PHASED ARRAY TESTING DATA GENERATION THROUGH GENERATIVE ADVERSARIAL NETWORKS AND ADAPTIVE DISCRIMINATOR AUGMENTATION

Non-destructive testing is used to detect faults and verify structuralproperties. Its main advantage is that it does not alter material properties. Currently,this testing and evaluation process is done manually. Using classification and featuredetection algorithms is a possibility to evaluate data obtained by sensors faster andmore efficiently. However, the lack of data volume to train and evaluate with safetymodels is one of the barriers to employing such methods. This work proposes usinggenerative adversarial neural networks to generate synthetic images of nondestructive tests obtained by ultrasonic sensors to improve the performance ofclassification models and assist the learning process.