A Novel Approach to Recognize Hidden Machinery Malfunctions Based on Operational Data

Abstract

This work introduces a novel approach that identifies hidden operation malfunctions of complex machinery equipment. The architecture of the proposed approach is based on the ?? ?????????????? ????????h?????? (kNN) unsupervised algorithm for anomaly detection, along with the ?? ?? correlation coefficient, and aims to provide information on whether systematic malfunctions exist on the equipment. The proposed approach can provide insights into the root cause of the malfunctions. For the evaluation of the proposed approach, a case of a food industry packaging machine is studied. The rejections of the packaged products can be used as a metric for defining the operational efficiency of the machine. The anomaly detection task is performed for each variable that monitors the operation of the machine. Then, the correlation coefficient between the detected anomalies and the rejections of the machine is calculated. The higher the correlation coefficient, the most probable cause for the rejections is a malfunction of the machine’s subsystem that the specific measured parameters monitor. A semi-synthetic dataset based on real operational data was created to conduct experiments, and high-accuracy results were obtained.