Anomaly Detection of Gas Turbine Hot Components Based on Deep Autoencoder and Support Vector Data Description

Anomaly detection of gas turbine hot components can ensure its operational safety and reliability. With the boom of artificial intelligence, data-driven fault diagnosis is becoming increasingly popular. However, in actual applications, fault data of gas turbines are rare or even unavailable. Aiming to solve the anomaly detection problem of gas turbine hot components in the case of only normal data available, this paper proposed an anomaly detection method based on the fusion of deep autoencoder and support vector data description. This method uses normal data to train deep autoencoder and then uses the reconstruction errors of deep autoencoder to train support vector data description. Experiments show that, compared with conventional anomaly detection methods, the proposed method can significantly improve the anomaly detection accuracy and realize more sensitive and robust anomaly detection of gas turbine hot components.