Event Modeller Data Analytic for Harmonic Failures

Abstract

The optimum performance of power plants has major technical and economic benefits. A case study in one of the Malaysian power plants reveals an escalating harmonic failure trend in their Continuous Ship Unloader (CSU) machines. This has led to a harmonic filter failure causing performance loss leading to costly interventions and safety concerns. Analysis of the harmonic parameter using Power Quality Assessment indicates that the power quality is stable as per IEEE standards; however, repetitive harmonic failures are still occurring in practice. This motivates the authors to explore whether other unforeseen events could cause harmonic failure. Usually, post-failure plant engineers try to backtrack and diagnose the cause of power disturbance, which in turn causes delay and disruption to power generation. This is a costly and time-consuming practice. A novel event-based predictive modelling technique, namely, Event Modeller Data Analytic (EMDA), designed to inclusive the harmonic data in line with other technical data such as environment and machine operation in the cheap computational effort is proposed. The real-time Event Tracker and Event Clustering extended by the proposed EMDA widens the sensitivity analysis spectrum by adding new information from harmonic machines' performance. The added information includes machine availability, utilization, technical data, machine state, and ambient data. The combined signals provide a wider spectrum for revealing the status of the machine in real-time. To address this, a software-In-the-Loop application was developed using the National Instrument LabVIEW. The application was tested using two different data; simulation data and industrial data. The simulation study results reveal that the proposed EMDA technique is robust and could withstand the rapid changing of real-time data when events are detected and linked to the harmonic inducing faults. A hardware-inthe-Loop test was implemented at the plant to test and validate the sensitivity analysis results. The results reveal that in a single second, a total of 2,304 input-output relationships were captured. Through the sensitivity analysis, the fault causing parameters were reduced to 10 input-output relationships (dimensionality reduction). Two new failure causing event/parameter were detected, humidity and feeder current. As two predictable and controllable parameters, humidity and feeder velocity can be regulated to reduce the probability of harmonic fluctuation.