Deep-learning-driven intelligent component-level energy prediction of ultra-precision machine tools with IoT platform

Abstract

This study aims to investigate the utilization of deep learning technology to accurately predict the energy consumption of ultra-precision machining tools (UPMT)s at the component level. First, the energy consumption characteristics were thoroughly evaluated to serve as the foundation for separating the power data of various components. The training dataset was then generated using a modified Discrete Wavelet Transform (DWT) technique that extracted the component power depending on its frequency characteristic. Next, a multi-outputs 1-Dimension Convolutional Neural Network - Long Short Term Memory (1DCNN-LSTM) model was established and deployed on the Internet of Things (IoT) platform to classify component status while also predicting component power. For better model performance, the Optuna framework was leveraged to find the optimal hyperparameters configuration. The results indicated that the accuracy of the classification model of working components could reach 99 %. Additionally, the power consumption predictions of 11 working components performed well. The R2 values of the regression model for 11 types of components varied from 0.975 to 0.996. Notably, this research has significant theoretical and practical implications for enhancing the accuracy of UPMT energy consumption predictions and supporting the development of intelligent manufacturing.