Optimizing System Design performance and its accuracy using Feature Importance

With continuous improvements in model accuracy and computational performance, system simulation can be seamless integrated into development tools and methodologies as a "virtual test bed" in synergy with experimental tests. However, virtual vehicle and powertrain thermal models still face significant obstacles in realizing their full benefits. One common challenge is modeling accuracy and its anomalies, particularly in models provided by internal automotive companies and external suppliers. The models developed for vehicle thermal management and its systems have been created using specific software, making it difficult to verify the accuracy of the models across different systems and their performance. Integrating different systems in a Co-simulation environment has affected the overall accuracy of the Vehicle Simulation Platform model. This situation significantly reduces the return on investment in model development. To address these issues, we propose deploying feature importance using machine learning techniques at the vehicle level simulation platform development. Feature importance techniques refer to scores used to determine the relative importance of each feature in generating the output of each system. The goal of this paper is to utilize feature importance to help identify key variables to focus on when calibrating and validating systems in relation to vehicle-level performance across different drive cycles, aiming to achieve good correlation with physical tests within ± 90% accuracy.