The state-of-charge (SOC) and state-of-energy (SOE) are critical parameters for lithium batteries, essential for extending battery life and enhancing system reliability. These parameters are highly correlated, yet there is limited research on their joint estimation. This paper proposes a joint deep learning framework, called Multi-channel Bidirectional Long Short-term Memory and Inverted Transformer (MLIT), for SOC and SOE prediction. MLIT utilizes an inverted transformer to separately extract the features of each physical quantity, allowing for better extraction of independent feature information and avoiding alignment issues caused by different sampling frequencies. Additionally, the unique encoder-decoder structure of MLIT enables the model to effectively integrate historical SOC and SOE data and learn the deep correlation between SOC and SOE. Experimental results demonstrate that MLIT achieves higher prediction accuracy compared to current mainstream models, and its joint estimation approach results in more stable prediction outcomes with smaller maximum errors.

Early detection of crop diseases is critical for safeguarding food security and improving agricultural productivity. In recent years, the integration of advanced imaging modalities—multispectral and hyperspectral sensors—with artificial intelligence (AI) has enabled unprecedented precision in early disease detection. This paper surveys state-of-the-art research published from 2020 to 2025 on crop disease prediction and detection using AI models trained on multispectral and hyperspectral images. We discuss data acquisition platforms (e.g., UAVs, satellites, ground-based systems), outline major AI architectures (including convolutional neural networks, capsule networks, and physics-informed generative adversarial networks), and highlight both promising results and remaining challenges. Future research directions to enhance early detection and management of crop diseases are proposed.

State-of-health (SOH), state-of-charge (SOC), and state-of-energy (SOE) co-estimation is vital for reliable operation and longevity of Li-ion batteries (LIBs). However, the intricately coupled changes occurring in multiple states across varying operating stages and differing timescales under cycle aging challenge flexible, accurate, and robust co-estimation. To tackle them, this article proposes a novel co-estimation method, leveraging a large-scale pretrained language model (PLM) empowered by multistate explicit-implicit prompt learning. Specifically, a state-wise contextual synthesizer is introduced to augment battery data into state-dependent explicit prompts, serving as situational directives for PLM to make flexible estimates over cycles and charge-discharge phases. A cross-state disentangling scheme is devised to capture coupling relationships among states and unique intricacies within states by structuring shared and individual implicit prompts. It allows PLM to perceive capacity decay for accuracy loss compensation under cycle aging. A multistate synergistic regulator is built to calibrate inter- and intrastate knowledge and interact them with PLM, strengthening co-estimation robustness to state dynamics at differing timescales. Experiments demonstrate that the proposed method yields accurate co-estimates of fluctuating SOH over the lifecycle, as well as variable SOC and SOE during charge-discharge phases, with root-mean-square error (RMSE) reductions exceeding 38%, 29%, and 29%, respectively, compared with conventional deep co-estimators.

Accurate joint estimation of state-of-charge (SOC) and state-of-health (SOH) is essential for maximizing the reliability and lifespan of Lithium-ion batteries. However, the complex coupled dynamics of SOC and SOH across diverse operational stages on differing timescales challenge the flexibility and robustness of joint estimation. Hence, this paper proposes a multi-task explicit-implicit prompt learning method that exploits a large-scale pre-trained language model (PLM) for joint estimation. A per-state task interpreter is introduced to elaborate battery data into explicit prompt texts, directing PLM to grasp measurement representations and task semantics for flexible joint state estimates across varying operational stages. A cross-state task adaptor is designed to refactor implicit prompt vectors into a base type that encodes inter-state coupling and a custom type that perceives state-specific intricacies, enhancing PLM adaptability for joint estimation. An adaptive gated integrator is constructed to discriminatively aggregate base and custom prompts, shaping calibrated multi-state knowledge. It further engages extensively with the PLM space to inject knowledge, strengthening joint-estimation robustness against coupled state changes on differing timescales. Experiments demonstrate that the proposed method delivers accurate joint estimates of the persistently changing SOC during charge-discharge processes and the fluctuating SOH over cycles, ensuring high flexibility and robust performance.

Fused deposition modeling (FDM) is a material-extrusion additive manufacturing technology that is widely used in rapid prototyping, complex product modeling, and functional part fabrication. However, process-parameter drift and environmental disturbances may induce underfilling, overfilling, warping, delamination, and other defects, thereby reducing part quality or interrupting the manufacturing process. Since FDM is characterized by point-wise extrusion and layer-by-layer deposition, layer-surface images naturally contain both spatial morphology and temporal evolution information. Existing image-based diagnostic methods often treat layer images as independent samples, and the selection of the image-stream length is still insufficiently supported by experimental evidence. Moreover, spatial compression in spatiotemporal neural networks may remove local defect information that is important for distinguishing similar process-parameter drifts. This study provides a deployment-oriented analysis of FDM image-stream diagnosis by systematically examining how layer-window length, spatial feature preservation, and strict data partitioning influence process-parameter drift recognition. To address these issues, this paper studies ConvLSTM-based FDM image-stream process-parameter drift diagnosis. Continuous region-of-interest image streams are constructed for one nominal condition and six process-parameter drift conditions. In this paper, the time step T denotes the number of consecutive layer-surface images, or, equivalently, the number of consecutive printed layers, contained in one diagnostic image stream. A ConvLSTM-Flatten baseline is first developed to preserve complete spatial feature maps and to evaluate the effect of different time-step lengths. Then, a ConvLSTM model with adaptive spatial pooling and temporal attention (ASP-TA) is constructed to analyze the influence of spatial pooling granularity and temporal feature fusion. The experiments show that the ConvLSTM-Flatten model achieves the highest average test accuracy of 0.7288 at T=9, whereas T=3 is identified as a practical optimal time step when test accuracy, image-frame computation, diagnosis latency, and convergence behavior are considered together. The paired trial-wise accuracy difference between T=9 and T=3 is small and not statistically significant over ten repeated trials. Thus, the diagnostic window corresponding to T=3 covers three consecutive deposited layers; after the initial window is available, stride-one stream construction allows the diagnosis to be updated with each newly acquired layer image. ASP-TA with a pooling size of eight consistently outperforms ASP-TA with a pooling size of four, but both are lower than the Flatten baseline, indicating that preserving sufficient spatial information is essential for distinguishing FDM process-parameter drift states. The results reveal the non-monotonic influence of time-step length and clarify the tradeoff between spatial feature preservation and model compactness in FDM image-stream process-parameter drift diagnosis.

Accurate and reliable capacity estimation is a critical factor in ensuring the safe and stable operation of lithium-ion batteries. However, most data-driven battery capacity estimation methods rely on data collected either during a complete charging process or within specific voltage ranges, which limits their applicability in real-world scenarios. Hence, this paper proposes a novel framework that utilizes only randomly sampled continuous fragments of charging data as input, without relying on specific data ranges. The framework comprises a multi-convolutional neural network (CNN) block, an attention-enhanced long short-term memory network, and a fully connected layer. The multi-CNN block consists of multiple depthwise and pointwise convolution layers, combined via gating mechanisms. By separately processing information extraction and feature learning for different variable types, the model can comprehensively capture both inter-variable temporal relationships and intra-variable temporal dependencies. In addition, the framework processes voltage data to obtain the rate of voltage change over time. The derived rate is then used as an input parameter, thereby enriching the representation of the input features. Experiments on the CALCE dataset demonstrate that the proposed framework achieves the greatest improvement on the best-performing battery cells, with the mean absolute error reduced by 6.3% and the root mean square error reduced by 4.8%. Overall, MCLA can effectively adapt to charging data in practical scenarios that are incomplete or irregular, thereby achieving accurate and reliable capacity estimation.

Accurate remaining useful life (RUL) prediction is essential for ensuring system reliability and guiding maintenance decisions in prognostics and health management, and graph neural network (GNN) is an effective approach to exploit the dependencies between multi-sensor data. Despite this advantage, most existing GNN-based studies overlook the uncertainty analysis of RUL prediction, which is essential for trustworthy decision-making. To address this limitation, we propose a graph-based uncertainty analysis framework (G-UA) for RUL prediction that includes uncertainty quantification, calibration, and decomposition. Specifically, G-UA models the predictive uncertainty considering both node- and edge-level uncertainties in graph data, leveraging temporal variations in sensor signals and prior domain knowledge from system structure. The obtained uncertainty is further calibrated to align with actual observations and decomposed into aleatory and epistemic components for reliable decision-making. Furthermore, G-UA performs health state prediction jointly via a multi-task learning framework to enhance its practicality. The effectiveness of the proposed G-UA is demonstrated by the case study of an aero-engine dataset.

Lithium-ion batteries (LIBs) are extensively deployed in industrial fields. To ensure the reliability and efficiency of LIBs, it is crucial to conduct capacity trajectory prediction. However, existing model-based and deep learning (DL) approaches for LIBs capacity trajectory prediction are constrained by imprecise modeling, insufficient analysis of internal states, and the over-reliance on single-source domain transfer learning (TL). To accurately predict LIBs capacity trajectory under varying environmental conditions, a novel electrochemical impedance spectroscopy-based multi-source domain TL framework is proposed in this study. The distribution of relaxation times technique and equivalent circuit model are employed to extract timescale and physical parameters for internal states investigation, and key features are selected as inputs for the DL models, which produce capacity estimation and the cycle intervals between knots corresponding to specific capacity levels. The capacity trajectory is subsequently predicted by applying the piecewise cubic Hermite interpolating polynomial to these knots. Besides, the two-stage weight allocation strategy is developed for the effective integration of multiple source domains, where the first stage involves weight allocation to LIBs within each source domain and the second stage focuses on assigning domain-specific weights. Experiments on two publicly available EIS datasets validate the effectiveness and generalization capability of the proposed framework across different temperatures and LIBs types. Comparison results with state-of-the-art TL methods further demonstrate the necessity of the two-stage weight allocation strategy and highlight the framework superiority in accurately and reliably predicting capacity trajectories.

Accurate wind power forecasting (WPF) is critical for maintaining supply-demand balance in smart grids. However, the inherent randomness of meteorological conditions and frequent operational data loss challenge robust trend prediction, especially for generalization to unseen wind farms. To address this, we propose an attentive residual prompt tuning approach, which constructs a multi-layer self-attention mechanism with shared attention parameters within the prompt tuning subspace, enabling LLMs to achieve zero-shot and robust WPF. A hard prompt generator is introduced to redefine WPF as a language modeling task, allowing the LLM to leverage its representation learning strengths for salient temporal feature extraction and anomaly detection. Meanwhile, a soft prompt adapter with attentive residual modules and bridge-shared parameters contextualizes the LLM for WPF tasks. The residual self-attention blocks model temporal dependencies both within and across wind farms, whereas the shared parameters facilitate feature transfer across temporal and spatial dimensions. This design mitigates reliance on complete input features, improving robustness to large-scale missing data. Furthermore, a hybrid soft-hard prompt fusion mechanism incorporates WPF-specific knowledge into the LLM, enhancing its zero-shot extrapolation capability under severe data loss. Extensive experiments on multi-region wind farms show our method surpasses seven state-of-the-art WPF approaches in robustness and accuracy under complex data loss conditions, and achieves superior zero-shot forecasting on unseen wind farms.

The integration of Industry 4.0 technologies into cattle farming, termed Cattle Farming 4.0, promises to revolutionize traditional livestock management. This paper explores the application of advanced technologies such as the Internet of Things (IoT), Artificial Intelligence (AI), Big Data analytics, and automation in cattle farming. Leveraging research from 2020 to 2025, we propose Cattle Farming 4.0, a smart, data-driven system to optimize breeding, feeding, health management, and more, enhancing productivity and sustainability.

Safety and stability are two critical issues in many practical switched control systems, particularly those used in industrial, transportation, medical, and other high-risk applications where failure could lead to serious consequences. This paper is concerned with the issue of proposing sufficient conditions for simultaneously verifying the stability and safety of time-varying switched nonlinear systems (TSNS) under a state-dependent but unsafe-region-independent switching signal. Specifically, based on multiple Lyapunov functions and multiple barrier functions, we at first present several sufficient conditions for simultaneously obtaining the (uniform) stability with safety, (uniform) asymptotic stability with safety, and (uniform) exponential stability with safety of TSNS with any form of unsafe set. Furthermore, considering the broad applications of finite-time stability in practical systems, a sufficient condition for simultaneously achieving finite-time stability and safety is presented. Note that the above conditions relax the requirements of the nonincreasing property of the multiple Lyapunov functions and multiple barrier functions along the trajectories of TSNS. In the end, the effectiveness of our results is illustrated by four examples.

Balise health prognostics is critical to proactively maintaining the reliable operation of high-speed railway wireless communication. However, the intricately evolving nature of balise health under coupled ground-train transmission dynamics across rail lines challenges generalized and accurate prognostics. To tackle it, this article proposes an explicit-implicit chain-of-thought (CoT) framework that endows a large pre-trained language model (PLM) with desired prognostic power via closed-loop reasoning comprising perception, adaptation, and feedback. Specifically, a stepwise instruction dispatcher is introduced to synthesize balise measurement data with situational context and logic rules, tailoring conditional directives for varying line scenarios. It can prompt PLM to progressively perceive distinctive health evolution signatures and deterioration cues throughout multi-step reasoning, shaping an explicit, causally traceable CoT for generalized prognostics. A dual-granularity adaptor is built to persistently capture coupling relationships among health deterioration factors and sequentially append their unique nuances, forming an implicit, highly adaptable CoT for PLM to improve prognostic accuracy. A regulatory routing mechanism is devised to harmonize explicit and implicit CoTs, while optimizing their collaboration via iterative feedback, mitigating drift of thought to boost prognostic performance. Experiments on hundreds of real-world balises across multiple lines demonstrate that the proposed method achieves accurate health prognostics with strong causal interpretability.

Accurate state of health (SOH) and remaining useful life (RUL) predictions are essential for battery health assessment, early fault detection, and ensuring system safety. However, existing methods struggle to effectively capture multiscale spatiotemporal characteristics, recognize intricate degradation patterns, and achieve synergy between SOH and RUL tasks due to independent architectures and limited information inheritance. To address these challenges, we propose a novel cascaded SOH-RUL assessment framework that integrates recursive hyperparameter transfer to enable deep coupling between SOH and RUL predictions. The framework employs a Triple-Orthogonal-Plane CNN to map battery data onto three orthogonal hyperplanes, extracting and fusing temporal-spatial features via an attention-based adaptive weighting mechanism. Additionally, a Dual-Coupled Memory-Learning LSTM with a novel gating interaction mechanism enhances temporal feature modeling by coupling forget and input gates and introducing peephole connections. Extensive experiments on multiple datasets, including NASA, Oxford, and CALCE, under diverse degradation scenarios, demonstrate significant improvements in prediction accuracy, robustness, and generalization. This framework offers a promising solution for advancing battery health management and system reliability.

Rule-based static analysis tools are widely utilized for their high customizability. However, the creation of effective rules presents significant challenges, including the considerable human effort to handle the complexity of rules, and the additional costs involved in developing rules across various programming languages or frameworks. To address the significant challenges in manual rule creation for static analysis, this paper proposes a novel framework that leverages large language models (LLMs) to automate the generation of static analysis rules. The framework is specifically designed to alleviate the substantial human effort typically required in constructing and maintaining rule sets. Furthermore, we introduce a natural language-mediated rule migration methodology, which ensures semantic consistency when transferring functionally similar rules across different programming languages or frameworks. By seamlessly integrating LLMs with existing static analysis tools, our approach not only enhances the scalability and adaptability of rule generation but also enables efficient vulnerability scanning without the need for extensive computational resources such as large GPU clusters. This integration aims to bridge the gap between natural language understanding and program analysis, thereby facilitating more intelligent and resource-efficient static analysis. The method achieved 81.98 % grammatical and 74.73 % functional validity in rule generation for Semgrep, while migrating rules across 4 languages (Python, Java, JavaScript, and Golang). On the real-world engineering evaluation, it uncovered 9 unknown vulnerabilities in the latest version of the Linux Kernel undetectable by now. This work highlights the potential of our LLM-driven framework provides better handling of corner cases and maintains compatibility with industry-standard tools.

In this article, we provide a comprehensive review of machine learning-based sports analytics in baseball. This review is primarily guided by the following three research questions: (1) What baseball analytics problems have been studied using machine learning? (2) What data repositories have been used? (3) What and how machine learning techniques have been employed for these studies? The findings of these research questions lead to several research contributions. First, we provide a taxonomy for baseball analytics problems. According to the proposed taxonomy, machine learning has been employed to (1) predict individual game plays; (2) determine player performance; (3) estimate player valuation; (4) predict future player injuries; and (5) project future game outcomes. Second, we identify a set of data repositories for baseball analytics studies. The most popular data repositories are Baseball Savant and Baseball Reference. Third, we conduct an in-depth analysis of the machine learning models applied in baseball analytics. The most popular machine learning models are random forest and support vector machine. Furthermore, only a small fraction of studies have rigorously followed the best practices in data preprocessing, machine learning model training, testing, and prediction outcome interpretation.

This article presents a systematic review on blockchain-facilitated cybersecurity solutions for Internet of Things (IoT) devices in space–air–ground integrated networks (SAGIN). First, we identify the objectives and the context of the blockchain-based solutions for SAGIN. Although, typically, the blockchain is primarily used to enhance the trustworthiness of some systems or operations, it is necessary to document exactly in what context the blockchain is used that is specific to the IoT and SAGIN. Second, we investigate how blockchain technology is used to achieve the objectives. Again, we want to report the technical details on how blockchain is used in this specific field instead of general discussion. Third, we provide a critique on the technical correctness of the blockchain-based solutions. As we elaborate in this article, there are serious technical issues in the proposed solutions. The most pervasive assumption made in many blockchain-based solutions is that higher-level trustworthiness can be achieved by using any form of blockchain. Fourth, we provide a guideline on when blockchain technology could be useful for IoT and SAGIN and what types of blockchain could be useful to enhance the security of ubiquitous IoT in SAGIN.

The aircraft engine is a core component of the airplane, and its operational condition is directly related to the safety of both the equipment and personnel. Accurate Remaining Useful Life (RUL) prediction can help identify potential failure risks in advance, prevent unexpected breakdowns, and ensure operational safety throughout the flight. This paper presents a novel approach for predicting the RUL of aircraft engines using deep Bayesian graph neural ordinary differential equations. The method integrates dynamic graph structure learning to capture time-varying sensor dependencies, Bayesian inference, and neural differential equations for continuous-time modeling of spatiotemporal characteristics. Uncertainty quantification is a central focus, employing Bayesian techniques and Monte Carlo Dropout to distinguish and estimate both epistemic model and aleatoric data uncertainty. Significantly improving the reliability of uncertainty estimates. Experiments conducted on the C-MAPSS dataset demonstrate the effectiveness of the approach. The model achieves optimal prediction accuracy with a specific input sequence length, leading to well-calibrated confidence intervals with coverage rates close to nominal levels. This methodology offers enhanced RUL prediction accuracy and robust uncertainty quantification, providing valuable support for predictive maintenance decisions.

The aircraft engine is a core component of the airplane, and its operational condition is directly related to the safety of both the equipment and personnel. Accurate Remaining Useful Life (RUL) prediction can help identify potential failure risks in advance, prevent unexpected breakdowns, and ensure operational safety throughout the flight. This paper presents a novel deep learning framework for predicting the RUL of aircraft engines, integrating Bayesian inference with Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and an attention mechanism. The proposed method first utilizes an attention mechanism to weigh the importance of different sensor inputs, followed by CNNs to extract salient spatial features from these readings, which are then fed into LSTMs to model the temporal dependencies inherent in engine degradation trajectories. Uncertainty quantification is a central focus, employing Bayesian techniques and Monte Carlo sampling at inference to distinguish and estimate both epistemic (model) and aleatoric (data) uncertainty. A key contribution is an innovative dual uncertainty calibration method that optimizes calibration parameters for both uncertainty types via maximum likelihood estimation, significantly improving the reliability of uncertainty estimates. Experiments conducted on the C-MAPSS dataset demonstrate the effectiveness of the approach. The model achieves optimal prediction accuracy with a specific input sequence length, and the dual calibration technique reduces calibration error substantially, leading to well-calibrated confidence intervals with coverage rates close to nominal levels. This methodology offers enhanced RUL prediction accuracy and robust uncertainty quantification, providing valuable support for predictive maintenance decisions.