Electrification System Modeling with Machine/Deep Learning for Virtual Drive Quality Prediction

A virtual 'model' is generally a mathematical surrogate of a physical system and when well correlated, serves as a basis for understanding the physical system in part or in entirety. Drive Quality (DQ) defines a driver's 'experience' of a blend of controlled responses to an applied input. The 'experience' encompasses physical, biological and bio- chemical perception of vehicular motion by the human body. In the automotive domain, many physical modeling tools are used to model the sub-components and its integration at the system level. Physical Modeling requires high domain expertise and is not only time consuming but is also very 'compute-resource' intensive. In the path to achieving 'vDQP (Virtual Drive Quality Prediction)' goal, one of the requirements is to establish 'well-correlated' virtual environments of high fidelity with respect to standard test maneuvers. This helps in advancing many developmental activities from a Analysis, Controls and Calibration standpoint. Recently, machine/deep learning have proven to be very effective in pattern recognition, classification tasks and human-level control to model highly nonlinear real world systems. This paper investigates the effectiveness of machine/deep learning with various algorithms for the modeling of an electric vehicle system, integration with virtual embedded controllers, drive quality analysis and correlation to vehicle data, thereby enabling scope for 'front- loading' iterative activity in the developmental cycle and enable laptop-based fine tuning of software calibrations.