Intelligent Tool wear prognosis and diagnosis: A cloud analytics approach based on measurement of machining power data

Abstract

Industry 4.0 is transforming manufacturing through the digitalization and integration of advanced technologies such as artificial intelligence, the Internet of Things, and cloud computing. These advancements create interconnected, intelligent systems with significant potential to enhance prognostics and health management of machine tool subsystems in smart manufacturing. An intelligent framework for tool wear prognosis and diagnosis during machining based on machining power signal as a single input feed to a suite of machine learning and deep learning models is presented in this paper. Six predictive models were trained and validated in the cloud using data from 2474 milling experiments and subsequently tested with data from 205 unseen experiments. The framework for evaluating tool wear prognosis, implemented with Microsoft Azure ML Studio^®, made use of machining power signals during the machining of Inconel 617, a nickel-based superalloy used in aerospace and nuclear applications. Experimental results demonstrate that the Random Forest Regression (RFR) model achieved the highest accuracy, with a mean absolute percentage error of 3.1%, outperforming other predictive models. Comparing Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) models revealed a 15% performance improvement for BiLSTMs, attributed to their bidirectional data processing. The proposed cloud-based tool wear prognosis and diagnosis system operates exclusively on machining power signal data, ensuring minimal sensor requirements, in-situ and non-contact operation, and robustness against environmental variations. Compared with acoustic emission – and machine vision – based approaches, the system can incorporate Role-Based Access Control (RBAC) to securely and efficiently manage data and models.