Surrogate modeling of waveform response using singular value decomposition and Bayesian optimization

In the early stage of vehicle development, it is required to implement a target cascading study by solving inverse problems. However, simulation costs dynamics predict transient responses and frequency make difficult. The purpose this paper propose method construct surrogate model which can waveform solution Bayesian optimization using posterior distribution trained responses. Replacement expensive more economical enhance study. paper, we vectorized training data matrix from be evaluated CAE simulations based on Design Experiments. proposed method, supervised unsupervised learning are introduced. singular value decomposition used as feature extraction (Unsupervised learning) applied data. Obtained vectors modes represent Gaussian Process introduced regression (Supervised each weight obtained projection modes. response predicted superposition prediction values By Process, Expected Improvement function in minimize cost mean waveform. feasibility illustrated an application for suspension design problem impact harshness phenomenon.