Decision support system for ultrasound non-destructive evaluation based on extreme learning machines embedded in microcontrollers.
Non-destructive Evaluation, Time-of-Flight Diffraction, Extreme Learning Machine, Microcontroller
This work aims to establish a technological basis for an ultrasound non-destructive evaluation decision support system, that may be portable to a microcontrolled platform. Extreme learning machines, a fast training class of artificial neural network, are used as pattern classifier. The purpose is to obtain a dedicated system capable of recording the input data, training and operating the classifier system, and finally producing an integrity status indication for supporting the operator decision. A case study on integrity evaluation of weld beads in steel plates was considered. The basics of ultrasonic non-destructive testing and artificial neural networks are introduced. Further aspects are considered in order to accomplish the referred objective, such as: a comparison between a fast-training neural network and traditional ones; a method to reduce the feature set dimensionality and the proposed approach for system realization in microcontrollers. The results indicate that the system realizes the training step of a classifier in the microcontroller to carry out the classification of weld defects in similar accuracy performance as reported works in the literature.