Machine Learning Based Monitoring of the Pneumatic Actuators’ Behavior Through Signal Processing Using Real-World Data Set

Abstract

Application of machine learning in smart manufacturing could enrich condition-based maintenance, enabling to monitor more complex signals describing the behavior of an equipment. Predicting equipment malfunction or failure could not only reduce manufacturing cost but could also improve product quality and supply reliability, aspects equally important these days. This research describes a machine learning based method to diagnose the condition of pneumatic actuators by processing complex signals from this type of equipment. It is using real-world data from a Hungarian multinational manufacturer of electrical components. The “Manufacturing Data Life Cycle” conceptual framework by Tao et al. [1] was followed for data processing, including data cleaning, pre-processing and classification by hierarchical clustering. The method was able to identify those signal patterns that indicate abnormal behavior of the equipment. Data volume was a challenge for this work, overcome by using parallel and GPU execution of computations.