Tiantian Liang

Monocular depth estimation (MDE) is a fundamental task for 3D scene understanding and provides an efficient perception solution for autonomous driving, robotic navigation, augmented reality, and other intelligent systems. Despite recent progress in supervised models and large-scale depth foundation models, accurate MDE in scenes containing potentially movable object categories remains challenging, since such foreground categories often introduce feature entanglement with static background regions and blurred object boundaries. In this paper, ‘dynamic’ refers to potentially dynamic semantic-category regions, such as vehicles and pedestrians, derived from semantic labels or pseudo-labels; it does not denote explicit frame-level motion estimated from optical flow, tracking, or multi-frame input. To address these challenges, we propose semantic-guided progressive network (SGP-Net), a semantically guided progressive refinement framework for MDE based on multi-task learning. SGP-Net is built upon DCDepth and extends it with semantic-aware branches and dynamic-category constraints. SGP-Net consists of two key components. First, a semantically-guided pyramid feature fusion module uses pixel-level semantic priors to reweight multi-scale depth features, thereby preserving object contours and enhancing geometric perception in dynamic-category regions. Second, a dynamically weighted multi-constraint loss with a phased weight scheduling strategy is designed to balance scale-invariant and edge-aware supervision during training while alleviating negative transfer between tasks. Experiments on KITTI , NYU-Depth-v2 , and TOFDC show that SGP-Net improves key relative-error and accuracy metrics over the DCDepth baseline and remains competitive with recent methods. On KITTI dynamic-category pixels, SGP-Net reduces AbsRel from 0.0477 to 0.0448 and improves δ 1 from 0.9786 to 0.9862. Qualitative results and ablation studies further verify the effectiveness of semantic-guided refinement and dynamic-category supervision in regions belonging to potentially movable categories.

The course Railway Signal Fundamental Equipment is a core professional course in rail transit–related engineering programs. Due to the large variety of signal equipment, complex operating mechanisms, and diverse fault modes, traditional assessment methods face significant challenges in accurately evaluating students’ learning outcomes and engineering competencies. In particular, conventional score-based evaluation approaches are insufficient to reflect students’ mastery of knowledge modules and fault analysis abilities. To address these issues, this paper proposes an adaptive assessment–driven teaching reform framework based on XGBoost, Item Response Theory (IRT), and Genetic Algorithm (GA). First, a multidimensional course objective and competency evaluation system is established, and relevant teaching and assessment data are collected. Then, an XGBoost-based learner model is constructed to estimate students’ ability profiles by learning the nonlinear relationships between assessment performance features and latent learning abilities. IRT is employed to model assessment item characteristics, while GA is introduced to optimize adaptive assessment strategies through dynamic item selection. Based on the assessment results, targeted teaching improvement strategies are further proposed. Experimental results indicate that the proposed framework can evaluate the attainment of course objectives in a more accurate, reasonable, and scientific manner, effectively overcoming the limitations of traditional assessment methods. The study provides practical support for data-driven teaching reform of railway signal courses and offers a feasible solution for intelligent assessment in engineering education.

With China's railway system entering a new stage of high-quality development, the electrical and control system has become a core system of electric multiple units (EMUs). Mastery of this system is essential for students majoring in Electrical Engineering and Automation in rail transit institutions to acquire cutting-edge industry knowledge and key technologies. The course titled “Electrical and Control Systems of EMUs” is a distinctive offering in the Electrical Engineering and Automation program at Dalian Jiaotong University. Calculating course achievement based on quality standards and knowledge frameworks is a crucial part of evaluating the effectiveness of teaching reforms and the quality of talent cultivation. This paper takes this course as the research subject. Firstly, quality standards for the course are established, and relevant teaching data are collected to provide a foundation for subsequent research. Secondly, an Improved Tornado Optimization with Coriolis force (ITOC) algorithm is proposed and then is used to enhance the back propagation neural network (BPNN), leading to the proposal of an intelligent course achievement evaluation method. Subsequently, suggestions for teaching improvement are put forward. The study demonstrates that the enhanced neural network technique can evaluate the attainment of course objectives reasonably, accurately, and scientifically, thereby contributing to the high-quality development of curricula in electrical engineering and automation.

This paper proposes a neural network-based optimized robust nonlinear Kalman filter for concurrent fault estimation of high-speed trains using descriptor systems. Considering the displacement and speed time delays and multi-source disturbances acting on every car, and regarding the concurrent actuator and sensor faults as the auxiliary variables of the train, a time-delay nonlinear descriptor system is established. Fuzzy and fuzzy inference layers are added to the hyper basis function neural network; meanwhile, an accelerated gradient algorithm is used for optimizing the basis function of the network. These strategies derive an improved fuzzy hyper basis function neural network to achieve a higher nonlinear multi-source disturbance approximation accuracy, and robust upper bounds are proposed to enhance the filtering accuracy of descriptor systems. Furthermore, a fusion intelligent optimization algorithm using a circle chaotic mapping method to improve the population initialization is proposed to better estimate the unknown system noise. These measures implement better Kalman filtering of descriptor systems to achieve better concurrent fault estimation, even under unknown noise. Simulation results show that, compared with the improved radial basis function neural network-based optimized robust nonlinear Kalman filter, using the proposed method, the displacement, speed, and fault estimation errors are comprehensively decreased, depending on more accurate multi-source disturbance and noise estimations.

Bearings are critical components of mechanical equipment, and predicting their remaining useful life (RUL) is important in industry. This paper proposes a RUL prediction method based on the assessment of a bearing’s health status. Features from the time, frequency, and time–frequency domains of the bearing’s vibration signal are extracted to construct a feature set. A multibranch encoder and restricted Boltzmann machine are used to improve the stacked autoencoder to reduce dimensionality. Local weights and health-sample means are introduced into the Mahalanobis distance to improve the health index. Subsequently, the weighted convolutional Euclidean distance serves as the distance metric in K -means clustering to achieve a more accurate health status assessment and provide historical data for RUL prediction. An improved self-attention (ISA) mechanism is proposed by incorporating depthwise separable convolutions and residual-like connections into self-attention mechanisms, enhancing the global and local dependencies of the temporal convolutional network (TCN). Thus, a more accurate RUL prediction is achieved. Comparative and ablation experiments confirm that the proposed ISA-TCN achieves superior predictive accuracy. Generalization experiments further demonstrate its strong adaptability, while anti-noise experiments demonstrate its strong robustness to uncertainties. Finally, experiments on multi-output RUL predictions validate the model’s effectiveness. This approach offers valuable insights for RUL prediction of rotating machinery under real-world scenarios involving variable operating conditions, noise interference, and multi-device environments.

In this paper, a spider monkey optimisation (SMO) algorithm is utilised to identify the parameters of the permanent magnet synchronous motor (PMSM), considering the parameters vary during the motor operation, which affects the sensorless control (SC) performance of the motor. An improved sliding mode observer (SOBS) is proposed for estimating the position and speed of the rotor. First, the SMO algorithm is used to identify the parameters of PMSM. Then, based on the identification results, an improved SOBS is proposed by a piecewise Sigmoid function. Furthermore, the stator position and speed are estimated by extended state observer (ESO) and phase-locked loop (PLL). Finally, a comparison simulation scenario is provided to demonstrate the efficacy of the suggested approach.