Real-Time Event Based Predictive Modelling for Industrial Control and Monitoring

Abstract

The purpose of this research work is to presents a novel event-based predictive modelling technique, namely, Event Modeller Data Analytic (EMDA), applicable for a large-scale real-time complex system. Borrowed from the Event Tracker and Event Clustering method, EMDA continuously estimates and builds a correlation map between system input (triggered data) and output (event data) parameters while predicting system failure based on machine performance metrics. With the aid of advanced machine learning models, EMDA can potentially predict linear and non-linear problems, thereby improving rapid decision-making for system engineering problems. For proof of concept, EMDA was used to analyse the mystery of an escalating harmonic failure trend in one of the Malaysian power plants. Analysis of the harmonic parameter in their Continuous Ship Unloader machines indicates that the power quality is stable as per IEEE standards; however, in practice, repetitive harmonic failures occur and the reasons remain unknown. The hypothesis associated with this research is that: "A fault in a power system distribution could be influenced not only by internal events but also by external events such as environment and climate change". In addition to the conventional method used by the power system engineers, we challenge the body of knowledge in the subject area by exploring and potentially incorporating external variables that may influence the state of the system. Software-In-the-Loop application was developed using the National Instrument LabVIEW. The purpose of this deployment was to test and validate the concept and to demonstrate whether the correlation analysis was in synchronisation with the latent knowledge (KPI) translated into the system. EMDA was also used as a tool to visualise the occurrences of the system parameters and its KPIs with a predictive analytical approach to data. This research conducts extensive experimental work on both industrial and synthetic data to evaluate the proposed method. The results of the study reveal that in addition to the known parameters that may affect harmonic filter performance, there is one new parameter that shows a reasonable correlation with performance. The previously unknown parameter is the humidity of the operational environment having a significant impact on the occurrence of harmonic failure. This proved the hypothesis set in the underpinning research endeavour presented in this thesis. By controlling the humidity of the operational environment and deploying EMDA, the state transition and trends were accurately predicted. The results of this research can help power generation plants to devise adaptive strategies to optimise the performance of plants.