An improved correlation-based anomaly detection approach for condition monitoring data of industrial equipment

An improved latent correlation anomaly detection (LCAD) method is proposed to detect anomalies from condition monitoring datasets of industrial equipment. Above all, original data were segmented to various work cycles. Then, latent correlation vector (LCV) was used to denote the latent correlation among different parameters. Based on a latent correlation probabilistic model (LCPM), an anomaly detection function (ADF) is formulated to determine the state of equipment. In order to compare this method with previously reported anomaly detection methods, simulated datasets were constructed to evaluate the effectiveness of this method. Another experiment was also conducted to test the applicability of this method based on real flight datasets. Both experiments demonstrated superior accuracy and much lower missing alarm rates of this improved LCAD method.

I. Introduction

With the rapid development of condition monitoring technologies, numerous data containing abundant information of equipment are generated every day. How to extract useful information by conducting a thorough analysis of these data is becoming a hot research subject. Anomaly detection, or outlier detection is an important part of data analysis and data mining technologies. It aims to detect observations which deviate so much from others that they are suspected of being generated by different mechanisms [1]. Accurate and timely anomaly detection is critical for reliable and affordable operation of major industrial equipment. Accurate and timely alerts for anomalies can let operators allocate additional monitoring resources, schedule preventive maintenance, minimize downtime, and reduce costly expenses for unexpected maintenance events [2].