Low-order Physical Modeling for Vehicle Acceleration Behavior Using Tensor Based Blind Source Separation

This paper proposes low-order physical modeling method using operational output-only data without inputs. Vehicle characteristics and the controllers are designed to be satisfied with a number of requests and constraints such as driveability and ride comfort. Those need dynamic models which can represent the corresponding motional behaviors. However, the detail physical modeling for various vehicles and the simplification, or the system identification with input measurement is usually difficult or expensive in terms of labor. Therefore, in this paper, firstly, time-frequency data array of output accelerations of vehicle components is constructed and then decomposed by one of tensor decomposition methods, parallel factor (PARAFAC) decomposition, to separate stationary vibrations and transient vibrations like vibration modes. Secondly, using the result and the structure of PARAFAC, trade-off decision can be made between the model accuracy and the simplicity regarding number of motions considered in modeling. The effectiveness of this approach is demonstrated for driveability by using operational tip-in acceleration data.