Faults happen inevitably in traction systems and thus place the security of the whole high-speed train at risk. In order to improve the safety and reliability of high-speed trains, this paper deals with fault detection and diagnosis (FDD) problem for traction systems. Because of high sampling frequency of equipped sensors, FDD strategies in the supervision system of high-speed trains should be of enough high computation efficiency, which is a great bottleneck for artificial intelligencebased FDD methods. For the sake of reducing the computational load while maintaining the satisfactory diagnostic accuracy, an enhanced FDD architecture using the modified principal component analysis (PCA) and broad learning system (BLS) is developed in this paper. Based on the proposed data driven design whose core is to extract fault information, fast and accurate FDD can be achieved without requirements for mathematical models or control mechanism of high-speed trains. The effectiveness and feasibility of the proposed online design are illustrated on the traction control platform of high-speed trains.

Due to the advanced development of sensor technology, the data deluge has begun in the complex systems of high-speed trains (HSTs) and, therefore, hastens the popularity of data-driven research. Among these activities, data-driven detection and identification of faults have received considerable attention to ensure the safe and reliable operations of HST, especially the deep learning-based methods. Up to now, these deep learning-based methods are effective only for static systems. It, hence, motivates us to develop the data-driven fault identification (FI) method for traction systems in HST. In this study, we will develop an FI method via the collaborative deep learning method, where the first neural network is used for eliminating dynamic behaviors, and the second neural network is responsible for identifying the fault amplitude. By the use of the proposed neural networks with a deep architecture, the FI task can be achieved in a collaborative fashion. Its successful application on the traction systems of HST illustrates the effectiveness of collaborative deep learning on the one hand and opens an avenue on the data-driven FI methods using neural networks on the other hand.

As one of the critical components of high-speed trains, the running gears system directly affects the operation performance of the train. This paper proposes a state-degradation-oriented method for fault diagnosis of an actual running gears system based on the Wiener state degradation process and multi-sensor filtering. First of all, for the given measurements of the high-speed train, this paper considers the information acquisition and transfer characteristics of composite sensors, which establish a distributed topology for axle box bearing. Secondly, a distributed filtering is built based on the bilinear system model, and the gain parameters of the filter are designed to minimize the mean square error. For a better presentation of the degradation characteristics in actual operation, this paper constructs an improved nonlinear model. Finally, threshold is determined based on the Chebyshev’s inequality for a reliable fault diagnosis. Open datasets of rotating machinery bearings and the real measurements are utilized in the case studies to demonstrate the effectiveness of the proposed method. Results obtained in this paper are consistent with the actual situation, which validate the proposed methods.

The stable operation of the high-speed train running gear is one of the key components to ensure the safety and reliability of high-speed trains. This paper proposes a research method based on the hidden Markov model and analytic hierarchy process (HMM-AHP) high-speed train running gear fault prediction. The multi-observation sequence is used to make the hidden Markov model more accurately model the system under study to describe the internal and external characteristics of the system. In order to ensure the more accurate and reliable results of the fusion of multi-observation sequence data, the analytic hierarchy process method is used to reasonably assign the feature weighting factor to predict the online fault of the high-speed train running gear, and form an online fault diagnosis framework based on HMM-AHP model. At last, this paper takes the high-speed train running gear as the research object. The proposed HMM-AHP model is used to simulate the high-speed of trains running gear, and the validity and accuracy of the model are verified.

The past two decades have witnessed the development of fault detection (FD) methods in the field of high-speed trains. This trend is the consequence of the increasing requirement for safe and reliable running of mechanical systems in trains. Despite numerous reported methods, data-driven frameworks require further improvement in analyzing dependencies of multi-source variables on system states to enhance distributed fault detection performance in trains. In this paper, the joint observer-based distributed FD scheme for the dynamic process is investigated for traction control systems of high-speed trains. To be specific, this paper proposed a joint observer model via joint functions aiming at improving the analysis ability for multi-source signals. Then, the existence of the joint state and output observer is proved. Additionally, to enhance the detection ability of traction systems, the data-driven distributed FD method is proposed based on subspace identification and the average consensus method. Finally, numerical simulations and physical experiments were conducted to validate the efficacy of the proposed method. Comparative results demonstrate superior dynamic performance of the proposed method, achieving false alarm rates (FAR) and missed alarm rates (MAR) below 1% in high-speed train traction systems.

In high-speed train traction systems, intense electromagnetic interference and varying operating conditions give rise to significant colored noise, posing substantial challenges to fault detection. A weighted elimination mechanism is introduced to realize residual generation under colored noise in a manner equivalent to generalized least squares. Furthermore, a precision metric is incorporated into kernel generation to mitigate correlated noise, thereby improving fault detection robustness. The proposed method is organized into an offline learning stage and an online monitoring stage. In the offline stage, Wiener projection is used to remove input explainable components and construct conditional residuals. The Schur complement is then used to characterize the second-order statistics of these residuals. Ledoit--Wolf shrinkage is further adopted to obtain stable precision matrices and derive whitening operators, thereby addressing covariance ill-conditioning caused by line spectrum colored noise. Based on this metric, a kernel residual is constructed under the precision inner product by annihilating the past dependent term in the stacked data equation. In the online stage, the learned kernel is applied to sliding window data, and the resulting residual statistic is compared with a healthy threshold for fault decision. Finally, the effectiveness of the proposed method under strong colored noise and varying operating conditions is validated through numerical simulations and traction drive testbed experiments.