Automated ultrasound-based flaw detection of medium-density polyethylene butt fused joints using chord transducers and implementation of artificial intelligence

Butt fusion pipe joints in a polyethylene gas and water distribution pipeline network requests high attention from the quality control inspection. As the joining process performed in-field, these joints are prone to various flaws. Thus, the infrastructure industry requires efficient, simple, inexpensive, and effective follow-up joint inspection technique. Ultrasound is proved method for NDE testing in many application areas; however, geometry of butt joint seam complicates use of standard pulse-echo technique. Advanced testing schemes such as tandem and chord reflection look more suitable. Present work focused on performance of chord transducers for detection flaws in PE pipes butt joints for the most common diameters 2", 4" and 6". Samples with artificially introduced flaws of various nature and sizes were fabricated and tested for statistical estimation of system performance. Large collection of obtained A-scans was utilized to be a base for development of deep learning algorithm of flaws detection and classification. The trained model was implemented on a prototype to automate the inspection procedure with the good performance with more than 90% accuracy in defect detection. (author)