A Collaborative Fault-Tolerant Strategy for Displacement Sensors in CCR-BIM: Dynamic Kernel Width Optimization of AGRNN and Ring-Shaped Redundancy

Displacement sensors are essential for closed-loop control and stable rotor suspension in the composite cage rotor bearingless induction motor. Considering their susceptibility to faults under complex operational conditions, this paper presents a collaborative fault detection and fault-tolerant strategy that integrates a dynamic kernel-width-optimized Adaptive Generalized Regression Neural Network (AGRNN) with a four-sensor ring-shaped redundant architecture. First, by dynamically optimizing the kernel width parameters and incorporating regularization mechanisms, the proposed AGRNN effectively mitigates the overfitting issues commonly associated with traditional Generalized Regression Neural Networks (GRNN). This enables high-precision displacement estimation based on features such as three-phase suspension currents, air-gap flux linkage, and unbalanced magnetic pull. Fault detection is realized through residual analysis combined with adaptive thresholding techniques. Second, a four-sensor ring-shaped redundancy structure is introduced, which utilizes coordinate transformation matrices to reconstruct displacement signals in the event of single or dual sensor failures. Simulation and experimental results confirm that the AGRNN-based method significantly outperforms conventional GRNN approaches in both single- and dual-fault scenarios. Overall, the strategy markedly improves CCR-BIM robustness under complex disturbances and multifault conditions and provides guidance for high-reliability, high-precision motor systems.