A deep learning product structure parameterization design optimization method and system

Abstract

This invention discloses a deep learning-based product structure parameterized design optimization method and system, relating to the field of computer-aided design technology. The key technical solutions include the following steps: collecting multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports; generating a parameter full-process influence correlation map based on the multi-source heterogeneous data; performing feature judgment on the parameter full-process influence correlation map to obtain a set of key constraint parameters; and generating an optimization target system based on the priority ranking and interactive constraint relationships of the key constraint parameter set. This optimization target system includes structural parameter targets in the design phase, structural manufacturing targets in the process phase, and structural stability targets in the usage phase. The effect is to avoid the problems of insufficient production adaptability and poor usage stability caused by focusing only on a single stage target, improving the parameter coordination throughout the product lifecycle, and shortening the product development iteration cycle.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, and more specifically, to a deep learning-based method and system for parametric design optimization of product structure. Background Technology

[0002] In the research and development of complex industrial products such as high-end equipment and precision components, structural parametric design is a core element that determines product performance, production feasibility, and reliability throughout the entire life cycle.

[0003] However, existing parametric design optimization methods for product structures are mostly single-stage goal-oriented. For example, the design stage focuses only on the theoretical realization of structural functions, the process stage considers the adaptability to production equipment separately, and the usage stage re-evaluates stability. The parameter decisions of each stage are isolated and require repeated iterations and modifications, leading to a prolonged R&D cycle. Moreover, because parameter design documents, CAD model data, finite element simulation results, production line process records, and failure case reports are stored in a scattered manner throughout the product lifecycle, and the data formats, coding rules, and measurement standards differ significantly, it is difficult to achieve effective data integration and correlation analysis. This results in the inability to reuse risk experience from historical data, and a high recurrence rate of similar failure problems. At the same time, traditional parameter optimization relies on preset fixed target weights and lacks dynamic response to actual performance verification data and risk assessment results. When the initial solution shows deviations in indicators during physical testing, it is impossible to accurately adjust the target priorities of the design, process, and usage stages. Parameters can only be modified manually through trial and error, which is not only inefficient but also fails to meet multi-dimensional indicator requirements simultaneously. Some solutions, although meeting the design functional thresholds, cannot be implemented due to insufficient process adaptability or excessive usage stability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for optimizing product structure parametric design using deep learning.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A deep learning-based product structure parametric design optimization method, comprising the following steps: Collect multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records and failure case reports, and generate a parameter full-process influence correlation map based on the multi-source heterogeneous data; The key constraint parameter set is obtained by performing feature analysis on the parameter impact correlation map throughout the entire process. An optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization objective system includes structural parameter objectives in the design stage, structural manufacturing objectives in the process stage, and structural stability objectives in the usage stage. An initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain the initial scheme. Obtain performance verification data and risk assessment results corresponding to the initial solution; The weight allocation of the target system is dynamically adjusted and optimized based on the verification data and risk assessment results. The initial scheme is optimized based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and the final product structure parameterized design optimization result is output.

[0006] Preferably, multi-source heterogeneous data on product structure design is collected. This multi-source heterogeneous data includes parameter design documents, simulation data, process execution records, and failure case reports. Based on this multi-source heterogeneous data, a full-process parameter impact correlation map is generated, specifically including the following steps: Collect detailed parameter documents, CAD model parameters, finite element simulation data, production line process parameter execution records, quality inspection reports, and cause analysis reports of historical failure cases during the product structure design process; Integrate the data formats, parameter encodings, and measurement standards of multi-source heterogeneous data to transform unstructured failure case reports into risk impact factors; Based on the structural parameters and performance impact dimensions, a mapping relationship between parameters and influencing factors throughout the entire process is generated. The performance impact dimensions include the design compliance performance dimension, process adaptability performance dimension, and stability performance dimension. A parameter impact correlation map throughout the entire process is generated based on the mapping relationship between parameters and influencing factors throughout the entire process.

[0007] Preferably, the key constraint parameter set is obtained by performing feature judgment on the correlation map of the entire parameter process, specifically including the following steps: The entire process of obtaining parameters affects the association edge attributes of each parameter node in the association graph. Based on the influence strength and transmission path length of the association edge attributes, the node levels of the parameter nodes are divided. The parameter node with the most influence edge coverage in each node level is extracted as a candidate constraint parameter. Determine the state fluctuation range of candidate constraint parameters at different process stages, select candidate constraint parameters whose state fluctuation range exceeds the preset range, and combine the impact of candidate constraint parameters on downstream process nodes to filter out candidate constraint parameters with constraint attributes to obtain a constraint parameter set. Based on the number of closed loops associated with each parameter node in the parameter full-process influence correlation graph, the core degree of each parameter node in the parameter full-process influence correlation graph is determined; the constraint parameter set that meets the preset standard of core degree and simultaneously satisfies the fluctuation amplitude and influence conditions is merged to form the key constraint parameter set.

[0008] Preferably, an optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. This optimization objective system includes structural parameter objectives in the design phase, structural manufacturing objectives in the process phase, and structural stability objectives in the usage phase. Specifically, it includes the following steps: Prioritize the impact of key constraint parameter sets on the entire product process based on the breadth and duration of their influence; identify the interactive constraint relationships between each key constraint parameter set, including the coupling and balancing degree of parameter values. Based on priority ranking, pre-set structural parameter targets for the highest priority set of key constraint parameters, and determine the implementation boundaries of the core functions of the product and the basic parameters based on the structural parameter targets; Based on the combination of structural manufacturing objectives and coupling degree, the production process is adapted to form processing parameter values; By combining the balance relationship of parameters, a structural stability target is established for the use phase, and the range of coordinated fluctuation of the balance parameters under the operating conditions is determined. Based on the dynamic change characteristics of the parameter interaction constraint relationship and the basic parameters, the processing parameter values ​​are adapted and corrected to obtain the corrected parameter values. The corrected parameter values ​​are matched and adjusted with the coordinated fluctuation range to obtain the coordinated fluctuation matching degree. Based on the coordinated fluctuation matching degree, the initial value range of the design stage target is constrained to form an optimization target system.

[0009] Preferably, an initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. This process specifically includes the following steps: The range of values ​​for key constraint parameters and the optimization objective system are combined to form candidate solutions, and the candidate solutions are integrated to form an initial solution pool. Historical case data from the parameter-related graph throughout the entire process is retrieved. The initial solution pool is matched and verified by combining the correspondence between historical case data and target achievement results, and candidate solutions that achieve the preset threshold in historical cases are selected. By combining the risk trigger records of candidate solutions in historical cases, we identify the potential risk association characteristics of candidate solutions in the initial solution pool, screen out candidate solutions with similar risk association characteristics, and finally obtain the basic constraints that meet the objectives of each stage. Based on the basic constraints, we adapt the candidate solutions in the historical effective cases to obtain the initial solution.

[0010] Preferably, obtaining the performance verification data and risk assessment results corresponding to the initial solution specifically includes the following steps: Obtain performance simulation data corresponding to the initial scheme, including indicators such as structural strength, process adaptability accuracy, and operational stability; Select an initial scheme to prepare physical test specimens, conduct full-condition physical tests according to standard test procedures, and collect actual performance data of the specimens and process execution data to obtain performance verification data. By comparing the deviations between simulation data and performance verification data, and combining the risk impact factors in the parameter full-process impact correlation graph, risk level assessment is performed on each initial scheme to generate risk assessment results.

[0011] Preferably, the weight allocation of the target system is dynamically adjusted and optimized based on verification data and risk assessment results. The initial scheme is then optimized based on the weight allocation until it meets the preset multi-dimensional optimization thresholds, and the final product structure parameterized design optimization result is output. This specifically includes the following steps: Based on the differences in the achievement of goals at each stage in the performance verification data, and the degree of impact corresponding to different risk levels in the risk assessment results, the weight allocation of structural parameter goals in the design stage, structural manufacturing goals in the process stage, and structural stability goals in the use stage in the optimization goal system is adaptively adjusted to obtain the weight allocation results. Based on the weight allocation results, the key constraint parameter set in the initial scheme is iteratively optimized, and the value range and interaction adaptation relationship of the key constraint parameter set are corrected to generate an optimized scheme. The performance verification results and risk assessment results corresponding to the optimization scheme are judged and the final product structure parameterized design optimization results are output.

[0012] Preferably, the final product structure parameterized design optimization result is output by judging the performance verification results and risk assessment results corresponding to the optimization scheme, specifically including the following steps: When the performance verification results of the optimization scheme meet the preset target requirements of each stage and the risk assessment results are within the preset risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results. If the performance verification results of the optimization scheme do not meet the preset target requirements of each stage, and the risk assessment results exceed the preset risk range, the performance verification and risk assessment process is repeated for the optimization scheme to obtain a new round of performance verification data and risk assessment results. Based on the risk assessment results, the weight allocation of the optimization target system is recalibrated to generate an iterative optimization scheme. When the performance verification results of the iterative optimization scheme meet the preset target requirements of each stage, and the risk assessment results are within the preset acceptable risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results.

[0013] A deep learning-based product structure parametric design optimization system includes: Data Acquisition Module: Collects multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports. Generates a full-process parameter impact correlation map based on the multi-source heterogeneous data. Judgment module: Performs feature judgment on the correlation graph of the influence of parameters throughout the entire process to obtain the set of key constraint parameters; Optimization module: Generates an optimization target system based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization target system includes structural parameter targets in the design phase, structural manufacturing targets in the process phase, and structural stability targets in the usage phase. Screening module: Based on the optimization target system, an initial parameter optimization scheme pool is generated. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. Acquisition module: Acquires performance verification data and risk assessment results corresponding to the initial solution; Output module: Based on the verification data and risk assessment results, dynamically adjust and optimize the weight allocation of the target system, optimize the initial scheme based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and output the final product structure parameterized design optimization results.

[0014] Compared with existing technologies, this invention has the following beneficial effects: By constructing a parameter full-process influence correlation map, the interactive constraint relationship of parameters in the design, process, and usage stages is clarified. Based on the parameter full-process influence correlation map, an optimization target system covering the three stages is generated. This can synchronously correlate the design compliance, process adaptability, and usage stability of structural parameters, avoiding the problems of insufficient production adaptability and poor usage stability caused by focusing only on a single stage target. This improves the parameter synergy throughout the product lifecycle and shortens the product R&D iteration cycle. By integrating heterogeneous data such as parameter documents, simulation data, process records, and failure cases, and transforming unstructured information into risk impact factors, a mapping relationship between parameters and full-process influence factors is constructed using deep learning. This reduces the recurrence rate of similar failure problems and improves the reuse rate of historical data, significantly reducing the design trial-and-error costs caused by data fragmentation. Based on performance verification results and risk assessment data, the weights of the optimization target system are dynamically adjusted, and parameter values ​​are corrected through iterative optimization. This facilitates parameter adjustment to reduce deviations, improves the compliance rate of multi-dimensional optimization thresholds, and improves the product's process qualification rate and usage stability indicators, significantly enhancing the feasibility and reliability of the solution. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the steps of a deep learning-based product structure parameterization design optimization method according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the steps in forming an optimization target system in a deep learning-based product structure parameterization design optimization method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a deep learning-based product structure parameterization design optimization system provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0019] Reference Figures 1-3 As shown.

[0020] Example 1 further illustrates the deep learning-based product structure parameterization design optimization method and system proposed in this invention.

[0021] A deep learning-based product structure parametric design optimization method, comprising the following steps: Collect multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records and failure case reports, and generate a parameter full-process influence correlation map based on the multi-source heterogeneous data; The key constraint parameter set is obtained by performing feature analysis on the parameter impact correlation map throughout the entire process. An optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization objective system includes structural parameter objectives in the design stage, structural manufacturing objectives in the process stage, and structural stability objectives in the usage stage. An initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain the initial scheme. Obtain performance verification data and risk assessment results corresponding to the initial solution; The weight allocation of the target system is dynamically adjusted and optimized based on the verification data and risk assessment results. The initial scheme is optimized based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and the final product structure parameterized design optimization result is output.

[0022] Collect multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports. Generate a full-process parameter impact correlation map based on the multi-source heterogeneous data, specifically including the following steps: Collect detailed parameter documents, CAD model parameters, finite element simulation data, production line process parameter execution records, quality inspection reports, and cause analysis reports of historical failure cases during the product structure design process; Integrate the data formats, parameter encodings, and measurement standards of multi-source heterogeneous data to transform unstructured failure case reports into risk impact factors; Based on the structural parameters and performance impact dimensions, a mapping relationship between parameters and influencing factors throughout the entire process is generated. The performance impact dimensions include the design compliance performance dimension, process adaptability performance dimension, and stability performance dimension. A parameter impact correlation map throughout the entire process is generated based on the mapping relationship between parameters and influencing factors throughout the entire process.

[0023] First, we comprehensively collect various data related to product structure design to form multi-source heterogeneous data. This data covers the entire process, including design, simulation, production, and quality. For example, it includes parameter details documents and CAD model parameters in the design phase, finite element simulation data in the simulation phase, production line process parameter execution records in the production process, as well as quality inspection reports and cause analysis reports of historical failure cases. Taking gearboxes as an example, we collect the tolerance requirements for gear module and tooth width, the three-dimensional geometric dimensions of each component, the stress distribution results of gears under rated load, the machine tool speed and tool feed rate during gear machining, the actual measurement data of gear dimensions, and records of past gear tooth breakage problems caused by module deviation.

[0024] Next, these multi-source heterogeneous data are processed uniformly. Since different data formats, encoding, and measurement standards vary, they need to be integrated first. For example, finite element simulation data and process execution records in different formats are converted into a unified numerical format; parameter names with different encoding rules are mapped to the same naming system; and unstructured failure case reports are transformed into risk impact factors. For instance, the description of a gear module deviation of 0.2mm leading to tooth breakage in a failure case is transformed into a correlation factor between module deviation and tooth breakage risk. For example, for every 0.1mm increase in module deviation, the risk of tooth breakage increases by 20%.

[0025] Finally, based on the mapping relationship between structural parameters and performance impact dimensions, a full-process parameter impact correlation map is generated. Performance impact dimensions include three dimensions: design compliance, process adaptability, and stability. Taking the gear module of a gearbox as an example, the design compliance dimension corresponds to whether it conforms to industry size standards; the process adaptability dimension corresponds to whether the module is suitable for the machining accuracy of the production line; and the stability dimension corresponds to the fatigue strength of the gear at that module. By clarifying the influence relationship between the gear module and other parameters such as bearing type and housing wall thickness under these three dimensions, the mapping between this parameter and factors in each stage of the entire process is identified. Finally, the mapping relationships of all parameters are integrated to form a full-process parameter impact correlation map for the gearbox.

[0026] The key constraint parameter set is obtained by performing feature analysis on the parameter impact correlation graph throughout the entire process, specifically including the following steps: The entire process of obtaining parameters affects the association edge attributes of each parameter node in the association graph. Based on the influence strength and transmission path length of the association edge attributes, the node levels of the parameter nodes are divided. The parameter node with the most influence edge coverage in each node level is extracted as a candidate constraint parameter. Determine the state fluctuation range of candidate constraint parameters at different process stages, select candidate constraint parameters whose state fluctuation range exceeds the preset range, and combine the impact of candidate constraint parameters on downstream process nodes to filter out candidate constraint parameters with constraint attributes to obtain a constraint parameter set. Based on the number of closed loops associated with each parameter node in the parameter full-process influence correlation graph, the core degree of each parameter node in the parameter full-process influence correlation graph is determined; the constraint parameter set that meets the preset standard of core degree and simultaneously satisfies the fluctuation amplitude and influence conditions is merged to form the key constraint parameter set.

[0027] First, the association edge attributes of each parameter node in the parameter influence relationship graph are analyzed to classify the node hierarchy. The association edge attributes include influence strength and transmission path length. Influence strength represents the degree to which one parameter affects another, while transmission path length represents the hierarchical span of the influence relationship between parameters. Taking the structural parameters of a gearbox as an example, for the gear module and tooth width parameter nodes, higher influence strength and shorter transmission path lengths indicate a more significant direct influence. Based on these two indicators, all parameter nodes can be divided into different levels. Then, the parameter node with the most influence edge coverage in each level is extracted as a candidate constraint parameter. For example, if the gear module has association edges with bearing type and gearbox wall thickness in multiple levels, it is selected as a candidate constraint parameter.

[0028] Next, the candidate constraint parameters are evaluated based on both state fluctuation and downstream impact to select those with constraint attributes. State fluctuation amplitude refers to the range of numerical changes of the candidate parameter across different process stages: design, manufacturing, and usage. If the fluctuation amplitude of a candidate parameter exceeds a preset range, its impact on downstream process nodes needs further evaluation. For example, if the initial value of the gear module in the design stage and the actual processed value in the manufacturing stage fluctuate beyond the preset tolerance range, and this fluctuation directly leads to insufficient load capacity of downstream bearings, then this gear module will be included in the constraint parameter set.

[0029] Finally, the coreity of the parameter nodes is combined to form a set of key constraint parameters. Coreity is determined by the number of closed loops in the parameter's influence graph throughout the entire process. A closed loop refers to the cyclical influence relationship formed between parameters; the more closed loops, the higher the coreity of the parameter. The formula for calculating coreity is: Coreity = Number of closed loops / Total number of closed loops in the graph × 100. When the coreity reaches a preset standard and simultaneously satisfies the fluctuation amplitude and influence conditions, the corresponding constraint parameters will be merged. For example, if the gear module has a large number of closed loops, reaches a preset threshold in coreity, and simultaneously satisfies the fluctuation and influence constraints, then it will form a set of key constraint parameters together with other qualified parameters, such as the housing wall thickness.

[0030] An optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. This optimization objective system includes structural parameter objectives in the design phase, structural manufacturing objectives in the process phase, and structural stability objectives in the usage phase. Specifically, it includes the following steps: Prioritize the impact of key constraint parameter sets on the entire product process based on the breadth and duration of their influence; identify the interactive constraint relationships between each key constraint parameter set, including the coupling and balancing degree of parameter values. Based on priority ranking, pre-set structural parameter targets for the highest priority set of key constraint parameters, and determine the implementation boundaries of the core functions of the product and the basic parameters based on the structural parameter targets; Based on the combination of structural manufacturing objectives and coupling degree, the production process is adapted to form processing parameter values; By combining the balance relationship of parameters, a structural stability target is established for the use phase, and the range of coordinated fluctuation of the balance parameters under the operating conditions is determined. Based on the dynamic change characteristics of the parameter interaction constraint relationship and the basic parameters, the processing parameter values ​​are adapted and corrected to obtain the corrected parameter values. The corrected parameter values ​​are matched and adjusted with the coordinated fluctuation range to obtain the coordinated fluctuation matching degree. Based on the coordinated fluctuation matching degree, the initial value range of the design stage target is constrained to form an optimization target system.

[0031] First, the key constraint parameters are prioritized and interactive constraint relationships are identified. Prioritization is based on the breadth and duration of the parameter's impact on the entire product process. Breadth of impact refers to the number of process stages affected by the parameter, and duration of impact refers to the length of the period during which the parameter's influence lasts. The priority calculation formula is: Priority = (Breadth of impact × 0.6 + Duration of impact × 0.4) / Maximum of both × 100, with higher values ​​indicating higher priority. Simultaneously, interactive constraint relationships between parameters are identified, including coupling degree and check / balance degree. Coupling degree refers to the degree of correlation between parameter values, such as the interdependence between the gear module and bearing type in a gearbox. Check / balance degree refers to the mutual restriction between parameters, such as how increasing the gearbox wall thickness limits its lightweighting.

[0032] Next, structural parameter targets are constructed based on priorities during the design phase. Taking a gearbox as an example, if the gear module is the highest priority key constraint parameter, structural parameter targets will be preset for it first. For example, the module will be set to a specific value to determine the realization boundary of the core function of gearbox transmission efficiency, while clarifying the value range of the basic parameter of gear tooth count.

[0033] Subsequently, the coupling degree is used to form the structural manufacturing target for the process stage. Based on the coupling degree between the structural manufacturing target and the parameters, the production process is adapted to determine the processing parameter values. For example, the coupling degree between gear module and tooth width is high. When determining the processing parameters of gear module, the processing accuracy of tooth width needs to be adapted simultaneously to ensure that the processing parameters of both can match the equipment capabilities of the production line and form feasible processing parameter values.

[0034] Subsequently, structural stability targets for the usage phase are established based on the balance relationships between parameters. By combining the balance relationships between parameters, the coordinated fluctuation range of parameters under operating conditions is determined. For example, there is a balance between the housing wall thickness and the gearbox weight. During the usage phase, the deformation range of the wall thickness under vibration conditions needs to be clearly defined, while simultaneously matching the fluctuation range of the weight to ensure that structural stability is maintained even when both fluctuate in tandem.

[0035] Finally, the parameters are dynamically adjusted to form an optimization target system. Based on the dynamic changes of the parameter interaction constraint relationship and the basic parameters, the processing parameter values ​​are adapted and adjusted. After obtaining the adjusted parameter values, they are matched with the cooperative fluctuation range, and the cooperative fluctuation matching degree is calculated. The formula for calculating the cooperative fluctuation matching degree is: matching degree = length of the overlap interval between the adjusted parameter value and the fluctuation range / total length of the fluctuation range × 100. When the matching degree reaches the preset requirement, it is used to constrain the initial value range of the design stage target, and finally form an optimization target system covering the design, process, and use stages.

[0036] An initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. This process includes the following steps: The range of values ​​for key constraint parameters and the optimization objective system are combined to form candidate solutions, and the candidate solutions are integrated to form an initial solution pool. Historical case data from the parameter-related graph throughout the entire process is retrieved. The initial solution pool is matched and verified by combining the correspondence between historical case data and target achievement results, and candidate solutions that achieve the preset threshold in historical cases are selected. By combining the risk trigger records of candidate solutions in historical cases, we identify the potential risk association characteristics of candidate solutions in the initial solution pool, screen out candidate solutions with similar risk association characteristics, and finally obtain the basic constraints that meet the objectives of each stage. Based on the basic constraints, we adapt the candidate solutions in the historical effective cases to obtain the initial solution.

[0037] First, an initial solution pool is constructed by combining the value range of the key constraint parameter set with the optimization objective system. The value range of the key constraint parameter set refers to the feasible range of core parameters, such as the gear module and housing wall thickness. The optimization objective system covers the target requirements of the design, process, and usage stages. Taking a gearbox as an example, the value range of the gear module (e.g., 2mm-3mm) and the value range of the housing wall thickness (e.g., 5mm-8mm) are combined with the objectives of each stage, such as transmission efficiency ≥95% in the design stage and processing qualification rate ≥98% in the process stage, to form multiple candidate solutions. These candidate solutions are then integrated to obtain the initial parameter optimization solution pool.

[0038] Next, historical case data is used to match and verify the initial solution pool. The historical case data in the correlation graph, which reflects the overall impact of parameters, includes the correspondence between past gearbox parameter solutions and the achieved target results. For example, in historical solutions, when the gear module is 2.5mm and the gearbox wall thickness is 6mm, the transmission efficiency reaches 96% and the processing qualification rate reaches 99%. At this point, preset thresholds for achieving the target are set, such as transmission efficiency ≥ 95% and processing qualification rate ≥ 98%. The candidate solutions in the initial solution pool are matched with historical cases, and solutions that meet the thresholds for achieving the target are selected. For example, if the parameter combination of a candidate solution corresponds to a transmission efficiency and processing qualification rate that both meet the preset requirements, this round of verification is passed.

[0039] Finally, high-risk solutions are screened out based on historical risk records to obtain the initial solution. Historical cases contain risk trigger records for candidate solutions; for example, parameter combinations may have caused abnormal noises due to mismatches between gear module and bearing type. Based on these records, potential risk association characteristics of candidate solutions in the initial solution pool are identified. For example, if the gear module and bearing type combination in the current solution matches the characteristics of historical risk solutions, that solution will be screened out. After risk screening, basic constraints that meet the objectives of each stage are obtained. These constraints are then used to adapt candidate solutions from historically valid cases to ultimately determine the initial solution for gearbox parameter design. For example, a parameter combination of 2.5mm gear module and 6mm gearbox wall thickness satisfies the objectives of each stage and has no historical risk association characteristics.

[0040] Obtaining performance verification data and risk assessment results corresponding to the initial solution includes the following steps: Obtain performance simulation data corresponding to the initial scheme, including indicators such as structural strength, process adaptability accuracy, and operational stability; Select an initial scheme to prepare physical test specimens, conduct full-condition physical tests according to standard test procedures, and collect actual performance data of the specimens and process execution data to obtain performance verification data. By comparing the deviations between simulation data and performance verification data, and combining the risk impact factors in the parameter full-process impact correlation graph, risk level assessment is performed on each initial scheme to generate risk assessment results.

[0041] First, performance simulation data corresponding to the initial design is obtained. This data covers key indicators such as structural strength, process adaptability accuracy, and operational stability. Taking the initial design of the gearbox as an example, for the parameter combination of a gear module of 2.5mm and a housing wall thickness of 6mm, simulation software is used to simulate its structural strength under rated load, such as the stress value at the gear tooth root; process adaptability accuracy, such as whether the dimensional deviation of gear machining meets the accuracy range of the production line equipment; and operational stability, such as the vibration amplitude after long-term operation. The corresponding simulation data are obtained, such as a tooth root stress of 120MPa, a machining dimensional deviation of 0.02mm, and a vibration amplitude of 0.1mm / s.

[0042] Next, physical test specimens were fabricated and full-condition tests were conducted to collect actual performance data. Based on the initial design of the gearbox, corresponding physical specimens were machined, and then full-condition physical tests were carried out according to standard testing procedures. For example, the gearbox was simulated to run continuously at rated speed and under full load. At the same time, actual structural strength data, such as the actual stress at the tooth root measured by stress sensors, and process execution data, such as the actual accuracy deviation of the machine tool during machining, were recorded. These data were integrated to obtain performance verification data, such as the actual tooth root stress being 125 MPa and the actual machining deviation being 0.03 mm.

[0043] Finally, the deviations between the simulation and verification data are compared, and a risk assessment is completed in conjunction with risk impact factors. The deviation is calculated as follows: Deviation value = (|Performance verification data - Simulation data| ÷ Simulation data) / 100. Taking the gear root stress of the gearbox as an example, the deviation value is: (|125 - 120| ÷ 120) / 100 ≈ 4.17%. Then, the risk impact factors in the parameter full-process influence correlation graph are considered. For example, in historical cases, a deviation of more than 5% in the gear root stress increases the risk of fracture. In this case, the risk threshold is 5%. A risk level assessment is performed on the initial scheme: since the current deviation of 4.17% < 5% does not exceed the risk threshold, the risk level can be determined as low. Finally, the corresponding risk assessment result is generated, clarifying the performance compliance and risk level of the initial scheme.

[0044] The weight allocation of the target system is dynamically adjusted and optimized based on the verification data and risk assessment results. The initial scheme is then optimized based on the weight allocation until it meets the preset multi-dimensional optimization thresholds, and the final product structure parameterized design optimization results are output. The specific steps include: Based on the differences in the achievement of goals at each stage in the performance verification data, and the degree of impact corresponding to different risk levels in the risk assessment results, the weight allocation of structural parameter goals in the design stage, structural manufacturing goals in the process stage, and structural stability goals in the use stage in the optimization goal system is adaptively adjusted to obtain the weight allocation results. Based on the weight allocation results, the key constraint parameter set in the initial scheme is iteratively optimized, and the value range and interaction adaptation relationship of the key constraint parameter set are corrected to generate an optimized scheme. The performance verification results and risk assessment results corresponding to the optimization scheme are judged and the final product structure parameterized design optimization results are output.

[0045] First, based on performance verification data and risk assessment results, the weight allocation of the target system is adjusted and optimized. The difference in the achievement of targets at each stage refers to the gap between the actual completion of the design, process, and usage stage targets and the preset standards. The degree of risk impact refers to the magnitude of the effect of different risk levels on the feasibility of the solution. Taking a gearbox as an example, if performance verification data shows that the structural parameter targets in the design stage, such as transmission efficiency, are achieved at 98%, the structural manufacturing targets in the process stage, such as processing qualification rate, are achieved at 92%, and the structural stability targets in the usage stage are achieved at 95%, while the risk assessment results show that the risk level in the process stage is medium and the other stages are low, then the weight of the targets in the process stage should be increased accordingly. The adjustment of the weight allocation can be calculated by adjusting the weight = original weight × target achievement / average achievement of each stage + risk impact coefficient. For example, if the original weight of the process stage is 0.3, the average achievement of each stage is 95%, and the risk impact coefficient corresponds to 1.2, then the adjusted weight is 0.3 × (92% / 95% + 1.2) ≈ 0.65, thus obtaining the new weight allocation result.

[0046] Next, based on the new weight allocation results, the key constraint parameter set of the initial scheme is iteratively optimized. Taking a gearbox as an example, the key constraint parameter set includes the gear module and the gearbox wall thickness. According to the results of the weight increase in the process stage, parameters related to process adaptation will be adjusted first. For example, the value range of the gear module will be slightly adjusted from 2.5mm to 2.4mm to correct its interaction adaptation relationship with the accuracy of the processing equipment. At the same time, the gearbox wall thickness will be appropriately adjusted to match the module change, thereby generating a new optimized scheme.

[0047] Finally, the performance and risks of the optimized scheme are verified, and the final result is output. For the adjusted gearbox parameter scheme, the performance verification results and risk assessment results are obtained again to determine whether it meets the preset multi-dimensional optimization thresholds, such as transmission efficiency ≥95%, processing qualification rate ≥98%, and risk level ≤0. If the processing qualification rate corresponding to the gear module of 2.4mm in the optimized scheme is improved to 98%, all stage targets are met, and the risk level is low, then this parameter combination will be output as the final product structure parameterized design optimization result.

[0048] The final product structure parameterized design optimization result is output by judging the performance verification results and risk assessment results corresponding to the optimization scheme, which includes the following steps: When the performance verification results of the optimization scheme meet the preset target requirements of each stage and the risk assessment results are within the preset risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results. If the performance verification results of the optimization scheme do not meet the preset target requirements of each stage, and the risk assessment results exceed the preset risk range, the performance verification and risk assessment process is repeated for the optimization scheme to obtain a new round of performance verification data and risk assessment results. Based on the risk assessment results, the weight allocation of the optimization target system is recalibrated to generate an iterative optimization scheme. When the performance verification results of the iterative optimization scheme meet the preset target requirements of each stage, and the risk assessment results are within the preset acceptable risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results.

[0049] First, the performance verification results and risk assessment results of the optimized scheme are preliminarily determined. The risk level is quantified into a numerical range of 0-3, where 0 corresponds to low risk, 1 to medium risk, 2 to high risk, and 3 to extremely high risk. The preset target requirements for each stage include a transmission efficiency ≥95% in the design stage, a processing qualification rate ≥98% in the process stage, and a vibration amplitude ≤0.15mm / s in the usage stage. The preset risk range is a risk level ≤0. Taking the optimized scheme of the gearbox as an example, if its performance verification results show a transmission efficiency of 96%, a processing qualification rate of 99%, and a vibration amplitude of 0.12mm / s, and the risk assessment result is low risk, then the scheme meets the preset multi-dimensional optimization thresholds, and this parameter combination is directly output as the final product structure parameterized design optimization result.

[0050] If the optimized solution does not meet the requirements, it will proceed to the iterative optimization process. For example, if the performance verification results of the gearbox optimization solution show that the processing qualification rate is only 95%, which is below the 98% standard, and the risk assessment results show that there is a medium risk, then the performance verification and risk assessment process needs to be repeated. First, the physical prototype of the solution is remade, and full-condition tests are conducted to obtain a new round of performance data. At the same time, the risk assessment results are updated in combination with new risk impact factors. Then, the weight allocation of the optimization target system is calibrated based on these new results. For example, the weight of the process stage target is further increased. Based on the new weights, key constraint parameters such as fine-tuning the value of the gear module are adjusted to generate an iterative optimization solution.

[0051] Finally, the iterative optimization scheme is evaluated again. Taking the adjusted gearbox iterative scheme as an example, if the performance verification results show that the processing qualification rate has increased to 98%, the targets of each stage meet the preset requirements, and the risk assessment results fall back to the low-risk range, then the iterative scheme meets the multi-dimensional optimization threshold. At this time, the parameter combination corresponding to the scheme is output as the final product structure parameterized design optimization result.

[0052] A deep learning-based product structure parametric design optimization system includes: Data Acquisition Module: Collects multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports. Generates a full-process parameter impact correlation map based on the multi-source heterogeneous data. Judgment module: Performs feature judgment on the correlation graph of the influence of parameters throughout the entire process to obtain the set of key constraint parameters; Optimization module: Generates an optimization target system based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization target system includes structural parameter targets in the design phase, structural manufacturing targets in the process phase, and structural stability targets in the usage phase. Screening module: Based on the optimization target system, an initial parameter optimization scheme pool is generated. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. Acquisition module: Acquires performance verification data and risk assessment results corresponding to the initial solution; Output module: Based on the verification data and risk assessment results, dynamically adjust and optimize the weight allocation of the target system, optimize the initial scheme based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and output the final product structure parameterized design optimization results.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based product structure parametric design optimization method, characterized in that, The method includes the following steps: Collect multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records and failure case reports, and generate a parameter full-process influence correlation map based on the multi-source heterogeneous data; The key constraint parameter set is obtained by performing feature analysis on the parameter impact correlation map throughout the entire process. An optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization objective system includes structural parameter objectives in the design stage, structural manufacturing objectives in the process stage, and structural stability objectives in the usage stage. An initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain the initial scheme. Obtain performance verification data and risk assessment results corresponding to the initial solution; The weight allocation of the target system is dynamically adjusted and optimized based on the verification data and risk assessment results. The initial scheme is optimized based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and the final product structure parameterized design optimization result is output.

2. The deep learning-based product structure parameterization design optimization method according to claim 1, characterized in that, Collect multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports. Generate a full-process parameter impact correlation map based on the multi-source heterogeneous data, specifically including the following steps: Collect detailed parameter documents, CAD model parameters, finite element simulation data, production line process parameter execution records, quality inspection reports, and cause analysis reports of historical failure cases during the product structure design process; Integrate the data formats, parameter encodings, and measurement standards of multi-source heterogeneous data to transform unstructured failure case reports into risk impact factors; Based on the structural parameters and performance impact dimensions, a mapping relationship between parameters and influencing factors throughout the entire process is generated. The performance impact dimensions include the design compliance performance dimension, process adaptability performance dimension, and stability performance dimension. A parameter impact correlation map throughout the entire process is generated based on the mapping relationship between parameters and influencing factors throughout the entire process.

3. The deep learning-based product structure parameterization design optimization method according to claim 2, characterized in that, The key constraint parameter set is obtained by performing feature analysis on the parameter impact correlation graph throughout the entire process, specifically including the following steps: The entire process of obtaining parameters affects the association edge attributes of each parameter node in the association graph. Based on the influence strength and transmission path length of the association edge attributes, the node levels of the parameter nodes are divided. The parameter node with the most influence edge coverage in each node level is extracted as a candidate constraint parameter. Determine the state fluctuation range of candidate constraint parameters at different process stages, select candidate constraint parameters whose state fluctuation range exceeds the preset range, and combine the impact of candidate constraint parameters on downstream process nodes to filter out candidate constraint parameters with constraint attributes to obtain a constraint parameter set. Based on the number of closed loops associated with each parameter node in the parameter full-process influence correlation graph, the core degree of each parameter node in the parameter full-process influence correlation graph is determined; the constraint parameter set that meets the preset standard of core degree and simultaneously satisfies the fluctuation amplitude and influence conditions is merged to form the key constraint parameter set.

4. The deep learning-based product structure parameterization design optimization method according to claim 3, characterized in that, An optimization objective system is generated based on the priority ranking and interaction constraint relationships of the key constraint parameter set. This optimization objective system includes structural parameter objectives in the design phase, structural manufacturing objectives in the process phase, and structural stability objectives in the usage phase. Specifically, it includes the following steps: Prioritize the impact of key constraint parameter sets on the entire product process based on the breadth and duration of their influence; identify the interactive constraint relationships between each key constraint parameter set, including the coupling and balancing degree of parameter values. Based on priority ranking, pre-set structural parameter targets for the highest priority set of key constraint parameters, and determine the implementation boundaries of the core functions of the product and the basic parameters based on the structural parameter targets; Based on the combination of structural manufacturing objectives and coupling degree, the production process is adapted to form processing parameter values; By combining the balance relationship of parameters, a structural stability target is established for the use phase, and the range of coordinated fluctuation of the balance parameters under the operating conditions is determined. Based on the dynamic change characteristics of the parameter interaction constraint relationship and the basic parameters, the processing parameter values ​​are adapted and corrected to obtain the corrected parameter values. The corrected parameter values ​​are matched and adjusted with the coordinated fluctuation range to obtain the coordinated fluctuation matching degree. Based on the coordinated fluctuation matching degree, the initial value range of the design stage target is constrained to form an optimization target system.

5. The deep learning-based product structure parameterization design optimization method according to claim 4, characterized in that, An initial parameter optimization scheme pool is generated based on the optimization target system. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. This process includes the following steps: The range of values ​​for key constraint parameters and the optimization objective system are combined to form candidate solutions, and the candidate solutions are integrated to form an initial solution pool. Historical case data from the parameter-related graph throughout the entire process is retrieved. The initial solution pool is matched and verified by combining the correspondence between historical case data and target achievement results, and candidate solutions that achieve the preset threshold in historical cases are selected. By combining the risk trigger records of candidate solutions in historical cases, we identify the potential risk association characteristics of candidate solutions in the initial solution pool, screen out candidate solutions with similar risk association characteristics, and finally obtain the basic constraints that meet the objectives of each stage. Based on the basic constraints, we adapt the candidate solutions in the historical effective cases to obtain the initial solution.

6. The deep learning-based product structure parameterization design optimization method according to claim 5, characterized in that, Obtaining performance verification data and risk assessment results corresponding to the initial solution includes the following steps: Obtain performance simulation data corresponding to the initial scheme, including indicators such as structural strength, process adaptability accuracy, and operational stability; Select an initial scheme to prepare physical test specimens, conduct full-condition physical tests according to standard test procedures, and collect actual performance data of the specimens and process execution data to obtain performance verification data. By comparing the deviations between simulation data and performance verification data, and combining the risk impact factors in the parameter full-process impact correlation graph, risk level assessment is performed on each initial scheme to generate risk assessment results.

7. The deep learning-based product structure parameterization design optimization method according to claim 6, characterized in that, The weight allocation of the target system is dynamically adjusted and optimized based on the verification data and risk assessment results. The initial scheme is then optimized based on the weight allocation until it meets the preset multi-dimensional optimization thresholds, and the final product structure parameterized design optimization results are output. The specific steps include: Based on the differences in the achievement of goals at each stage in the performance verification data, and the degree of impact corresponding to different risk levels in the risk assessment results, the weight allocation of structural parameter goals in the design stage, structural manufacturing goals in the process stage, and structural stability goals in the use stage in the optimization goal system is adaptively adjusted to obtain the weight allocation results. Based on the weight allocation results, the key constraint parameter set in the initial scheme is iteratively optimized, and the value range and interaction adaptation relationship of the key constraint parameter set are corrected to generate an optimized scheme. The performance verification results and risk assessment results corresponding to the optimization scheme are judged and the final product structure parameterized design optimization results are output.

8. The deep learning-based product structure parameterization design optimization method according to claim 7, characterized in that, The final product structure parameterized design optimization result is output by judging the performance verification results and risk assessment results corresponding to the optimization scheme, which includes the following steps: When the performance verification results of the optimization scheme meet the preset target requirements of each stage and the risk assessment results are within the preset risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results. If the performance verification results of the optimization scheme do not meet the preset target requirements of each stage, and the risk assessment results exceed the preset risk range, the performance verification and risk assessment process is repeated for the optimization scheme to obtain a new round of performance verification data and risk assessment results. Based on the risk assessment results, the weight allocation of the optimization target system is recalibrated to generate an iterative optimization scheme. When the performance verification results of the iterative optimization scheme meet the preset target requirements of each stage, and the risk assessment results are within the preset acceptable risk range, the scheme meets the preset multi-dimensional optimization threshold and outputs the final product structure parameterized design optimization results.

9. A deep learning-based product structure parametric design optimization system, applied to the deep learning-based product structure parametric design optimization method described in claims 1-8, characterized in that, include: Data Acquisition Module: Collects multi-source heterogeneous data on product structure design, including parameter design documents, simulation data, process execution records, and failure case reports. Generates a full-process parameter impact correlation map based on the multi-source heterogeneous data. Judgment module: Performs feature judgment on the correlation graph of the influence of parameters throughout the entire process to obtain the set of key constraint parameters; Optimization module: Generates an optimization target system based on the priority ranking and interaction constraint relationships of the key constraint parameter set. The optimization target system includes structural parameter targets in the design phase, structural manufacturing targets in the process phase, and structural stability targets in the usage phase. Screening module: Based on the optimization target system, an initial parameter optimization scheme pool is generated. The scheme pool is then preliminarily screened using historical case data from the parameter full-process influence correlation graph to obtain initial schemes. Acquisition module: Acquires performance verification data and risk assessment results corresponding to the initial solution; Output module: Based on the verification data and risk assessment results, dynamically adjust and optimize the weight allocation of the target system, optimize the initial scheme based on the weight allocation until it meets the preset multi-dimensional optimization threshold, and output the final product structure parameterized design optimization results.

Images