An Unsupervised Anomaly Detection Based on Self-Organizing Map for the Oil and Gas Sector

Abstract

Anomaly detection plays a crucial role in preserving industrial plant health. Detecting and identifying anomalies helps prevent any production system from damage and failure. In complex systems, such as oil and gas, many components need to be kept operational. Predicting which parts will break down in a time interval or identifying which ones are working under abnormal conditions can significantly increase their reliability. Moreover, it underlines how the use of artificial intelligence is also emerging in the process industry and not only in manufacturing. In particular, the state-of-the-art analysis reveals a growing interest in the subject and that most identified algorithms are based on neural network approaches in their various forms. In this paper, an approach for fault detection and identification was developed using a Self-Organizing Map algorithm, as the results of the obtained map are intuitive and easy to understand. In order to assign each node in the output map a single class that is unique, the purity of each node is examined. The samples are identified and mapped in a two-dimensional space, clustering all readings into six macro-areas: (i) steady-state area, (ii) water anomaly macro-area, (iii) air-water anomaly area, (iv) tank anomaly area, (v) air anomaly macro-area, (vi) and steady-state transition area. Moreover, through the confusion matrix, it is found that the algorithm achieves an overall accuracy of 90 per cent and can classify and recognize the state of the system. The proposed algorithm was tested on an experimental plant at Università Politecnica delle Marche.

Keywords

• machine learning
• unsupervised learning
• Industry 4.0
• anomaly detection
• smart industrial plant
• predictive maintenance