Fault Detection in the Gas Turbine of the Kirkuk Power Plant: An Anomaly Detection Approach Using DLSTM-Autoencoder

Developing a maintenance schedule for different parts of a power plant can help to prevent serious system damage while also having a direct effect on reducing maintenance time and costs. Moreover, the proposed procedure can provide a better understanding of the health state of the system. The occurrence of problems (anomalous behavior) in power plants usually takes time and happens gradually. Predicting the possibility of anomalous behavior arising can therefore result in the prediction of a failure. In this paper, the authors tried to propose a new method for reducing the volume of input data by extracting the latent layer output of an Autoencoder, which was then applied to identify features. We achieved this by keeping relevant features and learning the anomalous behavior's encoded representation. Furthermore, the latent layer output from the encoder was used to train a Deep Long-Short-Term Memory (DLSTM) model with three hidden layers for fault occurrence prediction by considering an anomaly detection approach. To this end, a threshold approach was used to determine the possibility of a fault occurring in the turbine, with data collected from the Kirkuk gas turbine power plant. Piezoelectric accelerometer CA 202 was utilized as a sensor to collect data in industry. Since most of the recorded data were normal and the amount of abnormal data was small, various techniques, i.e., flipping and jittering, were employed to create synthetic abnormal data and augment the applicable dataset. In summary, according to the proposed methodology, the system is modelled as a time series classification problem to mimic the anomaly detection problem. The results showed that the accuracy of the trained model for detecting turbine vibration as a system abnormality is over 86%. Also, the implementation of the proposed prediction model on the experimental dataset yielded satisfactory results for detecting and predicting the occurrence of faults in Kirkuk gas turbine power plant.