Grinding wheel specification cybernetic recommendation with multi-task multi-imbalanced learning in smart manufacturing system

Abstract

Over the years, the grinding wheels industry has played a crucial role in mechanical engineering. Grinding wheel specification is composed of various factors such as abrasive type, wheel grade, bond type, and more, which are derived from customer workpiece operation information. Traditionally, the determination process of grinding wheel specifications is complicated. Experienced engineers need to check out the limitation and consider the interaction of each factor. When the new specification order comes, engineers have to spend a lot of time on historical data search and discussion. Recently, the application of artificial intelligence attracts more and more attention in the manufacturing industry. Through collecting historical data, the model has the ability to extract the domain knowledge, predict the specification that does not exist, and reduce the decision time. However, when training the model, this grinding wheel specification faced multiple imbalanced targets and multi-task issues. In the past, research efforts primarily focused on studying single-task learning (STL) algorithms. Seldom of them has a systematic way to discuss the model determination of multiple targets and correlation influence, especially on the tabular-type dataset. In this study, we proposed a multiple resampling chain multi-task learning framework to address the issue of data science. To tackle the multi-task imbalance resampling problem, we also proposed a resampling strategy called Multiple Resampling Chain. Moreover, we conducted a sensitivity analysis to examine the correlation effect on model improvement. Finally, we proposed a cyber-recommendation system that integrates the training model and interface to provide a more intuitive way of specification prediction. This system not only can help engineers to improve the time spent on making decisions, but also make the whole manufacturing process to be more intelligent.