In order to satisfy the requirements of on-line monitoring of the suspension system under complex conditions, a fault detection method for maglev train suspension system based on multi-model switching is proposed. In the proposed method, the healthy samples are extracted through the moving time window, and then the features of the healthy samples are extracted by the Fast Walsh-Hadamard transform and filtered by the median filter. Then, the normalization is used to eliminate the difference of feature vectors, and then the principal component analysis method is used to reduce the dimension and de-correlation of the feature matrix contributing to a hyper-sphere space. Finally, the health threshold and the fault threshold are determined by Euclidean distance, then fault detection models are established for various operating conditions. Taking the positive line operation condition as an example, the proposed method is compared with the method based on the original feature and support vector data description based on the original feature. The results demonstrate that the proposed method is superior to the other two methods in terms of health detection rate and false positive rate. In addition, the proposed method is of the characteristics of low computational complexity, no-parameter optimization, and good robustness. It can be applied in practical engineering providing a certain research basis for data deep mining such as fault diagnosis and fault prediction.INDEX TERMS Maglev train suspension system, complex conditions, multi-model switching, fault detection models.

In this study, a Magnetic Levitation Energy Harvester (MLEH) was designed and fabricated. The magnetic field distribution and power generation performance of multiple cylindrical magnets were studied. The full factorial design (FFD) of L20 (22 × 5) test was carried out with the sliding magnet arrangement, coil arrangement and wiring method as the control factors, and the output power as target factor. Sweeping-frequency vibration tests and railroad spectrum random vibration tests were conducted to verify the power generation capacity of the prototype. Experimental results show that the device has a broadband response and the railroad vibration test proves the effectiveness of harvester in the application scenario for powering the rail-side sensors. The range of maximum output voltage, power and corresponding frequency in sweeping-frequency vibration tests with the amplitude of 1 to 10 mm and frequency of 5 to 50 Hz are 1.5 to 4.5 V; 1.80 to 17.0 mW and 9.7 to 30.8 Hz. The maximum output voltage and power are 1.33 V and 1.47 mW based on the measured railroad spectrum. Finally, a retrospective review in the efficiency, effectiveness and volume figure of merit is conducted to evaluate the performance of MLEH, indicating a high power density of the proposed harvester.

As the key sensor of Electric Vehicle Permanent Magnet Synchronous Machine (PMSM) drive system, the current sensor usually occurs gain and offset fault, and the fault will directly affect the performance and stability of PMSM drive system. Therefore, the inverter nonlinearity dead-time compensation method is first investigated allowing for parasitic capacitor effect, and the proposed method are compared with the traditional dead-time compensation method, moreover, the technical advantages are all validated via experimental results, the fault diagnosis accuracy problem of current sensor caused by the inverter nonlinearity induced offset voltage between inverter reference and output voltage are cleverly solved. Then, the differential algebraic based fault diagnosis method and adaptive fault-tolerant control strategy are presented. An integrated solution of PMSM drive system is innovatively proposed integrated the proposed dead-time compensation method with proposed differential algebraic based fault diagnosis method and adaptive fault-tolerant control strategy together. At last, system simulation and dSPACE based experimental test are implemented to validate the feasibility and effectiveness of proposed integrated solution of PMSM drive system and the conclusions are shown.

This paper investigates the distributed fault detection problem for linear discrete time-varying heterogeneous multi-agent systems under relative output information. Due to the lack of absolute outputs, an augmented model is built by stacking all local relative output information. Then, the fault detection problem consisting of residual-generation and residual-evaluation is handled using the H ∞ filtering framework. The residual-generation problem is actually a minimization problem of an indefinite quadratic form, and the Krein space-Kalman filtering theory is applied, which results in a low computational burden despite the time-varying characteristic. Using the Krein space theory, a necessary and sufficient condition for the minimum is derived, and a residual-generation algorithm is developed. Further, a residual-evaluation mechanism is designed by constructing an evaluation function and detecting faults by comparing it with a threshold. Finally, two illustrative examples are given to demonstrate the effectiveness of the proposed fault detection approach.

To overcome the influence of multiple operating conditions for fault detection, this paper proposes a method to detect fault for the suspension system of maglev trains. Firstly, the complex operating condition of the maglev train is divided into some simple conditions, and the operating samples are extracted through the time window in the same simple operating condition. Secondly, the features of the extracted samples are extracted by the Fast Walsh-Hadamard transform, and the noise is removed by the median filtering. Thirdly, after adopting principal component analysis to reduce the dimensionality of the feature, the mean of the feature is calculated and applied to calculate the Euclidean distance from the features of the new data obtained by the same processing in each time. Fourthly, the new Euclidean distance obeying the Gaussian distribution is obtained through the Box-Cox transformation. Finally, the fault threshold in each simple operating condition is established through the characteristics of Gaussian distribution. Taking a certain operating condition as an example, the method is compared with the three similar methods. The results reveal that the proposed method can detect the faults timely and has a low false alarm rate. Moreover, the proposed sub-healthy data are within an acceptable range. Based on these results, the proposed method can be applied in fault detection of the maglev train, providing a certain basis for fault diagnosis. INDEX TERMS Multiple operating conditions, principal component analysis, Euclidean distance, Box-Cox transformation, fault detection ZHIQIANG LONG. received the B.S. degree in automation from the