Real-time Bayesian model calibration method for C/SiC mechanical behavior considering model bias

The mechanical behavior of C/SiC composites has significant nonlinear and dispersion. To obtain the real mechanical behavior of C/SiC composites in actual structure, a real-time calibration method for mechanical behavior of C/SiC composites based on the dynamic Bayesian network (DBN) is proposed. The epistemic uncertainty of the mechanical behavior is introduced in the form of model bias. The method comprises offline and online phases. Firstly, the surrogate model of the mechanical behavior is built based on the proper orthogonal decomposition and the Gaussian process model offline. Then, based on the DBN, model parameters can be continuously calibrated by the sensor data online. The validity of the method is verified by the uniaxial tensile test of C/SiC composites. The results show that the proposed method can real-time accurately calibrate and predict the mechanical behavior of all the C/SiC samples and the introduction of the model bias is of great significance.