Condition monitoring of SSE gas turbines using artificial neural networks

摘要

SSE (formerly Scottish & Southern Energy) has been pursuing a path of continuous improvement in asset reliability, performance and productivity. The Equipment Performance Centre of SSE uses predictive analytics to monitor SSE business critical assets. Gas turbines are one of the most versatile prime mover technologies in the fleet. Power generation using gas turbines has become very popular because of the availability of gas and its relatively low emissions. Detecting impending failures on gas turbines reduces the risk of operational disruption, risk to personnel and loss of revenue, and maintains plant integrity. Effective monitoring and diagnosis requires a range of techniques. This paper outlines one of the diagnostic techniques used to detect combustor-related damage by the examination of gas turbine exhaust temperatures. The diagnostics methodology involves the use of an auto-associative neural network (AANN) to detect anomalies when the current condition deviates from normal behaviour. This type of neural network is a novel approach for the detection of combustor-related damages. The AANN was developed in MATLAB. This method has successfully detected several different combustor-related incidents in the SSE gas turbine fleet. Events such as this have the potential to release material into the gas flow path and cause damage to the turbine. The cost of such a failure can be in the region of several million pounds. The technique can also be used to detect anomalies on other gas turbine sections, ie compressor, fuel, cooling and mechanical systems. This paper illustrates how the technique can be applied to monitor any series of industrial gas turbines.