PROCESS CONTROL COMBINING MACHINE LEARNING AND FINGERPRINT APPROACHES

Abstract

Manufacturing operations in large machine tools often requires several hours per part. Ensuring output quality is vital to avoid time and financial losses. While quality assurance was always problematic and costly, the recent advent of Industry 4.0 brought a new perspective to the problem as cutting machines are now fully digitized. This paper proposes a process control framework that combines a fingerprint approach that detects deviations with respect to the validated process and a Long Short-Term Memory (LSTM) algorithm that predicts the upcoming signals. This paper demonstrates how combining these two methodologies surpasses the performance of previous purely learning-based algorithms.