Neural networks and genetic algorithms for the evaluation of coatings thicknesses in thermal barriers by infrared thermography data

摘要

In the context of using non-destructive thermal control methods for the coatings thicknesses evaluation in thermal barriers. We have treated the laser-pulsed thermography data with the neural networks to model the relationship between the thermal response and the coating thickness. The algorithms based on the error gradient computation are used during the learning step. Indeed, the initial weights of the network found and the number of data processed facilitated the convergence of these algorithms. In this work we presented a neural network training method using pre-processing of data by principal component analysis (PCA) to optimize the number of network inputs and the genetic algorithm for the optimum initial weights determination in the network training by the back propagation algorithm. The two algorithms recombination allowed the thicknesses evaluation with deviations less than 5%.