Intelligent diagnosis and early warning method for quality safety hazards in ship assembly construction

Abstract

The present invention relates to the field of shipbuilding technology, and discloses an intelligent diagnosis and early warning method for quality and safety hazards in ship assembly construction. The method receives real-time sensor data from multiple detection nodes in ship construction, analyzes it through a pre-trained quality feature extraction model, and generates a hidden danger feature vector and risk level. Based on this, a hidden danger diagnosis strategy is dynamically generated, including detection equipment scheduling instructions and defect repair priority sequence. The pre-trained hidden danger ship dynamics model is used to predict the defect diffusion trend, adjust the dynamic response threshold, and optimize through an adaptive genetic algorithm to update the repair priority sequence and output quality early warning instructions to the ship construction management system. The method can accurately diagnose and warn of quality and safety hazards, reasonably dispatch resources, improve the quality and safety of ship construction, reduce costs, and promote the intelligent development of the shipbuilding industry.

Description

Technical Field

[0001] The present invention relates to the technical field of shipbuilding, and in particular to an intelligent diagnosis and early warning method for quality safety hazards in ship assembly construction. Background Art

[0002] Shipbuilding is a complex and highly systematic project, involving numerous steps and a vast technical ecosystem. Effective diagnosis and early warning of quality and safety hazards are crucial during the shipbuilding process, directly impacting the ship's navigational safety, service life, and economic profitability. However, traditional shipbuilding quality and safety inspection and early warning methods have numerous limitations.

[0003] In the early days, shipbuilding quality inspections relied primarily on manual experience. Inspectors relied on visual inspections and simple measurement tools to determine structural defects and weld quality. This approach was not only inefficient but also significantly affected by human factors. Inspectors varied in their technical proficiency and work habits, making it difficult to standardize judgments and potentially overlook potential quality and safety hazards. For example, manual inspections can be difficult to accurately detect small cracks in a ship's internal structure or internal defects in welds. These hazards can gradually escalate over the long term, potentially leading to serious safety incidents.

[0004] With technological advancements, simple instruments and equipment, such as ultrasonic flaw detectors and magnetic particle detectors, have been applied to shipbuilding inspections. These devices have improved inspection accuracy and efficiency to a certain extent, enabling the detection of defects that are difficult to detect manually. However, they still have significant limitations. They can only detect specific types of defects and cannot comprehensively analyze the various parameters involved in the shipbuilding process. During shipbuilding, structural stress, welding deformation, material defects, and environmental factors are interrelated and mutually influential. A single piece of testing equipment cannot fully assess the overall impact of these factors on ship quality and safety.

[0005] Furthermore, traditional early warning methods are often based on fixed threshold settings. Once detection data exceeds a preset threshold, an early warning signal is issued. However, in actual shipbuilding, the development trends and severity of hidden dangers vary under different operating and environmental conditions. Fixed thresholds cannot be dynamically adjusted based on real-time conditions, making them prone to false alarms or missed alarms. For example, in severe weather conditions, the stress on the ship's structure will change, and the previously appropriate stress threshold may no longer apply. In such cases, fixed threshold early warning methods are unable to accurately identify quality and safety hazards.

[0006] As shipbuilding projects become increasingly large and complex, traditional methods for diagnosing and warning of quality and safety hazards are no longer sufficient to meet the demands of modern shipbuilding. On the one hand, large ships are long construction cycles and high costs, and any quality and safety issues can lead to significant economic losses. On the other hand, with the development of the shipping industry, the safety and reliability requirements for ships are becoming increasingly stringent. Therefore, there is an urgent need for an intelligent method that can comprehensively analyze multiple data, dynamically assess hazard risks in real time, and accurately warn and efficiently address quality and safety hazards to ensure the quality and safety of ship assembly and construction. Summary of the Invention

[0007] The purpose of the present invention is to provide an intelligent diagnosis and early warning method for quality safety hazards in ship assembly construction, so as to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent diagnosis and early warning method for ship assembly construction quality safety hazards, the method comprising:

[0009] Receive real-time sensor data from multiple detection nodes during the shipbuilding process, including structural stress, welding deformation, material defect parameters, and environmental monitoring indicators;

[0010] Based on the pre-trained quality feature extraction model, a multi-dimensional fusion analysis is performed on the real-time sensor data to generate the hidden danger feature vector and risk level of each detection node;

[0011] Dynamically generate a hidden danger diagnosis strategy based on the risk level, the strategy including inspection equipment scheduling instructions and defect repair priority sequence;

[0012] Based on the pre-trained hidden danger propagation dynamics model, the diffusion trend of defects in a preset time period in the future is predicted, and the dynamic response threshold in the hidden danger diagnosis strategy is adjusted;

[0013] The dynamic response threshold is optimized by an adaptive genetic algorithm, the defect repair priority sequence is updated, and the final quality warning instruction is output to the ship construction management system.

[0014] Preferably, the steps of constructing the quality feature extraction model include:

[0015] Collect historical ship construction defect data and construct a training dataset containing defect types, geometric dimensions, and environmental correlation factors;

[0016] Performing spatial feature extraction on the training data set through a residual neural network to generate an initial defect feature map;

[0017] Combining the attention mechanism to weight the initial defect feature map and generate multi-scale fusion features;

[0018] The multi-scale fusion feature is combined with a time series analysis module to generate the quality feature extraction model that can be updated online.

[0019] Preferably, the dynamically generated hidden danger diagnosis strategy includes:

[0020] Assigning an initial repair weight to each detection node based on the risk level;

[0021] Calculate the repair efficiency coefficient of each node based on real-time equipment availability data and personnel scheduling status;

[0022] Nonlinearly superimposing the repair efficiency coefficient and the initial repair weight to generate the defect repair priority sequence;

[0023] The detection device scheduling instruction is generated according to the priority sequence result and combined with the device load balancing constraint.

[0024] Preferably, the steps of constructing the hidden danger propagation dynamics model include:

[0025] Collect historical defect diffusion data and environmental disturbance data from ship block construction to construct a spatiotemporal evolution dataset;

[0026] Performing frequency domain feature decomposition on the spatiotemporal evolution data set using a wavelet transform algorithm to extract critical frequency components of defect diffusion;

[0027] Combined with finite element simulation data, a structural stress propagation matrix is ​​established to quantify the defect correlation strength between different detection nodes;

[0028] The critical frequency component and the defect correlation strength are input into a time-series graph convolutional network to generate the hidden danger propagation dynamics model.

[0029] Preferably, the method further comprises:

[0030] Identifying detection nodes whose diffusion rates exceed a preset threshold based on the prediction results of the hidden danger propagation dynamics model;

[0031] assigning high-risk labels to identified nodes in the hidden danger diagnosis strategy;

[0032] Based on the high-risk marker, a redundant detection resource allocation mechanism is triggered, and low-priority detection tasks are suspended.

[0033] Preferably, the calculation of the repair efficiency coefficient includes:

[0034] Obtain equipment failure rate, personnel skill level and environmental interference factors to build a dynamic performance evaluation matrix;

[0035] The improved Dijkstra algorithm is used to calculate the optimal resource scheduling path between nodes and generate the basic repair efficiency value;

[0036] Performing a tensor product operation on the dynamic performance evaluation matrix and the basic repair performance value to obtain the repair efficiency coefficient;

[0037] The calculation formula of the repair efficiency coefficient is:

[0038]

[0039] Where Q represents the repair efficiency coefficient, ρ j represents the inverse of the real-time failure rate of the jth type of equipment, σ j represents the skill level weight of the j-th type of personnel, ϵ b represents the basic repair effectiveness value, and m represents the total number of equipment categories.

[0040] Preferably, the generation of the multi-scale fusion feature includes:

[0041] performing channel dimension normalization processing on the initial defect feature map to eliminate dimensional differences;

[0042] Extract defect saliency features at different scales through a multi-head attention mechanism;

[0043] The feature pyramid network is used to fuse the salient features across scales and reconstruct hierarchical defect representation.

[0044] Preferably, the execution of the adaptive genetic algorithm includes:

[0045] Setting an objective function for defect repair priority optimization, which includes repair timeliness and resource consumption constraints;

[0046] The adaptability of individuals in the population is evaluated through the elite retention strategy, and the crossover and mutation probabilities are dynamically adjusted;

[0047] In each round of iteration, a non-inferior solution set is screened based on the Pareto front, and the dynamic response threshold is updated;

[0048] Outputting the quality warning instruction that meets the multi-objective convergence conditions;

[0049] Wherein, the objective function is:

[0050]

[0051] In the formula, G represents the objective function value, R t Indicates the repair time rating, C r represents the resource consumption score, λ and μ are the normalization coefficients of timeliness and resource consumption, respectively.

[0052] Preferably, the method further comprises:

[0053] After assigning the high-risk tag, monitor the defect repair progress of the corresponding node in real time;

[0054] If the repair progress lags behind the preset threshold, the dynamic resource reallocation mechanism will be activated to increase the number of detection equipment and adjust personnel configuration.

[0055] Preferably, the setting of the resource consumption constraint includes:

[0056] Collect energy consumption data and labor-hour costs of each testing device to build a resource consumption baseline model;

[0057] Calculate the upper limit of dynamic resource quota based on the real-time task load rate;

[0058] Taking the upper limit of the dynamic resource quota as a constraint condition of the objective function;

[0059] The calculation formula for the upper limit of the dynamic resource quota is:

[0060]

[0061] Where C max represents the upper limit of dynamic resource quota, η represents the energy consumption coefficient of the device per unit time, P d represents the real-time task load rate, τ is the labor cost conversion factor, and ω is the emergency resource reserve constant.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] The intelligent diagnosis and early warning method for quality and safety hazards in ship assembly construction proposed in the present invention has significant beneficial effects in many aspects. In terms of data processing and hazard diagnosis, the method receives real-time sensor data from multiple detection nodes during the ship construction process, including structural stress, welding deformation, material defect parameters and environmental monitoring indicators, and uses a pre-trained quality feature extraction model to perform multi-dimensional fusion analysis. This process can fully explore the potential connections between various types of data, and greatly improves the accuracy of extracting quality and safety hazard features compared to traditional methods that only rely on a single or a small amount of data for judgment. For example, traditional methods may only focus on single data of welding deformation, ignoring the influence of structural stress and environmental factors on it, while the present invention can comprehensively consider these factors and more accurately judge whether there are quality and safety hazards in the welding part. The generated hazard feature vector and risk level are also more accurate and reliable, providing a solid data foundation for subsequent processing.

[0064] In terms of diagnostic strategy generation and resource scheduling, dynamically generating hidden danger diagnostic strategies based on risk levels has great advantages. An initial repair weight is assigned to each detection node, and the repair efficiency coefficient is calculated in combination with real-time equipment availability data and personnel scheduling status, thereby generating a defect repair priority sequence. This approach fully considers the actual repair resources and personnel situation. Unlike the traditional method of uniform repair order or random repair arrangement, the present invention can give priority to nodes with high risk and high repair efficiency, effectively improving the overall efficiency of the repair work. At the same time, combined with the equipment load balancing constraint to generate detection equipment scheduling instructions, it avoids excessive use or idleness of equipment, improves the utilization rate of detection equipment, reduces equipment maintenance costs, and enables more reasonable allocation of resources.

[0065] In terms of hidden danger prediction and dynamic adjustment, a pre-trained hidden danger propagation dynamics model predicts the spread of defects within a preset time period and adjusts the dynamic response threshold, demonstrating strong forward-looking capabilities. By predicting the spread of defects in advance, appropriate measures, such as enhanced inspections and adjustments to construction processes, can be taken before hidden dangers escalate, effectively avoiding serious quality and safety issues caused by the spread of defects. Furthermore, the use of an adaptive genetic algorithm to optimize the dynamic response threshold further improves the accuracy and timeliness of early warnings. During the repair process, real-time monitoring and activation of a dynamic resource reallocation mechanism based on repair progress ensure smooth repair work, avoid delays due to insufficient resources, and safeguard the quality and progress of shipbuilding.

[0066] Overall, this approach improves the quality and safety of shipbuilding, reduces the risk of ship failures and accidents due to quality and safety issues, and extends the service life of ships. Economically, it effectively reduces the costs of shipbuilding and subsequent maintenance, avoids the additional costs of rework due to quality issues, and reduces operational losses due to ship failures. Furthermore, it promotes the development of intelligent and efficient shipbuilding technology, providing strong technical support for the sustainable development of the shipbuilding industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a working principle diagram of the intelligent diagnosis and early warning method for ship assembly construction quality safety hazards according to the present invention;

[0068] Figure 2 Diagram of the steps for constructing the hidden danger propagation dynamics model;

[0069] Figure 3 A flowchart of treatment measures based on prediction results;

[0070] Figure 4 A flowchart for monitoring the repair progress of high-risk nodes. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] See also Figure 1-Figure 4 The present invention provides an intelligent diagnosis and early warning method for quality and safety hazards in ship assembly construction, aiming to achieve efficient diagnosis and accurate early warning of quality and safety hazards in the ship construction process. The specific steps are as follows:

[0073] Real-time sensor data reception: During the shipbuilding process, sensors are deployed at multiple key inspection nodes to collect various real-time sensor data. These inspection nodes are located in various structural locations of the ship, covering areas that have a significant impact on the ship's quality and safety. The collected sensor data includes structural stress data, which monitors the magnitude and distribution of stress experienced by the ship's structure during construction; welding deformation data, which reflects the deformation caused by the welding process; material defect parameters, which detect the presence and characteristics of internal defects in the material; and environmental monitoring indicators, such as temperature, humidity, wind speed, and other environmental factors at the construction site. This real-time sensor data is transmitted to the data processing system via a data transmission network.

[0074] Hidden danger feature vector and risk level generation: Received real-time sensor data is processed using a pre-trained quality feature extraction model. This model, trained on historical shipbuilding data, possesses powerful feature extraction capabilities. It performs a multi-dimensional fusion analysis of real-time sensor data, comprehensively considering the correlations and characteristics between different types of data, to generate a hidden danger feature vector for each detection node. Furthermore, based on the characteristics of the hidden danger feature vector and pre-defined risk assessment rules, the risk level of each detection node is determined. Risk levels can be categorized as low, medium, or high, enabling targeted follow-up measures.

[0075] Dynamic Generation of Hidden Danger Diagnosis Strategies: Based on the determined risk level, a hidden danger diagnosis strategy is dynamically generated. This strategy consists of two key components: inspection equipment scheduling instructions and a defect repair priority sequence. Regarding inspection equipment scheduling instructions, based on the risk level of each inspection node and the current distribution and availability of inspection equipment, inspection equipment is rationally dispatched to the corresponding node for further inspection, ensuring efficient resource utilization. Regarding the defect repair priority sequence, a comprehensive consideration of factors such as risk level, repair difficulty, and impact on the overall quality of the ship is used to determine the repair order for each defective inspection node, ensuring that defects with the greatest impact on ship quality and safety are addressed first.

[0076] Defect Proliferation Trend Prediction and Strategy Adjustment: Leveraging a pre-trained defect propagation dynamics model, we predict the future defect propagation trend over a preset time period. This model combines historical defect propagation data with current shipbuilding practices to simulate and predict defect propagation under various conditions. Based on these predictions, we adjust the dynamic response threshold within the defect diagnosis strategy. If a defect is predicted to spread rapidly, the dynamic response threshold is lowered to enable more timely detection and resolution of potential issues. Conversely, if a defect is predicted to spread slowly, the dynamic response threshold is appropriately raised to optimize resource allocation.

[0077] Dynamic response threshold optimization and early warning output: An adaptive genetic algorithm optimizes the dynamic response threshold. Based on constraints such as repair timeliness and resource consumption, the algorithm continuously adjusts the dynamic response threshold to achieve optimal diagnostic and early warning results. During the optimization process, the defect repair priority sequence is updated to ensure a more balanced and optimal repair. Ultimately, the optimized results are generated into a quality early warning instruction and output to the shipbuilding management system. Upon receiving the early warning instruction, the shipbuilding management system can take timely measures, such as dispatching maintenance personnel to carry out repairs and adjusting the construction schedule, to ensure the quality and safety of shipbuilding.

[0078] The implementation of the present invention will be further described below with reference to Examples 1 to 6.

[0079] Example 1:

[0080] In this embodiment, the construction of the quality feature extraction model and the generation process of multi-scale fusion features are emphasized.

[0081] The construction of a quality feature extraction model is crucial for accurately diagnosing quality and safety hazards in shipbuilding. First, extensive historical shipbuilding defect data is collected. This data covers a wide range of defect types, including but not limited to welding defects and internal cracks in materials. The defect's geometric dimensions, such as length, width, and depth, are also recorded, along with environmental factors such as ambient temperature and humidity during construction, and welding current and voltage. This data is used to construct a training dataset encompassing defect types, geometric dimensions, and environmental factors.

[0082] Next, a residual neural network is used to extract spatial features from the training dataset. Residual neural networks have powerful feature learning capabilities and can effectively extract spatial features from data. During this process, they perform layer-by-layer convolution operations on the input training data, gradually extracting deep spatial features and generating an initial defect feature map. This initial defect feature map contains basic spatial feature information about the defects, but it is not comprehensive or accurate enough.

[0083] To further optimize features, an attention mechanism is incorporated to assign weights to the initial defect feature map. This mechanism automatically focuses on important features in the data, assigning higher weights to these key features. When processing shipbuilding defect data, it can highlight features that are more critical to defect diagnosis, such as defect edge features and abnormal features at specific locations, thereby generating multi-scale fused features.

[0084] When generating multi-scale fusion features, the specific operations are as follows: First, the initial defect feature map is normalized by channel dimension. Since data from different channels may have different dimensions and numerical ranges, normalization can eliminate dimensional differences and make different features comparable. Then, a multi-head attention mechanism is used to extract salient defect features at different scales. The multi-head attention mechanism analyzes data from multiple perspectives, extracting richer feature information. Finally, a feature pyramid network is used to fuse these salient features across scales to reconstruct a hierarchical defect representation. The feature pyramid network can effectively fuse features at different scales to form a more comprehensive and representative defect feature representation, providing more accurate data support for subsequent hidden danger diagnosis.

[0085] Combining multi-scale fusion features with a time series analysis module generates an online updateable quality feature extraction model. The time series analysis module captures data trends over time. Given the dynamic nature of shipbuilding, data changes at different time points are crucial for hazard diagnosis. By combining this module with multi-scale fusion features, the quality feature extraction model can continuously adapt to new data, enabling online updates and improving the accuracy and timeliness of shipbuilding quality and safety hazard diagnosis.

[0086] Example 2:

[0087] When dynamically generating a hidden danger diagnosis strategy, each detection node is first assigned an initial repair weight based on its risk level. Nodes with higher risk levels are assigned a higher initial repair weight, while nodes with lower risk levels are assigned a lower initial repair weight. This ensures that defects with higher risks are addressed first.

[0088] Then, based on real-time equipment availability data and personnel scheduling status, the repair efficiency coefficient for each node is calculated. Equipment failure rate, personnel skill level, and environmental interference factors are obtained to construct a dynamic performance evaluation matrix. The equipment failure rate reflects the likelihood of equipment failure under current usage. The lower the equipment failure rate, the higher the reliability of its normal operation during the repair process, and the greater the positive impact on repair efficiency. The personnel skill level weight reflects the contribution of personnel with different skill levels to the repair work. Personnel with higher skill levels are likely to have higher repair efficiency and quality. The environmental interference factor takes into account the impact of environmental factors at the construction site on the repair work. For example, severe weather conditions may reduce repair efficiency.

[0089] The improved Dijkstra algorithm is used to calculate the optimal resource scheduling path between each node and generate a basic repair efficiency value. The Dijkstra algorithm is a classic shortest path algorithm. In this scenario, it is used to find the optimal path from the resource supply point to each detection node to determine the optimal resource scheduling method, thereby obtaining the basic repair efficiency value. The basic repair efficiency value reflects the repair efficiency that can be achieved by resource scheduling under ideal conditions, without considering equipment failure rate, personnel skill level, and environmental interference factors.

[0090] Perform tensor product operation on the dynamic performance evaluation matrix and the basic repair performance value to obtain the repair efficiency coefficient. The calculation formula of the repair efficiency coefficient is:

[0091]

[0092] Among them, Q represents the repair efficiency coefficient, which comprehensively reflects the impact of multiple factors such as equipment, personnel and environment on the repair efficiency. j It represents the inverse of the real-time failure rate of the j-th type of equipment. The larger the inverse of the real-time failure rate, the more reliable the equipment is and the greater the improvement in repair efficiency. j Represents the skill level weight of the j-th type of personnel, which is used to measure the contribution of personnel with different skill levels to the repair efficiency. b represents the foundation repair effectiveness value, which is calculated by the improved Dijkstra algorithm. m represents the total number of equipment categories, covering various types of equipment used in the shipbuilding process.

[0093] The repair efficiency coefficient is nonlinearly superimposed with the initial repair weight to generate a defect repair priority sequence. This nonlinear superposition takes into account both repair weight and repair efficiency, making the repair priority sequence more reasonable. Nodes with high repair efficiency and large initial repair weights are positioned higher in the repair priority sequence.

[0094] Finally, based on the priority sequence results and combined with device load balancing constraints, we generate dispatch instructions for the detection devices. When generating dispatch instructions, we must prioritize testing high-priority nodes while also considering device load balancing to avoid overuse of some devices while leaving others idle, thereby improving the overall utilization efficiency of the detection equipment.

[0095] Example 3:

[0096] This embodiment describes in detail the process of constructing a hidden danger propagation dynamics model. The hidden danger propagation dynamics model is of great significance for predicting the diffusion trend of defects and taking preventive measures in advance.

[0097] We extensively collect historical defect diffusion data and environmental disturbance data from ship block construction. This historical defect diffusion data records the diffusion of defects over time and environmental changes during past ship construction processes, including information such as the diffusion rate, direction, and range of the defects. Environmental disturbance data includes data on the impact of environmental factors such as temperature fluctuations, humidity changes, and external forces on defect diffusion during the construction process. This data is used to construct a spatiotemporal evolution dataset.

[0098] The wavelet transform algorithm is used to decompose the spatiotemporal evolution data set into frequency domain features. The wavelet transform algorithm can decompose complex time series data into components of different frequencies and extract the critical frequency components for defect diffusion. These critical frequency components reflect the key frequency characteristics of the defect diffusion process and are important for understanding the inherent mechanisms and laws of defect diffusion.

[0099] A structural stress propagation matrix was established by combining finite element simulation data. Finite element simulation discretizes the ship structure to simulate the stress distribution and changes in the structure under various operating conditions. Based on the simulation results, the stress transfer relationship between different detection nodes was calculated, and the defect correlation strength between different detection nodes was quantified. The defect correlation strength indicates the degree to which a defect at one node affects other nodes. A greater correlation strength indicates a higher probability of defect propagation between the two nodes.

[0100] The critical frequency components and defect correlation strengths are fed into a time-series graph convolutional network. This network combines the strengths of graph convolutional networks and time series analysis, effectively processing data with spatiotemporal characteristics. In this embodiment, it uses the critical frequency components and defect correlation strengths to learn the propagation patterns of defects in both time and space, generating a dynamic model of potential danger propagation. This model can accurately predict the future spread of defects within a preset time period, providing a powerful basis for adjusting potential danger diagnosis strategies.

[0101] Example 4:

[0102] This example mainly introduces relevant treatment measures based on the prediction results of the hidden danger propagation dynamics model, as well as the monitoring of the repair progress of high-risk nodes and the dynamic resource reallocation mechanism. Specifically, it includes:

[0103] Based on the predictions from the hidden danger propagation dynamics model, we identify detection nodes whose diffusion rates exceed a preset threshold. This threshold is set based on extensive historical data and experience to determine the severity of the defect's spread. If the defect diffusion rate at a node exceeds the preset threshold, it indicates that the defect at that node poses a significant potential risk and warrants special attention.

[0104] Assign high-risk tags to identified nodes in the hidden danger diagnosis strategy. High-risk tags can clearly distinguish these high-risk nodes, allowing for special measures to be taken in subsequent processing. For example, these high-risk nodes can be prioritized when developing inspection and repair plans.

[0105] Based on high-risk flags, a redundant detection resource allocation mechanism is triggered, pausing low-priority detection tasks. This redundant detection resource allocation mechanism increases the frequency and detection methods of high-risk nodes, ensuring timely detection of defect changes. Furthermore, pausing low-priority detection tasks allows resources to be focused on addressing issues at high-risk nodes, improving resource utilization efficiency.

[0106] After assigning a high-risk tag, the corresponding node's defect repair progress is monitored in real time. This progress data can be obtained by installing monitoring equipment at the repair site or requiring maintenance personnel to regularly report on repair progress. The actual repair progress is compared with a preset threshold. If the repair progress lags behind the threshold, it indicates a problem with the repair work, possibly due to difficulties with repair or insufficient resources.

[0107] At this point, the dynamic resource reallocation mechanism is activated to increase the number of testing equipment and adjust staffing. Adding testing equipment can improve the accuracy and efficiency of testing and promptly identify new issues that may arise during the repair process. Adjusting staffing can, based on actual conditions, deploy more experienced or more appropriately skilled personnel to participate in the repair work, accelerating the repair process and ensuring the quality and safety of the ship's construction.

[0108] Example 5:

[0109] During the execution of the adaptive genetic algorithm, the objective function for defect repair priority optimization is first set. The objective function is:

[0110]

[0111] Among them, G represents the objective function value, which comprehensively considers two key factors: repair timeliness and resource consumption. tIndicates the repair timeliness score, which is used to measure the time efficiency of the repair work. The higher the repair timeliness score, the faster the repair work is completed. r λ represents the resource consumption score, reflecting the resources consumed during the repair process, such as the energy consumption of testing equipment and the labor cost of maintenance personnel. A lower resource consumption score indicates higher resource utilization efficiency. λ and μ are the normalization coefficients for timeliness and resource consumption, respectively. They are used to adjust the relative importance of repair timeliness and resource consumption in the objective function and can be set appropriately based on actual conditions.

[0112] An elite retention strategy evaluates the fitness of individuals in a population and dynamically adjusts crossover and mutation probabilities. This strategy ensures that individuals with high fitness in each generation are retained in the next generation during the genetic algorithm's iterations, preventing the loss of high-performing individuals. When evaluating the fitness of individuals in a population, the fitness value of each individual is calculated based on the objective function. Individuals with higher fitness values ​​have a greater probability of being selected for the next generation. Furthermore, the crossover and mutation probabilities are dynamically adjusted based on the fitness distribution of individuals in the population. If the fitness differences among individuals in a population are small, it indicates that the algorithm may be stuck in a local optimum. In this case, the mutation probability is appropriately increased to increase population diversity and help the algorithm escape the local optimum. If the fitness differences among individuals in a population are large, the mutation probability is appropriately reduced to accelerate the algorithm's convergence.

[0113] In each iteration, the Pareto front is used to screen the set of non-inferior solutions. In multi-objective optimization problems, the Pareto front is a set of solutions that cannot improve at least one objective by sacrificing other objectives. By screening the non-inferior solution set, a series of solutions that strike a good balance between repair timeliness and resource consumption can be obtained. Appropriate solutions are selected from these non-inferior solutions and the dynamic response threshold is updated.

[0114] Iterations continue until the multi-objective convergence criteria are met. These criteria can be customized to meet actual needs. For example, the algorithm is considered converged when the objective function value changes minimally within a certain number of iterations, or when the set of non-inferior solutions no longer changes significantly. Once the convergence criteria are met, a quality warning instruction is output, providing accurate decision-making for the shipbuilding management system.

[0115] Example 6:

[0116] This embodiment mainly introduces the process of setting resource consumption constraints and the calculation method and application of the upper limit of dynamic resource quota.

[0117] When setting resource consumption constraints, we first collect energy consumption data and labor-hour costs for each piece of testing equipment. Energy consumption data is collected by installing energy monitoring devices on the equipment, recording its energy consumption under different operating conditions. Labor-hour costs are calculated based on factors such as maintenance personnel wages and working hours. This data is used to construct a resource consumption baseline model, which reflects the resource consumption level required to complete a specific inspection and repair task under normal circumstances.

[0118] Then, the upper limit of the dynamic resource quota is calculated based on the real-time task load rate. The real-time task load rate reflects the busyness of the current shipbuilding task. The higher the task load rate, the more resources need to be invested. The calculation formula for the upper limit of the dynamic resource quota is:

[0119]

[0120] Among them, C max It represents the upper limit of dynamic resource quota, which limits the maximum amount of resources allowed to be consumed under the current task load rate. η represents the energy consumption coefficient of the device per unit time, which is used to measure the difference in energy consumption levels of different devices. d represents the real-time task load rate, calculated based on the actual situation of the current shipbuilding task. τ is the labor-hour cost conversion factor, used to convert labor-hour costs into a resource measurement unit consistent with equipment energy consumption. ω is the emergency resource reserve constant, which reserves a certain amount of resources for emergency response.

[0121] The dynamic resource quota cap is used as a constraint for the objective function. During the adaptive genetic algorithm optimization process, the generated repair solution is ensured to not exceed the dynamic resource quota cap in terms of resource consumption. This effectively controls resource usage, avoids excessive resource consumption, and ensures the economic and sustainable nature of the shipbuilding process. Furthermore, the dynamic resource quota cap is used as a basis for the development of inspection equipment scheduling instructions and defect repair plans, ensuring the rational allocation of resources and improving resource utilization efficiency.

[0122] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0123] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnosis and early warning method for ship assembly quality safety hazards, characterized by: include: Receive real-time sensor data from multiple detection nodes during the shipbuilding process, including structural stress, welding deformation, material defect parameters, and environmental monitoring indicators; Based on the pre-trained quality feature extraction model, a multi-dimensional fusion analysis is performed on the real-time sensor data to generate the hidden danger feature vector and risk level of each detection node; Dynamically generate a hidden danger diagnosis strategy based on the risk level, the strategy including inspection equipment scheduling instructions and defect repair priority sequence; Based on the pre-trained hidden danger propagation dynamics model, the diffusion trend of defects in a preset time period in the future is predicted, and the dynamic response threshold in the hidden danger diagnosis strategy is adjusted; The steps of constructing the hidden danger propagation dynamics model include: Collect historical defect diffusion data and environmental disturbance data from ship block construction to construct a spatiotemporal evolution dataset; Performing frequency domain feature decomposition on the spatiotemporal evolution data set using a wavelet transform algorithm to extract critical frequency components of defect diffusion; Combined with finite element simulation data, a structural stress propagation matrix is ​​established to quantify the defect correlation strength between different detection nodes; Inputting the critical frequency component and the defect correlation strength into a time-series graph convolutional network to generate the hidden danger propagation dynamics model; The dynamic response threshold is optimized by an adaptive genetic algorithm, a defect repair priority sequence is updated, and a final quality warning instruction is output to a shipbuilding management system. The execution of the adaptive genetic algorithm includes: Setting an objective function for defect repair priority optimization, which includes repair timeliness and resource consumption constraints; The adaptability of individuals in the population is evaluated through the elite retention strategy, and the crossover and mutation probabilities are dynamically adjusted; In each round of iteration, a non-inferior solution set is screened based on the Pareto front, and the dynamic response threshold is updated; Outputting the quality warning instruction that meets the multi-objective convergence conditions; Wherein, the objective function is: ； In the formula, G represents the objective function value, R t Indicates the repair time rating, C r represents the resource consumption score, λ and μ are the normalized coefficients of timeliness and resource consumption, respectively; The dynamically generated hidden danger diagnosis strategy includes: Assigning an initial repair weight to each detection node based on the risk level; Calculate the repair efficiency coefficient of each node based on real-time equipment availability data and personnel scheduling status; Nonlinearly superimposing the repair efficiency coefficient and the initial repair weight to generate the defect repair priority sequence; The detection device scheduling instruction is generated according to the priority sequence result and combined with the device load balancing constraint.

2. The intelligent diagnosis and early warning method according to claim 1, characterized in that: The steps of constructing the quality feature extraction model include: Collect historical ship construction defect data and construct a training dataset containing defect types, geometric dimensions, and environmental correlation factors; Performing spatial feature extraction on the training data set through a residual neural network to generate an initial defect feature map; Combining the attention mechanism to weight the initial defect feature map and generate multi-scale fusion features; The multi-scale fusion feature is combined with a time series analysis module to generate the quality feature extraction model that can be updated online.

3. The intelligent diagnosis and early warning method according to claim 1, characterized in that: Also includes: Identifying detection nodes whose diffusion rates exceed a preset threshold based on the prediction results of the hidden danger propagation dynamics model; assigning high-risk labels to identified nodes in the hidden danger diagnosis strategy; Based on the high-risk marker, a redundant detection resource allocation mechanism is triggered, and low-priority detection tasks are suspended.

4. The intelligent diagnosis and early warning method according to claim 1, characterized in that: The calculation of the repair efficiency coefficient includes: Obtain equipment failure rate, personnel skill level and environmental interference factors to build a dynamic performance evaluation matrix; The improved Dijkstra algorithm is used to calculate the optimal resource scheduling path between nodes and generate the basic repair efficiency value; Performing a tensor product operation on the dynamic performance evaluation matrix and the basic repair performance value to obtain the repair efficiency coefficient; The calculation formula of the repair efficiency coefficient is: ； Where Q represents the repair efficiency coefficient, ρ j represents the inverse of the real-time failure rate of the jth type of equipment, σ j represents the skill level weight of the j-th type of personnel, represents the basic repair effectiveness value, and m represents the total number of equipment categories.

5. The intelligent diagnosis and early warning method according to claim 2, characterized in that: The generation of the multi-scale fusion feature includes: performing channel dimension normalization processing on the initial defect feature map to eliminate dimensional differences; Extract defect saliency features at different scales through a multi-head attention mechanism; The feature pyramid network is used to fuse the salient features across scales and reconstruct hierarchical defect representation.

6. The intelligent diagnosis and early warning method according to claim 3, characterized in that: Also includes: After assigning the high-risk tag, monitor the defect repair progress of the corresponding node in real time; If the repair progress lags behind the preset threshold, the dynamic resource reallocation mechanism will be activated to increase the number of detection equipment and adjust personnel configuration.

7. The intelligent diagnosis and early warning method according to claim 6, characterized in that: The setting of the resource consumption constraint includes: Collect energy consumption data and labor-hour costs of each testing device to build a resource consumption baseline model; Calculate the upper limit of dynamic resource quota based on the real-time task load rate; Taking the upper limit of the dynamic resource quota as a constraint condition of the objective function; The calculation formula for the upper limit of the dynamic resource quota is: ； Where C max represents the upper limit of dynamic resource quota, η represents the energy consumption coefficient of the device per unit time, P d represents the real-time task load rate, τ is the labor cost conversion factor, and ω is the emergency resource reserve constant.

Images