Autoencoder-assisted Feature Ensemble Net for Incipient Faults

TL;DR

AE-FENet tackles incipient fault detection in the Tennessee Eastman Process by embedding an autoencoder-based feature transformer within the FENet framework, replacing PCA. It fuses outputs from seven basic detectors into a feature matrix and processes them through sliding-window analysis, PCA, and deep autoencoding to learn rich representations, culminating in a decision index $D_q$ derived from normal-data statistics. The approach achieves an average detection performance above $96\%$ on the hard faults 3, 9, and 15 and demonstrates extensibility to Sparse AE, Attention AE, and Variational Autoencoder variants, indicating strong generalization for unsupervised fault monitoring in complex processes. This provides a scalable, data-efficient pathway for reliable incipient fault detection in industrial settings.

Abstract

Deep learning has shown the great power in the field of fault detection. However, for incipient faults with tiny amplitude, the detection performance of the current deep learning networks (DLNs) is not satisfactory. Even if prior information about the faults is utilized, DLNs can't successfully detect faults 3, 9 and 15 in Tennessee Eastman process (TEP). These faults are notoriously difficult to detect, lacking effective detection technologies in the field of fault detection. In this work, we propose Autoencoder-assisted Feature Ensemble Net (AE-FENet): a deep feature ensemble framework that uses the unsupervised autoencoder to conduct the feature transformation. Compared with the principle component analysis (PCA) technique adopted in the original Feature Ensemble Net (FENet), autoencoder can mine more exact features on incipient faults, which results in the better detection performance of AE-FENet. With same kinds of basic detectors, AE-FENet achieves a state-of-the-art average accuracy over 96% on faults 3, 9 and 15 in TEP, which represents a significant enhancement in performance compared to other methods. Plenty of experiments have been done to extend our framework, proving that DLNs can be utilized efficiently within this architecture.

Figures (4)

• Figure 1: The general framework of AE-FENet
• Figure 2: The network structure of AE-FENet
• Figure 3: The structure of autoencoder
• Figure 4: Detection performance of basic detectors and AE-FENet for fault 15 in TEP