A Data-driven Deep Learning Approach for Remaining Useful Life in the ion mill etching Process

Abstract

Prognostics and Health Management (PHM) is regarded as an essential element in the scope of intelligent manufacturing. Precise forecasting of the remaining useful life (RUL) of an ion mill is crucial in order to enhance the overall efficiency of the ion mill etching (IME) procedure. This paper proposed a Data-driven Deep Learning (DL) framework that integrates a Temporal Convolution Network (TCN), Long Short-Term Memory (LSTM), and self-attention mechanism to improve the accuracy of RUL prediction in the ion mill etching Process. Initially, sensor input data is divided into two parallel paths - one with TCN blocks for capturing long-range dependencies, and the other with LSTM layers for extracting temporal patterns. The outputs from both paths are then merged and input into an LSTM layer for enhanced learning, followed by a self-attention mechanism to highlight important features then fully connected layer for predicting RUL. The efficacy of this suggested model was assessed through the utilization of the 2018 PHM Data Challenge Dataset and juxtaposed against various Deep Learning models to demonstrate its efficacy. The results from the experiments indicate that ATCN-LSTM serves as a robust option for estimating the RUL in the ion mill etching Process as it outperformed all other models that were compared. The source code is publicly accessible at https://github.com/ion-mill-etching-Process.