[1C4] A comparison of two welding indication detection methods of carbon steel weld phased array ultrasound data
R Rhéaume and Y-É Cossette
Université du Québec à Trois-Rivières, Canada
Training an artificial intelligence (AI) model can be a complex task. Several factors can influence the quality of the training and thus the performance of the AI model. For example, the objectives of the training must be clear, a dataset of sufficient size must be available and the type of data in the training set must reflect the actual data that will be analysed later. Even if those factors are respected, special care must be taken when choosing the data to avoid unwanted bias. Developing an efficient welding indication detection method is therefore a difficult task that involves insights from experts, hours of labelling data and a great deal of training and fine-tuning to reach industry-level performance.
This paper presents a novel deep learning approach that detects welding indications in 3D phased array ultrasound (PAUT) scans and compares the detection results of two indication detection models. It demonstrates the difficulty of developing an industry-level deep learning model. The results demonstrate the effectiveness of the latest model developed. A discussion on the steps, the difficulties that have been overcome and the knowledge acquired during the development of both models is also included.
Keywords: deep learning, indication detection, characterisation, phased array ultrasound scans.
This paper presents a novel deep learning approach that detects welding indications in 3D phased array ultrasound (PAUT) scans and compares the detection results of two indication detection models. It demonstrates the difficulty of developing an industry-level deep learning model. The results demonstrate the effectiveness of the latest model developed. A discussion on the steps, the difficulties that have been overcome and the knowledge acquired during the development of both models is also included.
Keywords: deep learning, indication detection, characterisation, phased array ultrasound scans.
