Long Short-Term Memory-Based Cutting Depth Monitoring System for End Milling Operation

Abstract

The technologies related to manufacturing processes monitoring, optimization, and control are becoming prevalent to achieve autonomous operations in Smart Manufacturing. The present work establishes an edge-level system based on the long short-term memory (LSTM) model for monitoring significant variations of cutting depths during end milling of near-net-shaped components. The proposed system consists of a trained LSTM model that decodes force data to identify cutting depths and an edge-level interface for displaying abnormal changes to the operator. The LSTM model development requires considerable labeled data consisting of cutting force sequences and corresponding depth classes generated using machining experiments. The present work proposes to develop the LSTM model using synthetic datasets generated using the mechanistic force model to minimize experimental efforts. The optimum configuration was derived by investigating the effect of network parameters and adaptive learning methods. The performance of an optimal network was substantiated by conducting tests using previously unseen synthetic datasets derived from the mechanistic model. The optimal network architecture was integrated with a dynamometer and an edge-level system to capture end milling force data and display cutting depth information. A set of end milling experiments are carried over a range of parameters to examine the efficacy of the proposed approach in estimating cutting depth deviations. It has been demonstrated that the approach can be effectively used as an edge-level system to capture significant cutting depth variations during the end milling and alert machine operators.