Deep learning algorithms have become prevalent in equipment Prognostics and Health Management (PHM) modeling, yet their deployment in safety-critical applications remains constrained by inherent epistemic uncertainty. This study advances uncertainty calibration methodologies for two fundamental PHM tasks: classification and time-series prediction. First, we proposed a regularization-free calibration framework that dynamically adjust labels to achieve accuracy-uncertainty alignment in classification models. Building on this, we presented the first comprehensive uncertainty calibration framework for time-series prediction models, along with pioneering evaluation metrics specifically designed for temporal uncertainty assessment. To validate our methodology, we acquired comprehensive fault and degradation datasets from aircraft actuators. The evaluation framework encompassed two critical dimensions: (1) classification calibration across four distinct neural architectures on three benchmark datasets, and (2) degradation calibration evaluated across three neural architectures utilizing eight publicly available datasets. Our experimental results consistently demonstrated statistically significant enhancements in calibration performance metrics. The complete implementation is publicly available at https://github.com/ppqweasd/uncertainty-calibration.

Quantifying predictive uncertainty is crucial for building trustworthy decision-making systems in signal processing. However, many existing embedded evaluation methods rely on prior assumptions about the distributions of network parameters, which can compromise model accuracy. To address this, we proposed a distribution-aware uncertainty quantification framework for Bayesian neural networks. Our method explicitly accommodates different parameter distributions, thereby reconciling model fidelity with reliability. Theoretically, distribution-specific update rules through variational inference were derived. As a practical enhancement, we introduce an efficient fine-tuning strategy that initializes Gaussian distribution means through deterministic pre-training, significantly accelerating computation while maintaining competitive uncertainty estimation quality. Experimental results on aerospace oxygen concentrators and standard benchmarks demonstrate that our method maintains high prediction fidelity while providing reliable uncertainty estimates. The implementation is publicly available at: https://github.com/ppqweasd/Uncertainty-Quantification.

Aircraft fault knowledge graphs serve as a critical knowledge base for the intelligent maintenance and operations of aviation equipment. However, the entity alignment tasks in their construction remain overly dependent on manual annotation, leading to issues such as inconsistent annotation quality and low annotation efficiency. Unsupervised methods provide a promising solution and have garnered significant research interest. However, existing unsupervised entity alignment approaches often overlook the impact of noisy entities, presenting a significant challenge for aligning entities in aviation fault data. This paper proposes a solution by incorporating a large language model (LLM) into the entity alignment process for aircraft fault knowledge graphs. By leveraging the world knowledge encoded in the LLM, the approach enhances the performance of unsupervised entity alignment models. Specifically, we introduce the Collaboratively Enhanced-based Large Language Model Entity Alignment (CELLMEA), which utilizes data from the aircraft flight control system manual, fault analysis manual, and typical fault cases. The model's architecture includes a multi-view semantic information embedding that integrates structural, relational, and semantic data. Additionally, we propose an adaptive method for mixing hard negative samples, which generates higher-quality negative entities by combining noisy negative samples with reliable ones. Furthermore, an incremental consistency regularization technique is introduced to progressively refine the robustness of pseudo-labeling within the CELLMEA model. Finally, experimental results on a flight control system entity alignment dataset demonstrate that CELLMEA outperforms all baseline models, achieving an MRR (Mean Reciprocal Rank) value of 0.917 +/- 0.011. These results validate the model's effectiveness in handling unlabeled data and lay the groundwork for the engineering of aircraft fault knowledge graphs.

In order to accurately extract features from the complex signals of coupled faults collected and improve the accuracy of fault diagnosis, this paper utilizes the hippo optimization algorithm to optimize the VMD, realize the adaptive extraction of parameter combinations, and make use of the timeconvolutional network for the capture of long-term dependence on the feature vectors, combined with the bidirectional gated cyclic unit’s bi-directional processing of the data and the suppression of extraneous information by the attention mechanism, and then input the extracted The extracted feature vectors are inputted into the TCN-BiGRU-AM model for classification to realize fault diagnosis under bearing coupling faults.

This study focuses on condition monitoring and life prediction of electromagnetic relays in power systems. To address the problems of offline operation and low efficiency in existing monitoring systems, a multi-channel online condition monitoring system for electromagnetic relays was designed. Without affecting the normal operation of relays, the system adopts high-precision sampling resistors and optocoupler isolation circuits to achieve real-time acquisition of key parameters (including contact resistance and pull-in time) from 30 relay channels, and ensures reliable data transmission and storage through Wi-Fi modules and customized storage formats. Based on the complete life-cycle data obtained by the system, feature parameters were extracted using Pearson correlation coefficient method as inputs to construct an LSTM (long short-term memory) life prediction model. Experimental results show that the LSTM model achieves MAPE (mean absolute percentage error) and RMSE (root mean square error) values of 0.0155 and 64,436 respectively, with a 93% agreement between predicted and measured values, effectively verifying the reliability of the proposed monitoring system and the applicability of the LSTM algorithm in life prediction. The designed system and method can achieve long-term online health monitoring and effective life prediction of relays.

Addressing the reliance on manual annotation in current aircraft failure knowledge graph modeling, which results in inconsistent annotation quality and low efficiency, we propose semi-supervised methods as a potential solution paradigm for this problem, which have attracted extensive research interest. However, existing work on semi-supervised relational extraction ignores the effect of data imbalance on semi-supervised relational extraction, posing a challenge for semi-supervised relational extraction in aviation fault data. To address this challenge, this paper introduces a semi-supervised relation extraction method. This method improves the performance of the relation extraction model by utilizing unlabeled data. Specifically, we propose an adaptive meta-learning semi-supervised relation extraction model utilizing aircraft flight control system manuals, fault analysis manuals, and typical fault cases as data sources. Firstly, the pseudo-label generation network and the relationship classification network are established separately, and the relationship classification network is used as a meta-objective to optimize the pseudo-label generation network by using meta-learning. Next, the adaptive thresholding method combining global thresholding and local thresholding is used to optimize the selection of pseudo-labels. Furthermore, the incremental learning approach is introduced to improve the robustness of the model. Finally, taking the flight control system fault dataset as an example, the experimental results show that the model's performance is higher than all the baseline models and achieves an F1 score of 77.803 +/- 0.614% with 5% labeled data, which verifies the validity of the model in the case of insufficient labeled data and lays a foundation for the engineering of the knowledge graph of aircraft faults. The code for AdaMetaSRE is available at https://github.com/lijh01/AdaMetaSRE.

To address the issues of resource contention, deadlock, and premature convergence in parallel test task scheduling for complex systems, this paper proposes an adaptive differential evolution algorithm based on population dissimilarity. The algorithm employs integer encoding to represent task scheduling sequences, designs a population initialization method based on task dependency constraints, and dynamically evaluates population diversity using Kendall’s Tau correlation coefficient. It adaptively adjusts the length of mutation subsequences to balance global exploration and local exploitation capabilities. Additionally, the algorithm integrates crossover-selection operators and timedriven fitness function optimization to ensure that scheduling schemes satisfy task priorities and resource constraints. Simulation results demonstrate that, compared to traditional differential evolution algorithms, the proposed algorithm significantly improves convergence speed, scheduling efficiency, and stability, effectively avoiding premature convergence. The proposed algorithm provides an efficient and reliable optimization method for parallel test task scheduling in complex systems.

The fault modes of rotating bearings include independent faults and complex coupling faults composed of independent faults. The coupling fault includes various types of fault modes and the fault degree change continuously. The dynamic vertex graph attention network is proposed for coupling fault diagnosis of bearing faults, and interpretable fault features are extracted based on multi-head attention mechanism. Simulation data sets of source domain and target domain with different fault types are established, and DVGAT is utilized for zero-shot coupling fault diagnosis, and the diagnosis recall rate of each fault is no less than 98.5%.

In this paper, a fast recognition algorithm based on Deep Convolutional Neural Network (DCNN) is proposed and studied for the speed and accuracy of infrared target recognition in missile terminal guidance system. This algorithm makes full use of the advantages of deep learning model in feature extraction and pattern recognition, combined with the unique characteristics of infrared imaging, and designs and trains a deep convolutional neural network model suitable for the infrared target recognition task of missile terminal guidance. By optimizing network structure, adopting advanced training strategy and fusing multi-scale and multi-view infrared image information, the constructed model can realize efficient and accurate recognition of infrared targets under complex background, and meet the real-time requirements. The experimental results show that the algorithm not only significantly improves the accuracy of target recognition, but also ensures the ability to process a large amount of infrared image data in real time under the condition of high-speed flight, so as to effectively improve the strike accuracy and combat efficiency of the missile terminal guidance system, which has important theoretical value and practical application significance for promoting the development of precision guidance weapon technology in China.

To tackle the problems of over-reliance on traditional experience, poor troubleshooting robustness, and slow response by maintenance personnel to changes in faults in the current aircraft health management field, this paper proposes the use of a knowledge graph. The knowledge graph represents troubleshooting in a new way. The aim of the knowledge graph is to improve the correlation between fault data by representing experience. The data source for this study consists of the flight control system manual and typical fault cases of a specific aircraft type. A knowledge graph construction approach is proposed to construct a fault knowledge graph for aircraft health management. Firstly, the data are classified using the ERNIE model-based method. Then, a joint entity relationship extraction model based on ERNIE-BiLSTM-CRF-TreeBiLSTM is introduced to improve entity relationship extraction accuracy and reduce the semantic complexity of the text from a linguistic perspective. Additionally, a knowledge graph platform for aircraft health management is developed. The platform includes modules for text classification, knowledge extraction, knowledge auditing, a Q&A system, and graph visualization. These modules improve the management of aircraft health data and provide a foundation for rapid knowledge graph construction and knowledge graph-based fault diagnosis.

This paper discusses in detail the design and implementation of the terminal guidance system based on the composite tracking strategy of fusion template matching and Harris corner detection technology. Through the complementary advantages and deep integration of these two mature visual tracking methods, the aim is to improve the target recognition accuracy, tracking stability and anti-interference ability of the missile in the complex environment, so as to optimize the strike efficiency in the final guidance stage. Firstly, the basic principles of template matching and Harris corner detection are analyzed in detail. Then, the paper describes how to organically merge the two to form a compound tracking strategy. Specifically, template matching is used to quickly lock the approximate location of the target, and its keen ability to capture the significant features of the target is used to improve the success rate of first capture and the speed of tracking initialization. Then, Harris corner detection is introduced to extract and describe fine features of the target, and a robust set of target feature points is constructed to effectively deal with complex conditions such as target deformation, rotation and occlusion. On this basis, an adaptive weight allocation mechanism is designed to dynamically adjust the contribution of template matching and corner detection according to the complexity and SNR of the current tracking environment, so as to ensure the optimal tracking performance under various working conditions. In the experimental part, the performance of the proposed composite tracking strategy is evaluated comprehensively with the help of simulation software and actual flight test data. The results show that compared with the single template matching or corner detection methods, the fusion strategy has a significant improvement in target recognition rate, tracking accuracy, tracking stability and anti-interference ability. In summary, this paper proposes a composite tracking strategy that integrates template matching and Harris corner detection, and successfully applies it to the missile terminal guidance system to achieve effective target identification and stable tracking, and significantly improve the strike efficiency and environmental adaptability of the missile during the terminal guidance stage. It provides a new idea and practical reference for the development of missile precision guidance technology.

Coupling faults that simultaneously occur during the operation of mechanical equipment are widespread. These faults encompass a diverse range of high-order coupling relationships, involving multiple base fault types. Based on the advantages of hypergraphs for higher-order relationship descriptions, two coupling fault diagnosis architectures based on the hypergraph neural network are proposed in this paper: 1. In the coupling fault diagnosis framework based on feature generation, the base faults serve as the hypergraph nodes, and each hyperedge connects the base faults. The generator, which consists of the hypergraph neural network, generates coupling faults as negative samples to enforce regularization constraints for the discriminator training. 2. In the coupling fault diagnosis framework based on feature extraction, each node represents a fault mode, and each hyperedge connects nodes with common failure modes. The multi-head attention mechanism extracts the features of base faults, and the common fault features in a hyperedge are aggregated via the hypergraph neural network. The inner product correlation is used to diagnose the fault modes. The results show that the diagnostic accuracy for coupling faults with the two frameworks reaches 88.6% and 86.76%, respectively. Both frameworks can be used for the diagnosis and analysis of high-order coupling faults.

This paper mainly discusses how to effectively use multi-core embedded digital signal processor (DSP) technology to realize parallel computation of image tracking algorithm and memory optimization in high precision missile terminal guidance system. With the increasing requirement of missile strike accuracy in modern war, the vision-based terminal guidance technology has been widely concerned because of its strong autonomy and anti-jamming ability. However, such image processing algorithms are usually highly computation-intensive and data dependent, which requires high real-time and hardware resources. In this paper, the key steps of image tracking algorithm are analyzed and reconstructed deeply, and the architecture suitable for parallel processing of multi-core DSP is designed by modularization. The parallel computing advantages of multi-core DSP are given full play through task division and scheduling strategies to improve the execution efficiency and real-time performance of the algorithm. Secondly, the research also focuses on memory optimization, including the design of data cache strategy, reducing data redundancy, improving memory access locality, etc., and strives to maximize the data reading speed and reduce the memory bandwidth pressure under the condition of limited embedded system resources. To sum up, this research is committed to building an efficient and stable multi-core embedded DSP environment, parallel computing model and memory optimization scheme for high-precision missile terminal guidance image tracking algorithm, which has important theoretical value and practical significance for improving the intelligence level and combat efficiency of missile terminal guidance system.

As the demand for enhanced maritime safety continues to grow, ship target detection technology has become increasingly crucial in the domain of ocean surveillance. Conventional target detection methods face challenges such as high computational demands and reduced detection efficiency in ship identification scenarios. This study introduces an optimized lightweight ship detection algorithm, building upon the YOLOv5 framework. By incorporating streamlined model architecture and parameter tuning, the model retains YOLOv5’s high accuracy while minimizing computational overhead and boosting detection speed. The proposed model integrates deep separable convolution and pruning techniques, effectively shrinking the network size without compromising detection precision. For system validation, a publicly available dataset featuring a variety of ships is employed in the experimental phase. The simulation outcomes indicate that the refined model sustains high detection accuracy while substantially decreasing computational load. Experimental findings reveal that the model achieves a mean Average Precision (mAP) of 92.3% and accelerates detection to 45 frames per second (FPS), marking a 15% improvement over the original YOLOv5 algorithm. This research presents a novel approach for ship detection in maritime monitoring, offering significant practical application potential.

Parameter-Efficient Fine-Tuning is widely used to transfer models between different domains. However, for some high-reliability-equipment, the degradation is at a slow rate and continually fluctuates, making it difficult to extract features effectively. Moreover, collecting an integrated source domain for high-reliability-equipment is tough due to the small sample of related datasets. Aiming at the transfer problem of the time-series prediction model, this research proposed an LSTM-fine-tune model, where the parameters of the model are explicitly trained and partly frozen, such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance on that task. The algorithm is then benchmarked on sinusoidal functions, where data were randomly generated with different phases and amplitudes. The results show that the LSTM-fine-tune model can learn knowledge from different sinusoidal data and fit a new one quickly with high accuracy. This paper also considers solving two actual problems. One is transferring oxygen concentrator data from the experimental condition to the actual service condition, the results show that the accuracy is largely improved. Moreover, this paper tried to extract more general degradation knowledge from the Wiener process and then transferred it to the degradation data of the oxygen concentrator. The results show that the model quickly achieved higher prediction accuracy with the knowledge. The code and data from the test bed are accessible at https://github.com/panjinxin123/Adaptive-finetuning-in-degradation-time-series-forecasting-via-generating-source-domain.

Mechanical equipment is composed of several parts, and the interaction between parts exists throughout the whole life cycle, leading to the widespread phenomenon of fault coupling. The diagnosis of independent faults cannot meet the requirements of the health management of mechanical equipment under actual working conditions. In this paper, the dynamic vertex interpretable graph neural network (DIGNN) is proposed to solve the problem of coupling fault diagnosis, in which dynamic vertices are defined in the data topology. First, in the date preprocessing phase, wavelet transform is utilized to make input features interpretable and reduce the uncertainty of model training. In the fault topology, edge connections are made between nodes according to the fault coupling information, and edge connections are established between dynamic nodes and all other nodes. Second the data topology with dynamic vertices is used in the training phase and in the testing phase, the time series data are only fed into dynamic vertices for classification and analysis, which makes it possible to realize coupling fault diagnosis in an industrial production environment. The features extracted in different layers of DIGNN interpret how the model works. The method proposed in this paper can realize the accurate diagnosis of independent faults in the dataset with an accuracy of 100%, and can effectively judge the coupling mode of coupling faults with a comprehensive accuracy of 88.3%.

针对寿命预测模型迁移问题,提出了一种长短周期记忆网络微调(long short-term memory fine tune,LSTM-fine-tune)的迁移模型,利用理想条件下的试验数据对模型进行训练.在迁移过程中,对部分LSTM网络层进行冻结,利用实际服役环境下的数据对网络其他部分进行修正.为验证模型的泛化能力,采用不同相位与幅值的正弦函数生成数据,通过学习数据获取正弦函数的经验知识,并应用至其他正弦函数的回归,结果表明LSTM-fine-tune 模型能够快速拟合,平均均方误差仅为 1.0335,明显低于直接预测误差1.5368.为通过实际监测数据检验本方法泛化能力,分别获取了试验条件下与实际服役环境下氧气浓缩器的数据,对模型的泛化能力进行验证.结果表明,迁移后训练集预测精度提高了 43.0%,测试集预测精度提高了20.2%.

In response to the current problems in the maintenance process of aircraft power systems, such as relying on traditional experience, poor troubleshooting robustness, and slow response of maintenance personnel to fault changes. In this paper, Knowledge graph is introduced into the field of aviation power supply health management, and troubleshooting experience is presented in the form of Knowledge graph to improve the correlation ability between fault data. This paper proposes a method to build a fault Knowledge graph for aircraft power system health management based on the data source of a certain aircraft power system manual and typical fault cases: first, build a power system data set according to the semi structural and unstructured data such as the aircraft power system manual and typical fault cases; Then, the Bert-BiLSTM-Ptr-Net deep learning algorithm was applied to extract power system fault entities from power system manuals and typical fault cases, and the superiority of this algorithm was verified by comparing it with the Bert-BiLSTM-CRF algorithm; Next, the BERT-BiLSTM deep learning algorithm is used to extract relationships between power system fault entities. Compared to other deep learning algorithms, this algorithm has a relationship extraction effect in aircraft power system manuals and typical fault case texts; Finally, Neo4j Graph database is used to visualize the fault Knowledge graph of aircraft power system, which effectively improves the troubleshooting ability of maintenance personnel and is convenient to guide the fault maintenance of aircraft power system.

Coupling fault is a common fault in mechanical equipment pattern. In this paper, a graph neural network method is proposed to study the multi-mode fault coupling. Firstly, the wavelet transform is used to preprocess the time series data, then the coupled fault topology is established according to the fault correlation as the input of the graph neural network. Finally, the graph convolutional neural network is used to diagnose the multi-mode coupling fault. The algorithm is tested on the XJTU Gearbox dataset.

The integrated-servo-actuator (ISA) is one of the most critical subsystems of aircraft flight-control system, which controls the speed, direction, displacement, and force of load for an aircraft. The degradation state of the ISA directly influences the safety of the aircraft. Hence, it is valuable to predict the degradation of the ISA to ensure its operating reliability. However, as a kind of complex system, the degradation mechanisms and the influences of the stress cannot be fully understood. Moreover, the nonlinear degradation process brings more difficulties to establish an accurate life prognosis model. To address these challenging issues, this article proposed a hybrid degradation prognosis method that fused the physics-based model and data-driven model for ISA. The nonlinear Wiener process (NWP) algorithm is utilized to characterize the physical degradation process of ISA. The effect of different stresses is quantitatively modeled. Furthermore, the data-driven echo-state-network (ESN) is optimized to describe the nonlinear degradation process of ISA. More degradation data are generated based on the NWP model for ESN model training. Therefore, the degradation trajectory predicting model is fused with the physics-based degradation prognosis model, so that both the degradation mechanisms and time-series features within the monitoring data are combined. The experimental results based on the real ISA data illustrate that the proposed method has higher prediction accuracy, which is meaningful to enhance the operating reliability of the ISA.