| Abstract Scope |
Artificial Intelligence and Machine Learning (AI/ML) provide new tools for welding process
development, quality assurance, and quality control. It is critical to remember that these tools are only as good as the data used to train them. Succes relies on (1) Carefully consideration of the data which should be collected (2) Development of datasets which span production relevant conditions (3) Storage of data so that it can be readily used (4) Continuous collection of data for continuous improvement of ML/AI. These elements of success will be demonstrated through real world welding use cases. First, the importance of sensor selection will be illustrated for real-time ML-based quality control of ultrasonic battery tab welding. Second, methods of using ML for efficient, production-spanning dataset development will be demonstrated for automated MIG welding. Third, methods used in the additive manufacturing community to develop datasets which follow the findable, accessible, interoperable, and reusable (FAIR) data practices will be described. Finally, the use of industrial internet of things to automatically collect data from the point of generation will be described. Throughout, the difference between “Big Data” and “The Right Data” will be emphasized. |