The integration path and practical effectiveness of artificial intelligence (AI) with maglev transportation technology were explored from the perspective of systems engineering thinking to tackle the complex control and management challenges arising from the coupling of multiple subsystems in maglev transportation and promote its upgrading towards intelligence and efficiency. Firstly, at the theoretical level, the core characteristics of different technical routes of maglev transportation were analyzed, and the technical logic of AI adapting to the core contradiction of "dynamic uncertainty and high precision control" in maglev transportation was clarified. Meanwhile, the internal mechanism of AI realizing subsystem collaborative optimization under the guidance of systems engineering thinking was explained, and the systems engineering value of AI in constructing an intelligent closed-loop system of "perception, decision, control, optimization" was defined. Secondly, in five core subsystems including suspension guidance, traction transmission, the power supply system, signal communication, and intelligent operation and maintenance scheduling, the technical breakthroughs and scene adaptability of AI algorithms such as the BP neural network, LSTM, and reinforcement learning were elaborated. Additionally, the improvement effect of AI technology on suspension accuracy, traction efficiency, power supply stability, and signal reliability was quantitatively analyzed, and the full-dimensional empowering effect of AI on maglev systems was verified. Thirdly, an AI empowerment path of "top-level data specifications, software and hardware iterative verification, compound talent cultivation, and cross-domain collaborative innovation" was proposed, and the technical development context of medium-low speed, high-speed and ultra-high speed vacuum pipeline maglev was sorted out, with the engineering practice results under different scenarios summarized. Finally, existing challenges were analyzed, such as data silos, insufficient reliability in extreme operating conditions, disconnections between industry, academia, research and application, shallow integration of AI with emerging technologies, and lack of cross-industry collaborative mechanisms. The future development direction was proposed from five aspects of full-domain data fusion, robust model R&D, deepening of collaborative innovation mechanisms, integration of emerging technologies, and intelligent collaboration for cross-industry multimodal transportation. The results indicate that the deep integration of AI and maglev transportation technology under the guidance of systems engineering thinking is the core support for solving the multi-subsystem coupling difficulties of maglev systems and realizing the leap from high speed to intelligence, and its subsystem collaborative optimization mechanism provides a theoretical paradigm and practical guidance for the intelligent development of maglev transit.

A full-state fault detection algorithm incorporating hybrid decision-making and reset logic was proposed to address the problem that the probes of maglev train air gap sensors are prone to failure in special operating conditions including strong magnetic fields, vibration, and rail joint crossing. By adopting fault identification, signal switching, and fault-tolerant strategies, system stability in special operating conditions was ensured. Firstly, the suspension control system based on a joint structure was taken as the research object to carry out system dynamics modeling, and the coupling relationship and stability between the left and right electromagnets were analyzed. Secondly, a new method capable of rapidly identifying faults and implementing timely fault tolerance was proposed by integrating a full-state feedback control algorithm based on a signal selection strategy, sequential probability ratio test (SPRT), and residual chi-square detection. Simulation results show that in cases where some sensor probes fail, the initial fault detection time is delayed by 0.02~0.2 s after fault occurrence. The missed detection rate during the fault duration is approximately 5%. For sensor probe faults of varying severity degrees, the proposed fault-tolerant control strategy reduces the average fluctuation of the left electromagnet to 0.27 mm and the right electromagnet to 0.13 mm. The SPRT algorithm advances the response time to fault termination by about 0.5 s, enabling rapid detection and recovery. In rail gap crossing conditions, faults can be identified instantly, and the fluctuation of the suspension gap is significantly suppressed by approximately 89%~91%. The proposed method significantly enhances the robustness and fault tolerance of the suspension control system against gap sensor probe failures, ensuring the dynamic stability and ride comfort of maglev trains under long-term service.

In response to inaccurate safety assessments in high-speed maglev-bridge systems due to the neglected cumulative train operation effect of the traditional methods, a probabilistic safety assessment method based on the tail distribution of driving safety indicators was developed. A high-speed maglev-track bridge coupled stochastic dynamic model was established using the square power good lattice point method, multibody dynamics, and the modal superposition method. The probabilistic characteristic parameters of driving safety indicators (mean and standard deviation) were determined using the equal probability method and extreme-value theory. An explicit mapping relationship between the probabilistic characteristic parameters and the extreme-value cumulative distribution function (CDF) was constructed. According to a predefined confidence level, the extreme-value estimate of the safety indicator was obtained. By comparing this with the current normative threshold, a driving safety assessment method based on the probabilistic tail distribution characteristics was proposed. Finally, taking the case of a high-speed maglev train passing over a 24.768 m track beam, a driving safety assessment was conducted under different operating conditions such as varying train speeds. The results indicate that by employing the square power good lattice point method to construct a highly uniform set of track irregularity samples, the computational requirements can be met with only 131 samples. Compared to the 5 000 samples required by the Monte Carlo method, this significantly improves both the computational efficiency and accuracy of the probabilistic characteristic parameters for driving safety indicators. Furthermore, the extreme-value distribution proposed can effectively characterize the tail distribution characteristics of driving safety indicators. Its distribution curve well fits the maximum value samples, and the estimated extreme values can fully envelope the results generated by the Monte Carlo method. The number of failed samples evaluated based on the tail distribution model is less than 10, whereas it is as high as 2 280 under the normal distribution assumption. The method effectively avoids the limitation of traditional methods in failing to account for the cumulative effect of train operations, thereby providing a scientific basis for the safety assessment of high-speed maglev-bridge systems.

To address such issues as uncertain internal and external disturbances, open-loop instability, and the difficulty of controller parameter tuning in the electromagnetic suspension systems of maglev trains, a suspension gap control method was proposed active disturbance rejection control (ADRC) based on particle swarm optimization. The internal disturbances (e.g., model linearization errors, parameter perturbations) and external disturbances (e.g., track irregularities, sudden load changes) were treated as a unified total disturbance. An extended state observer was employed to estimate and compensate for them in real time. While the dynamic linearization of the system was achieved, a novel integrated discrete time-optimal control function was designed, and the corresponding tracking differentiator was established to enhance signal tracking performance. In response to the difficulty in manually obtaining optimal solutions for key parameters such as observer bandwidth and controller gain in the active disturbance rejection controller, the particle swarm optimization algorithm was introduced. By constructing a multi-objective fitness function incorporating the integral of absolute error and control energy consumption, the global optimization was carried out for controller parameters to balance the fast tracking capability and the noise suppression ability of the system. The research results demonstrate that under conditions of suspension startup, sudden load changes, and random track noise, the proposed method exhibits significant performance advantages over traditional manually tuned ADRC. The settling time during the startup process is significantly shortened with no overshoot. Under sudden applied load disturbances, the cumulative absolute error of the suspension gap is reduced by 28.3%, and a steady state can be recovered more rapidly. Furthermore, in a noisy sensor environment, the control voltage curve remains smooth without high-frequency chattering. This method effectively resolves the conflict between the rapidity and disturbance rejection of the maglev system, and realizes the optimal tuning of control parameters, providing an effective technical solution for the stable suspension of maglev trains under complex operating conditions.

A scheme for detecting and evaluating vibration and noise data of maglev vehicles was proposed, and the self-developed portable on-board device was employed to collect vibration and noise data of maglev vehicles at low cost and high frequency. The attitude deviation of the inertial sensor coordinate system was corrected, and the vibration characteristics of maglev vehicles were analyzed. The running quality of maglev vehicles was evaluated by adopting the ISO 2631-1:1997 weighted acceleration, Sperling ride-comfort index, and A-weighted sound level, and the speed and mileage information from the measured data were autonomously estimated. The results show that the vertical correction angles of the on-board device for the Shanghai and Changsha maglev systems are 0.08° and 0.95°, respectively, and the horizontal correction angles are 0.82° and 3.45°, respectively. The dominant frequencies of longitudinal and lateral car-body vibrations for both systems are within 10 Hz, while those of vertical car-body vibration are within 40 Hz for the Shanghai maglev system and within 30 Hz for the Changsha maglev system. The sampling frequency meets the detection requirements. For the Shanghai maglev system, the maximum ISO 2631-1:1997 weighted acceleration is 0.9 m·s^-2, with exceedance ratios of 95.5% for level Ⅰ and 71.1% for level Ⅱ. The maximum Sperling ride-comfort indexes for lateral and vertical directions are 2.64 and 2.73, with level Ⅰ exceedance ratios of 0.96% and 69.8% respectively. The maximum A-weighted sound level is 73.8 dB(A), and the ratio exceeding 70 dB(A) is approximately 78.2%, indicating that the running quality of the vehicles in the Shanghai maglev system is good. For the Changsha maglev system, the maximum ISO 2631-1:1997 weighted acceleration, Sperling indexes for the lateral and vertical directions are 0.26 m·s^-2, 2.32, and 2.15, respectively, all below the level Ⅰ limits, which indicates that the running quality of maglev vehicles in the Changsha maglev system is excellent. Visualizing various evaluation indicators in the longitude-latitude coordinate system of the maglev line intuitively reflects the along line distribution of the maglev vehicles' running quality. This method can provide data support for the maglev track operation and maintenance department to precisely regulate the service status of maglev tracks at different mileage locations.

In response to low modeling accuracy and poor robustness in the suspension control systems of EMS-type maglev trains caused by high dimensionality, strong nonlinearity, and open-loop instability, a high-precision model identification and time series prediction method was proposed based on a long short-term memory (LSTM) improved by the Bayesian optimization (BO) algorithm. First, an experimental platform for a single-degree-of-freedom magnetic suspension system was built. A controller was used to track complex excitation signals, including sine, pulse, and white noise. Time series data of the control voltage and suspension gap covering the dynamic characteristics of the system were collected. The input and output sequences were reconstructed using the sliding window method. Second, an LSTM model was constructed to capture the time-dependent features of the suspension system. In response to the difficulty of manual parameter tuning in balancing efficiency and performance, the BO algorithm was introduced. A Gaussian process was selected as the probabilistic surrogate model, and the expected improvement was used as the acquisition function. The mean absolute error of the model on the validation set was set as the objective function. A global optimization was performed for the number of hidden layer neurons, initial learning rate, and L2 regularization coefficient of the LSTM model. Research results show that after BO, the model can accurately predict the key state changes of the suspension gap. The prediction error stabilizes within ±0.1 mm, and the error distribution conforms to a normal distribution, indicating that the structural deviation of the model is eliminated. Compared with the unoptimized LSTM model, the mean absolute error on the test set is reduced by 48.04%, and the root mean square error is decreased by 41.83%. This method effectively solves the difficulty of hyperparameter setting in suspension system modeling, and significantly improves the prediction accuracy and generalization ability of the model, providing a reliable model foundation for the design of high-precision suspension controllers and real-time condition monitoring of maglev trains.