A Novel Blade Fault Diagnosis Using a Deep Learning Model Based on Image and Statistical Analysis

Abstract

AbstractArtificial intelligence technology has a high potential for machinery fault detection and diagnosis. Blade component failure is the main type of failure that usually occur in gas turbine and this component tends to fail unexpectedly. Detection and diagnosis of blade components are different with gear and bearing as both components have a standard vibration analysis and the fault can be examined using frequency domain analysis. Due to the complex structure of the blade system, the informative feature from the vibration signal on the blade fault often obscure with the noise signal. Therefore, this paper proposed a system using a combination of time–frequency image analysis and a stacked sparse autoencoder (SSAE) model to tackle the challenge of blade fault detection and diagnosis. The experiment is carried out using a multi-stage blade system and the result showed that proposed system is able to provide more than 90% diagnosis performance.KeywordsDeep learningGas turbineBladeFault diagnosis