High sensitivity detection and classification of defects in austenitic weldments using cluster analysis and pattern recognition

Abstract

Problems encountered in the ultrasonic non-destructive evaluation of austenitic stainless steel weldments (greater than 12 mm thick) are highlighted in this paper. Ultrasonic studies are presented on 14 mm and 16.5 mm thick austenitic stainless steel weldments. These thicknesses were chosen, to initially apply and standardise the developed techniques for defect detection and characterisation, and to subsequently extend these methods for larger weld thicknesses. Two novel cluster analysis methods and a pattern recognition method, that are developed for the efficient detection of defects such as line defects, porosities etc. in these weldments by analysing ultrasonic signals, are presented. Limitations of conventional ultrasonic testing for detection of defects less than 5% of the thickness of the weld and the development and application of advanced signal analysis methodologies to overcome these difficulties are described and the results are compared for both natural defects and simulated (artificial) defects. Results show that even a 1% of thickness notch can be detected, which is indicative of the useful capability and sensitivity of these techniques for an early detection of defects in austenitic stainless steel weldments, particularly in critical applications. An integrated Expert System model, which can be implemented in a personal computer, along with the developed cluster and pattern analysis procedures, has been suggested to tackle the complex problem of detection and characterisation of defects in austenitic stainless steel welds.