Polyethylene (PE) pipes are widely used in gas distribution. Their joints are prone to various flaws and are the most problematic part of the pipeline, so the infrastructure industry requires an effective inspection technique. Butt-fusion (BF) is the most common method of joining PE pipes. In this research, we investigated the applicability of machine learning (ML) to automate the ultrasonic inspection of PE pipe BF joints. Flawless and defective joints were fabricated. A-scan signals were collected from each group of samples using a customized chord transducer, with the aim of developing and assessing the viability of ML approaches to the problem of joint classification. We compared several ML approaches to the problem and found that convolutional neural networks were most performant, classifying signals with an F1 score of 0.874 in a four-class problem (identifying defect presence and type) and of 0.912 in binary classification (defect presence/absence only). Our results show that an ultrasonic chord-type transducer approach can effectively resolve flawless samples versus those with coarse contaminants or cold fusions and that an ML approach can be used to effectively assess these ultrasonic signals. Our findings can be used to develop a portable, efficient, user-friendly, and inexpensive device for in-field joint inspections.