Explainable Intelligent Fault Diagnosis for Nonlinear Dynamic Systems: From Unsupervised to Supervised Learning

Hongtian Chen, Zhigang Liu, Cesare Alippi, Biao Huang, Derong Liu

Abstract

The increased complexity and intelligence of automation systems require the development of intelligent fault diagnosis (IFD) methodologies. By relying on the concept of a suspected space, this study develops explainable data-driven IFD approaches for nonlinear dynamic systems. More specifically, we parameterize nonlinear systems through a generalized kernel representation for system modeling and the associated fault diagnosis. An important result obtained is a unified form of kernel representations, applicable to both unsupervised and supervised learning. More importantly, through a rigorous theoretical analysis, we discover the existence of a bridge (i.e., a bijective mapping) between some supervised and unsupervised learning-based entities. Notably, the designed IFD approaches achieve the same performance with the use of this bridge. In order to have a better understanding of the results obtained, both unsupervised and supervised neural networks are chosen as the learning tools to identify the generalized kernel representations and design the IFD schemes; an invertible neural network is then employed to build the bridge between them. This article is a perspective article, whose contribution lies in proposing and formalizing the fundamental concepts for explainable intelligent learning methods, contributing to system modeling and data-driven IFD designs for nonlinear dynamic systems.

Introduction

As modern automation systems, such as high-speed trains and large-scale chemical plants, grow in complexity, the stakes for accurate fault diagnosis have never been higher. While deep learning has revolutionized Intelligent Fault Diagnosis (IFD), it faces significant hurdles: the inherent nonlinearity of real-world systems, the high cost of obtaining labeled fault data (the "labeling dilemma"), and the lack of interpretability in "black-box" AI models.

Schematic of a nonlinear dynamic system with potential actuator ($f_a$) and sensor ($f_s$) faults.

This paper proposes a theoretical framework to address these challenges by unifying unsupervised and supervised learning paradigms through a mathematical "bridge," ensuring that diagnostic models are both high-performing and explainable.

Methodology: Generalized Kernel Representations and the Suspected Space

The proposed framework shifts away from pure black-box modeling toward a "gray-box" approach. The authors parameterize nonlinear systems using Generalized Kernel Representations, which allow for a consistent description of system dynamics across both modeling and diagnostic tasks.

A key component of this approach is the definition of a "Suspected Space"—a mathematical domain where potential faults are likely to manifest. By comparing actual system outputs to this space, the framework generates residuals that quantify the difference between expected and observed behavior, facilitating precise fault detection.

The Theoretical Bridge: Unifying Learning Paradigms

The core contribution of this work is the mathematical proof of a bijective mapping, or "bridge" ($P_{sp/unsp}$), between unsupervised and supervised learning entities. This bridge demonstrates that information can be transformed between these two paradigms without loss of diagnostic utility.

The mathematical bridge ($P_{sp/unsp}$) establishing a bijective mapping between unsupervised ($\mathcal{E}_s^{unsp}$) and supervised ($\mathcal{E}_s^{sp}$) spaces.

To realize this bridge in practice, the authors employ Invertible Neural Networks (INNs). Because INNs are designed to be reversible, they provide the necessary mathematical rigor to ensure that the mapping between unsupervised and supervised residuals is lossless. This leads to a significant finding: given the right mathematical structure, both learning styles can achieve optimal and equivalent fault-detection performance.

Key Insights and Future Directions

  • Performance Parity: The framework proves that the choice between supervised and unsupervised learning can be based on data availability rather than inherent performance limits, as both can reach the same level of diagnostic power.
  • Structural Explainability: By grounding the learning process in kernel representations, the authors provide a quantitative link between AI outputs and the physical dynamics of the system.
  • Unified Framework: The research successfully unifies two previously distinct branches of IFD, providing a single set of notations and operators for the field.

Despite these advances, several areas for future work remain. These include optimizing network hyperparameters for unknown systems, quantifying the minimum data requirements for specific confidence levels, and addressing the computational complexity of scaling INNs to extremely high-dimensional systems. This perspective article sets the stage for more transparent and flexible intelligent learning methods in industrial automation.

Summary updated

Abstract

The increased complexity and intelligence of automation systems require the development of intelligent fault diagnosis (IFD) methodologies. By relying on the concept of a suspected space, this study develops explainable data-driven IFD approaches for nonlinear dynamic systems. More specifically, we parameterize nonlinear systems through a generalized kernel representation for system modeling and the associated fault diagnosis. An important result obtained is a unified form of kernel representations, applicable to both unsupervised and supervised learning. More importantly, through a rigorous theoretical analysis, we discover the existence of a bridge (i.e., a bijective mapping) between some supervised and unsupervised learning-based entities. Notably, the designed IFD approaches achieve the same performance with the use of this bridge. In order to have a better understanding of the results obtained, both unsupervised and supervised neural networks are chosen as the learning tools to identify the generalized kernel representations and design the IFD schemes; an invertible neural network is then employed to build the bridge between them. This article is a perspective article, whose contribution lies in proposing and formalizing the fundamental concepts for explainable intelligent learning methods, contributing to system modeling and data-driven IFD designs for nonlinear dynamic systems.
BibTeX
@article{chen2022explainable,
  author = {Hongtian Chen and Zhigang Liu and Cesare Alippi and Biao Huang and Derong Liu},
  title = {{Explainable Intelligent Fault Diagnosis for Nonlinear Dynamic Systems: From Unsupervised to Supervised Learning}},
  journal = {IEEE Transactions on Neural Networks and Learning Systems},
  year = {2022},
  doi = {10.1109/tnnls.2022.3201511},
}
Figure