Development and Implementation of Principal Component Analysis Method for Monitoring of Gas Turbine

Gas turbines are complex and expensive machines that the cost of repairing unexpected failures is very high. There are many sensors installed in each gas turbine that record and collect large amounts of data. With the data mining of such big data, failure prediction is possible before the occurrence. The data set for the present study is the recorded quantities of sensors mounted on a 9-frame gas turbine in one of the country's power plants. The one column of data matrix rows was first labeled to identify healthy and defective row in each data sample. Then, by using the Principal Component Analysis method, the dimensions of the data matrix were reduced from seven to four dimensions and the main features were extracted. Following this, a model was developed by applying Artificial Neural Network method that was able to identify fault rows in the data matrix and identify the class of the data samples as healthy or defective. Accuracy, precision, and convergence of the model for two-to-six-dimensional model reductions were studied after machine learning was performed on 80% of the data. After matrix dimensionality reduction, and feature extraction by using "Principal Component Analysis" method, our well-designed model was also able to identify and classify the fault by using "Artificial Neural Network" method. In this thesis, it was found that our mode l by combining "Principal Component Analysis" method with "Artificial Neural Network" was able to show more than 90% precision with good accuracy and maximum degree of data matrix convergence. Moreover, it was able to specify the gas turbine fault class.