摘要
Abstract
In permanent magnet synchronous motor(PMSM)drive systems,inverter-switch faults and current-sensor faults are the most common electrical failures.These two fault types often exhibit similar current distortion characteristics,making them difficult to distinguish.This diagnostic challenge is particularly pronounced in cost-sensitive or fault-tolerant applications where only two phase-current sensors are available,as traditional methods typically rely on complete three-phase current information or require additional hardware.Existing methods that operate with a reduced sensor set often struggle to distinguish between the two fault types or fail to identify the fault's specific nature.Therefore,this paper proposes a novel fault diagnosis strategy based on current prediction using only two-phase current measurements.
Firstly,a robust current prediction model for normal operation is established by integrating a speed-adaptive extended state observer(SA-ESO)into a deadbeat predictive current control(DPCC)architecture,providing an accurate baseline for both motor control and fault detection.Secondly,a three-step diagnostic procedure is executed.Faults are initially detected by monitoring the time-integrated residual of the dq-axis currents,computed as the difference between the predicted and measured values from the normal model.To isolate the faulty component,two distinct sets of predictive models are developed.One for inverter-switch open-circuit faults based on motor physics under fault conditions,and another for current-sensor faults,which leverage the SA-ESO to reconstruct system states using only a single healthy current.The faulty component is identified by selecting the healthy-phase current that deviates least from the measured value.Finally,once the component is identified,the specific fault type(e.g.,upper/lower switch fault,sensor open-circuit,gain/offset error)is determined by analyzing the characteristics of the faulty phase current signal.Techniques such as Recursive Least Squares(RLS)are used to estimate the parameters of gain and offset faults.
Experimental results on a PMSM test rig show the proposed method detects fault occurrence within 1～3 switching periods.Simpler faults,such as sensor open-circuits,are identified within 5 switching periods,whereas inverter open-circuit faults take up to 10 switching periods.Diagnosing sensor gain and offset errors requires approximately 1.5 ms to allow the RLS estimation algorithm to converge.To verify the robustness of the proposed model,the system was subjected to sudden changes in load torque and rapid speed variations,as well as significant parameter mismatches(+20%deviation in inductance,resistance,and flux linkage).The results show that the fault detection and isolation logic remains reliable and avoids false alarms.A comparison with other state-of-the-art methods shows that the proposed strategy enables high-speed diagnosis and distinguishes inverter open-circuit faults from multiple types of current sensor faults,while requiring only two current sensors.
The following conclusions can be drawn.(1)The proposed three-step diagnostic framework,which combines a normal operation model for detection with multiple dedicated fault models for isolation,proves to be an effective architecture for addressing the ambiguity between inverter and sensor faults in two-sensor systems.(2)The integration of the speed-adaptive extended state observer(SA-ESO)significantly enhances the robustness of the current prediction against system parameter variations and external disturbances,which is critical for preventing misdiagnosis during transient operating conditions.(3)The proposed model without additional hardware is practical for industrial applications,imposes a moderate computational load,and does not depend on large datasets for training.关键词
永磁同步电机驱动系统/电流预测控制/逆变器开关管/电流传感器/故障诊断/自适应扩张状态观测器/单相电流重构Key words
Permanent magnet synchronous motor(PMSM)drive system/predictive current control/inverter switch/current sensor/fault diagnosis/extended state observer/single-phase current reconstruction分类
信息技术与安全科学
