Lightweight Deep Learning Neural Networks for Analyzing Infrared Thermal Imaging Non-Destructive Inspection Results

Non-destructive testing is commonly employed for inspecting vessel quality, in particular infrared thermography testing. However, owing to the lack of clear criteria for interpreting the resulting images, inspections have been conducted subjectively based on the inspector's experience. Small vessels such as fishing and leisure boats are typically constructed using fiber-reinforced plastics, which feature thick hull materials mixed with fabric. In this study, specimens using both combined-fabric and single-fabric methods were prepared to detect defects using infrared thermography. To inspect the results objectively, defects were detected via a statistical-analysis method using a temperature-change-rate equation and deep learning, which is an image-classification technology. Results of applying statistical analysis and deep learning reveal differences in the defect-detection temperatures between the mixed-fabric.