An intelligent recommender system for tool selection in conventional machining

Abstract

ABSTRACT Cutting tool manufacturers face a tough challenge in developing custom solutions for specific customer requirements. Several trials are required, encompassing the selection of materials, tool configuration parameters, manufacturing of tools and testing under target conditions to arrive at an acceptable solution. In this work, the authors present a recommender system that utilizes a hierarchical deep learning-based machine learning model, handcrafted using domain knowledge, to predict top N tool configurations for a given target requirement with a probability score. The authors also discuss methods for data augmentation to deal with limited data as well as a probabilistic approach to predict the top N tool configurations from the trained models. The proposed system is applied to a case of centerless cylindrical grinding wheel selection problem. The outcomes indicate an overall accuracy of 92.4% for single best-fit specification, with 100% within the top five recommendations for past designs. Some of the alternatives proposed by the model are observed to be potentially superior to what was chosen earlier by experts. Without the selection hierarchy, a deep learning model achieved a single best-fit accuracy of 83.5% and the probabilistic model achieved a top-five recommendation accuracy of 89.8%, highlighting the merit of the hierarchical approach.