IR thermography & NN models for damaged component thickness detection.

To achieve rapid detection of damage thickness in metal components using infrared thermography, a combination of heat transfer theory and image theory was employed. This involved theoretical analysis, finite element numerical simulation, a BP neural network prediction model, and infrared thermography experiments. Infrared thermal wave experiments were conducted under different heating temperatures. By analyzing the obtained temperature data, the response characteristics of surface temperature distribution to component thickness were investigated. The COMSOL numerical simulation software was used to simulate the surface temperature of the metal components. The bevel-cut metal components were heated to 80 °C, 105 °C, and 130 °C, and the fitted experimental temperature data were analyzed in conjunction with the simulated temperature data of the bevel-cut metal components. It was found that the fitted experimental temperature rise curve aligned with the simulated temperature rise curve trend. A comparative analysis of the simulation results and experimental values showed that the simulated temperature rise curve was basically consistent with the fitted experimental temperature curve, validating the feasibility of using numerical simulation as a substitute for experiments. The numerical simulation data were divided into a training set and a prediction set in an 8:2 ratio. Through training with the BP neural network, the predicted data were found to be basically consistent with the experimental data, verifying the feasibility of using the BP neural network for rapid detection of damage thickness in metal components. This laid the foundation for the subsequent promotion and application of BP neural network technology for rapid detection of damage thickness in metal components. This study holds significant importance for the application of neural network-based rapid detection technology for metal component thickness in the engineering field.