Optimization of lock-in thermography applied for automatic identification of internal defects in 3D-printing polymer

This work presents a methodology for automatic detection of internal defects in parts manufactured from 3D printing with two of the most common materials for these purposes. Active Thermography has been used, specifically the Lock-in technique. To ensure reliable detection, the varying frequencies of the stimulation source have been compared to determine the optimal configuration. Results have been analysed to identify the parameters that most affect the identification of defects. Results show that the frequency and the type of material used are the most critical parameters that condition the detection though the influence of this latter was less clear, so it was necessary apply a novel analysis based on Machine Learning. Most effective algorithm was an ensemble model, which achieved an accuracy rate of 88.9%. A variation of Maximum Stable External Regions is presented to automatised detection.