A Diesel Engine Structural Design Parameter Optimization Method Based on Multi-Level Constraint Transfer

Abstract

This invention discloses a method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer, belonging to the field of computer-aided design technology. The method includes the following steps: collecting design information and inter-layer constraints to generate multi-layer design data; constructing hierarchical nodes and edges, configuring constraint attributes and source identifiers; inputting an improved CompGCN model to obtain upper and lower bound constraint states and source paths; comparing upper and lower bound states to determine conflict parameters and adjustment ranges; locally optimizing conflict parameters to obtain parameters to be verified; conducting structural simulation to match engineering constraints and generate verification deviations; reapplying deviations along the source paths to correct corresponding polarity constraints; and optimizing the simulation again according to the correction range to output parameters after satisfying the constraints. This invention improves the accuracy of diesel engine structural parameter conflict identification and closed-loop optimization by improving the CompGCN model to achieve multi-layer constraint transfer and directional reapplying of verification deviations.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, and in particular to a method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer. Background Technology

[0002] As diesel engines evolve towards higher power density, lighter weight, and higher reliability, the structural design of components such as the engine block, cylinder head, and connecting structures must simultaneously meet requirements in terms of structural strength, fatigue life, deformation control, weight, and spatial arrangement. Current diesel engine structural designs typically determine system technical specifications based on overall engine design requirements, then decompose these specifications to subsystem and component levels. Structural parameters such as wall thickness, stiffener dimensions, fillet radii, hole diameters, and structural spacing are adjusted, and structural performance is verified using finite element simulation or parameter optimization methods to obtain structural parameters that meet design requirements. With the increase in the number of structural parameters and the types of constraints, the relationships between design information at different levels become more complex, and the same structural parameter may also be simultaneously affected by multiple upstream engineering constraints.

[0003] Current technologies for optimizing diesel engine structural parameters often use indicators such as strength, deformation, fatigue safety, and weight as parallel constraints or optimization indicators. Candidate parameters are searched within a preset parameter range, and structural simulation is used to determine whether the candidate parameters meet design requirements. When the same structural parameter is subject to different engineering constraints, multiple constraints are typically calculated comprehensively or the parameter range is adjusted uniformly. This lacks continuous recording of the source relationships and transmission processes between design requirements, system specifications, subsystem engineering constraints, and component structural parameters, making it difficult to distinguish the upper and lower bounds of different upstream constraints on the same structural parameter. Consequently, when different constraints mutually compress the feasible range of parameters, it is difficult to accurately determine the source of conflict and the effective adjustment range, easily leading to repeated parameter adjustments or expanding the optimization search range.

[0004] Furthermore, when structural simulation results show deformation exceeding limits, insufficient fatigue safety margin, or weight not meeting constraints, existing methods typically readjust relevant parameters or re-execute overall optimization based on simulation results. It is difficult to locate the specific upstream constraints that cause verification deviations along the original constraint transmission relationship, and it is also difficult to only correct the parameter boundaries of the corresponding constraints while keeping the effective constraints on the other side unchanged. This can lead to local verification problems causing repeated adjustments to multiple structural parameters and constraint ranges, affecting the pertinence and iterative efficiency of diesel engine structural design parameter optimization.

[0005] Therefore, how to provide a method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer. This invention fully utilizes an improved CompGCN model, multi-layer constraint transfer, and diesel engine structural simulation technology. It details the constraint transfer process between diesel engine structural design requirements, system specifications, subsystem engineering constraints, and component structural parameters. By improving the CompGCN model, it achieves constraint polarity preservation propagation and verification deviation source chain re-attachment. Combined with conflict parameter identification, local optimization, and structural simulation, it completes closed-loop optimization of diesel engine structural design parameters. It has the advantages of accurate constraint transfer, precise conflict location, highly targeted parameter adjustment, and high optimization efficiency.

[0007] A method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to an embodiment of the present invention includes: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters and inter-layer constraint relationships to generate multi-layer design data; Nodes and edges are constructed from multi-layer design data, and constraint polarity, transformation relationships and source identifiers are configured to generate a multi-layer constraint transfer graph. The multi-layer constraint transfer graph is input into the improved CompGCN model to obtain the upper bound constraint state, the lower bound constraint state, and the constraint source path. The improved CompGCN model includes a node embedding layer, a relation embedding layer, a constraint polarity preservation combination layer, a multi-relation aggregation layer, a verification bias source chain backfitting layer, and an output layer. Compare the upper and lower bound constraint states of the same structural parameter, and identify conflicting parameters and adjustment ranges by combining the current parameter values; By locally optimizing and adjusting the conflicting parameters within the range, the parameters to be verified are obtained. Perform diesel engine structural simulation on the parameters to be verified to obtain deformation, fatigue safety and weight results, match subsystem engineering constraints and associate source identifiers to generate verification deviations; Input the verification bias into the improved CompGCN model, paste it back to the corresponding upper or lower bound constraint state along the constraint source path according to the source identifier, and keep the other polarity constraint state unchanged to generate the correction range; Optimize and simulate within the correction range, and output the optimized parameters after the constraints are satisfied.

[0008] Optionally, the diesel engine structural design requirements include structural reliability requirements, lightweight requirements, structural life requirements, and space layout requirements; the system specifications include overall machine weight limits, structural strength indicators, fatigue life indicators, and external dimension limits; the subsystem engineering constraints include deformation constraints, stress constraints, fatigue safety constraints, and weight constraints; the component structural parameters include wall thickness, stiffener dimensions, fillet radius, hole diameter, and structural spacing; and the interlayer constraint relationships include the source correspondence between adjacent layer data, the direction of constraint action, and parameter conversion rules.

[0009] Optionally, generating multi-layer design data includes: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters and inter-layer constraints, establish data items according to their respective design levels, record the current values ​​of component structural parameters, and configure object identifiers and level identifiers for each data item; According to the object identifier, the data items of adjacent levels are matched accordingly, the source correspondence, constraint direction and parameter conversion rules in the inter-level constraint relationship are extracted, and the constraint transmission record between adjacent level data items is established; Arrange the data items according to the hierarchical identifier, associate the constraint transfer records with the corresponding data items, and form multi-layer design data that includes data items, current parameter values, hierarchical identifiers, and inter-layer constraint relationships.

[0010] Optionally, generating the multi-layer constraint transfer graph includes: According to the hierarchy of diesel engine structural design requirements, system specifications, subsystem engineering constraints, and component structural parameters, the data items in the multi-level design data are set as corresponding level nodes. Match adjacent level nodes based on inter-level constraints, and establish edges pointing from upstream nodes to downstream nodes according to the source of constraints; Determine the upper or lower bound constraint polarity based on the direction of the constraint action corresponding to the edge, configure the transformation relationship according to the parameter transformation rules, and configure the source information of the upstream constraint as the source identifier; By associating nodes at each level with the configured edges, a multi-layer constraint transfer graph is generated, which transmits the diesel engine structural design requirements to the component structural parameters layer by layer.

[0011] Optionally, obtaining the upper bound constraint state, lower bound constraint state, and constraint source path includes: Based on the CompGCN model, the node representation, relation representation, node and relation combination propagation, multi-relation aggregation and result output process are set up with a node embedding layer, a relation embedding layer, a multi-relation aggregation layer and an output layer. A constraint polarity preservation combination layer is added at the output convergence position of the node embedding layer and the relation embedding layer, and a verification bias source chain back-fitting layer is added between the multi-relation aggregation layer and the output layer to form an improved CompGCN model. The nodes in the multi-layer constraint transfer graph are input into the node embedding layer. Embedding mapping is performed based on the node's level and data to generate a node embedding representation. The edge input relationship embedding layer in the multi-layer constraint transfer graph is used to perform relationship embedding based on constraint polarity, transformation relationship and source identifier, and generate a relationship embedding representation carrying source identifier; The node embedding representation and relation embedding representation are combined with the input constraint polarity preservation layer. Adjacent nodes and corresponding relations are combined according to the constraint polarity of the edges. The combination results are passed layer by layer along the upper and lower polarity paths, and the source identifier is passed at the same time to generate upper and lower propagation features. The upper and lower bound propagation features are input into the multi-relationship aggregation layer. Propagation features of the same polarity are aggregated according to the component structure parameter nodes. The aggregation results of different polarities are kept independent of each other. The upper and lower bound aggregation states are generated, and the edges that have been propagated are concatenated according to the source identifier to generate the source path candidate set. Input the upper bound aggregation state, lower bound aggregation state, and source path candidate set into the verification deviation source chain back-paste layer, and associate the source path with the corresponding upper bound aggregation state or lower bound aggregation state according to the source identifier to establish the source chain back-paste index. Input the upper bound aggregation state, lower bound aggregation state, and source chain backlink index into the output layer to generate the upper bound constraint state and lower bound constraint state of each component's structural parameters, and generate the corresponding constraint source path based on the source chain backlink index. Training samples were constructed by collecting multi-layer design data, parameter constraint records, structural simulation verification records, and parameter correction records from historical diesel engine structural design cases. The constraint state identification error, source path matching error, and verification deviation re-attachment error were used as joint training objectives to iteratively train the improved CompGCN model until the joint training objectives met the preset convergence conditions, thus obtaining the trained improved CompGCN model.

[0012] Optionally, the step of identifying conflicting parameters and adjusting their ranges based on their current values ​​includes: Read the upper and lower bound constraint states of the structural parameters of the same component, extract the upper and lower bound parameter boundaries respectively, and compare the relative positions of the parameter boundaries on both sides. The current value of the parameter is compared with the upper and lower parameter boundaries respectively. The structural parameters of the component that do not have a common feasible interval between the two parameter boundaries or whose current value exceeds the common feasible interval are identified as conflict parameters. For conflicting parameters that share a common feasible interval, the parameter adjustment range is determined by the common feasible interval; for conflicting parameters that do not share a common feasible interval, the parameter adjustment range is determined by the interval between the boundaries of the two parameters.

[0013] Optionally, the conflict parameters within the scope of the local optimization adjustment, to obtain the parameters to be verified, include: Read the current value, adjustment range, and upper and lower bound constraint states of the conflicting parameters, keep the non-conflicting parameters unchanged, generate local candidate values ​​for each conflicting parameter within the adjustment range, and combine them to form a local candidate parameter group; The local candidate parameter groups are matched with the corresponding upper and lower bound constraint states respectively. The deviation of the candidate values ​​from the upper and lower bound parameter boundaries and the adjustment amount relative to the current parameter values ​​are calculated. The local candidate parameter groups are sorted and filtered according to the boundary deviation and parameter adjustment amount to obtain the target parameter group. Update the corresponding conflicting parameters with the candidate values ​​in the target parameter group, and keep the current values ​​of the remaining component structural parameters unchanged, and combine them to generate the parameters to be verified.

[0014] Optionally, the generation of verification bias includes: Write the parameters to be verified into the corresponding diesel engine component structural model, perform structural deformation, fatigue safety and structural weight simulation, and obtain the structural simulation results; Based on the structural simulation results of the subsystem engineering constraint matching, read the corresponding constraint boundary, compare the structural simulation results with the constraint boundary, and determine the deviation direction, deviation amount and verification margin. Screen subsystem engineering constraints that have boundary deviations or insufficient verification margins, obtain the corresponding source identifiers, and combine the source identifiers, deviation directions, and deviation amounts to generate verification biases.

[0015] Optionally, the generated correction range includes: Input the validation bias into the improved CompGCN model, match the constraint source path based on the source identifier carried by the validation bias, and locate the corresponding conflict parameters and upper or lower bound constraint states. Read the deviation direction and deviation amount of the verification deviation, determine the boundary correction direction and correction amount according to the transformation relationship of each edge in the matching constraint source path, and paste it back to the located upper or lower constraint state to obtain the corrected constraint state. Keep the parameter boundary corresponding to the other polarity constraint state of the conflict parameter unchanged, compare the parameter boundary corresponding to the correction constraint state with the other parameter boundary, and limit the correction range with the parameter boundaries on both sides.

[0016] Optionally, the optimization and simulation according to the correction range, and the output optimization parameters after satisfying the constraints, include: Within the correction range, candidate values ​​for conflicting parameters are regenerated, while keeping the parameter boundaries corresponding to non-conflicting parameters and uncorrected polarity constraints unchanged. Local optimization is then performed on the candidate values ​​to obtain corrected candidate parameters. The modified candidate parameters are written into the structural model of the diesel engine components, and the structural deformation, fatigue safety and structural weight simulations are performed. The results are then compared with the subsystem engineering constraints to obtain the verification status. When the verification status meets the subsystem engineering constraints, the candidate parameters for correction are determined as optimization parameters; when there are unmet constraints, a verification deviation carrying a source identifier is generated and input into the improved CompGCN model. The corresponding polarity constraint status and correction range are updated along the constraint source path according to the source identifier, while keeping the other polarity constraint status unchanged. Local optimization and structural simulation are re-executed until the subsystem engineering constraints are met.

[0017] The beneficial effects of this invention are: This invention unifies the processing of diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters, and inter-layer constraint relationships. It constructs a multi-layer constraint transfer diagram composed of nodes and constraint relationships at different levels, enabling constraint information dispersed across different design levels during the diesel engine structural design process to be transferred layer by layer along clearly defined source relationships. By configuring constraint polarity, transformation relationships, and source identifiers for inter-layer relationships, this invention maintains the differentiated effects of different upstream constraints on the same structural parameter. It reduces the problem of unclear parameter adjustment directions caused by unclear constraint sources or direct mixing of multiple types of constraints in traditional parameter optimization processes, thereby improving the completeness and accuracy of the constraint expression for diesel engine structural design parameters.

[0018] This invention further employs an improved CompGCN model to process the multi-layer constraint transfer graph. By using a constraint polarity-preserving combination layer, different constraints are respectively formed into upper and lower bound constraint states, while retaining the corresponding constraint source paths. This allows different constraints on the same component's structural parameters to propagate and aggregate separately. By comparing the upper and lower bound constraint states and combining them with the current values ​​of the structural parameters to identify conflicting parameters, local optimization is implemented within a limited adjustment range. This avoids large-scale repeated searches of all diesel engine structural parameters, reduces the number of invalid parameter combinations, and improves the targeting and efficiency of conflicting parameter identification and local parameter optimization under multiple constraint conditions.

[0019] This invention also verifies the structural deformation, fatigue safety, and weight of optimized parameters through diesel engine structural simulation. It utilizes an improved verification deviation source chain backtracking mechanism in the CompGCN model to backtrack verification deviations arising from unmet engineering constraints along the original constraint source path to the corresponding upper or lower constraint state, while maintaining the other polarity constraint state unchanged. This directional correction of conflicting parameter ranges allows for the re-performing of local optimization and structural simulation. This reduces repeated adjustments to overall parameters caused by local simulation verification failures, ensuring that structural parameter corrections correspond to specific constraint sources, and improving the convergence efficiency, verification reliability, and traceability of design results in diesel engine structural design parameter optimization. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a diesel engine structural design parameter optimization method based on multi-layer constraint transfer proposed in this invention; Figure 2 This is a schematic diagram of the improved CompGCN model of the diesel engine structural design parameter optimization method based on multi-layer constraint transfer proposed in this invention. Figure 3 This is a schematic diagram of the verification deviation source chain re-attachment for a diesel engine structural design parameter optimization method based on multi-layer constraint transfer proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figure 1 , Figure 2 and Figure 3 A method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer, comprising: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters and inter-layer constraint relationships to generate multi-layer design data; Nodes and edges are constructed from multi-layer design data, and constraint polarity, transformation relationships and source identifiers are configured to generate a multi-layer constraint transfer graph. The multi-layer constraint transfer graph is input into the improved CompGCN model to obtain the upper bound constraint state, the lower bound constraint state, and the constraint source path. The improved CompGCN model includes a node embedding layer, a relation embedding layer, a constraint polarity preservation combination layer, a multi-relation aggregation layer, a verification bias source chain backfitting layer, and an output layer. Compare the upper and lower bound constraint states of the same structural parameter, and identify conflicting parameters and adjustment ranges by combining the current parameter values; By locally optimizing and adjusting the conflicting parameters within the range, the parameters to be verified are obtained. Perform diesel engine structural simulation on the parameters to be verified to obtain deformation, fatigue safety and weight results, match subsystem engineering constraints and associate source identifiers to generate verification deviations; Input the verification bias into the improved CompGCN model, paste it back to the corresponding upper or lower bound constraint state along the constraint source path according to the source identifier, and keep the other polarity constraint state unchanged to generate the correction range; Optimize and simulate within the correction range, and output the optimized parameters after the constraints are satisfied.

[0023] In this embodiment, the diesel engine structural design requirements include structural reliability requirements, lightweight requirements, structural life requirements, and spatial layout requirements; the system specifications include overall machine weight limits, structural strength indicators, fatigue life indicators, and external dimension limits; the subsystem engineering constraints include deformation constraints, stress constraints, fatigue safety constraints, and weight constraints; the component structural parameters include wall thickness, stiffener dimensions, fillet radius, hole diameter, and structural spacing; and the interlayer constraint relationships include the source correspondence between adjacent layer data, the direction of constraint action, and parameter conversion rules.

[0024] In this embodiment, generating multi-layer design data includes: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters, and inter-layer constraints. Establish data items according to their respective design levels, record the current values ​​of component structural parameters, and configure object and level identifiers for each data item. Specifically, configuring object and level identifiers for each data item involves: From the diesel engine structural design task book, read the requirements for structural reliability, lightweighting, structural life, and spatial layout. From the overall design parameter table, read the overall machine weight limits, structural strength indicators, fatigue life indicators, and external dimension limits. From the structural design constraint table, read the deformation constraints, stress constraints, fatigue safety constraints, and weight constraints. From the current design version's diesel engine component 3D model and structural parameter table, read the wall thickness, stiffener dimensions, fillet radius, hole diameter, and structural spacing, and record the corresponding current values. From the design index decomposition table and parameter association records, read the source correspondence between adjacent levels of data, the direction of constraint action, and parameter conversion rules. Establish data items according to the design level to which the data belongs, including data originating from the diesel engine structure. Data items from the design task specification are marked as requirement-level data; data items from the overall design parameter table are marked as system specification-level data; data items from the structural design constraint table are marked as engineering constraint-level data; and data items from the component 3D model and structural parameter table are marked as structural parameter-level data. Requirement-level identifiers, system specification-level identifiers, engineering constraint-level identifiers, and structural parameter-level identifiers are written as the hierarchical identifiers for the corresponding data items. The original entry number of each data item is read, and the hierarchical identifier is combined with the original entry number to form an object identifier. If no original entry number exists, the data items at the same level are numbered consecutively starting from 1 according to the recording order, and the hierarchical identifier is combined with the consecutive number to form an object identifier. Based on the object identifier, data items at adjacent levels are matched accordingly. The source correspondence, constraint direction, and parameter transformation rules in the inter-level constraint relationships are extracted. Constraint transfer records between adjacent levels are then established. Specifically, establishing constraint transfer records between adjacent levels involves: The upstream data item identifier, downstream data item identifier, source correspondence, constraint direction, and parameter conversion rules are retrieved from the inter-layer constraint relationships. The upstream data item identifier is compared item by item with the object identifiers of each data item in the corresponding upstream level, and the downstream data item identifier is compared item by item with the object identifiers of each data item in the adjacent downstream level. A unique correspondence is established between upstream and downstream data items for data items with completely identical identifiers. The source correspondence is recorded as an association pair between the upstream and downstream data item object identifiers, and the constraint direction is recorded as the increasing or decreasing effect of upstream constraint changes on downstream data items. The conversion type corresponding to the parameter conversion rules is retrieved from the design index decomposition table and parameter association records. The conversion parameters are as follows: When the conversion type is proportional conversion, the constraint value of the upstream data item is multiplied by the conversion coefficient to obtain the downstream constraint value; when the conversion type is linear conversion, the constraint value of the upstream data item is multiplied by the conversion coefficient and then the conversion offset is added to obtain the downstream constraint value; when the conversion type is interval mapping, the interval where the upstream constraint value is located is matched with the interval correspondence table in the parameter association record and the corresponding downstream constraint interval is read. The conversion coefficient, conversion offset, and interval boundary are all read from the design index decomposition table or parameter association record. The upstream data item object identifier, downstream data item object identifier, constraint direction, parameter conversion rule, and the converted constraint value or constraint interval are combined to form a constraint transfer record between adjacent level data items. The data items are arranged according to their hierarchical identifiers, and constraint transfer records are associated with their corresponding data items to form multi-layered design data containing data items, current parameter values, hierarchical identifiers, and inter-layer constraint relationships. Specifically, this multi-layered design data includes data items, current parameter values, hierarchical identifiers, and inter-layer constraint relationships as follows: The corresponding data items are organized in the hierarchical order of requirement layer, system specification layer, engineering constraint layer, and structural parameter layer. The data content, object identifier, and hierarchy identifier are stored in each data item. The upstream data item object identifier and downstream data item object identifier in the constraint transfer record are read, and the constraint transfer record is attached to the data item with the matching object identifier. The association index of the corresponding constraint transfer record is recorded. The current value of the component structural parameter is stored synchronously in the structural parameter layer data item. The data items and association indexes of each layer are collected to form multi-layer design data.

[0025] In this embodiment, generating the multi-layer constraint transfer graph includes: Based on the hierarchy of diesel engine structural design requirements, system specifications, subsystem engineering constraints, and component structural parameters, the data items in the multi-level design data are set as corresponding level nodes. Specifically, the data items in the multi-level design data are set as corresponding level nodes as follows: Read the data content, object identifier, and hierarchy identifier of each data item, and establish nodes according to the requirement layer, system specification layer, engineering constraint layer, and structural parameter layer respectively. Write the data content, object identifier, and hierarchy identifier into the corresponding node. For the structural parameter layer node, synchronously write the current value of the component structural parameter, so that each node corresponds one-to-one with the corresponding data item in the multi-layer design data. Based on inter-level constraints, adjacent level nodes are matched, and edges are established from upstream nodes to downstream nodes according to the source of constraints. Specifically, this involves matching adjacent level nodes based on inter-level constraints and establishing edges from upstream nodes to downstream nodes according to the source of constraints. Read the upstream data item object identifier and downstream data item object identifier from the constraint transfer record. Compare the upstream data item object identifier with the object identifier of the corresponding upstream level node item by item. Compare the downstream data item object identifier with the object identifier of the adjacent downstream level node item by item. For nodes where the object identifiers on both sides are consistent, establish a directed edge from the upstream node to the downstream node. When one upstream node corresponds to multiple downstream nodes, establish directed edges pointing to each downstream node respectively. When one downstream node corresponds to multiple upstream nodes, retain all directed edges from each upstream node to the downstream node and write the association index of the corresponding constraint transfer record into each directed edge. The upper or lower bound constraint polarity is determined based on the direction of the constraint action corresponding to the edge, and the transformation relationship is configured according to the parameter transformation rules. The source information of the upstream constraint is configured as the source identifier. Specifically, the determination of the upper or lower bound constraint polarity based on the direction of the constraint action corresponding to the edge is as follows: Read the constraint direction from the constraint transfer record corresponding to the edge. When the upstream constraint limits the structural parameters of the component to not be higher than the corresponding parameter boundary, or the constraint change requires the structural parameters to be adjusted in the decreasing direction, mark the edge as the upper bound constraint polarity. When the upstream constraint limits the structural parameters of the component to not be lower than the corresponding parameter boundary, or the constraint change requires the structural parameters to be adjusted in the increasing direction, mark the edge as the lower bound constraint polarity. Read the parameter conversion rules from the constraint transfer record, and write the conversion coefficients corresponding to proportional conversion, the conversion coefficients and conversion offsets corresponding to linear conversion, and the interval boundaries and mapping intervals corresponding to interval mapping into the corresponding edges to form the conversion relationship of the edges. Read the upstream data item object identifier from the constraint transfer record, and write the upstream data item object identifier into the corresponding edge as the source identifier, so that each edge contains the constraint polarity, conversion relationship, and source identifier simultaneously. By associating nodes at each level with the configured edges, a multi-layer constraint transfer graph is generated, which progressively transmits the diesel engine structural design requirements to the component structural parameters. Specifically, this multi-layer constraint transfer graph, which transmits the diesel engine structural design requirements to the component structural parameters, is generated as follows: The nodes are arranged in the hierarchical order of requirement layer, system specification layer, engineering constraint layer, and structural parameter layer. The directed edges established between adjacent level nodes based on constraint transfer records are retained. The data content, object identifier, level identifier, and current value of structural parameters stored in each node are used as node attributes. The constraint polarity, transformation relationship, source identifier, and constraint transfer record association index stored in each edge are used as edge attributes. The corresponding nodes are connected according to the pointing relationship of the directed edges to form a multi-layer constraint transfer diagram in which the diesel engine structural design requirements are transferred layer by layer to the component structural parameters through system specifications and subsystem engineering constraints.

[0026] In this embodiment, obtaining the upper bound constraint state, the lower bound constraint state, and the constraint source path includes: Based on the CompGCN model, the node representation, relation representation, node and relation combination propagation, multi-relation aggregation and result output process are set up with a node embedding layer, a relation embedding layer, a multi-relation aggregation layer and an output layer. A constraint polarity preservation combination layer is added at the output convergence position of the node embedding layer and the relation embedding layer, and a verification bias source chain back-fitting layer is added between the multi-relation aggregation layer and the output layer to form an improved CompGCN model. This invention structurally optimizes the traditional CompGCN model by modifying the constraint propagation and verification feedback process, thereby improving the ability to express and transmit multi-layer design constraints. To address the issue of the same structural parameter being constrained by engineering constraints in opposite directions, a constraint polarity preservation combination layer is added at the output convergence point of the node embedding layer and the relation embedding layer. This allows the upper and lower bound constraints to propagate separately, reducing the mixing of constraints of different polarities during the aggregation process. In order to preserve the source relationships of constraints at different levels, the source identifier is preserved and the constraint source path is formed simultaneously during the multi-relationship aggregation process, so that the upper bound constraint state and lower bound constraint state of the structural parameters correspond to specific upstream constraint sources respectively. To address the problem of difficulty in directional feedback of verification deviations generated by diesel engine structural simulation, a verification deviation source chain back-attachment layer is added between the multi-relation aggregation layer and the output layer. This allows the verification deviation to act on the corresponding polarity constraint state along the matched constraint source path, while keeping the other polarity constraint state unchanged. The improved CompGCN model achieves independent propagation and directional feedback of multi-layer constraints by maintaining constraint polarity and verifying the source chain of deviation, thereby improving the pertinence and traceability of diesel engine structural parameter conflict identification and closed-loop optimization. The nodes in the multi-layer constraint transfer graph are input into the node embedding layer. Embedding mapping is performed based on the node's level and data to generate a node embedding representation. The node embedding layer includes: Node content encoding unit: Reads the data content stored in the node, distinguishes between numerical data and categorical data, standardizes the numerical data, indexes and embeds the categorical data, and generates node content features; Hierarchical encoding unit: Reads the hierarchical identifier of the node, maps the requirement layer, system specification layer, engineering constraint layer and structural parameter layer to hierarchical indexes respectively, and generates hierarchical features through a trainable hierarchical embedding table; Object Identifier Encoding Unit: Reads the hierarchy identifier and entry number from the node object identifier, converts the hierarchy identifier and entry number into indexes respectively, performs embedding mapping, and generates object identifier features; Parameter state encoding unit: Reads the current value of the parameter stored in the structural parameter layer node, standardizes the current value of the parameter, generates parameter state features, and sets the parameter state features of non-structural parameter layer nodes to zero vector; Node feature fusion unit: splices node content features, hierarchical features, object identification features and parameter state features, and generates a node embedding representation of a unified dimension through linear mapping and nonlinear activation; In the node embedding layer, the data content, object identifier, hierarchical identifier, and current values ​​of structural parameters stored in each node of the multi-layer constraint transfer graph are read. The node content encoding unit performs standardization on numerical data using the mean and standard deviation of training samples of the same type of historical diesel engine structural design. For categorical data, a category index is established according to the category name in the training samples, and node content features are generated through the trainable embedding table. The hierarchical encoding unit and the object identifier encoding unit generate hierarchical features and object identifier features, respectively. The parameter state encoding unit performs standardization on the current value of the parameter using the mean and standard deviation of training samples of the same type of structural parameters. When the standard deviation is zero, the standardization result is zero. The node feature fusion unit concatenates the node content features, hierarchical features, object identifier features, and parameter state features in that order. The concatenation result is converted into a preset embedding dimension through linear mapping and nonlinear activation is performed using the ReLU function. The preset embedding dimension is consistent with the output dimension of the relation embedding layer. The linear mapping parameters and each trainable embedding table are updated during the training of the improved CompGCN model to generate node embedding representations. The edges in the multi-layer constraint transfer graph are input into the relation embedding layer. Relation embedding is performed based on constraint polarity, transformation relationship, and source identifier to generate a relation embedding representation carrying the source identifier. The relation embedding layer includes: Constraint polarity encoding unit: Reads the upper bound constraint polarity or lower bound constraint polarity stored in the edge, converts the two types of constraint polarity into different polarity indices, and generates constraint polarity features through a trainable polarity embedding table; Transformation Relationship Encoding Unit: Reads the transformation type and transformation parameters stored on the edge, converts proportional transformation, linear transformation and interval mapping into corresponding type indices, and performs numerical encoding on the transformation coefficients, transformation offsets and interval boundaries to generate transformation relationship features; Source identifier encoding unit: Read the source identifier stored by the edge, extract the hierarchical identifier and entry number from the source identifier and perform index embedding respectively to generate source identifier features; Relationship Feature Fusion Unit: It splices constraint polarity features, transforms relationship features and source identifier features, generates relationship features through linear mapping and nonlinear activation, and preserves the association between relationship features and source identifiers; In the relation embedding layer, the constraint polarity encoding unit generates constraint polarity features according to the constraint polarity of the edges; the transformation relation encoding unit determines the transformation type according to the parameter transformation rules, reads the transformation coefficients for proportional transformation, reads the transformation coefficients and transformation offsets for linear transformation, reads the upper and lower limits of the input interval and the upper and lower limits of the output interval for interval mapping, and writes zero values ​​for parameters that do not participate in the current transformation type. Each numerical parameter is standardized using the mean and standard deviation of the training samples of the same type of historical diesel engine structure design. When the standard deviation is zero, the standardization result is set to zero. The embedding result of the transformation type index is concatenated with the standardized transformation parameters to form the transformation relation features; the source identifier encoding unit converts the source level and source object into embedding vectors respectively and concatenates them to form the source identifier features; the relation feature fusion unit concatenates the constraint polarity features, transformation relation features and source identifier features in the order of constraint polarity features, transformation relation features and source identifier features, transforms them to the same dimension as the node embedding representation through linear mapping and uses the ReLU function for nonlinear activation. The linear mapping parameters and the trainable embedding table are updated during the training of the improved CompGCN model to generate a relation embedding representation carrying the source identifier; The node embedding representation and relation embedding representation are input into the constraint polarity-preserving combination layer. Adjacent nodes and their corresponding relations are combined according to the constraint polarity of the edges. The combination results are then propagated layer by layer along the upper and lower bound polarity paths, simultaneously transmitting the source identifier, generating upper and lower bound propagation features. The constraint polarity-preserving combination layer includes: Polarity splitting unit: Reads the constraint polarity stored in the edge, assigns the edge with the upper bound constraint polarity to the upper bound polarity path, and assigns the edge with the lower bound constraint polarity to the lower bound polarity path. Node Relationship Combination Unit: Reads the node embedding representation of the starting node of the edge and the relationship embedding representation of the corresponding edge, performs element-wise multiplication on the feature values ​​of the same dimension, and generates node relationship combination features through linear mapping and non-linear activation; Hierarchical transmission unit: The node relationship combination feature is transmitted in the direction from the upstream node to the downstream node according to the edge, and cross-layer propagation is completed in the hierarchical order of demand layer, system specification layer, engineering constraint layer and structural parameter layer; Source synchronization unit: reads the source identifier associated with the relation embedding representation, and synchronously writes the source identifier when the node relation combination feature is passed along the corresponding edge, maintaining the correspondence between the propagation feature and the source identifier; In the constraint polarity preservation combination layer, the polarity splitting unit determines the propagation path based on the constraint polarity of each edge in the multi-layer constraint propagation graph; the node relationship combination unit reads the node embedding representation and corresponding relationship embedding representation of the starting node of the edge, performs element-wise multiplication on the same feature dimension, and generates node relationship combination features by linear mapping and non-linear activation using the ReLU function; the hierarchical propagation unit writes the node relationship combination features corresponding to the upper bound constraint polarity only into the upper bound polarity path, and writes the node relationship combination features corresponding to the lower bound constraint polarity only into the lower bound polarity path. When the node relationship combination features reach the intermediate layer node, they are added element-wise with the corresponding node embedding representation, and the addition result is used as the input for the node relationship combination of the adjacent downstream edge until it propagates to the structural parameter layer node; the source synchronization unit synchronously writes the source identifier of the corresponding edge during each cross-layer propagation, and maintains the propagation order of the source identifier according to the edge through which the propagation has passed. The linear mapping parameters are updated during the training of the improved CompGCN model, and the propagation results of the upper bound polarity path and the lower bound polarity path are collected respectively to generate upper bound propagation features and lower bound propagation features carrying source identifiers; The upper and lower bound propagation features are input into the multi-relation aggregation layer. Propagation features of the same polarity are aggregated according to the component structural parameter nodes, maintaining the independence of aggregation results of different polarities. This generates upper and lower bound aggregation states, and the edges traversed by the propagation are concatenated according to the source identifier to generate a candidate set of source paths. The multi-relation aggregation layer includes: Polarity grouping unit: The upper bound propagation features and lower bound propagation features are grouped according to the component structural parameter nodes and constraint polarity. Propagation features that point to the same component structural parameter node and belong to the same polarity path are assigned to the same aggregation group. Aggregation state generation unit: The propagation features within each aggregation group are accumulated dimension by dimension, and the accumulated result is divided by the number of propagation features in the corresponding aggregation group to generate upper bound aggregation state and lower bound aggregation state respectively; Source path concatenation unit: Reads the source identifier, start node object identifier and end node object identifier written by the propagation features on each propagation edge, forms a source identifier sequence according to the propagation order and concatenates the corresponding edges to generate a source path candidate set; In the multi-relation aggregation layer, the polarity grouping unit reads the target component structural parameter nodes corresponding to the upper and lower bound propagation features, assigns upper bound propagation features pointing to the same component structural parameter node to the upper bound aggregation group, and assigns lower bound propagation features pointing to the same component structural parameter node to the lower bound aggregation group, while keeping the two types of aggregation groups independent of each other; the aggregation state generation unit performs dimension-wise summation on the propagation features in each upper and lower bound aggregation group, and calculates the average value of each dimension using the number of propagation features within the aggregation group as the divisor, forming the upper and lower bound aggregation states of the corresponding component structural parameter nodes; the source path concatenation unit reads the propagation path of each feature during propagation... The process sequentially writes source identifiers, as well as the starting node object identifiers and ending node object identifiers corresponding to each edge; when the ending node object identifier of the current edge is consistent with the starting node object identifier of the next edge, the corresponding edges are connected according to the transmission order of the requirement layer, system specification layer, engineering constraint layer, and structural parameter layer, and the source identifiers of each edge are written into the source identifier sequence in the same order until they are connected to the corresponding component structural parameter node; for paths where the source identifier sequence, edge sequence, constraint polarity, and target component structural parameter node identifier are all consistent, one record is retained, and the remaining paths are retained separately, generating a source path candidate set corresponding to the upper bound aggregation state and the lower bound aggregation state; The upper bound aggregation state, lower bound aggregation state, and candidate source path set are input into the verification deviation source chain backtracking layer. The source path is associated with the corresponding upper bound aggregation state or lower bound aggregation state according to the source identifier, and a source chain backtracking index is established. The verification deviation source chain backtracking layer includes: Source Identifier Matching Unit: Reads the source identifier sequence, target component structural parameter node identifier and constraint polarity stored in each source path in the source path candidate set, and establishes a correspondence between each source identifier in the source identifier sequence and its corresponding source path; Polarity State Association Unit: Based on the constraint polarity of the source path, associate the source path of the upper bound constraint polarity with the upper bound aggregation state of the corresponding component structural parameter node, and associate the source path of the lower bound constraint polarity with the lower bound aggregation state of the corresponding component structural parameter node. Path index generation unit: Reads the source identifier sequence and edge sequence in the source path, configures the path number for each source path, records the edge sequence position corresponding to each source identifier in the source identifier sequence, and combines the source identifier sequence, target component structural parameter node identifier, constraint polarity, path number, edge sequence, corresponding position of source identifier and corresponding aggregation state to form the source chain backtracking index item; Indexing Unit: Compares the source identifier sequence, target component structural parameter node identifier, constraint polarity, and edge sequence in the source chain re-paste index entries. For duplicate index entries with identical fields, one entry is retained. For index entries with different source identifier sequences, target nodes, constraint polarity, or edge sequences, each entry is retained separately to form the source chain re-paste index. In the source chain back-paste layer for verifying deviations, the source identifier matching unit reads the candidate set of source paths, reads the source identifier sequence stored in the source path one by one, and establishes an association record between the source identifier, path number, and corresponding edge position for each source identifier in the source identifier sequence; the polarity state association unit determines the aggregation state association object according to the constraint polarity, establishes an association between the upper bound source path and the corresponding upper bound aggregation state, and establishes an association between the lower bound source path and the corresponding lower bound aggregation state; the path index generation unit records the edge sequence according to the transmission order of the edges in the source path, and configures the path number starting from 1 according to the arrangement order of the source path, and writes the source identifier sequence, target component structure parameter node identifier, constraint polarity, path number, edge sequence, edge position corresponding to each source identifier, and corresponding aggregation state into the same index item; the index sorting unit deletes all duplicate index items with the same fields, and retains the independent index items corresponding to different source identifier sequences or different edge sequences, and generates the source chain back-paste index; The upper bound aggregation state, lower bound aggregation state, and source chain backtracking index are input to the output layer to generate the upper bound constraint state and lower bound constraint state of each component's structural parameters. The corresponding constraint source path is generated based on the source chain backtracking index. The output layer includes: Constraint State Mapping Unit: Reads the upper bound aggregate state and the lower bound aggregate state, converts them into upper bound constraint state vector and lower bound constraint state vector respectively through independent linear mapping matrices, and records the boundary components in the mapping results as the upper bound parameter boundary and lower bound parameter boundary of the corresponding component structural parameters; Source path extraction unit: Reads the target component structural parameter node identifier, constraint polarity, path number and edge sequence from the source chain backlink index, extracts the edge sequence corresponding to the upper or lower bound constraint state according to the target node and constraint polarity, and generates the constraint source path; State path association unit: According to the target component structural parameter node identifier and constraint polarity, establishes association between the upper bound constraint state, lower bound constraint state and the corresponding constraint source path, and forms the constraint output result of each component structural parameter; In the output layer, the constraint state mapping unit reads the upper and lower bound aggregate states of the same component structural parameter node, respectively, and performs dimensional transformation through mutually independent upper and lower bound linear mapping matrices. The boundary components in the mapped state vector are used as the parameter boundaries of the corresponding polarity constraint states, and the remaining components are used as the state representations of the corresponding aggregate features. The upper and lower bound linear mapping matrices are initialized with Xavier uniform distribution, the bias term is initialized to zero, and they are updated during the training of the improved CompGCN model. The source path extraction unit filters index items according to the target component structural parameter node identifier in the source chain backlink index, reads the edge sequences in the upper and lower bound index items according to the constraint polarity, keeps different source paths under the same target node independent according to the path number, and connects each edge according to the recording order of the edge sequence to generate the corresponding constraint source path. The state path association unit associates the upper bound constraint state with the upper bound constraint source path and the lower bound constraint state with the lower bound constraint source path, and outputs the upper bound constraint state, lower bound constraint state, and constraint source path corresponding to each component structural parameter. Training samples were constructed by collecting multi-layer design data, parameter constraint records, structural simulation verification records, and parameter correction records from historical diesel engine structural design cases. Constraint state identification error, source path matching error, and verification deviation re-attachment error were used as joint training objectives to iteratively train the improved CompGCN model until the joint training objective met the preset convergence condition, resulting in the trained improved CompGCN model. Specifically, the trained improved CompGCN model is as follows: Training samples were divided according to the diesel engine structural design case numbers, with 80% of the cases used as the training set and 20% as the validation set. For each training sample, corresponding hierarchical nodes and directed edges were established for the multi-layer design data, and constraint polarities, transformation relationships, and source identifiers were configured to generate a multi-layer constraint transfer graph for training. The upper and lower bounds of the component structural parameters and their constraint polarities were read from the parameter constraint records as constraint state labels. These labels were then concatenated according to the node connection relationships and edge source identifiers in the multi-layer constraint transfer graph for training to form source path labels. The source identifier, deviation direction, and deviation amount of the verification deviation were read from the structural simulation verification records, and the corresponding structural parameters were read from the parameter correction records. The corrected polarity and corrected parameter boundaries are used as validation bias back-labels. The multi-layer constraint transfer graph used for training is input into the improved CompGCN model, which outputs the upper bound constraint state, lower bound constraint state, and constraint source path. The boundary correction results for the corresponding polarity constraint state are obtained by back-labeling the layer based on the validation bias source chain. The upper and lower bound parameter boundaries output by the model are subtracted from the corresponding label parameter boundaries. The absolute value of the difference is divided by the absolute value of the corresponding label parameter boundary. When the label parameter boundary is 0, 0.000001 is used as the divisor. The average of the results is used to obtain the constraint state identification error. The constraint source path output by the model is compared item by item with the source path label according to the edge sequence position, and the corresponding position... The number of inconsistent edges is obtained by adding the absolute value of the difference between the lengths of the two paths. The number of inconsistent edges is then divided by the larger of the two path lengths. When both path lengths are 0, the source path matching error is recorded as 0. The boundary correction result is compared with the verification deviation label. If the polarity is inconsistent, the polarity error is recorded as 1; if it is consistent, it is recorded as 0. The absolute value of the difference between the corrected parameter boundary and the label parameter boundary is divided by the absolute value of the label parameter boundary. When the label parameter boundary is 0, 0.000001 is used as the divisor to obtain the boundary correction error. The average of the polarity error and the boundary correction error is used as the verification deviation labeling error. The constraint states are weighted according to weights of 0.4, 0.3, and 0.3 respectively. The identification error, source path matching error, and validation bias re-attachment error are weighted and summed to obtain the joint training error. The Adam optimizer is used to update the trainable parameters in the node embedding layer, relation embedding layer, constraint polarity preservation combination layer, multi-relation aggregation layer, validation bias source chain re-attachment layer, and output layer. The learning rate is set to 0.001, and 32 training samples are input in each batch. After each round of training, the joint training error is calculated using the validation set. The iteration stops when the difference between the maximum and minimum values ​​of the joint training error of the validation set for 5 consecutive rounds is less than 0.001, or when the number of training rounds reaches 200. The model parameters corresponding to the round with the lowest joint training error of the validation set are saved, and the improved CompGCN model is obtained after training.

[0027] In this embodiment, the step of identifying conflicting parameters and adjusting their ranges based on their current values ​​includes: Read the upper and lower bound constraint states of the structural parameters of the same component, extract the upper and lower bound parameter boundaries respectively, and compare the relative positions of the parameter boundaries on both sides. The current value of the parameter is compared with both the upper and lower parameter boundaries. Component structural parameters whose current values ​​exceed a common feasible interval between the two parameter boundaries are identified as conflicting parameters. Specifically, the comparison of the current value with both the upper and lower parameter boundaries is as follows: The lower bound parameter boundary is used as the minimum allowable value of the component structural parameter, and the upper bound parameter boundary is used as the maximum allowable value of the component structural parameter. When the lower bound parameter boundary is less than the upper bound parameter boundary, the interval between the lower bound parameter boundary and the upper bound parameter boundary is determined as a common feasible interval. When the current value of the parameter is less than the lower bound parameter boundary or greater than the upper bound parameter boundary, the corresponding component structural parameter is determined as a conflicting parameter. When the lower bound parameter boundary is equal to the upper bound parameter boundary, the interval between the corresponding boundary value and the single point with the same boundary value is determined as a common feasible interval. When the current value of the parameter is not equal to the corresponding boundary value, the corresponding component structural parameter is determined as a conflicting parameter. When the lower bound parameter boundary is greater than the upper bound parameter boundary, it is determined that there is no common feasible interval, and the corresponding component structural parameter is determined as a conflicting parameter. For conflicting parameters that share a common feasible interval, the parameter adjustment range is determined by the common feasible interval; for conflicting parameters that do not share a common feasible interval, the parameter adjustment range is determined by the interval between the boundaries of the two parameters. Specifically, determining the parameter adjustment range involves: For conflicting parameters where the lower bound is less than the upper bound and the current value exceeds the common feasible interval, the lower bound is used as the lower limit of the adjustment range, and the upper bound as the upper limit, forming the parameter adjustment range. For conflicting parameters where the lower bound is equal to the upper bound and the current value is not equal to the corresponding boundary value, the corresponding boundary value is used as both the lower and upper limits of the adjustment range, forming a single-point parameter adjustment range. For conflicting parameters where the lower bound is greater than the upper bound, the smaller of the upper and lower bounds is used as the lower limit of the adjustment range, and the larger is used as the upper limit. The interval between the two parameter boundaries is defined as the conflict coordination range and used as the parameter adjustment range.

[0028] In this embodiment, the conflict parameters within the local optimization adjustment range yield the following parameters to be verified: Read the current value, adjustment range, and upper and lower bound constraint states of the conflicting parameters, keep the non-conflicting parameters unchanged, generate local candidate values ​​for each conflicting parameter within the adjustment range, and combine them to form a local candidate parameter group. Specifically, generating local candidate values ​​for each conflicting parameter within the adjustment range and combining them to form a local candidate parameter group involves: Read the current value, lower limit, upper limit, upper bound, and lower bound of each conflicting parameter. For conflicting parameters with a common feasible interval and whose current value is outside the adjustment range, divide the length of the range obtained by subtracting the lower limit from the upper limit into 10 equal parts. Use the resulting length as the candidate step size. Starting from the adjustment range boundary closest to the current value, take values ​​sequentially towards the center of the adjustment range according to the candidate step size, while retaining the two sides of the adjustment range, generating a total of 11 local candidate values. For conflicting parameters without a common feasible interval and whose current value is within the conflict coordination range, divide the distance from the current value to the lower limit of the adjustment range into 5 equal parts, and divide the distance from the current value to the upper limit of the adjustment range into 5 equal parts. Five parts are divided into five equal parts. The current value of the parameter is taken sequentially from the current value towards the two side boundaries according to the corresponding equal division length. The current value of the parameter and the candidate values ​​obtained from both sides are merged to generate a total of 11 local candidate values. For conflicting parameters that do not have a common feasible interval and whose current value is outside the conflict coordination range, the adjustment range length is divided into ten equal parts. The adjustment range boundary closest to the current value of the parameter is taken sequentially towards the other side boundary to generate a total of 11 local candidate values. When the lower limit of the adjustment range is equal to the upper limit of the adjustment range, the corresponding boundary value is used as the unique local candidate value. Cartesian combinations are performed according to the local candidate values ​​of each conflicting parameter. Each combination replaces the current value of the corresponding conflicting parameter, while keeping the current value of the non-conflicting parameters unchanged, forming a local candidate parameter group. The local candidate parameter groups are matched with their corresponding upper and lower bound constraint states, respectively. The deviation of the candidate values ​​from the upper and lower bound parameter boundaries and the adjustment relative to the current parameter values ​​are calculated. The local candidate parameter groups are then sorted and filtered according to the boundary deviation and parameter adjustment to obtain the target parameter group. Specifically, the sorting and filtering of the local candidate parameter groups according to the boundary deviation and parameter adjustment is as follows: According to the conflict parameter identifier, match the candidate values ​​in the local candidate parameter group with the corresponding upper and lower bound constraint states, and read the upper bound parameter boundary, lower bound parameter boundary, and current parameter value; when the candidate value is lower than the lower bound parameter boundary, subtract the candidate value from the lower bound parameter boundary to obtain the lower bound deviation; when the candidate value is not lower than the lower bound parameter boundary, record the lower bound deviation as 0; when the candidate value is higher than the upper bound parameter boundary, subtract the upper bound parameter boundary from the candidate value to obtain the upper bound deviation; when the candidate value is not higher than the upper bound parameter boundary, record the upper bound deviation as 0; match the lower bound deviation with the upper bound deviation. The boundary deviation of the corresponding conflicting parameter is obtained by summing the boundary deviations of all conflicting parameters in the same local candidate parameter group. The parameter group boundary deviation is obtained by summing the parameter adjustment of all conflicting parameters. The local candidate parameter groups are sorted in ascending order of parameter group boundary deviation. If the parameter group boundary deviations are the same, they are sorted in ascending order of parameter group adjustment. The local candidate parameter group with the first position in the sort is determined as the target parameter group. Update the corresponding conflicting parameters with the candidate values ​​in the target parameter group, and keep the current values ​​of the remaining component structural parameters unchanged, and combine them to generate the parameters to be verified.

[0029] In this embodiment, the generation of verification deviation includes: The parameters to be verified are written into the corresponding diesel engine component structural model, and structural deformation, fatigue safety, and structural weight simulations are performed to obtain structural simulation results. Specifically, the structural simulation results are as follows: Match the corresponding geometric parameters in the diesel engine component structural model according to the component identification and structural parameter names, and replace the current values ​​of the corresponding parameters with the parameters to be verified; read the elastic modulus, Poisson's ratio, material density and fatigue performance parameters from the material parameter table, and read the in-cylinder load, connection preload and structural constraint conditions from the current diesel engine design conditions and simulation task configuration records, and write them into the updated component structural model; perform structural deformation analysis, read the model nodal displacement and element equivalent stress, take the maximum nodal displacement as the structural deformation result, and take the maximum element equivalent stress as the structural stress result; perform fatigue analysis using the structural stress result and fatigue performance parameters, and read the minimum fatigue safety factor or the corresponding fatigue life as the fatigue safety result; calculate the product of the volume of each component and the corresponding material density and sum them to obtain the structural weight result; combine the structural deformation result, structural stress result, fatigue safety result and structural weight result to form the structural simulation result; Based on the structural simulation results of the subsystem engineering constraint matching, the corresponding constraint boundaries are read, and the structural simulation results are compared with the constraint boundaries to determine the deviation direction, deviation amount, and verification margin. Specifically, determining the deviation direction, deviation amount, and verification margin involves: Read the constraint object identifier, constraint index name, constraint type, and constraint boundary from the subsystem engineering constraints. Based on the constraint object identifier and constraint index name, match deformation constraints with the maximum nodal displacement in the structural simulation results, stress constraints with the maximum element equivalent stress, fatigue safety constraints with the minimum fatigue safety factor or fatigue life, and weight constraints with the structural weight results. For upper limit constraints, if the structural simulation result exceeds the constraint boundary, record the deviation direction as exceeding the upper limit, use the difference between the structural simulation result and the constraint boundary as the deviation amount, and record the verification margin as 0. When the deviation is not greater than the constraint boundary, the deviation is recorded as 0, and the verification margin is obtained by dividing the difference between the constraint boundary and the structural simulation result by the absolute value of the constraint boundary. For the lower limit constraint, when the structural simulation result is less than the constraint boundary, the deviation direction is recorded as below the lower limit, and the deviation is obtained by dividing the difference between the constraint boundary and the structural simulation result as the deviation, and the verification margin is recorded as 0. When the structural simulation result is not less than the constraint boundary, the deviation is recorded as 0, and the verification margin is obtained by dividing the difference between the structural simulation result and the constraint boundary by the absolute value of the constraint boundary. When calculating the verification margin, the constraint boundary is 0, and 0.000001 is used as the divisor. Screen subsystem engineering constraints with boundary deviations or insufficient verification margins, obtain corresponding source identifiers, and combine the source identifier, deviation direction, and deviation amount to generate verification bias. Specifically, the generation of verification bias involves: Read the safety margin requirements in the subsystem engineering constraints. Safety margin requirements recorded as proportions are directly used as the minimum verification margin. Safety margin requirements recorded as absolute values ​​are divided by the absolute value of the corresponding constraint boundary to obtain the minimum verification margin. When the constraint boundary is 0, 0.000001 is used as the divisor. When no safety margin requirement is recorded, the minimum verification margin is set to 0.05. When the deviation is greater than 0, the corresponding subsystem engineering constraint is determined as a boundary deviation constraint, and the deviation direction and deviation amount are retained. When the deviation is 0 and the verification margin is lower than the minimum verification margin, the corresponding subsystem engineering constraint is determined as a margin deficiency constraint. The margin deficiency ratio is obtained by subtracting the verification margin from the minimum verification margin. The margin deficiency ratio is multiplied by the absolute value of the corresponding constraint boundary to obtain the equivalent deviation amount. When the constraint boundary is 0, it is multiplied by 0.000001. The deviation direction of the upper limit constraint is recorded as approaching the upper limit, and the deviation direction of the lower limit constraint is recorded as approaching the lower limit. Read the source identifier of the corresponding edge of the subsystem engineering constraint. Combine the source identifier, deviation direction, and corresponding deviation amount to generate the verification deviation.

[0030] In this embodiment, the generated correction range includes: The validation bias is input into the improved CompGCN model. Based on the source identifier carried by the validation bias, the constraint source path is matched to locate the corresponding conflicting parameters and upper or lower bound constraint states. Specifically, locating the corresponding conflicting parameters and upper or lower bound constraint states by matching the constraint source path based on the source identifier carried by the validation bias involves: Read the source identifier from the verification deviation and compare it item by item with the source identifier sequence stored in each index item of the source chain re-attachment index. When the source identifier matches any identifier in the source identifier sequence, read the target component structural parameter node identifier, constraint polarity, path number, edge sequence, and edge position corresponding to the matching source identifier of the corresponding index item. Match the corresponding conflict parameter according to the target component structural parameter node identifier, and determine the re-attachment object according to the constraint polarity. The upper bound constraint polarity corresponds to the upper bound constraint state, and the lower bound constraint polarity corresponds to the lower bound constraint state. Starting from the edge position corresponding to the matching source identifier, extract the continuous edge sequence to the target component structural parameter node as the re-attachment edge sequence. When the same source identifier corresponds to multiple index items, retain the corresponding re-attachment edge sequence according to the path number, without merging different constraint source paths, to form a matching relationship between the verification deviation and the conflict parameter, the corresponding polarity constraint state, and the re-attachment edge sequence. Read the deviation direction and amount of the verification deviation, determine the boundary correction direction and amount based on the transformation relationship of each edge in the matching constraint source path, and paste it back to the located upper or lower bound constraint state to obtain the corrected constraint state. Specifically, obtaining the corrected constraint state is as follows: Read the deviation direction and deviation amount from the verification deviation, and read the transformation type and transformation parameters edge by edge according to the back-paste edge sequence; for proportional transformation, multiply the current deviation amount by the absolute value of the transformation coefficient to obtain the transmitted deviation amount. When the transformation coefficient is positive, the current deviation direction is maintained; when the transformation coefficient is negative, the current deviation direction is reversed. For linear transformation, multiply the current deviation amount by the absolute value of the transformation coefficient to obtain the transmitted deviation amount. The transformation offset is used to maintain the reference mapping position of the constraint values ​​at both ends of the edge and does not participate in the deviation amount scaling. When the transformation coefficient is positive, the current deviation direction is maintained; when the transformation coefficient is negative, the current deviation direction is reversed. For interval mapping, read the upper and lower limits of the input interval and the upper and lower limits of the output interval. Divide the current deviation amount by the absolute value of the input interval width to obtain the deviation ratio. When the input interval width is 0, the deviation ratio is recorded as 1; when the deviation ratio is greater than 1, it is recorded as 1. Multiply the absolute value of the output interval width by the deviation... The deviation is obtained by multiplying the proportions and the direction of constraint action is read from the corresponding constraint transfer record. Increasing the action maintains the current deviation direction, while decreasing the action reverses the current deviation direction. After performing edge-by-edge conversion according to the backing edge sequence, the deviation transferred to the component structural parameter node is used as the boundary correction amount. If the backing object is in the upper bound constraint state and the verification deviation exceeds or approaches the upper bound direction, the upper bound parameter boundary is corrected in the decreasing direction if the transfer direction has not been reversed. If the backing object is in the lower bound constraint state and the verification deviation is below or approaches the lower bound direction, the lower bound parameter boundary is corrected in the increasing direction if the transfer direction has not been reversed. When the transfer direction is reversed, the corresponding opposite correction direction is used. The boundary correction amount is increased or decreased on the original parameter boundary according to the boundary correction direction, and the corrected parameter boundary is written into the located upper bound constraint state or lower bound constraint state to generate the corrected constraint state. Keeping the parameter boundary corresponding to the other polarity constraint state of the conflicting parameter unchanged, compare the parameter boundary corresponding to the modified constraint state with the other parameter boundary, and limit the modification range with the parameter boundaries on both sides. Specifically, the modification range is limited by the parameter boundaries on both sides as follows: Read the corrected parameter boundary corresponding to the corrected constraint state and the uncorrected parameter boundary corresponding to the other polarity constraint state. When the correction object is a lower bound constraint state, compare the corrected lower bound parameter boundary with the unchanged upper bound parameter boundary. If the corrected lower bound parameter boundary is less than or equal to the upper bound parameter boundary, use the corrected lower bound parameter boundary as the lower limit of the correction range and the upper bound parameter boundary as the upper limit of the correction range to form a common feasible interval and determine it as the correction range. If the corrected lower bound parameter boundary is greater than the upper bound parameter boundary, use the smaller value between the upper bound parameter boundary and the corrected lower bound parameter boundary as the lower limit of the correction range and the larger value as the upper limit of the correction range, defining the interval between the two parameter boundaries. The new conflict coordination range is determined and used as the correction range. When the object to be corrected is an upper bound constraint state, the unchanged lower bound parameter boundary is compared with the corrected upper bound parameter boundary. If the lower bound parameter boundary is less than or equal to the corrected upper bound parameter boundary, the lower bound parameter boundary is used as the lower limit of the correction range and the corrected upper bound parameter boundary is used as the upper limit of the correction range to form a common feasible interval and determine it as the correction range. If the lower bound parameter boundary is greater than the corrected upper bound parameter boundary, the smaller value between the corrected upper bound parameter boundary and the lower bound parameter boundary is used as the lower limit of the correction range and the larger value is used as the upper limit of the correction range. The interval between the two parameter boundaries is determined as the new conflict coordination range and used as the correction range.

[0031] In this embodiment, the optimization and simulation according to the correction range, and the output of optimization parameters after satisfying the constraints, include: Within the correction range, candidate values ​​for conflicting parameters are regenerated while keeping the parameter boundaries corresponding to non-conflicting parameters and uncorrected polarity constraints unchanged. Local optimization is then performed on the candidate values ​​to obtain corrected candidate parameters. Specifically, this process involves regenerating candidate values ​​for conflicting parameters within the correction range while keeping the parameter boundaries corresponding to non-conflicting parameters and uncorrected polarity constraints unchanged, and then performing local optimization on the candidate values. Replace the original adjustment range of the corresponding conflicting parameter with the correction range, keeping the current value of the non-conflicting parameter, the current value of the conflicting parameter that has not obtained the correction range in this round, and the parameter boundary corresponding to the uncorrected polarity constraint unchanged; generate local candidate values ​​of the conflicting parameter that have obtained the correction range in this round according to the lower limit of the correction range, the upper limit of the correction range, and the current value of the corresponding conflicting parameter, and combine them to form a correction candidate parameter group; calculate the boundary deviation of each correction candidate parameter group relative to the upper and lower bound parameter boundaries and the parameter adjustment relative to the current value of the parameter, sort them according to the boundary deviation in ascending order, and if the boundary deviation is the same, sort them according to the parameter adjustment in ascending order, and determine the correction candidate parameter group with the first position in the sort as the correction candidate parameter; The modified candidate parameters are written into the structural model of the diesel engine components, and simulations of structural deformation, fatigue safety, and structural weight are performed. These simulations are then compared with subsystem engineering constraints to obtain the verification status. Specifically, the verification status is as follows: According to the component identification and structural parameter name, the modified candidate parameters are written into the corresponding geometric parameter positions. The material parameters, diesel engine design load, and structural constraints corresponding to the structural simulation task are read. Structural deformation, fatigue safety, and structural weight simulations are performed to obtain structural deformation results, structural stress results, fatigue safety results, and structural weight results. The corresponding simulation results are matched according to the constraint object identification and constraint index name in the subsystem engineering constraints. The constraint type, constraint boundary, source identification, and safety margin requirements are read. The boundary deviation, deviation direction, and verification margin are determined according to the boundary comparison method of the upper and lower limits of the constraints. When a boundary deviation occurs, the deviation direction of the upper limit constraint is recorded as exceeding the upper limit, and the deviation direction of the lower limit constraint is recorded as below the lower limit. No deviations were found. When a boundary deviation occurs, the deviation is recorded as 0. Safety margin requirements recorded in proportional form are directly used as the minimum verification margin. Safety margin requirements recorded in absolute form are divided by the absolute value of the corresponding constraint boundary to obtain the minimum verification margin. When the constraint boundary is 0, 0.000001 is used as the divisor. When no safety margin requirement is recorded, the minimum verification margin is set to 0.05. When the deviation is 0 and the verification margin is lower than the minimum verification margin, the deviation direction of the upper limit constraint is recorded as approaching the upper limit, and the deviation direction of the lower limit constraint is recorded as approaching the lower limit. When the deviation of all system engineering constraints is 0 and the verification margin is not lower than the corresponding minimum verification margin, the verification status is recorded as constraint satisfied. When there is any boundary deviation or insufficient verification margin, the verification status is recorded as constraint unsatisfied. When the verification status meets the subsystem engineering constraints, the corrected candidate parameters are determined as the optimization parameters. When there are unmet constraints, a verification deviation carrying a source identifier is generated and input into the improved CompGCN model. The corresponding polarity constraint status and correction range are updated along the constraint source path according to the source identifier, while keeping the other polarity constraint status unchanged. Local optimization and structural simulation are re-executed until the subsystem engineering constraints are met. Specifically, closed-loop iteration is performed based on the verification status to determine the optimization parameters, as follows: When the verification status is that the constraints are met, the current candidate parameter for correction is determined as the optimization parameter and the iteration ends. When the verification status is that the constraints are not met, the boundary deviation, deviation direction, verification margin, minimum verification margin, and source identifier corresponding to the unmet constraint are read. When the boundary deviation is greater than 0, the boundary deviation is directly used. When the boundary deviation is 0 and the verification margin is lower than the minimum verification margin, the deviation is obtained by multiplying the difference between the minimum verification margin and the verification margin by the absolute value of the corresponding constraint boundary. When the constraint boundary is 0, 0.000001 is used as the multiplier. The source identifier, deviation direction, and deviation are combined to generate the verification bias. The verification bias is input into the improved CompGCN model. The source identifier sequence in the source chain backlink index is matched according to the source identifier to determine the corresponding constraint source path, conflict parameter, and upper or lower bound constraint status. The transformation relationship in the backlink edge sequence is used to convert the source path into a new constraint path. Calculate the boundary correction direction and amount, update the corresponding polarity constraint state, and keep the parameter boundary corresponding to the other polarity constraint state unchanged; compare the corrected upper and lower parameter boundaries. If the lower parameter boundary is not higher than the upper parameter boundary, the common feasible interval between the two parameter boundaries is taken as the correction range. If the lower parameter boundary is higher than the upper parameter boundary, the conflict coordination range between the two parameter boundaries is taken as the correction range; within the correction range, regenerate the conflict parameter candidate values ​​to obtain the correction range of this round, perform local optimization and execute diesel engine structure simulation, and obtain the verification state again; if there are still unmet constraints in the verification state, continue to generate verification deviations and update the corresponding polarity constraint state and correction range until all system engineering constraints have no boundary deviations and the verification margins are not lower than the corresponding minimum verification margins. The correction candidate parameters that meet the conditions are determined as optimization parameters.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to the optimization of structural parameters of a certain type of high-strength 6-cylinder diesel engine. The optimization object was a digital model composed of the engine block, main bearing housing, cylinder bore reinforcement structure, and oil passage area, with an initial design weight of 187.6 kg. During the design phase, 12 requirements were established for structural reliability, lightweighting, structural lifespan, and space layout; 18 system specifications; 27 subsystem engineering constraints; 42 component structural parameters; and 96 inter-layer constraints. Component structural parameters included cylinder wall thickness, main bearing housing reinforcement rib thickness, cylinder bore fillet radius, oil passage diameter, connection area thickness, and structural spacing. The current values ​​for the main bearing housing reinforcement rib thickness are 9.5 mm, cylinder wall thickness, and cylinder bore fillet radius are 4.0 mm.

[0033] The system organizes design data according to the requirements layer, system specification layer, engineering constraint layer, and structural parameter layer. It reads structural reliability and lightweight requirements from the design task book; overall machine weight limits, structural strength indicators, and external dimension limits from the overall design parameter table; deformation constraints, stress constraints, fatigue safety constraints, and weight constraints from the structural design constraint table; and current values ​​of component structural parameters from the current version of the machine body 3D model and parameter table. Each data item is written with an object identifier and a hierarchy identifier, and a constraint transfer record between adjacent levels is established based on the design index decomposition table and parameter association records, forming multi-layered design data.

[0034] A multi-layer constraint transfer graph was established based on multi-layer design data, resulting in 69 nodes and 96 directed edges. Each edge stores the constraint polarity, transformation relationship, source identifier, and constraint transfer record index. Among them, 54 edges belong to the upper bound constraint polarity, and 42 edges belong to the lower bound constraint polarity. Taking the thickness of the main bearing housing stiffener as an example, the lightweight requirement forms the upper bound through the body mass specifications and weight engineering constraints, while the structural reliability requirement forms the lower bound through the fatigue life specifications and fatigue safety constraints. The two constraint source paths ultimately converge to the same structural parameter node.

[0035] The improved CompGCN model was trained using 240 historical diesel engine structural design cases, with 192 cases used as the training set and 48 cases as the validation set. Both node embedding and relation embedding dimensions were set to 64, the Adam optimizer learning rate was set to 0.001, and the batch size was set to 32. The weights for constraint state identification error, source path matching error, and validation bias reverting error were set to 0.4, 0.3, and 0.3, respectively. When the model reached 86 rounds of training, the difference between the maximum and minimum joint training errors of the validation set for five consecutive rounds was less than 0.001. The model parameters corresponding to the round with the lowest joint training error on the validation set were saved.

[0036] After the improved CompGCN model is trained by inputting a multi-layer constraint transit graph, the node embedding layer converts requirement, specification, engineering constraint, and structural parameter nodes into 64-dimensional node embedding representations, and the relation embedding layer converts constraint polarity, transformation relations, and source identifiers into 64-dimensional relation embedding representations. The constraint polarity preservation and combination layer establishes upper and lower bound polarity paths according to the constraint polarity of the edges, performs element-wise combination of node and relation embedding representations, and synchronously transmits the source identifier. The multi-relation aggregation layer aggregates the same polarity propagation features according to the target structural parameter nodes, keeps the different polarity aggregation states independent, and concatenates the edge sequence according to the source identifier to form a source path candidate set.

[0037] The model outputs the following parameters: Lower bound for the main bearing housing stiffener thickness is 9.8mm, upper bound is 10.7mm, and the current value of 9.5mm is below the common feasible range; lower bound for the cylinder wall thickness is 6.4mm, upper bound is 6.6mm, and the current value of 6.8mm is above the common feasible range; lower bound for the cylinder bore fillet radius is 4.6mm, upper bound is 4.3mm, and there is no common feasible range between the two parameter boundaries. The system identifies these three structural parameters as conflict parameters and establishes parameter adjustment ranges of 9.8-10.7mm, 6.4-6.6mm, and 4.3-4.6mm, respectively.

[0038] The system divides each adjustment range equally into 11 local candidate values. Three conflicting parameters are combined according to Cartesian rules to form 1331 local candidate parameter groups, while 39 non-conflicting parameters retain their current values. The system calculates the deviation of each candidate value from its upper and lower bounds and the adjustment relative to the current parameter value. Parameter groups are sorted by their deviation from the boundary values ​​in ascending order; if the deviations are the same, they are sorted by their adjustment values ​​in ascending order. The first parameter group in the sorted list has a main bearing housing reinforcing rib thickness of 9.8 mm, a cylinder wall thickness of 6.6 mm, and a cylinder bore fillet radius of 4.3 mm, forming the parameters to be verified.

[0039] After the parameters to be verified were written into the 3D structural model of the diesel engine block, high-strength cast iron was selected as the material, with an elastic modulus of 168 GPa, a Poisson's ratio of 0.28, and a density of 7.2 g / cm³. The simulation task applied a rated in-cylinder pressure of 18.5 MPa and a preload on the main bearing cap bolts, and adopted the installation constraints recorded in the current design task. The subsystem engineering constraints stipulated that the maximum structural deformation should not exceed 0.32 mm, the maximum equivalent stress should not exceed 320 MPa, the minimum fatigue safety factor should not be less than 1.45, and the structural weight should not exceed 186 kg, corresponding to verification margin requirements of 5%, 5%, 4%, and 0.4%, respectively. The first simulation yielded a maximum nodal displacement of 0.29 mm, a maximum equivalent stress of 298 MPa, a minimum fatigue safety factor of 1.39, and a structural weight of 184.7 kg. The fatigue safety constraint showed boundary deviation, while the other constraints met the corresponding verification margin requirements.

[0040] The system records the deviation direction of the fatigue safety constraint as below the lower limit with a deviation amount of 0.06, and obtains the corresponding source identifier. After verifying the deviation input into the improved CompGCN model, the verification deviation source chain backlay layer matches the constraint source path of "structural reliability requirements - fatigue life specifications - fatigue safety constraints - main bearing housing stiffener thickness" based on the source identifier, and locates the lower limit constraint state of the stiffener thickness. According to the transformation relationship saved on each edge in the path, the lower limit parameter boundary correction amount of stiffener thickness is obtained by 0.4mm, the lower limit parameter boundary is corrected from 9.8mm to 10.2mm, and the upper limit parameter boundary remains unchanged at 10.7mm, forming a correction range of 10.2-10.7mm.

[0041] The system regenerates 11 candidate values ​​only for the main bearing housing stiffener thickness within the corrected range, while keeping the cylinder wall thickness (6.6 mm) and cylinder bore radius (4.3 mm) unchanged. After sorting by boundary deviation and parameter adjustment, corrected candidate parameters of stiffener thickness 10.2 mm, cylinder wall thickness 6.6 mm, and cylinder bore radius 4.3 mm are obtained. The second structural simulation yields a maximum structural deformation of 0.28 mm, a maximum equivalent stress of 286 MPa, a minimum fatigue safety factor of 1.53, and a structural weight of 185.2 kg, corresponding to verification margins of 12.5%, 10.6%, 5.5%, and 0.43%, respectively. All of these margins are not lower than the corresponding minimum verification margins, and the system determines the current corrected candidate parameters as the optimized parameters.

[0042] To verify the closed-loop optimization effect, under the same initial structural model, material parameters, load conditions, and engineering constraints, this invention was compared with the traditional unified multi-objective parameter optimization method. The traditional method requires 8 rounds of structural simulation, evaluating a total of 4260 candidate parameters, with the final maximum structural deformation being 0.30 mm, the minimum fatigue safety factor being 1.51, and the structural weight being 185.9 kg. This invention completes closed-loop optimization in 2 rounds of structural simulation, evaluating a total of 1342 candidate parameters, with the maximum structural deformation being 0.28 mm, the minimum fatigue safety factor being 1.53, and the structural weight being 185.2 kg. Compared to the traditional method, this invention reduces the number of structural simulation iterations from 8 to 2, reduces the number of candidate parameters evaluated by 68.5%, and satisfies all system engineering constraints and corresponding minimum verification margins.

[0043] As can be seen from Example 1, the present invention can transform the hierarchical source relationship between diesel engine structural design requirements, system specifications, subsystem engineering constraints, and component structural parameters into a computable multi-layer constraint transfer diagram. It uses an improved CompGCN model to maintain the independent propagation of upper and lower bound constraints, and corrects the boundary of the specified polarity parameter along the corresponding constraint source path when verification deviations occur in structural simulation. This reduces the repeated adjustment of irrelevant parameters and improves the pertinence and computational efficiency of conflict parameter identification, parameter range correction, and closed-loop optimization of diesel engine structural design.

[0044] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer, characterized in that, include: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters and inter-layer constraint relationships to generate multi-layer design data; Nodes and edges are constructed from multi-layer design data, and constraint polarity, transformation relationships and source identifiers are configured to generate a multi-layer constraint transfer graph. The multi-layer constraint transfer graph is input into the improved CompGCN model to obtain the upper bound constraint state, the lower bound constraint state, and the constraint source path. The improved CompGCN model includes a node embedding layer, a relation embedding layer, a constraint polarity preservation combination layer, a multi-relation aggregation layer, a verification bias source chain backfitting layer, and an output layer. Compare the upper and lower bound constraint states of the same structural parameter, and identify conflicting parameters and adjustment ranges by combining the current parameter values; By locally optimizing and adjusting the conflicting parameters within the range, the parameters to be verified are obtained. Perform diesel engine structural simulation on the parameters to be verified to obtain deformation, fatigue safety and weight results, match subsystem engineering constraints and associate source identifiers to generate verification deviations; Input the verification bias into the improved CompGCN model, paste it back to the corresponding upper or lower bound constraint state along the constraint source path according to the source identifier, and keep the other polarity constraint state unchanged to generate the correction range; Optimize and simulate within the correction range, and output the optimized parameters after the constraints are satisfied.

2. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The diesel engine structural design requirements include structural reliability requirements, lightweight requirements, structural life requirements, and spatial layout requirements; the system specifications include overall machine weight limits, structural strength indicators, fatigue life indicators, and external dimension limits; the subsystem engineering constraints include deformation constraints, stress constraints, fatigue safety constraints, and weight constraints; the component structural parameters include wall thickness, stiffener dimensions, fillet radius, hole diameter, and structural spacing; the interlayer constraint relationships include the source correspondence between adjacent layer data, the direction of constraint action, and parameter conversion rules.

3. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The generation of multi-layer design data includes: Collect diesel engine structural design requirements, system specifications, subsystem engineering constraints, component structural parameters and inter-layer constraints, establish data items according to their respective design levels, record the current values ​​of component structural parameters, and configure object identifiers and level identifiers for each data item; According to the object identifier, the data items of adjacent levels are matched accordingly, the source correspondence, constraint direction and parameter conversion rules in the inter-level constraint relationship are extracted, and the constraint transmission record between adjacent level data items is established; Arrange the data items according to the hierarchical identifier, associate the constraint transfer records with the corresponding data items, and form multi-layer design data that includes data items, current parameter values, hierarchical identifiers, and inter-layer constraint relationships.

4. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The generation of the multi-layer constraint transfer graph includes: According to the hierarchy of diesel engine structural design requirements, system specifications, subsystem engineering constraints, and component structural parameters, the data items in the multi-level design data are set as corresponding level nodes. Match adjacent level nodes based on inter-level constraints, and establish edges pointing from upstream nodes to downstream nodes according to the source of constraints; Determine the upper or lower bound constraint polarity based on the direction of the constraint action corresponding to the edge, configure the transformation relationship according to the parameter transformation rules, and configure the source information of the upstream constraint as the source identifier; By associating nodes at each level with the configured edges, a multi-layer constraint transfer graph is generated, which transmits the diesel engine structural design requirements to the component structural parameters layer by layer.

5. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The process of obtaining the upper bound constraint state, the lower bound constraint state, and the constraint source path includes: Based on the CompGCN model, the node representation, relation representation, node and relation combination propagation, multi-relation aggregation and result output process are set up with a node embedding layer, a relation embedding layer, a multi-relation aggregation layer and an output layer. A constraint polarity preservation combination layer is added at the output convergence position of the node embedding layer and the relation embedding layer, and a verification bias source chain back-fitting layer is added between the multi-relation aggregation layer and the output layer to form an improved CompGCN model. The nodes in the multi-layer constraint transfer graph are input into the node embedding layer. Embedding mapping is performed based on the node's level and data to generate a node embedding representation. The edge input relationship embedding layer in the multi-layer constraint transfer graph is used to perform relationship embedding based on constraint polarity, transformation relationship and source identifier, and generate a relationship embedding representation carrying source identifier; The node embedding representation and relation embedding representation are combined with the input constraint polarity preservation layer. Adjacent nodes and corresponding relations are combined according to the constraint polarity of the edges. The combination results are passed layer by layer along the upper and lower polarity paths, and the source identifier is passed at the same time to generate upper and lower propagation features. The upper and lower bound propagation features are input into the multi-relationship aggregation layer. Propagation features of the same polarity are aggregated according to the component structure parameter nodes. The aggregation results of different polarities are kept independent of each other. The upper and lower bound aggregation states are generated, and the edges that have been propagated are concatenated according to the source identifier to generate the source path candidate set. Input the upper bound aggregation state, lower bound aggregation state, and source path candidate set into the verification deviation source chain back-paste layer, and associate the source path with the corresponding upper bound aggregation state or lower bound aggregation state according to the source identifier to establish the source chain back-paste index. Input the upper bound aggregation state, lower bound aggregation state, and source chain backlink index into the output layer to generate the upper bound constraint state and lower bound constraint state of each component's structural parameters, and generate the corresponding constraint source path based on the source chain backlink index. Training samples were constructed by collecting multi-layer design data, parameter constraint records, structural simulation verification records, and parameter correction records from historical diesel engine structural design cases. The constraint state identification error, source path matching error, and verification deviation re-attachment error were used as joint training objectives to iteratively train the improved CompGCN model until the joint training objectives met the preset convergence conditions, thus obtaining the trained improved CompGCN model.

6. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The method of identifying conflicting parameters and adjusting their ranges based on the current values ​​of the parameters includes: Read the upper and lower bound constraint states of the structural parameters of the same component, extract the upper and lower bound parameter boundaries respectively, and compare the relative positions of the parameter boundaries on both sides. The current value of the parameter is compared with the upper and lower parameter boundaries respectively. The structural parameters of the component that do not have a common feasible interval between the two parameter boundaries or whose current value exceeds the common feasible interval are identified as conflict parameters. For conflicting parameters that share a common feasible interval, the parameter adjustment range is determined by the common feasible interval; for conflicting parameters that do not share a common feasible interval, the parameter adjustment range is determined by the interval between the boundaries of the two parameters.

7. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The conflict parameters within the local optimization adjustment range yield the following parameters to be verified: Read the current value, adjustment range, and upper and lower bound constraint states of the conflicting parameters, keep the non-conflicting parameters unchanged, generate local candidate values ​​for each conflicting parameter within the adjustment range, and combine them to form a local candidate parameter group; The local candidate parameter groups are matched with the corresponding upper and lower bound constraint states respectively. The deviation of the candidate values ​​from the upper and lower bound parameter boundaries and the adjustment amount relative to the current parameter values ​​are calculated. The local candidate parameter groups are sorted and filtered according to the boundary deviation and parameter adjustment amount to obtain the target parameter group. Update the corresponding conflicting parameters with the candidate values ​​in the target parameter group, and keep the current values ​​of the remaining component structural parameters unchanged, and combine them to generate the parameters to be verified.

8. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The generated verification bias includes: Write the parameters to be verified into the corresponding diesel engine component structural model, perform structural deformation, fatigue safety and structural weight simulation, and obtain the structural simulation results; Based on the structural simulation results of the subsystem engineering constraint matching, read the corresponding constraint boundary, compare the structural simulation results with the constraint boundary, and determine the deviation direction, deviation amount and verification margin. Screen subsystem engineering constraints that have boundary deviations or insufficient verification margins, obtain the corresponding source identifiers, and combine the source identifiers, deviation directions, and deviation amounts to generate verification biases.

9. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The generated correction range includes: Input the validation bias into the improved CompGCN model, match the constraint source path based on the source identifier carried by the validation bias, and locate the corresponding conflict parameters and upper or lower bound constraint states. Read the deviation direction and deviation amount of the verification deviation, determine the boundary correction direction and correction amount according to the transformation relationship of each edge in the matching constraint source path, and paste it back to the located upper or lower constraint state to obtain the corrected constraint state. Keep the parameter boundary corresponding to the other polarity constraint state of the conflict parameter unchanged, compare the parameter boundary corresponding to the correction constraint state with the other parameter boundary, and limit the correction range with the parameter boundaries on both sides.

10. The method for optimizing diesel engine structural design parameters based on multi-layer constraint transfer according to claim 1, characterized in that, The optimization and simulation based on the correction range, after satisfying the constraints, outputs the following optimization parameters: Within the correction range, candidate values ​​for conflicting parameters are regenerated, while keeping the parameter boundaries corresponding to non-conflicting parameters and uncorrected polarity constraints unchanged. Local optimization is then performed on the candidate values ​​to obtain corrected candidate parameters. The modified candidate parameters are written into the structural model of the diesel engine components, and the structural deformation, fatigue safety and structural weight simulations are performed. The results are then compared with the subsystem engineering constraints to obtain the verification status. When the verification status meets the subsystem engineering constraints, the candidate parameters for correction are determined as optimization parameters; when there are unmet constraints, a verification deviation carrying a source identifier is generated and input into the improved CompGCN model. The corresponding polarity constraint status and correction range are updated along the constraint source path according to the source identifier, while keeping the other polarity constraint status unchanged. Local optimization and structural simulation are re-executed until the subsystem engineering constraints are met.

Images