Inter-turn short circuit fault, insulation fault, eccentricity fault, and demagnetization fault are the most common faults in permanent magnet synchronous motor (PMSM). As these faults degrade reliability and cause serious catastrophes, it is necessary to diagnose these faults. However, in current methods, a fault indicator (FI) can often diagnose only one or two faults, which means that multiple FIs are required for diagnosis above four faults. To diagnose multiple faults with fewer FIs, a fault diagnosis method based on stator tooth flux (STF) of multiple teeth and multiscale kernel parallel residual convolutional neural network (PR-CNN) is proposed in this article. First, the STF of multiple teeth under four faults is analyzed. Then, FI is proposed based on the characteristics of the STF. Finally, the proposed PR-CNN is compared with state-of-the-art CNNs highlighting the superiority in this application. The results indicate that the proposed method can diagnose the above four faults by a single FI with an accuracy of 98.8% and a training data ratio of 40%. This work provides a significant reference for the multiple fault diagnosis of PMSM.

Inter-turn short circuit fault, insulation fault, eccentricity fault, and demagnetization fault are the most common faults in permanent magnet synchronous motors (PMSM). As these faults degrade their reliability, an integrated fault diagnosis method that can diagnose multiple faults is necessary. Fault indicator (FI) is the basis of fault diagnosis. In current methods, a single FI is difficult to diagnose the above four faults. In this article, a FI composed of stator tooth flux (STF) with multiple teeth is proposed. The fault features extracted by deep learning networks from the proposed FI can diagnose the investigated four faults. First, the STF of multiple teeth under four faults is analyzed to understand the effects of these faults on flux at different positions. Then, the FI is constructed based on analyzed characteristics. The proposed FI exhibits excellent performance in relatively few STFs (different teeth). Finally, the effectiveness and generality of the proposed FI are verified by a deep learning network based on simulation data and experimental data. The proposed FI provides a significant reference for the diagnosis of multiple faults in PMSM.

Conventionalpermanent magnet (PM) machine temperature assessment approaches require physical, material, and loss distribution information. In practice, accurate structural information is rarely available, a lot of thermal parameters can only be calibrated based on extensive experiments, and the determination of various machine losses is also a demanding task. To solve these problems, an online temperature prediction method is proposed in this article. A general mathematical model that describes the relationship between machine temperature and loss variation is derived through analyzing a PM machine three-dimensional lumped parameter thermal network. With distributed sensors, this model can be determined purely from the measured data, which eliminates the need to acquire machine physical and material information. The analyses of loss distribution are also avoided by utilizing two measured temperatures. The evolution of all sensors can be concerned, so the prediction of the time and spatial distribution of temperature can be realized. The proposed method is validated by the experiments processed on a 10-kW PM machine under various working conditions. The influences of key parameters and the comparison with a neural network model are also demonstrated. The results suggest that the proposed method has the advantage of possessing closed-form expression and being computationally efficient.

In magnetic bearing systems, the air gap between the stator and the rotor is generally measured with an eddy current displacement sensor. However, the inductive displacement sensor (IDS) is a preferred choice due to its structural consistency with magnetic bearings, which contributes to a compact system design. While most of the reported IDS research studies are based on three-dimensional finite element method (3D FEM) simulation and the IDS of such studies is designed with a half-bridge structure, in this paper, a higher sensitivity IDS with a full-bridge structure is proposed. To optimize the response of the sensor, an accurate theoretical analysis method is presented for the sensor based on the Schwarz–Christoffel transformation, which is used for compensating fringe effects of sensor inductance. Compared to traditional 3D FEM simulation, fast sensor design and optimization can be achieved with the proposed method. Experimental results agree well with theoretical predictions. The sensitivity of the proposed sensor is nearly 15.5 mV/μm, and the displacement resolution is better than 1 µm.

Interturn short circuit (ITSC) fault is a common fault in electric machines, which may severely damage the machines if no protective measure is taken in time. There are numerous fault diagnosis methods under a steady-state condition. However, there is relatively limited research on fault diagnosis under dynamic conditions. The dynamic operation of motors, such as in electric cars, is a very common scenario. Hence, this paper proposes a search-coil based online method for detecting ITSC fault in permanent magnet synchronous machine (PMSM) under a dynamic condition. The search coils are placed on the stator circumference at equal intervals. Each search coil reflects the information about the magnetic field in its vicinity and also contains the fault information. In this paper, the voltage induced by the odd sideband harmonics around the even carrier (2ωc±ω0) is selected as the fault characteristic to be used in effectively improving the detected signal-to-noise ratio by excluding the interference of the counter-potential of the permanent magnet. Since two adjacent search coils are placed one pole apart, a set of quadrature signals can be acquired. The Digital Lock-In Amplifier (DLIA) technology is applied to extract the amplitude of the characteristic voltage, which overcomes the shortcomings of the traditional spectrum analysis in applying to non-stationary conditions. The amplitudes of the voltage at different search coils can be compared to further determine the occurrence of a fault and also its rough location if occurred. Experiments were conducted with a six-phase PMSM for demonstrating the effectiveness of the proposed method. The obtained results show that the proposed method can accurately determine the occurrence of a fault.