Generation of Uncorrelated Residual Variables for Chemical Process Fault Diagnosis via Transfer Learning-based Input-Output Decoupled Network

TL;DR

This work tackles fault diagnosis in chemical processes by generating uncorrelated residuals through a transfer-learning framework. The core idea is a transfer learning-based input-output decoupled network (TDN) that combines an input-output decoupled network (IDN) with a pre-trained variational autoencoder (VAE) to migrate faulty observations toward the normal domain, yielding residuals suitable for both fault detection and fault estimation. The training uses VAE-based losses on normal and simulated faulty data plus a maximum mean discrepancy (MMD) term to align domains, with the VAE parameters frozen during transfer to guide IDN learning. Experimental results on a nonlinear numerical example and a Three-Tank System show that TDN achieves superior fault detection (lower AFAR and AMDR) and fault estimation accuracy (lower RMSE) compared to baseline methods, driven by the decoupled residual structure that yields near-identity covariance across residual components. The approach offers a scalable, data-driven pathway for robust fault diagnosis in complex chemical processes where high-dimensional nonlinearities challenge conventional decoupling methods.

Abstract

Structural decoupling has played an essential role in model-based fault isolation and estimation in past decades, which facilitates accurate fault localization and reconstruction thanks to the diagonal transfer matrix design. However, traditional methods exhibit limited effectiveness in modeling high-dimensional nonlinearity and big data, and the decoupling idea has not been well-valued in data-driven frameworks. Known for big data and complex feature extraction capabilities, deep learning has recently been used to develop residual generation models. Nevertheless, it lacks decoupling-related diagnostic designs. To this end, this paper proposes a transfer learning-based input-output decoupled network (TDN) for diagnostic purposes, which consists of an input-output decoupled network (IDN) and a pre-trained variational autocoder (VAE). In IDN, uncorrelated residual variables are generated by diagonalization and parallel computing operations. During the transfer learning phase, knowledge of normal status is provided according to VAE's loss and maximum mean discrepancy loss to guide the training of IDN. After training, IDN learns the mapping from faulty to normal, thereby serving as the fault detection index and the estimated fault signal simultaneously. At last, the effectiveness of the developed TDN is verified by a numerical example and a chemical simulation.

Figures (11)

• Figure 1: The model structure of VAEs.
• Figure 2: Calculation flow of IDN for an individual sample. The notation "$\otimes$" marks the zero elements; "$\oslash$" stands for the non-adopted elements.
• Figure 3: The detailed structure of IDN.
• Figure 4: The overall model structure and forward propagation of TDN.
• Figure 5: FD curves of TDN-$\mathcal{D}_2$-$\mathcal{V}_4$ on the numerical example

Theorems & Definitions (3)

• Remark 1
• Remark 2
• Remark 3