Functional Gradient Plate Intrinsic Frequency Analysis Based on Fitting Deviation Optimization Physical Information Neural Networks

Aiming at the difficulty of solving the heat frequency control equation and the problem of boundary data points amplifying neural network bias, this paper proposes a fitting deviation optimization physical information neural network (FDO-PINN) for solving the intrinsic frequency of three-dimensional functional gradient plates in a thermal environment. An function is given to fit the neural network training data, then the training data is gradually sliced from the outside and its fitting error is calculated, and the data region with the smallest fitting error is found to be the physical solution region of the control equations. In this paper, an experimental simulation analysis of the comsol data is carried out, and the results show that the FDO-PINN is able to solve the problem of solving the intrinsic frequency in the thermal environment of the functional gradient plate, and the highest solving error is only 1.56% when the test temperature exceeds the training temperature by 100 degrees. The average error within the training temperature is only 0.62%.