Machine Learning-Based Model Bias Correction by Fusing CAE Data With Test Data for Vehicle Crashworthiness

Abstract Physics-based simulation and analysis have emerged as promising techniques for optimizing the number of physical prototypes for vehicle crashworthiness evaluation in frontal impact with rigid barriers. Nonetheless, one of the hurdles for vehicle crashworthiness virtual certification is the potential differences between the computer simulation predictions and physical test results. In this regard, this study aims at improving the prediction capability of the Computer-Aided Engineering (CAE) model for crashworthiness performance evaluation at speeds beyond those defined by current regulations and public domain testing protocols. One way of achieving this is by integrating data from a number of physical crash tests with the CAE data using machine learning models. A novel approach is proposed in the displacement domain (deceleration vs. displacement) to enable data fusion to help recover missing physics associated with the CAE model. A nonlinear spring-mass model is used in this study to simulate rigid-barrier vehicle frontal impact. The deceleration response is transformed from a function of time to a function of displacement, and a Gaussian process regression (GPR) model is applied to capture the model bias of the nonlinear spring constant under a dynamic analysis scheme. The training data for the GPR model are split into multiple clusters by a Gaussian mixture model to capture bias patterns under different speed regimes. After clustering a GPR model is trained for each group of data. The optimal GPR model, trained by a specific cluster exhibiting the highest probability of new data belonging to it, is utilized for prediction. This selected GPR model is integrated with the original CAE model to predict vehicle deceleration under a new crash speed during a vehicle deceleration dynamic analysis. The proposed approach is validated using physical vehicle crash tests and demonstrated improved accuracy of CAE model results and predictions.