Electrical and operational anomaly detection in energy intensive manufacturing industries

Abstract

Significant part of manufacturing sector in India lacks in transparency of energy flow and have a low awareness on energy efficiency measures. Manpower working in industries is mostly semi-skilled. Production output, its quality and cost are affected by energy wastage and device failure causing downtimes resulting from electrical faults and erroneuos operations. Such abnormality must be detected and reported proactively to the facility manager, who can act to avoid major losses. This paper describes the concept of detecting electrical and operational abnormality (anomaly) of loads through observing changes in electrical parameters collected by installation of energy meters. The paper proposes classification of anomalies based on their origin. Further, essential feature sets required for accurate detection of anomalies are described. To verify the concept, load data is collected from a pilot small scale manufacturing facility by installing energy meters at different load points in process lines. After preprocessing the raw data, necessary features are extracted and are subjected to classification algorithms for detecting possible anomalies. Results for two loads at test site are presented with comparison of support vector data descriptor and support vector machine algorithms for classification as normal or anomalous.