Traditional fault diagnosis methods often suffer from performance degradation under new working conditions due to distribution shifts between the source and target domains. To bridge this gap in cross-domain fault diagnosis (CDFD), the domain adaptation (DA) technique leverages transfer learning to align feature distributions, which facilitates knowledge transfer from labeled source domains to unlabeled target domains. Although existing studies on DA have demonstrated efficacy, they still face significant challenges due to abrupt domain shifts and insufficient feature discrimination. To overcome these problems, this study proposes a dynamic evolution mechanism to construct a sequence of hybrid domains that gradually evolves from the source to the target domain. This strategy establishes a smooth transition path to mitigate abrupt domain shifts. Additionally, a dual-path feature extraction structure empowered by wavelet packet transform (WPT) is introduced. This structure decomposes input signals into high-frequency and low-frequency components to enhance discriminative feature representation. The experimental results on rolling bearing and gearbox datasets demonstrate the effectiveness and generalization performance of the proposed method.

Under complex dynamic conditions, noise often masks minor faults in electric vehicle lithium-ion batteries. Motivated by this challenge, a fault diagnosis framework based on Long Short-Term Memory networks and Piecewise Asymmetric Denoising is proposed. This framework is distinctively denoted as LSTM-SAD. The method utilizes LSTM to capture the nonlinear dynamics of voltage. It decouples the influence of SOC and load current on terminal voltage. This process produces detrended residuals. A piecewise asymmetric denoising strategy is adopted. The training stage constructs a baseline with a high signal-to-noise ratio using median filtering. An adaptive threshold is set based on compressed quantiles. The online detection stage retains the original noise form. This approach preserves transient features when faults occur. Semi-physical experimental results show that this framework exhibits robustness under the Urban Dynamometer Driving Schedule (UDDS) condition. The method detects distinct faults in the early stage of discharge. It also identifies minor faults in the middle and later stages. The system generates no false alarms under normal operating conditions. This resolves the contradiction between sensitivity and reliability found in traditional threshold methods.

Iterative learning control applies to applications in which the same finite-duration task is repeated, with each instance termed a trial. The objective is to track a specified reference trajectory over a finite duration, termed the trial length. In some applications, such as multi-agent systems, tracking at each instant or point over the trial length is not required; only at selected points is it required, known as point-to-point iterative learning control. This article develops a new point-to-point design in which the points requiring tracking vary from trial to trial, and the solution minimizes energy, which is relevant to systems with a limited power budget. Also, an algorithm is developed to improve computational efficiency by sharing the burden among the agents forming the system. A numerical case study highlights the benefits of the new design.

Within object detection in remote sensing, targets often appear at extremely small scales, frequently resulting in missed detections or false positives due to the limited visual information and complicated background interference. To solve these issues, this study proposes a novel framework named Edge Feature Information Enhancement-You Only Look Once (EFIE-YOLO). First, a Cross-Scale and Detail-Enhanced Bidirectional Detection (CSDEBD) head is designed to enhance detail perception and feature discrimination. In addition, the integration of Dynamic Snake Convolution (DSConv) enables the network to adaptively focus on elongated and curved features, improving the extraction of edge information for small objects. Second, the Multi-Scale Edge Information Select (MEIS) module is employed in the C3k2 module, which consists of an Edge Enhancement Module (EEM) and a Dual-Domain Selection Module (DSM). EEM explicitly extracts multi-scale edge features, while DSM selectively emphasizes spatially and frequency-salient regions through spatial and frequency selection mechanisms, thereby filtering out irrelevant information and enhancing feature representation. Third, a C2TSSA module is incorporated to capture global statistical dependencies across tokens, which replaces standard self-attention with a Token Statistics Self-Attention (TSSA) mechanism to enable efficient long-range modeling at linear computational complexity. Compared to the baseline YOLO11-s, experimental results on the DOTA dataset demonstrate that EFIE-YOLO achieves 4.1% and 3.4% improvements in mAP@50 and mAP@50:95, while maintaining real-time inference speed and reducing model parameters and FLOPs.

This paper develops new results on data-driven iterative learning control for nonlinear batch processes. The dynamic linearization approach is used to obtain linearized local dynamical models utilizing only the collected process input and output data. As a result, no dynamic structure of the nonlinear model is required for the control design. Additionally, the design problem is formulated within the repetitive process framework, which simplifies the design procedure, facilitates the integrated synthesis of feedback and learning controllers, and aids in the adjustment of control parameters. The convergence of the new data-driven control method is demonstrated by the stability of the resulting repetitive process, which ensures that the tracking error decreases along both the time and iteration (batch) axes. Stability properties can be effectively checked using linear matrix inequality techniques. A numerical example is included to highlight the application of the new results.

The detection of small objects in remote sensing imagery remains a formidable challenge due to their minimal pixel occupancy, blurred structural boundaries, and susceptibility to environmental interference. To solve these problems, this paper proposes a novel network architecture named multi-dimensional information feature fusion-you only look once (MIFF-YOLO), which integrates several specialized modules. To address the challenge of small objects being obscured by complex environmental factors, we propose a multidimensional information fusion (MIF) module for the neck network, which leverages a 3D convolution and a full-domain transformer (FDT) to create cross scale dependencies and integrate global contextual information with local details. For the purpose of preserving the spatial and edge information of small objects, an efficient front end module (EFEM) is embedded into the C3k2 architecture. The EFEM module employs a parallel, learnable dual-path architecture that collaboratively integrates a Sobel convolution stream for explicit edge detection and a spatial information stream max-pooling for detail preservation, enabling simultaneous extraction of structural boundaries and contextual textures. These complementary features undergo an adaptive fusion via omni-dimensional dynamic convolution (ODConv), thereby enriching the capabilities of the feature representation. In order to address the loss of critical details in small object features during enlargement, dynamic upconvolution block (DUB) is introduced to replace standard upsampling module. Adaptive feature sampling is achieved through content-aware dynamic offsets, mitigating detail loss during resolution recovery. Compared with the original baseline algorithm, the improved network achieved a 3.7% improvement on mAP@50 and a 3.9% improvement on mAP@50:95, with the FPS reaching 120 on the DOTA dataset. This shows that the improved algorithm effectively enhances small object detection performance in remote sensing images while maintaining excellent real-time detection efficiency.

Existing fault estimation methods for Markov jump linear systems, particularly the interacting multiple model approaches, heavily rely on pre-defined fault model banks. This reliance renders them ineffective against faults with unknown or time-varying dynamics. Conversely, emerging Bayesian estimation approaches are typically confined to single-mode systems and fail to accommodate stochastic mode switching. To bridge this gap, this paper proposes a novel Bayesian framework for the real-time joint estimation of system states, hidden modes, and mode-dependent faults without requiring prior fault dynamic knowledge. In particular, an adaptive fault update mechanism based on innovation moments is introduced to address the trend discontinuity of fault signals caused by mode transitions, a challenge not addressed by standard recursive Bayesian estimators. Results from a numerical simulation example and a fermentation process simulation case study show improved robustness and estimation accuracy over traditional methods. The adaptive fault update also tracks abrupt fault changes while reducing fluctuations when the fault signal varies slowly.

To deal with unknown non-repetitive disturbances, a disturbance observer-enhanced model-free adaptive iterative learning control (MFAILC) scheme is proposed in this article for nonlinear batch processes. By utilizing the iterative dynamic linearization (IDL) approach, an equivalent linear data model is established in the iteration domain for the purpose of handling the nonlinear process characteristics. A partial-form disturbance observer (PDO) is developed under the IDL framework based on a conceptual nonlinear disturbance-observer representation, and an adaptive updating algorithm is introduced to update the observer gain vector. Then, a PDO-based MFAILC scheme is developed using the estimated disturbance information to compensate for non-repetitive disturbances. The convergence properties of the proposed scheme are analyzed via the contraction mapping principle. The effectiveness of the proposed data-driven control scheme is demonstrated through a numerical example and a heat-exchanger simulation. The results show that the proposed PDO-MFAILC scheme reduces the sum of root mean square errors by approximately 22% compared with the conventional MFAILC method.

An inherent assumption of perfect tracking in iterative learning control (ILC) is that there exists an ILC input such that the generated output can track the desired trajectory reference. This assumption may fail in practice, which gives rise to desired but untrackable tasks. This paper gives an end-to-end ILC design for repetitive untrackable tasks in closed-loop systems. The reference input is trial-to-trial updated together with the ILC feedforward input based on the measurement data. This two-player behavior of the closed-loop ILC system is investigated from a cooperative game perspective. A sufficient condition for the two-player end-to-end ILC to have a lower cost than the one-player norm optimal ILC (NOILC) is discovered. Finally, a numerical example is given to verify the effectiveness of the developed method.

In rotating machinery fault diagnosis, domain adaptation performance is frequently hindered by class-incomplete training data and distribution shifts between operating conditions–scenarios under which conventional methods tend to breakdown. To address this, we introduce a synergistic two-stage framework for multi-source domain adaptation. First, to resolve the critical absence of fault classes and enhance data diversity, the Incomplete Class Sample Completion (ICSC) framework synthesizes high-fidelity pseudo-samples for the missing classes. Subsequently, the Prototype-aware Class Conditional Adversarial Network (PCCAN) performs multi-granularity feature alignment, using global, class-conditional, and prototype-based constraints to enforce intra-class compactness and inter-class separability. The synergy between these class-completion and feature-alignment mechanisms enhances cross-domain recognition accuracy. The method’s efficacy is validated across the JNU, BJTU, and SDUST datasets. The experimental results confirm the framework’s effectiveness in handling cross-domain fault diagnosis tasks under class-incomplete conditions.

This paper presents a novel high-order error-based data-driven adaptive iterative learning control (HOE-DDAILC) strategy for nonlinear nonaffine systems. Using iterative dynamic linearization (IDL), the nonlinear system is first reformulated into an iterative linear data model (iLDM), enabling data-driven control design. The control learning law is obtained by minimizing a performance index including high-order error terms, ensuring asymptotic tracking of the reference trajectory. In addition, for systems with input constraints, a constrained HOE-DDAILC is developed via the combination of an unconstrained solution and interval projection, which guarantees feasible control inputs while preserving tracking accuracy. Simulation results on nonlinear systems validate the effectiveness and superiority of the proposed methods.

Iterative learning control (ILC) can significantly reduce the tracking error between the repetitive reference trajectory and the output by refining the ILC input with plenty of trials. However, re-learning is usually necessary in the presence of trajectory switching. To address this issue, this paper develops an experience transfer-based ILC method for nonaffine nonlinear systems by employing the radial basis function (RBF) neural network. First, a data-driven ILC algorithm that integrates feedback control is designed to acquire high-precision control performance of a nominal trajectory, which serves as the offline priors. Then, an RBF network is used to project the known experience information from the finite time-indexed domain into a state-dependent feature space for obtaining an equivalent controller. The developed method can extract the generalized inverse dynamics of the considered system by tracking on the same trajectory repetitively, which can be used for a new trajectory without re-iteration. Theoretical analysis is given, and a numerical case study demonstrates the effectiveness of the ILC design.

Iterative learning control (ILC) is an intelligent control methodology for tackling iteration-invariant exogenous inputs. It is of great significance to develop its extrapolation for more general repetitive tasks with mutual similarity, e.g., tasks with different time scales. In practice, discrete-time ILC with sampling behavior for time-scale-varying tasks suffers from the failure of perfect corresponding learning and environment-dependent iteration-varying disturbances. This paper develops a novel direct data-based ILC algorithm using off-policy Q-learning for tasks with varying time scales, enabling the robust learning of an optimal ILC policy from experimental input/output (I/O) data. From a two-player zero-sum game perspective, the iteration-varying disturbance generated from the varying time scales of repetitive tasks is tackled quantitatively with a preset disturbance attenuation level. Further, to emphasize the importance of theoretical guarantees of reinforcement learning (RL)-based ILC designs, the data efficiency of the developed algorithm is enhanced based on Willems' Fundamental Lemma, and a rigorous convergence analysis is given. The simulation model of an F-16 aircraft autopilot is employed to show the effectiveness of the developed approach. (c) 2025 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Due to the presence of increasing computation demands in telematics, RSUs are proposed to play a critical role in Vehicular Edge Computing (VEC). However, how to simultaneously improve the communication quality and reduce the service latency becomes a severe challenge due to the resource shortage. To tackle these issues, we explore how to utilize Unmanned Aerial Vehicles (UAVs) in VEC to facilitate the task offloading performance, i.e., the latency of the service and the stability of the task queues. A Genetic Algorithm (GA)-based Lyapunov optimization framework is proposed for task scheduling optimization. It aims to minimize system cost and stabilize edge server task queues by obtaining the optimal decision. The proposed algorithm optimizes the Lyapunov drift plus penalty function in each time slot. Finally, simulations verify that proposed LyGA scheme is able to achieve the trade-off between minimizing the system cost and maintaining queue stability compared with the benchmark methods.

During real-time production in industrial Internet of Things systems, equipment changes its operating speed due to changing operating conditions. And dynamic speed changes of rotating machinery under fluctuating workloads often lead to domain changes of vibration signals, which will directly lead to degradation of fault diagnostic model performance. Furthermore, the acquisition of data from multiple domains in real industrial scenarios is challenging due to the expense of collecting data from all possible working conditions. Consequently, applying diagnostic models trained using a single-source domain directly to an unknown target domain is a very challenging single domain generalization problem. Therefore, a generic single-source domain generalization framework via wavelet packet augmentation (WPA) and pseudo-domain generation (PDG) for fault diagnosis under unknown operating conditions is proposed in this article. PDG involves augmenting single-source domain by integrating data generation model, thereby enhancing prediction accuracy. Furthermore, a WPA method is proposed. Initially, the original signal is decomposed to obtain high- and low-frequency information. Subsequently, the high- and low-frequency information within the batch are linearly interpolated, respectively. Consequently, the interpolated high- and low-frequency information is then reconstructed to yield enhanced samples. The experimental results on four datasets show that the proposed framework can effectively improve the robustness of the generalization ability of fault diagnosis under unknown operating environments.

This paper proposes a predictor based data-driven anti-disturbance control scheme for nonlinear nonaffine systems with unknown dynamics, by only using the measured system input and output data. A partial-form dynamic linearization (PFDL) data model with disturbance predictor is firstly established for the data-driven control system design. Then, an anti-disturbance control law is developed by solving a constrained optimization problem, where the unknown pseudo-gradient (PG) vector at the current and future time instants are, respectively, estimated by a projection algorithm and another autoregressive prediction algorithm. Meanwhile, the unknown residual term at the current and future time instants are approximated by a high-order extended state observer (HOESO). The bounded convergence of the PG estimation, HOESO, and output tracking error is analyzed with proof. Finally, a case study is adopted to validate the effectiveness and advantage of the proposed method over the recently developed ones.

The utilization of transfer learning strategies to solve cross-domain fault diagnosis problems has achieved significant results. However, most existing multi-source domain generalization fault diagnosis methods use a single classifier or introduce auxiliary classifiers, focusing on learning domain-invariant features or global feature distribution matching. Furthermore, since the data distributions of different source domains may be significantly different, this may lose the data distribution information specific to each source domain. In addition, how to reduce the variation in risk between samples within the same domain training is also a challenging issue. Finally, it is also crucial to balance the predictive outputs of multiple classifiers to adapt them to the data distribution of the target domain. Based on the above challenges, this paper proposes a multi-domain weakly decoupled domain generalization network for fault diagnosis under unknown operating conditions. Feature weakly decoupled mechanism is achieved by employing multiple classifiers and incorporating the variance of samples within the same sample domain as a penalty term. This reduces the model's sensitivity to changes in the extreme distribution of samples within the domain. Classifier weakly decoupled mechanism, on the other hand, reduces the inter-domain risk variance by minimizing the loss of variance in the predicted output of the source domain classifiers. This improves the robustness of the model to inter-domain distributional changes and covariate changes. Experimental results on three datasets validate the effectiveness and general applicability of the proposed approach.

Iterative learning control (ILC) is typically applied in practice combined with a feedback controller for time-domain stability. In this closed-loop design with actuator constraints, existing constrained ILC designs suffer from determining the exact input constraint on the ILC controller. This issue brings in an important gap between the existing constrained ILC designs and their real-world applications. This paper gives a systematic consideration of the input constraint problem in the closed-loop ILC design with actuator saturation. A constraint-aware ILC is developed to autonomously determine the constraint on the feedforward controller. The convergence of the constrained ILC process is proved under the framework of alternating projection. Finally, the effectiveness of the developed method is verified on a numerical simulation.

A robust indirect-type iterative learning control scheme is developed for batch processes with state delays, time-varying uncertainties, and disturbances. In contrast to direct-type designs, the new scheme consists of two control loops, each of which can be designed independently. In the inner loop, a control law that is the sum of a generalised extended state observer-based state feedback and proportional plus integral control action acting on an error signal is designed for stability and robustness. The outer loop is designed to update the set-point command for the resulting closed-loop system. Finally, the stability theory for linear repetitive processes ensures robust tracking error convergence for the resulting dynamics in the presence of non-repetitive uncertainties and disturbances. Two numerical examples demonstrate the attributes of the new design.