TSFNN-Based Sensorless Diagnosis of Inter-Turn Short Circuits in Linear-Motor-Driven Steering Actuator

Permanent magnet linear synchronous motor (PMLSM) is emerging as an innovative alternative to hydraulic actuators in articulated steering systems of heavy-duty loaders, offering fast response, high precision, and inherent self-sensing potential. However, conventional current-based indicators usually trigger only after severe degradation, limiting their ability to capture incipient inter-turn short-circuit (ITSC) faults. To address this limitation, this study proposes a thrust-fluctuation-based sensorless diagnostic framework for early fault awareness. Thrust responses are decomposed using variational mode decomposition (VMD) and processed through statistical feature extraction to construct a multidimensional thrust-response index, thereby enhancing the detectability of weak faults. Furthermore, a Takagi–Sugeno fuzzy neural network (TSFNN) with an adaptive rule-clustering compression mechanism is developed, which reduces model complexity while improving generalization under varying operating conditions. Experimental validation on a full-scale loader platform confirms the effectiveness of the proposed method. Compared with four benchmark neural models, such as feedforward neural network (FNN), convolutional neural network (CNN), long short-term memory (LSTM), and gated recurrent unit (GRU), it achieves over 40% reduction in detection latency while maintaining an F1-score above 90% under mild ITSC conditions. These results are based on 30 repeated trials, indicating both consistent performance and statistical reliability. Overall, the results demonstrate that real-time thrust-based self-sensing is feasible and that intelligent health monitoring for PMLSM-driven steering systems can be effectively deployed in industrial settings.