Method and system for repairing sound force performance of covering layer

Abstract

The invention provides a sound force performance repairing method and system for a covering layer. The method comprises the steps that material aging basic parameters, the initial sound absorption rate and the initial anti-explosion strength of a target covering layer are obtained; based on the data, obtaining a material aging model and constructing a first full-life digital twinborn model to obtain a first sound force performance predicted value, and if a performance threshold value smaller than a preset first performance threshold value exists, determining the located area as a sound force performance weak area; constructing an original agent based on the target covering layer; acquiring a state space based on the original agent; and based on the data, obtaining an intelligent agent to generate a first path repairing strategy, obtaining a second sound force performance predicted value, and if the first path repairing strategy and the second sound force performance predicted value are greater than or equal to a preset second performance threshold, realizing repairing. According to the method for repairing the sound force performance of the covering layer, the sound force performance of the covering layer can be accurately predicted, dynamic self-repairing is carried out on a weak area, and it is guaranteed that the sound force performance of the whole life cycle of the covering layer stably reaches the standard.

Description

Technical Field

[0001] This invention relates to the field of acoustic performance monitoring and evaluation and intelligent self-repair technology for cover layers, and in particular to a method and system for repairing the acoustic performance of cover layers. Background Technology

[0002] Currently, the monitoring and maintenance of the acoustic-mechanical properties (such as sound absorption rate and blast resistance) of functional surface coatings on ships and underwater equipment mainly rely on two types of technologies: First, in terms of performance degradation prediction, existing technologies are mostly based on a single material aging model (such as aging equations based on temperature, humidity, and load) to predict performance degradation, or only obtain discrete data through periodic sampling inspections of actual ships, lacking integrated analysis of material aging models and actual operational monitoring data. Second, in terms of self-repair, existing technologies mostly adopt preset microcapsule repair agent release paths (such as fixed-area trigger release), without considering the dynamic distribution differences of coating performance degradation, resulting in a lack of real-time and targeted repair decisions.

[0003] Under the current technological background, there are problems such as low accuracy in predicting performance degradation and low self-repair efficiency: On the one hand, the aging model of a single material does not reflect the actual impact of complex working conditions of the actual vessel (such as water flow impact, acoustic vibration, etc.), and the data from periodic sampling tests are highly discrete, making it difficult to construct a continuous and accurate acoustic absorption rate-blast resistance attenuation curve, which in turn leads to a large deviation in the evaluation of the full life performance of the coating (the error often exceeds 15%). On the other hand, the preset release path cannot dynamically adapt to the changes in the weak areas of the coating, and the release of the repair agent is not targeted enough, which not only leads to a repair efficiency that is generally lower than 85%, but also easily results in repeated repairs or missed repairs. Summary of the Invention

[0004] The present invention aims to provide a method and system for repairing the acoustic performance of a cover layer to solve the above-mentioned technical problems. It can accurately predict the acoustic performance of the cover layer, dynamically self-repair weak areas of the cover layer, and ensure that the acoustic performance of the cover layer is stable and meets the standards throughout its entire life cycle.

[0005] To address the aforementioned technical problems, this invention provides a method for repairing the acoustic performance of a covering layer, comprising: Acquire several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance of the target coating layer; Based on the material aging parameters, initial sound absorption rate, and initial blast resistance, a material aging model is obtained. Based on the material aging model and the preset full life cycle, a first full life digital twin model is constructed and several first acoustic performance prediction values ​​are obtained. If there is a first acoustic performance prediction value that is less than the preset first performance threshold, it is determined as the first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as the acoustic performance weak area. Based on the target coverage layer, preset grid size, and preset reward parameters, an initial intelligent agent is constructed; based on the initial intelligent agent and the preset policy, a first action is obtained and executed, and the state space is obtained; Based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network, obtain the initial agent and action value loss set; Based on the action value loss set and the initial agent, obtain an agent to generate the first repair path strategy; Based on the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the covering layer is repaired.

[0006] In the above scheme, a first full-lifetime digital twin model is constructed by combining a material aging model with a preset full-lifetime cycle. This enables continuous prediction of the acoustic performance of the overlay layer throughout its entire lifetime. The model is then compared with a preset first performance threshold to identify the first weak acoustic performance value and locate the weak acoustic performance area, thus pinpointing the key region of acoustic performance degradation in the overlay layer and clarifying the target direction for repair decisions. Next, an initial agent is constructed using the target overlay layer, preset grid size, and preset reward parameters. This agent executes a preset strategy to obtain the first action and state space, building the basic framework for generating the first repair path strategy. Subsequently, combining the initial agent, state space, reward function, and other multi-dimensional parameters, the initial agent is trained using a preset experience replay pool and an initial evaluation network to obtain the action value loss set. This optimizes the action decision logic of the initial agent, improving its adaptability to the target overlay layer and the rationality of its decisions. Furthermore, by obtaining the first repair path strategy, intelligent and precise planning of the repair path is achieved, avoiding the rigidity problem of traditional preset paths and improving the targeting of repair decisions. Finally, the second acoustic performance prediction value is obtained through the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model. The repair effect is verified by using the preset second performance threshold as the standard. This achieves accurate prediction of the acoustic performance of the cover layer, enabling dynamic self-repair of the weak areas of the cover layer and ensuring that the acoustic performance of the cover layer meets the standards throughout its entire life cycle.

[0007] Furthermore, it also includes: Based on the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If there is a second acoustic performance prediction value that is less than a preset second performance threshold, then an acoustic performance recovery area is obtained based on the first repair path strategy and the weak acoustic performance area. After a preset time period, the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery zone were re-acquired. Based on the first acoustic absorption rate and the first blast resistance strength, a second repair path strategy is regenerated, and several third acoustic performance prediction values ​​are re-acquired until all the third acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold.

[0008] In the above scheme, by obtaining the second acoustic performance prediction value, prediction values ​​that do not reach the preset second performance threshold are filtered out. Combined with the first repair path strategy and the weak acoustic performance area, the acoustic performance recovery area is located, clarifying the area for secondary repair and avoiding omissions or ineffective repair areas. Next, after a preset time period, the first acoustic absorption rate and first blast resistance strength of the acoustic performance recovery area are re-collected. This allows for the acquisition of actual acoustic performance status data of the area after one repair, providing real data source support for the generation of the second repair path strategy. This ensures that the subsequently generated second repair path strategy conforms to actual working conditions, improving the targeted nature of the repair. Then, based on the updated first acoustic absorption rate and first blast resistance strength, the second repair path strategy is regenerated, and the third acoustic performance prediction value is re-acquired and compared with the preset second performance threshold again until all third acoustic performance prediction values ​​meet the standard. This ensures that the acoustic performance of the repaired cover layer fully meets the preset standard, achieving dynamic self-repair of weak areas in the cover layer and ensuring stable acoustic performance compliance throughout the entire life cycle of the cover layer.

[0009] Furthermore, the step of obtaining a material aging model based on the material's basic aging parameters, initial sound absorption rate, and initial blast resistance strength includes: An initial material aging model is established based on the aforementioned basic material aging parameters; Based on the initial sound absorption rate and initial blast resistance strength, the initial material aging model is corrected to obtain the material aging model.

[0010] In the above scheme, an initial material aging model is established based on fundamental material aging parameters. The acquired aging-related basic data is transformed into a quantifiable performance degradation analysis framework, providing initial theoretical model support for predicting the acoustic performance degradation of the covering layer. Then, the initial material aging model is calibrated using initial sound absorption rate and initial blast resistance strength. This allows the model to closely match the actual initial performance state of the covering layer, correcting the deviation between the theoretical model and the actual scenario, thereby improving the adaptability and prediction accuracy of the material aging model to the actual performance degradation process of the covering layer.

[0011] Furthermore, the step of correcting the initial material aging model based on the initial sound absorption rate and initial blast resistance strength to obtain the material aging model includes: Outlier removal and smoothing filtering are performed on the initial acoustic absorption rate and initial blast resistance strength to obtain the acoustic absorption rate and blast resistance strength. Based on the sound absorption rate and blast resistance strength, obtain the data confidence level; Based on the data confidence level, sound absorption rate, and blast resistance, the initial material aging model is corrected to obtain the material aging model.

[0012] In the above scheme, outlier removal and smoothing filtering of the initial acoustic absorption rate and initial blast resistance strength can eliminate interference information in the initial data, obtaining more accurate acoustic absorption rate and blast resistance strength data, providing reliable data source support for subsequent model calibration. Next, data confidence scores are used to quantify the reliability of the obtained acoustic absorption rate and blast resistance strength data, providing a quantitative basis for weight allocation in the subsequent model calibration process and ensuring the scientific nature of the calibration logic. Then, combining data confidence scores, acoustic absorption rate, and blast resistance strength, the initial material aging model is specifically calibrated, allowing the model parameters to better match the actual performance state of the covering layer, thereby improving the prediction accuracy and adaptability of the material aging model.

[0013] Further, the step of constructing a first full-lifetime digital twin model based on a material aging model and a preset full-lifetime cycle, and obtaining several first acoustic performance prediction values, wherein if any first acoustic performance prediction value is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the region where the first weak acoustic performance value is located is determined as an acoustic performance weak zone; includes: Based on the material aging model and the preset full life cycle, an acoustic absorption rate decay curve and an explosion resistance strength decay curve are constructed, and a first full life digital twin model is constructed based on the acoustic absorption rate decay curve and the explosion resistance strength decay curve. Based on the first full-life digital twin model, several first acoustic performance prediction values ​​are obtained. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area.

[0014] In the above scheme, by constructing sound absorption rate attenuation curves and blast resistance strength attenuation curves, and building a first full-lifetime digital twin model based on these curves, the abstract performance degradation law can be transformed into a visualized digital model, realizing dynamic simulation of the changes in the acoustic performance of the cover layer throughout its entire life cycle. Next, by obtaining several predicted first acoustic performance values ​​through the first full-lifetime digital twin model and comparing them with preset first performance thresholds, the first weak acoustic performance values ​​can be screened out and their locations identified as acoustic performance weakness areas. This allows for precise location of areas where the acoustic performance of the cover layer fails to meet standards throughout its entire life cycle, providing a clear and accurate target range for subsequent targeted repair decisions and avoiding blind repairs.

[0015] Furthermore, the construction of the original intelligent agent based on the target coverage layer, preset grid size, and preset reward parameters includes: Based on the target overlay layer, preset grid size, and preset reward parameters, establish the initial state space and reward function; Based on the initial state space, preset action space, and preset network parameters, an initial evaluation network and an initial target network are constructed. Based on the initial state space, preset action space, reward function, initial evaluation network, and initial target network, a primitive intelligent agent is constructed.

[0016] In the above scheme, an initial state space and reward function are established based on the target coverage layer, preset grid size, and preset reward parameters. The initial state space clarifies the perception dimension of the target coverage layer, and the reward function clarifies the decision optimization orientation, providing a foundation for building the initial intelligent agent. Next, using the initial state space, preset action space, and preset network parameters, an initial evaluation network and an initial target network are constructed, providing the intelligent agent with a carrier for action value evaluation and decision output, laying the network foundation for the agent's autonomous learning and decision-making. Finally, using the initial state space, preset action space, reward function, initial evaluation network, and initial target network, the initial intelligent agent is built, forming an intelligent decision-making subject with basic environmental perception, action execution, and value evaluation capabilities, providing a core execution unit for subsequent training optimization and repair path planning.

[0017] Furthermore, the step of obtaining the initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network includes: Based on the original agent and state space, obtain the initial agent; Calculate the reward value based on the initial agent and reward function; The initial state space, the first action, the reward value, and the state space are encapsulated and stored in a preset experience replay pool. The experience replay pool is then accessed. A predetermined number of samples are drawn from the experience buffer pool and input into the initial evaluation network to obtain a set of evaluation action values; Based on the state space and the initial target network, obtain the set of target action values; Based on the set of evaluated action values ​​and the set of target action values, calculate the set of action value losses.

[0018] In the above scheme, the original agent is combined with a state space 1 that covers the actual performance of the overlay layer to initially form an initial agent adapted to the repair scenario, providing a basic execution carrier for subsequent training and value calculation. Next, by calculating the reward value, the repair effect after the initial agent performs its first action can be quantified, providing clear feedback for the agent's iterative optimization. Then, by encapsulating the initial state space, the first action, the reward value, and the state space and storing them in a preset experience replay pool, complete experience samples of agent decision-making and environmental interaction can be accumulated, providing rich and effective data support for subsequent network training. Subsequently, a preset number of samples are extracted from the experience buffer pool and input into the initial evaluation network to obtain the set of evaluation action values, realizing the evaluation of the value of historical decision actions and providing an evaluation benchmark for calculating action value loss. Furthermore, the target action value set is obtained through the state space and the initial target network, establishing a benchmark reference standard for action value, providing a basis for comparison between the calculated evaluation value and the target value. Finally, the action value loss set is calculated by using the set of evaluation action values ​​and the target action value set, which can be used to quantitatively characterize the prediction bias of the initial evaluation network, providing optimization goals and directions for subsequent agent network parameter optimization.

[0019] Furthermore, the step of obtaining an agent based on the action value loss set and the initial agent to generate a first repair path strategy includes: Construct a total loss function based on the action value loss set and the preset loss function; The initial agent is trained with the goal of minimizing the total loss function until the preset convergence condition is met. The network parameters are then obtained, and the agent is obtained based on the network parameters to generate the first repair path strategy.

[0020] In the above scheme, a total loss function is constructed, which integrates the action value loss set with the preset loss rules to form a unified optimization objective for agent training, providing a clear quantitative guide for the parameter adjustment of the initial agent. Then, the initial agent is trained with the goal of minimizing the total loss function, continuously optimizing the agent network parameters until a preset convergence condition is met. Based on the optimized network parameters, the agent is obtained, and a first repair path strategy is generated to achieve precise intelligent planning of the acoustic performance repair path for the overlay layer.

[0021] Further, based on the first repair path strategy, the weak acoustic performance area, and the first full-lifetime digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to a preset second performance threshold, then the acoustic performance of the overlay layer is repaired; this includes: Based on the first repair path strategy, the weak acoustic performance area is repaired to obtain the acoustic performance recovery area, and the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery area are re-collected after a preset time period. Based on the first full-life digital twin model, the first sound absorption rate, and the first blast resistance strength, the first full-life digital twin model is updated to obtain the second full-life digital twin model. Based on the second full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the covering layer is repaired.

[0022] In the above scheme, a repair operation is performed on the weak acoustic performance area through the first repair path strategy. The repaired weak acoustic performance area is then identified as the acoustic performance recovery area. After a preset time period, the first acoustic absorption rate and the first blast resistance strength of this area are re-collected to obtain the actual performance data after repair, providing real and real-time data source support for subsequent model updates and effect verification. Next, the first full-lifetime digital twin model is updated using the first full-lifetime digital twin model, the first acoustic absorption rate, and the first blast resistance strength to obtain a second full-lifetime digital twin model that closely reflects the current actual performance state of the cover layer, improving the model's representation accuracy and prediction reliability of the cover layer's true performance. Finally, several second acoustic performance prediction values ​​are obtained through the second full-lifetime digital twin model. Using a preset second performance threshold as the judgment standard, the repair effect is verified to see if it meets the requirements. If all second acoustic performance prediction values ​​meet the requirements, it indicates that the acoustic performance of the cover layer has been effectively repaired, ensuring that the acoustic performance of the cover layer remains stable and meets the standards throughout its entire life cycle.

[0023] This invention provides a system for repairing the acoustic performance of an overlay layer, comprising a performance acquisition module, a model building module, a weak area identification module, an agent building module, an agent training module, a repair path generation module, and a repair effect verification module, specifically: The performance acquisition module is used to acquire several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance strength of the target coating layer. The model building module is used to obtain a material aging model based on the material aging basic parameters, initial sound absorption rate, and initial blast resistance strength. The weak area identification module is used to construct a first full-life digital twin model based on the material aging model and a preset full life cycle and obtain several first acoustic performance prediction values. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area. The agent building module is used to build an original agent based on the target coverage layer, preset grid size and preset reward parameters; and to obtain and execute a first action based on the original agent and preset policy, thereby obtaining the state space. The agent training module is used to obtain an initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool and initial evaluation network. The repair path generation module is used to obtain an agent based on the action value loss set and the initial agent, so as to generate a first repair path strategy. The repair effect verification module is used to obtain several second acoustic performance prediction values ​​based on the first repair path strategy, the weak acoustic performance area and the first full life digital twin model. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the cover layer is repaired.

[0024] This invention provides a system for repairing the acoustic performance of a cover layer. In practical applications, it only requires a weak area identification module. Combined with a material aging model and a preset full-lifecycle digital twin model, it constructs a first full-lifecycle digital twin model. This enables continuous prediction of the cover layer's acoustic performance throughout its entire lifecycle. The prediction is compared with a preset first performance threshold to filter out the first weak acoustic performance value, thus locating the weak acoustic performance area and pinpointing the key region of acoustic performance degradation in the cover layer, providing a clear target direction for repair decisions. Then, an agent building module is used to construct an initial agent using the target cover layer, preset grid size, and preset reward parameters. The initial agent executes a preset strategy to obtain the first action and state space, building the basic framework for generating the first repair path strategy. Subsequently, an agent training module is used, combining the initial agent, state space, reward function, and other multi-dimensional parameters. Using a preset experience replay pool and initial evaluation network, the initial agent is trained, and an action value loss set is obtained. This optimizes the initial agent's action decision logic, improving its adaptability to the target cover layer and the rationality of its decisions. Furthermore, a repair path generation module is employed to obtain a first repair path strategy, enabling intelligent and precise planning of the repair path. This avoids the rigidity of traditional preset paths and improves the targeted nature of repair decisions. Finally, a repair effect verification module is used to obtain a second predicted acoustic performance value through the first repair path strategy, weak acoustic performance areas, and a first full-lifetime digital twin model. The repair effect is verified using a preset second performance threshold as a standard, achieving accurate prediction of the acoustic performance of the cover layer. This allows for dynamic self-repair of weak areas in the cover layer, ensuring stable and compliant acoustic performance throughout the entire lifespan of the cover layer. Attached Figure Description

[0025] Figure 1 A flowchart illustrating a method for repairing the acoustic performance of a cover layer according to an embodiment of the present invention; Figure 2 This is an architectural diagram of a method for repairing the acoustic performance of a cover layer according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This embodiment provides a method for repairing the acoustic performance of a cover layer; please refer to the flowchart below. Figure 1 ,include: Step S1: Obtain several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance strength of the target coating layer; Step S2: Based on the material aging parameters, initial sound absorption rate, and initial blast resistance, obtain the material aging model; Step S3: Based on the material aging model and the preset full life cycle, construct the first full life digital twin model and obtain several first acoustic performance prediction values. If there is a first acoustic performance prediction value that is less than the preset first performance threshold, it is determined as the first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as the acoustic performance weak area. Step S4: Based on the target overlay layer, preset grid size, and preset reward parameters, construct the original intelligent agent; based on the original intelligent agent and the preset policy, obtain and execute the first action, and obtain the state space; Step S5: Based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network, obtain the initial agent and action value loss set; Step S6: Based on the action value loss set and the initial agent, obtain an agent to generate the first repair path strategy; Step S7: Based on the first repair path strategy, the weak acoustic performance area and the first full-life digital twin model, obtain several second acoustic performance prediction values. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, then the acoustic performance of the covering layer is repaired.

[0028] In this embodiment, the initial acoustic absorption rate (500Hz~5kHz frequency band) and initial blast resistance (maximum bearing stress under impact load) of the target covering layer are acquired in real time using sensors (such as sonar detectors and stress sensors) mounted on the actual vessel. Several basic material aging parameters of the target covering layer are obtained, and a material aging database and a performance database of the actual vessel's covering layer under different working conditions are constructed. A first full-lifetime digital twin model is constructed by combining the material aging model with a preset full-lifetime cycle, enabling continuous prediction of the covering layer's acoustic performance throughout its entire lifetime. This model is then compared with a preset first performance threshold to identify the first weak acoustic performance value and locate the weak acoustic performance area, pinpointing the key area of ​​acoustic performance degradation in the covering layer and clarifying the target direction for repair decisions. Then, an initial intelligent agent is built using the target covering layer, preset grid size, and preset reward parameters. The initial intelligent agent executes a preset strategy to obtain the first action. Specifically, the preset strategy involves: adopting... Strategy selection, first action a: exploration rate (High Exploration), has Probabilistic random selection of actions (exploring new strategies). The action with the highest Q-value in the network output is selected probabilistically (using the optimal policy), and its value decreases by 0.01 every 100 iterations until... (Stable utilization). Then, the first action is executed to obtain the state space, building the basic framework for generating the first repair path strategy. Subsequently, combining the original agent, state space, reward function, and other multi-dimensional parameters, the initial agent is trained using a pre-set experience replay pool and an initial evaluation network, obtaining the action value loss set. This optimizes the initial agent's action decision-making logic, improving its adaptability to the target coverage layer and the rationality of its decisions. Furthermore, by obtaining the first repair path strategy, intelligent and precise planning of the repair path is achieved, avoiding the rigidity problem of traditional pre-set paths and improving the targeting of repair decisions. Finally, the second acoustic performance prediction value is obtained through the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model. The repair effect is verified by using a preset second performance threshold (re-testing the second acoustic performance prediction value (including sound absorption rate and blast resistance strength) of the weak acoustic performance area to see if it recovers to 95% of the original unattenuated sound absorption rate and 95% of the original unattenuated blast resistance strength). This achieves accurate prediction of the acoustic performance of the cover layer, enabling dynamic self-repair of the weak areas of the cover layer and ensuring that the acoustic performance of the cover layer meets the standards throughout its entire life cycle.

[0029] Furthermore, it also includes: Based on the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If there is a second acoustic performance prediction value that is less than a preset second performance threshold, then an acoustic performance recovery area is obtained based on the first repair path strategy and the weak acoustic performance area. After a preset time period, the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery zone were re-acquired. Based on the first acoustic absorption rate and the first blast resistance strength, a second repair path strategy is regenerated, and several third acoustic performance prediction values ​​are re-acquired until all the third acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold.

[0030] In this embodiment, by acquiring the second acoustic performance prediction value, prediction values ​​that do not reach the preset second performance threshold are filtered out. Combined with the first repair path strategy and the acoustic performance weakness area, the acoustic performance recovery area is located, clarifying the area for secondary repair and avoiding omissions or ineffective repair areas. Next, after a preset time period, the first acoustic absorption rate and first blast resistance strength of the acoustic performance recovery area are re-acquired. This allows for the acquisition of actual acoustic performance status data of the area after one repair, providing real data source support for the generation of the second repair path strategy. This ensures that the subsequently generated second repair path strategy conforms to actual working conditions, improving the targeted nature of the repair. Then, based on the updated first acoustic absorption rate and first blast resistance strength, the second repair path strategy is regenerated, and the third acoustic performance prediction value is re-acquired and compared with the preset second performance threshold again until all third acoustic performance prediction values ​​meet the standard. This ensures that the acoustic performance of the repaired cover layer fully meets the preset standard, achieving dynamic self-repair of the weak areas of the cover layer and ensuring stable acoustic performance compliance throughout the entire life cycle of the cover layer.

[0031] Furthermore, the step of obtaining a material aging model based on the material's basic aging parameters, initial sound absorption rate, and initial blast resistance strength includes: An initial material aging model is established based on the aforementioned basic material aging parameters; Based on the initial sound absorption rate and initial blast resistance strength, the initial material aging model is corrected to obtain the material aging model.

[0032] In this embodiment, several basic material aging parameters of the target coating layer (including water flow impact strength (F, unit)) are obtained through laboratory tests. Underwater acoustic radiation intensity (S, unit) And mechanical vibration amplitude (A, unit mm). And combined with the aging coefficient of the covering material (such as rubber-based composite material) under different temperatures (-20℃~60℃), humidity (30%~90%), and hydrostatic pressure (0~10MPa). (e.g., elastic modulus attenuation rate, acoustic impedance change rate) Establish an initial material aging model: In the formula, Let be the material aging parameters at time t. Here, k represents the basic parameter for material aging, α represents the laboratory-based basic aging coefficient, and α represents the environmental impact factor. Standard water flow impact intensity Standard underwater acoustic radiation intensity =10 Standard mechanical vibration amplitude =0.5mm. This transforms the acquired aging-related basic data into a quantifiable performance degradation analysis framework, providing initial theoretical model support for predicting the acoustic performance degradation of the covering layer. Next, a weighted fusion algorithm is used to combine the model's theoretical predictions with actual vessel monitoring data using initial acoustic absorption rate and initial blast resistance strength. This calibrates the initial material aging model, allowing it to better reflect the actual initial performance state of the covering layer, correcting the deviation between the theoretical model and the actual scenario, thereby improving the adaptability and predictive accuracy of the material aging model for the actual performance degradation process of the covering layer.

[0033] Furthermore, the step of correcting the initial material aging model based on the initial sound absorption rate and initial blast resistance strength to obtain the material aging model includes: Outlier removal and smoothing filtering are performed on the initial acoustic absorption rate and initial blast resistance strength to obtain the acoustic absorption rate and blast resistance strength. Based on the sound absorption rate and blast resistance strength, obtain the data confidence level; Based on the data confidence level, sound absorption rate, and blast resistance, the initial material aging model is corrected to obtain the material aging model.

[0034] In this embodiment, by performing outlier removal and smoothing filtering on the initial acoustic absorption rate and initial blast resistance strength, interference information in the initial data can be eliminated, obtaining more accurate acoustic absorption rate and blast resistance strength data, providing reliable data source support for subsequent model calibration. Next, the reliability of the obtained acoustic absorption rate and blast resistance strength data is quantified using data confidence levels, providing a quantitative basis for weight allocation in the subsequent model calibration process and ensuring the scientific nature of the calibration logic. Then, combining data confidence levels, acoustic absorption rate, and blast resistance strength, the initial material aging model is specifically calibrated: the model's weight coefficients can be adjusted in real time according to the data confidence level; for example, when the data confidence level is ≥90%, the weight ratio of acoustic absorption rate and blast resistance strength is 70%. That is, the fusion value = initial material aging model × 30% + actual boat monitoring value (acoustic absorption rate and blast resistance strength) × 70%, thereby making the model parameters more closely match the actual performance state of the covering layer, thus improving the prediction accuracy and adaptability of the material aging model.

[0035] Further, the step of constructing a first full-lifetime digital twin model based on a material aging model and a preset full-lifetime cycle, and obtaining several first acoustic performance prediction values, wherein if any first acoustic performance prediction value is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the region where the first weak acoustic performance value is located is determined as an acoustic performance weak zone; includes: Based on the material aging model and the preset full life cycle, an acoustic absorption rate decay curve and an explosion resistance strength decay curve are constructed, and a first full life digital twin model is constructed based on the acoustic absorption rate decay curve and the explosion resistance strength decay curve. Based on the first full-life digital twin model, several first acoustic performance prediction values ​​are obtained. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area.

[0036] In this embodiment, by constructing sound absorption rate attenuation curves and blast resistance strength attenuation curves (the horizontal axis represents service time, and the vertical axes represent sound absorption rate and blast resistance strength, respectively), and pre-setting the full life cycle to 0-20 years, a first full life cycle digital twin model is built based on this. This transforms the abstract continuous prediction of performance degradation into a visualized digital model, realizing the dynamic simulation of the changes in the acoustic performance of the covering layer throughout its entire life cycle. The unattenuated original sound absorption rate and unattenuated original blast resistance strength can be obtained. Next, several first acoustic performance prediction values ​​are obtained through the first full life cycle digital twin model and compared with preset first performance thresholds (preset first sound absorption rate threshold is 60%, preset first blast resistance strength threshold is 80MPa). This allows for the selection of the first weak acoustic performance values ​​and the identification of their corresponding regions as weak acoustic performance areas (any region with performance below the threshold, such as a sound absorption rate <60% or a blast resistance strength <80MPa). This accurately locates the areas where the acoustic performance of the covering layer fails to meet standards throughout its entire life cycle, providing a clear and precise target range for subsequent targeted repair decisions and avoiding blind repairs.

[0037] Furthermore, the construction of the original intelligent agent based on the target coverage layer, preset grid size, and preset reward parameters includes: Based on the target overlay layer, preset grid size, and preset reward parameters, establish the initial state space and reward function; Based on the initial state space, preset action space, and preset network parameters, an initial evaluation network and an initial target network are constructed. Based on the initial state space, preset action space, reward function, initial evaluation network, and initial target network, a primitive intelligent agent is constructed.

[0038] In this embodiment, microcapsule repair agents are arranged in the target covering layer, each containing embedded distributed microcapsules (50-100 μm in diameter) encapsulating a repair agent (such as epoxy resin-based repair liquid). The trigger threshold of the microcapsules is linked to the blast resistance strength threshold (release is triggered when the blast resistance strength is ≤ 80% of the initial blast resistance strength). A deep reinforcement learning (DQN) algorithm is used to establish an initial state space and a reward function based on the target covering layer (the acoustic absorption attenuation rate (current value / initial value), blast resistance strength attenuation rate, and remaining microcapsule quantity in each region of the target covering layer), a preset grid size, and preset reward parameters. The initial state space s includes: the acoustic absorption attenuation rate of each grid cell. Explosion resistance attenuation rate of each grid cell Remaining amount of microcapsules in each grid unit And a preset regional importance weight W, such as W=1.2 near sonar, and W=1 for ordinary areas. Where: and Values ​​can range from 0 to 100%. ); This represents the sound absorption rate / blast resistance before performance degradation. This represents the attenuation rate of sound absorption rate / attenuation rate of blast resistance strength. This represents the sound absorption rate / explosion resistance after performance degradation; The value can range from 0 to 100% (initially 100%, deducted based on usage after release); W can range from 1.0 to 1.5 (preset based on the actual boat structure). The reward function r is: ;in: ;in, To simulate the actual performance value 24 hours after the repair is performed based on the first action a, The target performance value is the simulated 24h after the repair is performed according to the first action a. When the release amount exceeds the required amount, for every 10... Deduct 1 point. This indicates that repeated repairs were performed on the same area within 24 hours; 2 points will be deducted for each instance. This indicates that the area is not repaired, and 3 points will be deducted every hour; total reward (scope: The initial state space clarifies the perception dimension of the target coverage layer, and the reward function clarifies the decision optimization orientation, providing a foundation for building the original intelligent agent. Next, the initial state space and the preset action space a (including: 1. Repair agent release area (Aarea) (coordinate accuracy ±0.1m) 2. Repair agent release amount (Adose) where: Areaa is the coordinates of 2000 grid cells (e.g., (3,2) represents the 3rd row, 2nd column grid), and Adose contains 5 levels (... Indicates no release, 20 / 50 / 80 / 100 ()) and preset network parameters (preset network parameters include: learning rate) An initial evaluation network and an initial target network were constructed (using a 3-layer fully connected neural network as the initial evaluation network (Q-network) and the initial target network (Target Q-network). The number of neurons in the input layer = the number of grid units × 3 (the attenuation rate of sound absorption rate, the attenuation rate of blast resistance strength, and the remaining amount of microcapsules in each grid). The hidden layer has 256 + 128 neurons (with ReLU activation function). The number of neurons in the output layer = the number of actions (the number of release regions × the release amount level). This provides a carrier for the agent to evaluate the value of actions and make decision-making outputs, laying the network foundation for the agent's autonomous learning and decision-making. Finally, through the initial state space, the preset action space, the reward function, the initial evaluation network, and the initial target network, a primitive agent was built, forming an intelligent decision-making subject with basic environmental perception, action execution, and value evaluation capabilities, providing a core execution unit for subsequent training optimization and repair path planning.

[0039] Furthermore, the step of obtaining the initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network includes: Based on the original agent and state space, obtain the initial agent; Calculate the reward value based on the initial agent and reward function; The initial state space, the first action, the reward value, and the state space are encapsulated and stored in a preset experience replay pool. The experience replay pool is then accessed. A predetermined number of samples are drawn from the experience buffer pool and input into the initial evaluation network to obtain a set of evaluation action values; Based on the state space and the initial target network, obtain the set of target action values; Based on the set of evaluated action values ​​and the set of target action values, calculate the set of action value losses.

[0040] In this embodiment, the original agent is combined with a state space 1 that encompasses the actual performance of the overlay layer to initially form an initial agent adapted to the repair scenario, providing a basic execution carrier for subsequent training and value calculation. Next, by calculating the reward value, the repair effect after the initial agent performs the first action can be quantified, providing clear feedback for the iterative optimization of the agent. Then, the initial state space, the first action, the reward value, and the state space are encapsulated and stored in a preset experience replay pool (initialized capacity of 10). 5Used for storage<s,a,r,s',done> The initial evaluation network uses a set of empirical samples (s represents the initial state space, a represents the first action, r represents the reward value, s' represents the state space, and done represents the terminator; done=1 indicates the end of the one-year cycle, done=0 indicates continuation). This allows for the accumulation of complete empirical samples of agent decision-making and environmental interaction, providing rich and effective data support for subsequent network training. Subsequently, when the number of samples in the empirical pool is ≥1000, a preset number of samples (up to 32) are extracted from the empirical buffer pool and input into the initial evaluation network to obtain the set of evaluated action values. This achieves the evaluation of the value of historical decision actions, providing an evaluation benchmark for calculating action value loss. Furthermore, the target action value set is obtained through the state space and the initial target network. ,in The discount factor (emphasizing long-term rewards) is used to establish a benchmark for action value, providing a basis for comparing the deviation between the evaluated value and the target value. Finally, by calculating the action value loss set using the evaluated action value set and the target action value set, the set can be used to quantitatively characterize the prediction bias of the initial evaluation network, providing optimization goals and directions for subsequent optimization of agent network parameters.

[0041] Furthermore, the step of obtaining an agent based on the action value loss set and the initial agent to generate a first repair path strategy includes: Construct a total loss function based on the action value loss set and the preset loss function; The initial agent is trained with the goal of minimizing the total loss function until the preset convergence condition is met. The network parameters are then obtained, and the agent is obtained based on the network parameters to generate the first repair path strategy.

[0042] In this embodiment, the total loss function is constructed by using the mean squared error loss function ( This method integrates the action value loss set with preset loss rules to form a unified optimization objective for agent training, providing clear quantitative guidance for adjusting the parameters of the initial agent. Next, the initial agent is trained with the goal of minimizing the total loss function. A digital twin platform is used to simulate overlay performance degradation scenarios, with a training cycle of 1000 iterations. Initialization is performed before each training round: the overlay state (initial performance, full microcapsule capacity) is reset, the simulation start time is set to the first year of service, and each training round simulates a one-year service cycle (36 time steps, each step being one day). The agent network parameters are continuously optimized until the preset convergence conditions are met (meeting any of the following conditions: 1. The average total reward value fluctuation over 50 consecutive training rounds is ≤5% (e.g., the average reward stabilizes from 8.2 to 8.5, a fluctuation of 3.6%); 2. The overlay full-cycle repair efficiency (the percentage of areas meeting performance standards after repair) is consistently ≥95% (consecutive rounds ≥95.2%); 3. The unit performance recovery amount of the repair agent (repair agent dosage / performance recovery value) is consistently ≤2. (That is, for every 1% restoration of sound absorption rate, the energy consumption is ≤2) (Repair agent). Based on the optimized network parameters, an agent is obtained (the evaluation network parameters are updated through the Adam optimizer, and every 200 iterations (approximately 0.5 years of simulation time), the parameters of the evaluation network are copied to the target network to avoid excessive fluctuations in the target Q value), thereby generating the first repair path strategy to achieve precise intelligent planning of the acoustic performance repair path of the overlay layer. The generation rules of the first repair path strategy are as follows: starting from the central grid of the weak acoustic performance area, a spiral outward path is adopted (clockwise or counterclockwise, selected according to the microcapsule distribution density), moving 1 grid at each step, prioritizing the coverage of areas with the same priority, and avoiding cross-regional jumps (reducing path loss); repeated repair avoidance: the repair time of each grid is recorded, and if a grid has been repaired within 24 hours and The path will automatically skip that grid; if performance rebounds after the fix ( If the performance exceeds the preset second performance threshold again, it will be marked as a high-risk area, and the re-inspection frequency of this area will be increased in the path (from once every 7 days to once every 3 days).

[0043] Further, based on the first repair path strategy, the weak acoustic performance area, and the first full-lifetime digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to a preset second performance threshold, then the acoustic performance of the overlay layer is repaired; this includes: Based on the first repair path strategy, the weak acoustic performance area is repaired to obtain the acoustic performance recovery area, and the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery area are re-collected after a preset time period. Based on the first full-life digital twin model, the first sound absorption rate, and the first blast resistance strength, the first full-life digital twin model is updated to obtain the second full-life digital twin model. Based on the second full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the covering layer is repaired.

[0044] In this embodiment, a first repair path strategy is used to perform repair operations on areas with weak acoustic performance (prioritizing areas with attenuation rates P ≥ 30% for any performance characteristic, with a path planning response time ≤ 1s). Specifically, this involves a priority ranking strategy for areas with weak acoustic performance (since blast resistance has a greater impact on structural safety, blast resistance has a weight of 0.6, and acoustic absorption rate has a weight of 0.4); high priority (P ≥ 30): >25% or >30% of critical areas (such as near sonar) should have repair agent released within 1 hour; medium priority (20≤P<30): =15%-25% or =20%-30% of the normal area, processed synchronously with the area of ​​the same priority within 24 hours); low priority (P<20): <15% and Areas with less than 20% acoustic performance will not be repaired initially, but will be reassessed every 7 days. Based on the attenuation rate and material properties of areas with weak acoustic performance, the repair agent is calculated using the following formula: ;in,( The coefficient for the amount of repair agent used in the butyl rubber composite layer is denoted by the coefficient (the specific value can be determined experimentally). 20 represents the minimum release amount, and 100 represents the maximum release amount (to avoid ineffective release due to excessively low release or waste due to excessively high release). W represents the preset regional importance weight. By performing the repair operation, the original weak acoustic performance area after repair is identified as the acoustic performance recovery area. After a preset time period, the first acoustic absorption rate and the first blast resistance strength of this area are re-collected to obtain the actual performance data after repair, providing real and real-time data source support for subsequent model updates and effect verification. Next, the first full-lifetime digital twin model is updated using the first full-lifetime digital twin model, the first acoustic absorption rate, and the first blast resistance strength to obtain a second full-lifetime digital twin model that closely reflects the current actual performance state of the cover layer, improving the model's accuracy in representing the true performance of the cover layer and its predictive reliability. Finally, several second acoustic performance prediction values ​​are obtained through the second full-life digital twin model. The preset second performance threshold is used as the judgment standard to verify whether the repair effect meets the standard. If all the second acoustic performance prediction values ​​meet the requirements, it means that the acoustic performance of the cover layer has been effectively repaired, ensuring that the acoustic performance of the cover layer is stable and meets the standard throughout its entire life cycle.

[0045] This embodiment provides a system for repairing the acoustic performance of an overlay layer, including a performance acquisition module, a model building module, a weak area identification module, an agent building module, an agent training module, a repair path generation module, and a repair effect verification module, specifically: The performance acquisition module is used to acquire several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance strength of the target coating layer. The model building module is used to obtain a material aging model based on the material aging basic parameters, initial sound absorption rate, and initial blast resistance strength. The weak area identification module is used to construct a first full-life digital twin model based on the material aging model and a preset full life cycle and obtain several first acoustic performance prediction values. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area. The agent building module is used to build an original agent based on the target coverage layer, preset grid size and preset reward parameters; and to obtain and execute a first action based on the original agent and preset policy, thereby obtaining the state space. The agent training module is used to obtain an initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool and initial evaluation network. The repair path generation module is used to obtain an agent based on the action value loss set and the initial agent, so as to generate a first repair path strategy. The repair effect verification module is used to obtain several second acoustic performance prediction values ​​based on the first repair path strategy, the weak acoustic performance area and the first full life digital twin model. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the cover layer is repaired.

[0046] This embodiment provides a system for repairing the acoustic performance of a cover layer. In practical applications, it only requires a weak area identification module. Combined with a material aging model and a preset full-lifecycle digital twin model, it constructs a first full-lifecycle digital twin model. This enables continuous prediction of the cover layer's acoustic performance throughout its entire lifecycle. The model is then compared with a preset first performance threshold to filter out the first weak acoustic performance value, thus locating the weak acoustic performance area and pinpointing the key region of acoustic performance degradation in the cover layer, providing a clear target direction for repair decisions. Next, an agent construction module is used to build an initial agent using the target cover layer, preset grid size, and preset reward parameters. The initial agent executes a preset strategy to obtain the first action and state space, building the basic framework for generating the first repair path strategy. Subsequently, an agent training module is used, combining the initial agent, state space, reward function, and other multi-dimensional parameters. Using a preset experience replay pool and initial evaluation network, the initial agent is trained, and an action value loss set is obtained. This optimizes the initial agent's action decision logic, improving its adaptability to the target cover layer and the rationality of its decisions. Furthermore, a repair path generation module is employed to obtain a first repair path strategy, enabling intelligent and precise planning of the repair path. This avoids the rigidity of traditional preset paths and improves the targeted nature of repair decisions. Finally, a repair effect verification module is used to obtain a second predicted acoustic performance value through the first repair path strategy, weak acoustic performance areas, and a first full-lifetime digital twin model. The repair effect is verified using a preset second performance threshold as a standard, achieving accurate prediction of the acoustic performance of the cover layer. This allows for dynamic self-repair of weak areas in the cover layer, ensuring stable and compliant acoustic performance throughout the entire lifespan of the cover layer.

[0047] To verify that the proposed acoustic performance repair method for the covering layer can accurately predict the acoustic performance of the covering layer, enabling dynamic self-repair of weak areas and ensuring stable acoustic performance throughout the entire life cycle of the covering layer, an embodiment is provided: an underwater section covering layer of a certain type of ship (material: butyl rubber composite layer, initial sound absorption rate 90%, initial blast resistance 20MPa), with dimensions of 10m × 5m. The material aging model is as follows: aging model of sound absorption rate. Aging model of blast resistance strength Where t is the service time (in years). After 3 years of service, the sound absorption rate collected by the sensor is 82%, and the blast resistance is 17 MPa. Through weighted fusion (70% weight of actual ship monitoring data), the corrected predicted sound absorption rate after 3 years is 81.5%, and the blast resistance is 16.8 MPa, with an error of ≤1% compared to the actual detected value. Self-repair decision-making is implemented: Weak acoustic performance areas are identified by comparing the sound absorption rate attenuation curve and the blast resistance attenuation curve. The sound absorption rate in the central area of ​​the covering layer (coordinates 3-6m × 2-3m) is found to have attenuated to 75% (attenuation rate 16.7%), and the blast resistance has attenuated to 15 MPa (attenuation rate 25%), thus identifying it as a weak acoustic performance area. Then, a first repair path strategy is planned through reinforcement learning: the DQN algorithm outputs a repair path that spirals outwards from the center of the weak acoustic performance area, with a release amount of 50... The final repair results were as follows: the sound absorption rate of the weak acoustic performance area was restored to 89%, the blast resistance was restored to 19MPa, and the repair efficiency reached 96.5%, which met the requirements.

[0048] This embodiment obtains a material aging model by integrating basic material aging parameters with actual vessel monitoring data, overcoming the limitations of single models or discrete data. It reduces the prediction errors of both the sound absorption rate attenuation curve and the blast resistance strength attenuation curve to within 5%, thus accurately reflecting the performance changes of the coating throughout its entire lifespan. This embodiment uses reinforcement learning to generate dynamic path planning, making the release of the repair agent more targeted, improving repair efficiency to over 95%, and reducing waste of the repair agent (reducing usage by 15-20%). This embodiment, combined with digital twin technology, achieves closed-loop management of the coating performance from passive monitoring to active prediction and dynamic repair, extending the coating's service life by 10-15%.

[0049] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for repairing the acoustic performance of a covering layer, characterized in that, Applied to the target overlay layer, including: Acquire several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance of the target coating layer; Based on the material aging parameters, initial sound absorption rate, and initial blast resistance, a material aging model is obtained. Based on the material aging model and the preset full life cycle, a first full life digital twin model is constructed and several first acoustic performance prediction values ​​are obtained. If there is a first acoustic performance prediction value that is less than the preset first performance threshold, it is determined as the first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as the acoustic performance weak area. Based on the target coverage layer, preset grid size, and preset reward parameters, an initial intelligent agent is constructed; based on the initial intelligent agent and the preset policy, a first action is obtained and executed, and the state space is obtained; Based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network, obtain the initial agent and action value loss set; Based on the action value loss set and the initial agent, obtain an agent to generate the first repair path strategy; Based on the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the covering layer is repaired.

2. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, Also includes: Based on the first repair path strategy, the weak acoustic performance area, and the first full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If there is a second acoustic performance prediction value that is less than a preset second performance threshold, then an acoustic performance recovery area is obtained based on the first repair path strategy and the weak acoustic performance area. After a preset time period, the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery zone were re-acquired. Based on the first acoustic absorption rate and the first blast resistance strength, a second repair path strategy is regenerated, and several third acoustic performance prediction values ​​are re-acquired until all the third acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold.

3. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, The process of obtaining a material aging model based on the material's basic aging parameters, initial sound absorption rate, and initial blast resistance strength includes: An initial material aging model is established based on the aforementioned basic material aging parameters; Based on the initial sound absorption rate and initial blast resistance strength, the initial material aging model is corrected to obtain the material aging model.

4. The method for repairing the acoustic performance of a covering layer according to claim 3, characterized in that, The process involves correcting the initial material aging model based on the initial sound absorption rate and initial blast resistance strength to obtain the material aging model; including: Outlier removal and smoothing filtering are performed on the initial acoustic absorption rate and initial blast resistance strength to obtain the acoustic absorption rate and blast resistance strength. Based on the sound absorption rate and blast resistance strength, obtain the data confidence level; Based on the data confidence level, sound absorption rate, and blast resistance, the initial material aging model is corrected to obtain the material aging model.

5. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, The process involves constructing a first full-lifetime digital twin model based on a material aging model and a preset full-lifetime cycle, and obtaining several first acoustic performance prediction values. If any of the first acoustic performance prediction values ​​is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the region where the first weak acoustic performance value is located is determined as an acoustic performance weak zone; including: Based on the material aging model and the preset full life cycle, an acoustic absorption rate decay curve and an explosion resistance strength decay curve are constructed, and a first full life digital twin model is constructed based on the acoustic absorption rate decay curve and the explosion resistance strength decay curve. Based on the first full-life digital twin model, several first acoustic performance prediction values ​​are obtained. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area.

6. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, The process of constructing the original intelligent agent based on the target coverage layer, preset grid size, and preset reward parameters includes: Based on the target overlay layer, preset grid size, and preset reward parameters, establish the initial state space and reward function; Based on the initial state space, preset action space, and preset network parameters, an initial evaluation network and an initial target network are constructed. Based on the initial state space, preset action space, reward function, initial evaluation network, and initial target network, a primitive intelligent agent is constructed.

7. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, The method involves obtaining the initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool, and initial evaluation network. include: Based on the original agent and state space, obtain the initial agent; Calculate the reward value based on the initial agent and reward function; The initial state space, the first action, the reward value, and the state space are encapsulated and stored in a preset experience replay pool. The experience replay pool is then accessed. A predetermined number of samples are drawn from the experience buffer pool and input into the initial evaluation network to obtain a set of evaluation action values; Based on the state space and the initial target network, obtain the set of target action values; Based on the set of evaluated action values ​​and the set of target action values, calculate the set of action value losses.

8. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, The process of obtaining an agent based on the action value loss set and the initial agent to generate a first repair path strategy includes: Construct a total loss function based on the action value loss set and the preset loss function; The initial agent is trained with the goal of minimizing the total loss function until the preset convergence condition is met. The network parameters are then obtained, and the agent is obtained based on the network parameters to generate the first repair path strategy.

9. The method for repairing the acoustic performance of a covering layer according to claim 1, characterized in that, Based on the first repair path strategy, the acoustic performance weak area, and the first full-lifetime digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to a preset second performance threshold, the acoustic performance of the overlay layer is repaired; including: Based on the first repair path strategy, the weak acoustic performance area is repaired to obtain the acoustic performance recovery area, and the first acoustic absorption rate and the first explosion resistance of the acoustic performance recovery area are re-collected after a preset time period. Based on the first full-life digital twin model, the first sound absorption rate, and the first blast resistance strength, the first full-life digital twin model is updated to obtain the second full-life digital twin model. Based on the second full-life digital twin model, several second acoustic performance prediction values ​​are obtained. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the covering layer is repaired.

10. A system for repairing the acoustic performance of a covering layer, characterized in that, It includes a performance acquisition module, a model building module, a weak area identification module, an agent building module, an agent training module, a repair path generation module, and a repair effect verification module, specifically: The performance acquisition module is used to acquire several basic material aging parameters of the target coating layer, the initial sound absorption rate of the target coating layer, and the initial blast resistance strength of the target coating layer. The model building module is used to obtain a material aging model based on the material aging basic parameters, initial sound absorption rate, and initial blast resistance strength. The weak area identification module is used to construct a first full-life digital twin model based on the material aging model and a preset full life cycle and obtain several first acoustic performance prediction values. If there is a first acoustic performance prediction value that is less than a preset first performance threshold, it is determined as a first weak acoustic performance value, and the area where the first weak acoustic performance value is located is determined as an acoustic performance weak area. The agent building module is used to build an original agent based on the target coverage layer, preset grid size and preset reward parameters; and to obtain and execute a first action based on the original agent and preset policy, thereby obtaining the state space. The agent training module is used to obtain an initial agent and action value loss set based on the original agent, state space, reward function, initial state space, first action, preset experience replay pool and initial evaluation network. The repair path generation module is used to obtain an agent based on the action value loss set and the initial agent, so as to generate a first repair path strategy. The repair effect verification module is used to obtain several second acoustic performance prediction values ​​based on the first repair path strategy, the weak acoustic performance area and the first full life digital twin model. If all the second acoustic performance prediction values ​​are greater than or equal to the preset second performance threshold, the acoustic performance of the cover layer is repaired.

Images