Toward Robust Fault Identification of Complex Industrial Processes Using Stacked Sparse-Denoising Autoencoder With...

Area of Science:

• Industrial Process Monitoring
• Artificial Intelligence
• Machine Learning

Background:

• Complex industrial processes (CIPs) require accurate fault identification for safe and efficient operation.
• Traditional methods struggle with noise and complex data patterns in CIP monitoring data (CIPMD).
• Deep learning offers potential for robust feature extraction from noisy industrial data.

Purpose of the Study:

• To propose a robust end-to-end deep learning fault recognition scheme for CIPs.
• To develop a stacked sparse-denoising autoencoder (SSDAE)-Softmax model for accurate fault identification.
• To optimize the model using the state transition algorithm (STA) for enhanced performance.

Main Methods:

• Integrating sparse autoencoder (SAE) and denoising autoencoder (DAE) into sparse denoising autoencoder (SDAE) for feature representation.
• Stacking multiple SDAEs with layerwise pretraining and a Softmax classifier for the SSDAE-Softmax model.
• Employing the state transition algorithm (STA) for hyperparameter optimization of the SSDAE-Softmax model.

Main Results:

• The SSDAE-Softmax model demonstrated effective identification of various process faults in simulation and real-world industrial systems (TEP and CCP).
• The proposed method showed strong robustness and adaptability against noise interference in CIPMD.
• Deep learning-based feature representation combined with STA optimization enabled adaptive learning of intrinsic data characteristics.

Conclusions:

• The SSDAE-Softmax model provides an effective solution for fault recognition in complex industrial processes.
• The method exhibits superior robustness and adaptability to noise, crucial for real-world industrial monitoring.
• This deep learning approach enhances the accuracy and reliability of fault identification in CIPs.