Finite Element Supported Data Augmentation for a Deep Learning Driven Intelligent Failure Analysis System Based on Infrared Thermography

One of the biggest hurdles in developing an innovative, intelligent failure analysis system that caters to the needs of the manufacturing industry, is the lack of manufacturing data to train a deep learning driven system. The Finite Element Method (FEM) has the potential to bridge this data gap by developing simulations that encapsulate the underlying physical phenomena in the form of complex differential equations and solve them numerically. This work presents a detailed study comparing the results obtained for pulse thermography experiments and the FEM. Samples with artificially introduced defects in the form of flat bottom holes are fabricated and used for this study. The concept of defect signatures based on 2D axisymmetric FEM is introduced to significantly reduce the computation time for transient thermal simulations. The method for parameter estimation and boundary conditions for the FEM is explained. The evaluation of the temperature evolution of sound and defect regions, surface temperature distribution and thermal contrast shows, that the FEM results are an excellent estimate of the experimental setup and are a good fit for generating varied data for data augmentation to aid the development of a deep learning based image segmentation model.