Analysis and Synthesis of Frequency-Diverse Ultrasonic Flaw Detection System using Order Statistics and Neural Network Processors

Ultrasonic imaging has been an essential tool for nondestructive evaluation of materials and flaw detection. However, flaw detection in the presence of microstructure scattering noise is a challenging problem. This chapter presents frequency-diverse ultrasonic detection algorithms which are essential to decorrelate the microstructure scattering noise and to enhance the visibility of echoes associated with defects in materials. In particular, the performance of ranked order statistics processors (such as minimum, median and maximum detectors) is examined using both theory and ultrasonic experimental measurements. Furthermore, this chapter gives emphasis to the concept of split-spectrum processing combined with neural networks as post processors to achieve improved flaw detection. Neural networks, because of trainability, offer an exceptionally robust performance and are capable of outperforming conventional detection techniques such as minimum, median, average, geometric mean, and polarity threshold detectors. An FPGA-based case study is presented for demonstrating the real-time operation of the ultrasonic flaw detection algorithms. Architecture details and implementation results with various Hardware/Software partitioning schemes are discussed.+