Filter-Based Intelligent Output-Constrained Control of Uncertain MIMO Nonlinear Systems With Sensor and Actuator Faults

Ning Zhou, Canyang Zhao, Xiaodong Cheng, Yuanqing XiaIEEE Trans. SMC: Systems 2025 · DOI

Abstract

This paper presents a novel Adaptive-Neural-Backstepping Fault-Tolerant Constrained (ANBFTC) control algorithm for Multi-Input Multi-Output (MIMO) nonlinear systems. The proposed method addresses the critical challenge of maintaining system stability and tracking performance under simultaneous sensor and actuator faults, while strictly adhering to asymmetric output constraints. By utilizing a filter-based adaptation coordinate transformation (FBACT) and radial basis function neural networks (RBFNNs), the controller eliminates the need for prior knowledge of fault magnitudes, significantly enhancing robustness for real-world applications such as robotics and industrial automation.

Introduction

Multi-Input Multi-Output (MIMO) nonlinear systems, including quadcopters, robotic arms, and chemical reactors, are foundational to modern engineering but present significant control challenges. These systems are often subject to three primary difficulties: model uncertainty, strict output constraints for safety, and hardware faults. In harsh environments, sensors may provide scaled or biased data, while actuators may lose effectiveness.

Most existing control strategies handle these issues in isolation. However, managing simultaneous sensor and actuator faults while strictly enforcing asymmetric constraints—without knowing the exact size or timing of those faults beforehand—remains a major gap. This research bridges that gap by providing a unified approach that enforces safety limits even when multiple components fail concurrently.

Methodology

The proposed ANBFTC algorithm integrates several advanced control strategies to achieve fault tolerance and constraint satisfaction.

Core Technical Pillars

  • Filter-Based Adaptation Coordinate Transformation (FBACT): This core innovation allows the controller to adapt to faults without needing a priori information about their bias or scale. It transforms the coordinates through a filter, treating faults as part of the system's total uncertainty.
  • Radial Basis Function Neural Networks (RBFNNs): RBFNNs are employed to approximate unknown nonlinear dynamics in real-time, removing the requirement for the system to be "locally Lipschitz," a common but restrictive mathematical assumption.
  • Dynamic Surface Control (DSC): To prevent the "explosion of complexity" common in traditional backstepping methods, DSC uses first-order filters at each step to simplify the calculation of derivatives, making the algorithm computationally efficient.
  • Barrier Lyapunov Functions (BLFs): Asymmetric BLFs are used to ensure that system outputs never violate predefined safety boundaries. If the system approaches a constraint, the mathematical "cost" goes to infinity, forcing the controller to keep the system within safe limits.
Block diagram of the proposed ANBFTC control architecture showing the iterative steps from Step 1 to Step n.

The adaptive nature of the controller is driven by several parameters designed to estimate external disturbances, neural network weights, and fault coefficients. These parameters allow the system to remain "fault agnostic," adapting to changes without explicit fault detection hardware.

Summary of adaptive parameters and their corresponding design functions within the control law.

Evaluation and Results

The algorithm was evaluated using a numerical simulation of a complex mechanical system: two inverted pendulums connected by a spring and a damper. This benchmark model is highly nonlinear, coupled, and sensitive to disturbances, making it an ideal testbed for fault-tolerant control.

Key Findings

  • Robust Constraint Satisfaction: Despite significant sensor and actuator faults introduced during the simulation (e.g., a sensor losing 30% of its scaling at $t=20s$), the system outputs never violated the predefined safety boundaries.
  • Tracking Accuracy: All tracking errors converged to a small neighborhood of zero, demonstrating high precision even under failure conditions.
  • Theoretical Stability: The researchers mathematically proved that all signals in the closed-loop system are Semi-Globally Uniformly Ultimately Bounded (SGUUB), guaranteeing that the system remains stable and will not "blow up" under the proposed control law.

Conclusion and Future Work

The ANBFTC algorithm represents a significant advancement in fault-tolerant control for complex MIMO systems. By combining intelligent neural approximation with filter-based adaptation, it provides a safety-first solution that is both robust and computationally manageable. Future research will focus on validating these results on physical hardware, such as quadcopters or robotic manipulators, to account for real-world noise and latency, and extending the framework to handle more complex failure modes like intermittent "stuck-at" faults.

Summary updated

BibTeX
@article{zhou2025filter,
  author = {Ning Zhou and Canyang Zhao and Xiaodong Cheng and Yuanqing Xia},
  title = {{Filter-Based Intelligent Output-Constrained Control of Uncertain MIMO Nonlinear Systems With Sensor and Actuator Faults}},
  journal = {IEEE Trans. Syst. Man Cybern. Syst.},
  year = {2025},
  doi = {10.1109/tsmc.2025.3559694},
}
Figure