Application of artificial neural networks for detecting compressor fouling in industrial gas turbines: a case study of an aero-derivative unit at an oil and gas facility in the Niger Delta, Nigeria

This study investigates the application of artificial neural networks for the detection of compressors fouling degradation in industrial gas turbines during operation to mitigate the loss in engine performance. An Artificial Neural Network (ANN)-based model was developed to monitor and predict compressor fouling degradation in an aero-derivative gas turbine derived from the Siemens SGT 400 class of gas turbines. Performance data from a Siemens SGT 400 gas turbine unit were obtained and used for the investigation. The obtained engine data represent all faults indicative of compressor performance. For the baseline, data were collected after maintenance actions had taken place, while the degraded case covers historical engine performance from 01 January 2013 to 28 February 2013, accounting for approximately 1,392 Equivalent Operating Hours (EOH). The dataset, encompassing variables such as temperature, pressure, gas flow, power, compressor discharge temperature, and compressor discharge pressure, was processed to eliminate irrelevant and redundant parameters before usage. A Multi-Layer Perceptron (MLP) was chosen as the architecture for the ANN. The outcomes of the training phase showed that the ANN achieved a classification accuracy of 96.2 % in proficiently distinguishing between “fouling” and "other factors" conditions. Additionally, the validation performance plot demonstrates that the network achieved its best performance with a value of 0.077507 at 18 epochs out of 24 training iterations. Finally, the confusion matrix demonstrates the model's capability to predict both fouling and non-fouling scenarios with a minimal rate of misclassification.