Ultrasonic sensor signals and self organized mapping with nearest neighbors for the microstructural characterization of thermally-aged Inconel 625 alloy

Non-destructive ultrasonic testing is a widely used technique in industry to evaluate the microstructure in a metal alloy. In this work, we present a novel approach to investigate microstructural deterioration in three different hybrid welds by thermal aging treatments at two different temperatures for four different time periods. Transverse and longitudinal ultrasonic waves with transducers with frequencies of 4 and 5 MHz were used to determine the ultrasonic attenuation and velocity for each sample. Backscatter and echo signals were used to automatically recognize the microstructural changes in welded samples submitted to aging. Furthermore, we assessed the efficiency of the unsupervised Self-Organizing Map algorithm together with the k-Nearest Neighbor classifier (SOM-kNN) to characterize the morphology of the precipitates. Three values of k for the classifier kNN and 18 different values of neurons for the SOM network, totalizing 54 combinations, were used to evaluate the microstructural deterioration. The results of the experiments showed that the best topology at a temperature of 650 °C obtained 100% accuracy, 5% data usage, 6.2E-5s test time, and 1.4E+3s training time. At 950 °C, the best accuracy was 98.3%, with 38.1% data usage, 3.6E-5s test time, and 2.3E+1s training time. The proposed approach reduced the computational cost by sample selection while maintaining high accuracy.