Development of a predictive maintenance algorithm for welding guns splash

In parallel with the evolution of the technologies in every industrial field, also the maintenance has been involved in a rapid development and improvement of all that it concerns. Maintenance encloses various and different areas of industrial sectors, from the economy to the security, from the management to the optimization of the use of resources, and therefore over the years it has held a greater and greater importance until it has been established as one of the major strategies to focus on. The increase of reliability and accuracy of algorithms in the Artificial Intelligence field has led several companies to experiment the application of such algorithms in solving maintenance issues. The ISI-Welding company is a leading company in resistance spot welding. In order to improve the quality of their products and to offer better services to their customers, they had begun a research to implement an intelligent approach to the maintenance of spot resistance welding guns. This thesis work represents the attempt to predict and prevent the occurrence of the splash, a frequent disturbance defined as a spillage of melted metal from the designed welding spot. A strong basis of the mentioned topics has been gained through the study of the state-of-art of maintenance and welding guns, then a deep knowledge of the system has been acquired analysing the physical phenomena involved in a welding process. An identification of the model and an estimate of a physical one have been carried out according to the theoretical bases, with good results. Finally, Machine Learning algorithms have been implemented in order to classify the presence of the splash during the welding process and to predict it. In particular, Neural Net classification and Long Short-Term Memory algorithms have been used. Results are encouraging and prove the feasibility of this approach, a further collection of larger datasets and the measurement of new critical variables may lead to the development of a complete and exhaustive model of Predictive Maintenance.