Wireless sensor network is composed of multiple micro-sensor nodes， and positioning technology is one of the important applications of WSN. At present， many localization algorithms have high localization accuracy in line of sight （LOS） environment， but poor localization accuracy in non-line-of-sight （NLOS） environment. An improved maximum entropy fuzzy probability data association algorithm based on arrival time was proposed. The grouping idea was utilized to divide N measure- ment values into L groups， and each group obtained the corresponding mobile node position estimation， model probability and covariance matrix through the interactive multi model （IMM） algorithm. Afterwards， the obtained L position estimation was subjected to non-line-of-sight detection through a validation gate. The position estimation contaminated by non line of sight errors was discarded， and the corresponding correlation probabilities was used to weight the correct position estimates to obtain the final position estimation. Simulation and experimental results show that the proposed algorithm can reduce the influence of non line of sight errors and achieve higher positioning accuracy than the existing methods.

In systems such as aerospace with high-reliability requirements， parallel forms were often used in engineering to improve system reliability. When a component in the parallel system fails， the working environment of the remaining normal components deteriorates due to the load redistribution. In order to solve this problem，a reliability evaluation model of the parallel repairable system based on load-sharing was established. The model was suitable for parallel repairable systems with the same and different component failure rates. When the load-sharing factor was one， the model will be degenerated into the traditional Markov model. The numerical results show that the variation trend of transient avai-lability obtained by the proposed and traditional method is same. Compared to the results evaluated by traditional methods， the reduction in steady-state availability of the system is not huge when considering load-sharing， while the MTBF will decrease significantly. Therefore， it should be cautious to use MTBF obtained by the traditional Markov model to measure system reliability and determine maintenance intervals.

Inertial/satellite/visual integrated navigation can obtain highly accurate continuous positioning information in complex urban environments， and therefore has become increasingly widely used. However， sensor gross errors caused by changes in the external environment can lead to a decrease in the accuracy of integrated navigation， and even cause divergence in the results of integrated navigation. In response to the problem of sensor gross errors caused by environmental factors in the low-cost inertial/satellite/visual integrated navigation process， an inertial/satellite/visual integrated navigation algorithm based on an improved graph optimization algorithm was proposed. A robust kernel function was introduced to the cost function based on the traditional graph optimization algorithm to filter the sensor gross errors caused by environmental factors， and earth rotation acceleration compensation to the IMU （inertial measurement unit） pre-integration was added to improve the accuracy of pre-integration， thus， the accuracy of integrated navigation was improved. Finally， a comparative experiment was designed based on the kaist urban driving dataset， and the results show that the improved graph optimization method proposed can effectively filter sensor gross errors caused by external environmental factors， and the navigation accuracy improves by 7% compared to the traditional graph optimization method.

Because of its efficient learning and characterization capabilities， the deep convolutional neural network （DCNN） model is widely used in the auxiliary diagnosis of various types of diseases， including lung disease. Most of the existing DCNN models are self-supervised，which have high requirements for the quantity and quality of data sets. However， the privacy of medical data and the high cost of data labeling greatly limit the performance and effectiveness of the model. A phased self-superivised diagnostic model based on comparative learning was proposed.The model extracted the features of unlabeled lung CT images augmented by random data which did not require too many labeled COVID-19 data sets for training， which effectively solved the problem that the small labeled data sets made it difficult to train supervised models. At the same time， four ablation experiments were designed to verify the performance of this model. The experimental results show that the model has excellent performance in the self-supervised diagnosis task of lung CT images of COVID-19.

The C/SiC quadrangular pyramid lattice structure composite has the functions of lightweight，high strength load bearing，heat insulation， and so on， which can be used in the thermal protection system of hypersonic aircraft， but its mechanical properties decline significantly in high-temperature environments. In order to improve the mechanical properties of C/SiC quadrangular pyramid lattice structure composites at high-temperature，ZrC was introduced with high melting point into the matrix of C/SiC quadrangular pyramid lattice structure composites， and the C/SiC-ZrC quadrangular pyramid lattice structure composite was prepared by precursor impregnation and pyrolysis （PIP）. The compressive mechanical properties of the C/SiC-ZrC quadrangular pyramid lattice structure composite were tested at room temperature after high-temperature treatment at room temperature， 1 200 ℃ and 1 600 ℃ respectively， and the corresponding elastic modulus and compressive strength of the material were obtained. Scanning electron microscope （SEM） and X-ray diffractometer （XRD） were used to observe the surface characteristics of C/SiC-ZrC quadrangular pyramid lattice structure composites after high-temperature treatment， the phase and composition of the materials before and after heat treatment were studied， the reasons for the improvement of mechanical properties of C/SiC-ZrC quadrangular pyramid lattice structure composites after high-temperature treatment were analyzed， and simultaneously a comparison was made with the mechanical properties of C/SiC quadrangular pyramid lattice structure composites. The results show that the mechanical properties of C/SiC-ZrC quadrangular pyramid lattice structure composites are significantly improved after high-temperature heat treatment.

To optimize the valuing system of flight dispatcher’s job qualification and increase the dispatcher group’s entire quality，the G1 and CRITIC combination weighting method-based valuing model was adapted and dispatcher’s job characteristics was integrated into this system which covered four first-level indicators and twenty-one second-level indicators. Subjective and objective combination weighting methods was used to confirm weight according to different valuing indicators， finally，a formula of flight dispatcher’s competence valuing model was built. The standard for valuing dispatcher’s qualification was classified and suggestions for evaluating and training dispatcher were put forward. This article provides theoretical references for improving dispatcher’s qualification and enhancing the operating management quality and the safety ensuring capacity of airlines.

In the study of abnormal state identification of aircraft engine starting systems， the parameters of the starting system have characteristics such as strong correlation between data， high data dimensions and a lot of redundant information in data. In order to reduce the data dimension and improve the accuracy of abnormal state identification， an improved ReliefF algorithm was combined with probabilistic neural network（PNN） and improved ReliefF-PNN was proposed for abnormal state identification of aircraft engine starting system， which effectively reduced the dimension of parameters and improved the performance of the abnormal state identification model. The obtained model was used for identification verification and analysis of the starting system.The results show that the accuracy of using the improved ReliefF-PNN algorithm to identify abnormal states is better than before， further improving the performance of the model.

In response to the severe lighting conditions and complex background in the production site of power plant， an improved YOLOv5s abnormal state detection method for slag extractor was proposed to ensure the safe and efficient operation of slag extractor in complex power plant environments. On the basis of YOLOv5s network， the ShuffleNet was introduced to replace the original backbone network and to achieve network lightweight by reducing network parameters. At the same time， an improved convolutional attention module was added to the ShuffleNet， and more attention was paid to the target features of the slag extractor scraper by concatenating space and channel attention mechanisms. The weighted bidirectional feature pyramid Bi-FPN （Bilateral Feature Pyramid Network） and bounding box regression loss SIoU （Scaled IoU） function were introduced to obtain more effective feature maps for feature information to improve target detection accuracy. The research results show that the improved model significantly reduces the number of parameters， reduces the model volume by 15.2%， improves the average accuracy of mAP （mean Average Precision） by 2.2%， and reduces detection time by 58.0%. While ensuring detection accuracy， real-time and accurate detection of abnormal states of the slag extractor is achieved.

Aiming at the problems that the initial trend of rolling bearing performance degradation was not obvious and the early fault was difficult to detect， a performance degradation evaluation method based on the combination of variational modal decomposition （VMD） signal preprocessing and support vector data description （SVDD） were proposed. Firstly， the original signal was decomposed by variational mode. Secondly， a new screening index P was proposed for the selection of modal components （IMF）. The calculation formula of this index consists of kurtosis of envelope spectrum and Wasserstein distance， and the modal components with P value greater than the threshold M were selected for signal reconstruction. Finally， the root mean square value， waveform factor and peak-to-peak value of the reconstructed signal were extracted to construct a feature vector representing the degradation of bearing performance， and the SVDD performance degradation evaluation model was established with the degradation feature vector of healthy samples as input， which was verified by the full-life sample feature vector. The experimental results show that this method is more sensitive to early faults and can accurately detect early faults.

A new type of automatic transmission line crimping system was designed to solve the problems of long crimping time and low crimping accuracy in traditional manual wire crimping. The STM32 microcontroller was used as the main controller，which could achieve automatic wire movement，automatic crimping，and automatic measurement functions. To solve the problems of nonlinearity，time-varying parameters，and susceptibility to external environmental interference in automatic crimping systems，a fuzzy PID intelligent crimping control system based on LabVIEW was designed. The experimental results show that the designed intelligent control system for automatic crimping of wires has the characteristics of small overshoot，fast response speed，and high control accuracy，greatly improving production efficiency and operational safety，which can provide technical support for the practical engineering application of automatic crimping systems for transmission lines.

It is a key issue for national and regional development to explore the motivation of scientific and technological talents agglomeration and create a talent innovation highland. The ecological environment of talents is the key factor affecting the agglomeration of scientific and technological talents. Based on the theory of talents ecological environment， 14 cities in Liaoning Province were taken as examples and used qualitative comparative analysis to explore the influence path of talents ecological environment on the scientific and technological talents agglomeration. The results show that there are three paths to generate high-tech talents agglomeration， including the economy-technology-driven path dominated by public services， the economy-culture-driven path dominated by public services， and the economy-technology-driven path dominated by public services， culture and education. Five paths to generate non-high-tech talents agglomeration are found in this paper. From the perspective of configuration， this paper studied the multiple interaction paths of ecological environment factors affecting the agglomeration of scientific and technological talents， which has important theoretical significance and practical value for the construction of talents highland and the talent strategy of China.

In order to prepare high-quality graphene more efficiently and cost-effectively， expanded graphite was prepared by heating a commercial graphene precursor in a microwave oven， and then functionalized graphene （D2000-g-GnPs） was prepared by grafting polyetheramine （D2000） onto the graphene （GnPs） surface. Epoxy resin （EP）/graphene composite binders were then prepared by ultrasonic dispersion and magnetic stirring. The effects of graphene on the improvement of the mechanical properties of epoxy resin and the surface morphology of the tensile fracture surface before and after modification were investigated. The experiments showed that the tensile strength， Young's modulus， fracture toughness and fracture energy release rate of epoxy resin increased and then decreased with the increase of graphene content， and the mechanical properties of composites reached the maximum value when the mass fraction of modified graphene reached 0.3%. The tensile strength， Young's modulus and fracture toughness reached 76.269 MPa， 1.315 GPa and 1.79 MPa.m^0.5 repectively， which were 24.95%， 10.97% and 16.23% higher than those of graphene/epoxy composites before modification. In addition， the addition of D2000-g-GnPs significantly improved the electrical conductivity of the epoxy binder and a low pre-permeation value of 1.0% was achieved.

Rolling bearings fault diagnosis problem based on fractional-order BP neural network was researched. The working characteristics of rolling bearings was reflected in a more comprehensive way by extracting five state types characteristic signals. The characteristic signals were normalized and used as the input of neural network， the fault type of rolling bearing was used as the output of neural network. Condition monitoring and fault diagnosis were performed on rolling bearing by applying fractional-order BP neural network in order to determine which fault type it belonged to. Compared to integer-order BP neural network， the accuracy of fractional-order BP neural network was higher and met the error requirements faster. The simulation experiment results show that the fractional-order BP neural network can acquire running conditions of rolling bearings accurately.

Established a coupling analysis numerical simulation platform for air-cooled turbine blades， and a three-dimensional conjugate heat transfer numerical simulation for radial air-cooled MARK II blades was carried out using the conjugate heat transfer method. The multi-field coupling characteristics of the air-cooled turbine blades were analyzed， and the numerical results were compared with the experimental results. The results show that the choice of turbulence model affects the temperature distribution but has little influence on the pressure distribution on the blade surface； Among the six turbulence models selected， the SST k- \omega turbulence model can simulate the flow field better， and the temperature and pressure on blade surface are the closest to the test results， which meets the engineering calculation requirements. When the total temperature at the turbine inlet is non-uniform in the radial direction， a local high-temperature zone is formed at the trailing edge of the blade， which increases the temperature gradient on the blade. Therefore， the cooling method can be improved to enhance the blade strength.

During the use of UAV for pedestrian and non-motorized vehicle detection in motorway， the problem of low accuracy and poor performance in object detection was found. To solve these problems， a pedestrian and non-motorized vehicle detection algorithm YOLOv5s-P2S was proposed for UAV perspective. Firstly， the neck part of the YOLOv5s model was extended based on the original PAFPN feature fusion scheme， and a detection layer specifically for small object was added. Then， a small object detection head was added to the prediction part to predict the output feature map of the small object detection layer. Finally， the localization loss function of YOLOv5s was modified to SIOU to improve the detection accuracy and regression efficiency of the anchor box. The experimental results showed that compared with the YOLOv5s model， the average accuracy mean mAP₅₀ of YOLOv5s-P2S increased by 0.05， and the parameters only increased by 0.2M. YOLOv5s-P2S can meet the accuracy and real-time requirements of pedestrian and non-motorized vehicle object detection for UAV perspective.

The classic BBR congestion control algorithm lacks timely adjustment of sending behavior and packet quantity during link congestion， which can easily exacerbate the degree of congestion and result in significant delays. Through analysis， the main root cause of this issue is the hysteresis in BBR congestion detection. To address this problem， the BBR congestion prediction and avoidance （BBR-CPA） algorithm was proposed. This algorithm started from the round-trip time update mechanism of BBR， dynamically detected the bottleneck path， and predicted the congestion state of the link based on RTT data， thus reduced the packet quantity in advance to alleviate potential congestion in the link. During operation， it records the bottleneck path’s bandwidth data and performed mean processing on bandwidth estimation， accelerated the convergence of the link to the optimal state. Experimental results show that compared to the classic BBR algorithm and the latest BBR-S and BBR-ACD algorithms， BBR-CPA achieve an average delay reduction of 56%， 44%， and 8%.It effectively eliminates link congestion and reduces the resulting delays.

In order to improve the capacity of emergency rescue services， an emergency graphic information transmission technology based on BeiDou-3 short message was studied， and an emergency graphic information transmission platform based on BeiDou-3 was designed. In order to improve the efficiency of BeiDou short message information transmission， text and image compression coding algorithms were studied respectively， and the text and image information were compressed into packets with smaller information capacity， which reduced the length of packets required for information transmission. The experimental results show that the compression rate of the algorithm reaches 50% for text messages and 2% for image messages. The test results of the platform show that the information interaction between the terminal and the commanding machine can be carried out efficiently with limited communication resources， which helps to respond to the emergency situation more quickly. The research results have a certain reference value for the emergency management department to improve the capacity of emergency rescue services.

The performance health indicator of rolling bearings constructed from a single model can only describes the performance degradation states of rolling bearings from a single perspective， which has certain limitations.To solve this problem， a method for constructing HI based on PCA multi-model fusion was proposed. The idea was to use SVDD， AAKR， and GMM models to construct the corresponding single model HI， and then fuse them through PCA. The first principal component was selected as SAG-HI， containing "multi angle" performance degradation information. Experimental results shows that compared to the HI of each single model， SAG-HI achieved 98.06% grey confidence level in maintaining reliability with rolling bearings， and its correlation， monotonicity， and robustness were the best. Envelope spectrum analysis verified its ability to accurately monitor early fault occurrences.

In order to analyze the influence of fuel supply law of aero-engine in the fuel spiking surge test， a hybrid exhaust turbofan engine with a small bypass ratio was taken as the research object. The engine was modeled as a typical system composed of essential components and built an equilibrium equation considering the rotor dynamics equation and volume effect. The fuel spiking surge process of the engine was simulated. The error between the simulation result and experimental data was within 4%， which verifies the accuracy of the calculation method. The key influencing factors of the step fuel supply law were obtained through the simulation. The simulation results can provide a theoretical basis for the fuel spiking test of the engine and reduce the risk of overtemperature and overturning during the fuel spiking surge process， which has certain engineering significance.

A clustering ant colony hybrid algorithm was proposed for the vehicle routing problem with capacity constraints， which divided the vehicle routing problem into several traveling salesman problems for solution. Firstly， the generation method of pheromones and paths in the ant colony algorithm was improved to effectively split and solve the vehicle routing problem； Then， by grading the population， the convergence speed of the ant colony algorithm was accelerated， and three neighborhood search operators were set to avoid the ant colony algorithm falling into local optima； Finally， simulation experiments were designed to reasonably design some parameters of the algorithm， and 50 Solomon benchmark examples were selected for experimental verification of the algorithm. The experimental results show that the algorithm proposed in this paper has fast convergence speed， high stability， and good solution results.

In order to improve the effectiveness of feature extraction and accuracy of fault identification of rolling bearing， a signal reconstruction method based on joint indicators and a fault diagnosis method based on CWT-2DCNN were proposed. First， a joint indicator was constructed according to kurtosis and cross-correlation number to screen and reconstruct the intrinsic mode fuction（IMF） components obtained by ensemble empirical mode decomposition（EEMD）.Secondly， continue wavelet transform（CWT） was used to extract the features of the reconstructed signal in time-frequency domain. Finally， a fault recognition model based on convolutional neural network（CNN） was constructed with time-frequency feature diagram as input， so as to realize the intelligent fault diagnosis of rolling bearing. The experimental results show that the fault diagnosis accuracy of the proposed signal reconstruction and fault diagnosis method is 99.48%， and it still has a high correct recognition rate under strong noise， indicating that it has a strong generalization ability.

The method of fuzzy set qualitative comparative analysis （fsQCA） was used to explore the path of high concentration of scientific and technological talents in Northeast China. It is proposed that the opportunity should be seized to carry out collaborative innovation of talent chain， industrial chain and education chain. It was suggested that start from the high-quality development of characteristic industries and the strengthening of scientific and educational advantages， accurately focus on characteristic industries， in order to accelerate the transformation and upgrading of equipment manufacturing industry. By building advanced manufacturing industry clusters， giving full play to the leading role of "chain masters"， stimulating small and medium-sized enterprises to introduce talents， and giving full play to the advantages of colleges and universities， improving the awareness of government services， and optimizing the ecological environment of talents， it can promote the Northeast China to adjust the strategy of scientific and technological talents，gather scientific，technological talents under the background of comprehensive revitalization in the new era.

In practical engineering occasions， the collected fault signals of rolling bearing are from multiple sources. The multiple fault signals are coupled to form the composite fault signals in the propagation path， which makes the fault diagnosis problem more complex. If the composite fault signal is directly analyzed， the extracted fault feature contains multiple source fault frequencies， which may result in a failure to determine the location of the fault. In order to resolve the problem， variational Bayesian independent vector analysis（VBIVA） algorithm was proposed and applied to fault diagnosis. Simulation results show that the proposed algorithm solved the problem of blind source separation and fault diagnosis in the comparison with IVA and VBICA.

Traditional manufacturing methods of thick-plate curved surface parts have problems， such as high cost and low efficiency， and the multi-point forming technology provides a solution. In the multi-point forming technology， it is difficult to determine the process parameters because of springback， which increases the number of experimental operations and reduces the forming efficiency. In order to improve the forming efficiency， the simulation through ABAQUS， one finite element software， was combined with experiments to study and analyze the sheet metal forming in this paper. Firstly， three-dimensional modeling and simulation of the sheet-forming process was carried out using finite element software. Then compensation modification of the mold curved surface was carried out by analyzing the simulation results. Finally， experiments were performed on the modified mold curved surface to obtain the parts that meet the requirements. The forming experiments on the modified mold surface can reduce the number of operations and improve the forming efficiency， which provides a reference for subsequent research.

In order to improve the flutter velocity of a lifting surface with embedded mass，the lifting surface was simplified as a rectangular cantilever plate with concentrated mass.Based on the assumed mode method and the first-order piston theory，a flutter model was obtained，and flutter characteristics of the rectangular cantilever plate were investigated.Firstly，the structural dynamic model of the rectangular cantilever plate was established based on the Kirchhoff thin plate theory，and the modal properties of the plate with embedded mass were obtained using the assumed mode method； secondly，the first-order piston theory was introduced to approximate the modal aerodynamic force under supersonic conditions，and the time-domain expression of the flutter equation was obtained by substituting it into the structural dynamics model； finally，the flutter equation was transformed into the frequency domain by introducing the modal displacement in exponential form，and the flutter velocities and flutter frequencies were solved by the p-k method.By changing the position，mass，number，and other parameters of embedded objects，the law of their influence on the flutter characteristics of the supersonic cantilever plate were obtained.The results show that the flutter velocity of the lifting surface can be effectively increased by embedding small mass objects near the leading edge of the cantilever plate near the wing tip，and the flutter velocity is insensitive to the change of the embedded mass near the wing root.

In order to solve the problems of high missed detection rate and low detection success rate in UAV small target detection， a small target detection algorithm based on YOLOv5 was proposed.Firstly， the swin transformer module was integrated into the backbone structure and the neck structure respectively， which improved the accuracy of target detection on the basis of reducing the computational cost， and could adapt to the detection of small target in UAV aerial photography.Secondly， the convolutional attention module （CBAM） was introduced to enhance the network’s attention for small target features.Finally， the original loss function CIoU was replaced by the SIOU loss function， and the weights of high-quality samples were emphasized to accelerate convergence and improve the regression accuracy.Experimental results show that the detection accuracy on Visdrone2019 dataset is 35.3% after model optimization， which is 5.2% higher than that of YOLOv5.Compared with other classical and advanced algorithms，SWCBSI-YOLO algorithm performs well and meets the detection requirements of small targets for UAV aerial photography.

The multi-depots vehicle routing problem with time windows （MDVRPTW） was studied. The MDVRPTW model was designed，and an adaptive large neighborhood search algorithm combined with the Gaussian mixture model （GMM） clustering algorithm was proposed. By classifying the customer set before the neighborhood transformation， the initial solution was optimized， and the algorithm efficiency was improved. Six different transformation factors were used. A scoring system was used to enable the algorithm to select appropriate transformation factor adaptively at different stages of the iteration. The rationality of parameter was analyzed and three simulation experiments were designed. The experimental results verifies the efficiency of the algorithm.

Tracing learners’mastery of knowledge is a pivotal research direction in the realm of wisdom education.Traditional deep knowledge tracing methods predominantly focus on recurrent neural networks，facing challenges such as the lack of interpretability and handling long sequence dependencies.Additionally，many methods overlook the influence of learner characteristics and exercise features on experimental results.Addressing these issues，a cross-attention mechanism knowledge tracing model was proposed.The model integrated knowledge points and exercise features information to obtain a question feature embedding module.Subsequently，improvements were made to the attention mechanism based on learner responses，resulting in a dual attention mechanism module.To account for real exercise-solving situations，a guess-error module based on attention mechanisms was introduced.firstly，the model took in exercise features information，obtaining a learner response with integrating exercise information through the exercise features embedding module.Following processing by the guess-error module，authentic learner responses were derived.Finally，the prediction module yielded the probability of a learner answering correctly in the next instance.Experimental results demonstrate that the cross-attention knowledge tracing model，incorporating exercise features，outperformed the traditional dynamic keyhole transformer （DKT） model，with 3.13% increase in AUC and 3.44% increase in ACC.This model proves effective in handling long sequence dependencies while exhibiting enhanced interpretability and predictive performance.

In order to meet the requirements of hailstone ingestion test verification for domestic turbofan engines and obtain effective hailstone ingestion capability verification，according to the relevant requirements of hailstone ingestion test in GJB241A-2010 and CCAR-33R2 standards，the design method of hailstone ingestion test scheme of turbofan engine was studied，and the experience of the design process was summarized.Based on the analysis of the connotation of standards related to hailstone ingestion tests and practical application scenarios，the design methods of the test parameters，such as hailstone quantity，hailstone speed，engine working state，and impact position were studied.By studying the hailstone projection scheme in relevant foreign tests，the design and evaluation method of the hailstone projection scheme was determined and the corresponding simulation analysis method was defined.The test scheme designed according to the method was verified on a certain engine，which can provide reference and guidance for the design of the hailstone ingestion test scheme of the turbofan engine.

Aiming at the current target detection methods based deep learning for transmission line，the feature extraction ability is poor for small target，easy to misdetection leakage detection，detection accuracy is low， detection speed is slow.A transmission line target detection method was proposed based on an improved neural network model YOLOv7.Firstly，the MobileNetV2 network was used as the feature extraction part of YOLOv7 to achieve lightweight processing of the model.Secondly，the CA mechanism and ASPP module were introduced to improve the accuracy and perception of the model.Finally，the self-drawn transmission line obstacle data set was used for training.Improved YOLOv7 network andare compared with the original YOLOv7 model.The results show that the algorithm proposed has significantly improved the accuracy and recall rate，which meets the fault detection in complex scenarios and is more conducive to model deployment of mobile devices and embedded systems.

Considering that the causal mechanism of aviation accidents is complicated and has many causal factors with strong grey characteristics， the traditional gray prediction model is only applicable to univariate prediction and has the defect of low prediction accuracy.A method of aviation safety prediction was proposed based on a multivariate grey model optimized by genetic algorithm. Firstly， the analysis method of fishbone diagram was applied from the perspective of SHEL model to determine the factors affecting aviation safety， and the correlation coefficient matrix visualization graph was used to further screen the key causative factors. Secondly， a multivariate grey aviation safety prediction model was constructed with human factors， environmental factors， equipment and facility factors， external influencing factors and as the strong input indexes of the prediction model， and the optimal solution of the model’s undetermined parameter r was searched globally and parallel by genetic algorithm. Finally， simulation experiments were conducted utilizing Chinese civil aircraft accident rate of 10 000 and aviation unsafe event statistics from 2007 to 2016. Predictive comparisons were then made between two gray prediction models， GM（1，1）and MGM（1，n）. The findings indicate that compared to the traditional gray model， the proposed method demonstrates an average prediction error of around 1.6% in the aviation safety short-time prediction， showcasing the effectiveness and high accuracy of the proposed method.

Thermal contact resistance between aircraft components is unavoidably.It is very important to acquire thermal contact resistance accurately for detailed design of thermal structure.To overcome the measurement difficulty under high temperature and high pressure，a measuring device was independently developed based on static heat flow method.The device can effectively obtain the thermal contact resistance between solid structures under given interface pressure and hot surface temperature up to 1 500 ℃.By using this device，three kinds of thermal contact resistance tests were carried out successfully.The influence law of pressure，temperature and interface roughness on the thermal contact resistance were figured out.Experimental results show that the measurement device is stable and reliable with uncomplicated operation，while the simulation accuracy for temperature and pressure load is satisfied.This device can provide supports for the development and engineering application of aircraft thermal structures.

In response to the crack at the installation side of a certain engine pipeline bracket， the research direction for solving the failure problem was clarified through analysis of the fracture at the bracket crack and the static strength stress of the bracket components. From aspects of part design， processing and assembly， the cause of the problem was determined to be the interaction stress between the bracket and the pipeline， which made the bracket work under original vibration stress and additional stress introduced during assembly. Due to the strong rigidity of the bracket components and weak buffering under stress load， the stress generated from bracket elastic deformation would be transmitted to the hole edge， probably causing the formation of cracks. The improvement with floating structure was designed which has the ability of axial deformation compensation and can significantly reduce the stress transmission. The proposed optimization design was validated by experiments and tests.

Using the carbon fiber-reinforced composite tank can remarkably reduce the weight of the launch vehicle.However，the analysis method for cryogenic tanks subjected to mechanical-thermal loads remains to be studied，especially to accurately consider the microthermal stress produced between fiber and matrix in the cryogenic environment.A representative volume model containing multiple fibers was adopted，combined with the matrix and fiber failure criteria，to establish a microscopic stress field and failure prediction model.The k-means clustering method was used for dimensionality reduction calculation，and an efficient and high-fidelity trans-scale analysis method for composite tanks was proposed.The results of illustrative examples show that the proposed method can accurately predict the elastic constants and failure strength of the composite single-layer plate according to the thermal and mechanical constants of fiber and matrix.The leakage failure process of a composite tank subjected to mechanical-thermal loads was simulated，and the critical load and failure state were given.

The strength degradation law of composite materials is very important for studying the structural fatigue life. However， some parameters involved in existing residual strength models are confirmed using residual strength test data， so the cost of model construction is high. By exploring the relationship between fatigue life and residual strength of composite materials， taking the cumulative distribution function of fatigue life as the starting point， a damage degree with clear physical meaning was constructed，and then an improved residual strength model was proposed. The improved residual strength model didn’t need residual strength test data but only initial static strength and fatigue life data. A fatigue life prediction model called S-N-φ model was constructed based on the improved residual strength model， which can consider the influence of initial static strength on fatigue life. The results show that the prediction accuracy of the improved residual strength model is satisfactory， and the S-N-φ model has better prediction accuracy than the classical S-N curve model.

The changes in stiffness and damping of the bolted joint surface can cause the changes of dynamic characteristics of the whole bolted joint structure，so it is of great practical value to accurately obtain the dynamic parameters of the joint surface in engineering.Based on the bolted structure，an improved equivalent model of the stiffness of the joint surface with uneven distribution in the range of bolt preload was proposed，and the distribution of different stiffness matrix elements in the finite element modeling was analyzed.The stiffness parameter identification of bolted joint surface was carried out by combining experiment and finite element analysis method.The results show that the increase of the number of stiffness matrix elements can improve the accuracy of natural frequency solutions to some extent.At the same time，considering the uneven stiffness distribution of the joint surface in the range of bolt preload，the equivalent modeling accuracy of the bolted joint structure can be effectively improved.

In order to solve the problem that the classical K-means clustering algorithm reguired users to know the number of clusters in advance and the clustering results were sensitive to initialization of the algorithm， a comprehensive scheme was proposed to improve the random initial partitioning of K-means algorithm and visually determine the number of clusters. Firstly， the data was standardized to make it obey normal distribution， and the most important features were extracted by principal component analysis to achieve dimensionality reduction of high-dimensional data. Then， the farthest centroid selection and min-max distance rule were used to modify the random initialization of K-means algorithm to avoid empty clusters and ensure data separability. Based on these， the statistical empirical rule was used to estimate the range of the number of clusters， and the optimal number of clusters was assessed by searching the elbow of sum-of-squared-error curve within this range. Finally， by calculating and comparing the silhouette coefficients of each cluster， the clustering quality of the algorithm was evaluated， thereby ultimately determining the inherent number of clusters in the data. The simulation results show that the proposed scheme can not only visually determine the potential number of clusters in the data， but also provide an effective method for high-dimensional data analysis in the era of big data.

Based on the NNW-PHengLEI software，for surfaces with uniformly distributed rough elements，rough surface boundary condition corrections and transition momentum thickness Reynolds numbers were introduced into the k-ω SST turbulence model and the γ-Re_θt transition model respectively.At the same time，some rough surface settings were added to achieve local/global rough surface boundary layer flow simulation capability.Test results show that when an airfoil is in a fully turbulent state，rough surfaces can reduce the stall angle of attack by 4°，and an equivalent roughness height of 1×10^-3 of chord length can cause a decrease in maximum lift coefficient by approximately 36%.Additionally，rough surfaces can advance the boundary layer transition，leading to an increase in skin friction drag.The computational analysis proves that the rough surface boundary condition proposed by Hellsten et al.can reflect the roughness effect more accurately than the dissipation rate boundary condition.The computational aerodynamic forces are consistent with the experimental data，and the flow field simulation results conform to the flow characteristics of rough surfaces.In addition，Hellsten’s model makes it possible to ensure stable coupling simulation with the transition model and its rough surface correlation function.The simulation results for the transition onset location and its trend with roughness level and freestream turbulent intensity are consistent with the experimental data.Still，the transition length needs further verification and correction through more rigorous experiments.

The damage mechanism of fiber reinforced resin matrix composites is complex.To ensure long-term stable application，advanced health monitoring technology must be used to monitor the damage.A sensor based on carbon nanopaper can sensitively monitor resistance changes and impact damage on carbon fiber reinforced polymer （CFRP）composite.A damage monitoring system based on an artificial neural network （ANN） deep-learning algorithm was designed.Through data analysis，the system could effectively monitor the occurrence and location of CFRP damage for a long time，and the damage location accuracy was as high as 92%.It can be inferred that the damage monitoring system can evaluate the health status of composite materials.

To study the evacuation characteristics in the Airbus 330 cabin crew under different hatch opening conditions，the Pathfinder software was used to simulate the evacuation characteristics.The evacuation characteristics of personnel were explored in four situations： all cabin doors opened， emergency doors could not be opened， front cabin doors could not be opened， and rear cabin doors could not be opened.The results show that the simulation time for evacuation is 63.3 s when all hatches are open，while 118.3 s when the emergency hatch cannot be opened.The per-unit-time flow of people at No.3，No.4，No.7，and No.8 hatches decreases by 0.09 people，0.2 people 0.09 people，0.04 people respectively.It takes 64s to evacuate people when the front cabin door could not be opened.The number of people evacuated from No.3，No.4，No.5，and No.6 doors are 47，48，52，and 50 respectively，indicating that the failure to open the front cabin door does not have no significant impact on the evacuation time.When the rear cabin door could not be opened，the evacuation time is 112.3s.The time interval from the first person to the last person from the No.3，No.4，No.5 and No.6 doors is 67.7 s，67.6 s，109.9 s and 107.1 s，respectively.This indicates that more passengers in the economy class choose to escape from the hatches No.5 and No.6，which increases the evacuation time.The research results provide a reference for understanding the evacuation characteristics of personnel under different hatch opening conditions.