An improved ensemble fusion autoencoder model for fault diagnosis from imbalanced and incomplete data

Skewed distribution and incompleteness of monitored data might cause feature submergence and information loss, rendering the fault diagnosis from imbalanced and incomplete data commonly existing in industrial systems is still an intractable problem. Therefore, in order to improve the accuracy of fault diagnosis from imbalanced and incomplete data, this paper proposes a fusion autoencoder (FAE) network and an ensemble diagnosis scheme. A designed multi-level denoising strategy and a variable-scale resampling strategy are adopted as compensation for information loss and skewed distribution. The multiple FAE networks are constructed by combining the advantages of improved sparse autoencoder (SAE) with denoising autoencoder (DAE) to enhance the adaptability. Different hyper parameters are configured for each FAE to ameliorate the diagnostic flexibility, and Bagging strategy is employed to integrate each network into a complete FAE fault diagnosis model. Furthermore, evaluation criteria are suggested and the application range of the model is tested. Finally, different experiments are conducted to verify the effectiveness and practicability of the proposed method.