Neural decision support system for ultrasound nondestructive evaluation embedded in a DSP

Abstract

Ultrasound non-destructive testing is widely applied to identify defects in structures and equipment as it combines simplicity, fast execution and high efficiency in the detection of flaws. However, in many practical evaluations the accuracy of the obtained results relies on the experience of the operator. Some automatic decision support systems have been proposed in the literature aiming at helping the operator in the decision making process. These systems usually comprise different signal processing steps in order to extract relevant information from the measured signals and perform automatic classification. For applications which require instantaneous feedback to the operator, a limitation is that usually the decision support system is designed to operate in personal computers with recorded experimental data. This paper presents an electronic signal processing system (based on a digital signal processor architecture) that was designed to be used in parallel with the ultrasonic evaluation equipment in order to provide instantaneous decision support information for the operator during the testing process. The proposed system comprises different modules such as analog to digital conversion, signal feature extraction (using the discrete Fourier, Wavelet and Cosine transforms), information compaction (through Principal Component Analysis) and automatic classification (using a Neural Network classifier). These routines are embedded in a digital signal processor. An LCD (Liquid Crystal Display) is used to indicate the system decision. The proposed system efficiency is evaluated in a practical industrial problem and the results point out that it is possible to provide valuable information to the operator during the evaluation procedure.