AI, Machine Learning and Deep Learning in Nuclear Manufacturing for ITER components

Maria Ortiz De Zuniga - Fusion for Energy

Abstract:

Artificial Intelligence (AI) has been applied to many different fields, whereas in others it is still to be explored. In general, this is the case of the nuclear manufacturing and, specifically, the case related to welding success rate prediction and the analysis of outputs from the phased-array ultrasonic (PAUT) nondestructive testing (NDT).The aim of this work is the development and analysis of Al tools for welding success rate prediction and the posterior output processing of PAUT applied to welding defects detection in the ITER Vacuum Vessel manufacturing. Thanks to its complexity, the manufacturing of this large equipment-based on the French nuclear design and manufacturing code (RCC-MR)-has generated a large amount of data. Since the Vacuum Vessel is the first confinement barrier of the nuclear fusion installation, ensuring the quality of its welds is a serious challenge. Each of the five European sectors has approximately one kilometer of welding to be performed. Any defect in these welds results in a large disruption on a quality schedule and on a mechanical level, which has to be recovered, within feasibility limits. The first tool, based on the Random Forest Regression method, is able to predict the success rate of electron beam welding (EBW) and of Tungsten Inter Gas (TIG) welding, with up to 100% accuracy.

Once the weld is performed, the Vacuum Vessel double-wall nature results in un-inspectable welds during the last stages of the segment manufacturing on the full weld depth or from both sides through conventional non-destructive testing methods, such as radiographic examination as accepted by the RCC-MR; resulting in the need to qualify a more advanced ND technique, such as PAUT. PAUT data processing and interpretation has to be carried out by a human expert and requires up to one week per weld on average, due to the coarse grain material of austenitic stainless steel used in the Vacuum Vessel-316LN-IG-and the complexity of the qualified PAUT procedures. This process is long and costly, affecting performance and requiring a large number of resources noting that the cost of training alone to develop a suitably qualified NDE personnel who can do UT examination can be considerable. The second tool tackles the processing operation of the PAUT output. PAUT output can be generated in two formats: an A-scan and an S-scan (similar to a scan in the medical field). Due to the complexity of the task and large amount of data, Deep Learning was promptly identified as the correct Al subset to use for this development. On one hand, a convolutional neural network CNN) has been chosen to analyze data through its perceptron’s and process data as an image, similar to the processing of images performed in the human visual cortex. On the other hand, a long short-term memory (LSTM) model was selected to process the same data in a two-dimensional wave representation.. Due to the sensitive nature of the information provided and the large consequences of false positives or false negatives, a conservative approach was chosen for the final output. This double gate approach not only increases the accuracy of the result but also increases the human confidence in this tool. This development shows that Al is an appropriate tool to process PAUT data, allowing prompter data availability and giving an additional information set in order for projects to take informed decisions. The subjective interpretation and human error factors can be decreased through this automation, as is the large time required to process each PAUT