To address the inherent nonlinearity and strong coupling among rotor displacement, speed, and flux linkage in the composite cage rotor bearingless induction motor (CCR-BIM), an inverse system decoupling control strategy based on a support vector machine (SVM) optimized by the improved simulated annealing-genetic algorithm (ISA-GA) is proposed. First, based on the structure and working principle of CCR-BIM, the mathematical model of CCR-BIM is derived, and its reversibility is rigorously analyzed. Subsequently, an SVM regression equation is established, and the SVM kernel function parameters are optimized using the ISA-GA to train a high-precision inverse system decoupling control model. Finally, the inverse system is cascaded with the original system to construct a pseudo-linear system model, achieving linearization and decoupling control of CCR-BIM. To verify the effectiveness and practicability of the proposed decoupling control strategy, the proposed control method is compared with the traditional inverse system decoupling control strategy through simulation and experimentation. Both simulation and experimental results demonstrate that the proposed decoupling control strategy can effectively achieve decoupling control of rotor displacement, rotational speed, and flux linkage in CCR-BIM.

The bearingless induction motor (BIM) relies on accurate displacement feedback to maintain stable magnetic suspension, but sensor faults, degradation, and noise can distort feedback and induce transients during branch switching. This paper proposes a confidence-fusion-based fault-tolerant displacement measurement method for the BIM suspension feedback chain. A four-channel asymmetric redundant sensor configuration is developed, and channel state evaluation functions are constructed from sampling-difference terms and geometric-consistency residuals. A decreasing Sigmoid mapping with first-order smoothing generates continuous confidence coefficients to represent channel health. Combined with discrete fault flags of the primary channels, four reconstruction branches, AB, BC, AC, and CD, are adaptively weighted to obtain the reconstructed displacement, which is connected to the original suspension controller through a smooth feedback access mechanism. A MATLAB/Simulink closed-loop suspension model is used to evaluate the method under fault-free operation, an abrupt fault of primary channel A, simultaneous and sequential faults of primary channels A and B, abrupt and gradual degradation, constant bias, intermittent signal dropouts, and noise disturbance of primary channel B. Results show that the method identifies abnormal primary channels, redistributes reconstruction weights according to sensor conditions, and maintains a fallback path through the CD branch under dual-primary-channel failure. Under channel-B degradation, the confidence coefficient tracks the deterioration and supports the subsequent AB-to-AC branch transfer, whereas under noise disturbance, the fault flag remains inactive and unnecessary branch switching is avoided. The method improves feedback continuity without changing the main suspension controller.