused by resistance inside the battery, which is a second-order function of the current. As the current demand drops, the battery heat load drops by the square of the drop in the current.
While the reduction in peak heat loads was expected because of limited accelerations, we were surprised to see the drop in cycle average battery heat load even with an added electrical load on the battery from the computer and sensors. This was a result of the reduced current spikes from constant charging and discharging. The cycle aberage heat load also dropped by nearly 50%. This reduction in heat load reduces chiller size for the battery thermal management system and further affects the compressor power.
Figure 4: Difference on heat loads between the autonomous driver (red line) and human driver (green line). (Image source: Mentor)
Was this effect just for that drive cycle that we picked, or was it actually prevalent across multiple drive cycles, and, did it represent general driving? CSEG collected numerous real-world drive cycles in Michigan, and we assessed the impact on AV technology on a few of these drive cycles with different accelerations, average speeds and driving distances. The trends remained similar, with the quantity of drop varying depending on the drive cycle. Figure 5 shows the numbers from PowertrainLive calculations for a city and a highway drive cycle.
|Figure 5: Results from the PowertrainLive calculations. (Image source: Mentor)|
From this study, we were able to analyze and compare the effects on driving range and thermal requirements of autonomous driving on electric cars. We found that the battery range dropped significantly in city driving conditions because of the power consumption of the computer. While the overall power consumption from the battery was higher, the peak and cycle average heat loads from the battery were lower due to reduced aggressiveness in the driving and the spikes in current draw.
You can conduct similar experiments on your EV designs using these same tools. The vehicle-simulation model PowertrainLive and the correlated vehicle models can be accessed through the Internet browser and subscription. The vehicle model can be integrated with your current design tools such as FloMaster. You can also take Flomaster for a free test drive on the cloud.
For further details and data from the study, please email the author at firstname.lastname@example.org
Sudhi Uppuluri is the technical director at CSEG, the maker of PowertrainLive. He has 20 years of experience in the automotive powertrain simulation. He is also a lecturer at University of Wisconsin Madison on Vehicle Level Modeling and has various technical publications on related subjects in SAE and AIAA journals.
Doug Kolak is a business development manager for the Mechanical Analysis Division at Mentor, A Siemens Business. He joined the Flomaster team in 2007 as a CFD engineer, during which time he has worked heavily with top companies in the aerospace, automotive, and process sectors worldwide to understand emerging challenges and develop software tools to better meet those requirements.