A method and system for optimizing casting process parameters based on digital twins

Abstract

This invention discloses a method and system for optimizing casting process parameters based on digital twins, comprising the following steps: Step 1: Acquire casting production records and twin prediction records to generate a process evidence set; Step 2: Establish a casting twin model and form a closure-bearing structure; Step 3: Establish closure mapping edges and generate a set of casting twin focal elements; Step 4: Calculate state residuals to generate twin residual records; Step 5: Backtrack to generate twin residual conflict retracement fronts; Step 6: Construct an improved Dubois-Prade algorithm; Step 7: Assign conflict quality to generate parameter correction records; Step 8: Update process parameter combinations and output optimization results. This invention improves the traceability of casting parameter correction and the stability of closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method and system for optimizing casting process parameters based on digital twins. Background Technology

[0002] With the increasing demands for digital transformation of casting equipment and improved process quality stability, digital twin modeling, state prediction, and process parameter optimization technologies for casting production processes have received widespread attention. Existing casting process optimization systems primarily rely on on-site sensor data, simulation prediction results, and quality inspection results for model calibration, generating parameter adjustment schemes through neural networks, particle swarm optimization, and statistical experimental analysis. However, these systems commonly suffer from the following problems in practical applications: The casting production records and twin prediction records come from different sources, and the data are prone to misalignment in terms of process sequence, acquisition time, and status fields, making it difficult to establish a stable correspondence between the measured and predicted states, thus affecting the usability of process evidence. In the twin model, there is a lack of structured relationships between process parameters, process states, defect evolution, and quality objectives, making it difficult to trace residual results back to specific parameter nodes along the casting process path. When conflicts occur among multiple sources of evidence, existing fusion methods mostly remain at the level of risk assessment and credibility evaluation, making it difficult to transform conflict quality into corrective records oriented towards process parameter combinations, resulting in a lack of interpretable closed-loop update basis for parameter optimization results.

[0003] Therefore, how to provide a method and system for optimizing casting process parameters based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a casting process parameter optimization method and system based on digital twins. This invention constructs a casting twin model, a closure bearing structure, and an improved Dubois-Prade algorithm to form a process evidence set by combining casting production records and twin prediction records. Based on the twin residual conflict rewind front, conflict quality is allocated to parameter adjustment focal elements, thereby achieving closed-loop updating of casting process parameter combinations. This invention has the advantages of traceable parameter correction, high accuracy of conflict fusion, and strong stability of process optimization.

[0005] A method for optimizing casting process parameters based on digital twins according to an embodiment of the present invention includes the following steps: Step 1: Obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; Step 2: Establish a casting twin model based on the casting process path, connect parameter nodes to state nodes, connect state nodes to defect nodes, and connect defect nodes to target nodes to form a closed-loop support structure. Step 3: Establish a closure mapping edge from the parameter node to the target node in the closure bearing structure, and map the process evidence set into the closure mapping edge to generate a set of casting twin focal elements carrying basic probability assignments; Step 4: Compare the measured states in the process evidence set with the predicted states in the twin prediction records, calculate the state residuals corresponding to the same closure mapping edge, and generate twin residual records; Step 5: Based on the twin residual record, trace backward along the closure mapping edge to the parameter node to generate the twin residual conflict retrace front; Step 6: Write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to construct the improved Dubois-Prade algorithm; Step 7: Use the improved Dubois-Prade algorithm to process the set of casting twin coke elements, distribute the conflict mass along the twin residual conflict rollback front to the parameter-adjusting coke elements, and generate parameter correction records; Step 8: Update the process parameter combination in the casting twin model based on the parameter correction record, and output the casting process parameter optimization results.

[0006] Optionally, step one specifically includes: Collect process execution records, process parameter records, and quality inspection records generated during the casting production process. Extract casting batch identifiers, process identifiers, and collection times from the process execution records, extract parameter values ​​from the process parameter records, and extract measured states from the quality inspection records to generate casting production records. Read the prediction output record generated during the casting twin simulation process, extract the casting batch identifier, process identifier, prediction time, and prediction status from the prediction output record, and generate twin prediction record; The casting production records and twin prediction records are matched based on the casting batch identifier and process identifier. The matching results are aligned according to the acquisition time and prediction time to generate process time sequence association records. Arrange the process sequence records according to the casting process sequence, and write the parameter values, actual measured status and predicted status into the same evidence unit, and collect the evidence units to generate a process evidence set.

[0007] Optionally, step two specifically includes: Extract casting batch identifiers, process identifiers, and collection times from the process evidence set; arrange the process identifiers under the same casting batch identifier according to the collection time to generate casting process paths. A casting twin model is established based on the casting process path. The casting twin model is a directed mapping model that organizes the process evidence set according to the casting process path, and node generation rules and node connection rules are configured. According to the node generation rules, parameter values ​​are extracted from the process evidence set and parameter nodes are generated according to the process identifier. Measured and predicted states are extracted from the process evidence set and state nodes are generated according to the process identifier. Defect nodes are generated according to the process identifier corresponding to the state nodes. Target nodes are generated according to the casting batch identifier. According to the node connection rules, the parameter nodes are connected to the status nodes corresponding to the same process identifier, the status nodes are connected to the defect nodes corresponding to the same process identifier, and the defect nodes are connected to the target nodes corresponding to the same casting batch identifier, forming a closed-loop bearing structure.

[0008] Optionally, step three specifically includes: Perform direction calibration on the node connection relationships in the closure bearing structure, organize the connections from parameter nodes to state nodes into parameter state edges, organize the connections from state nodes to defect nodes into state defect edges, and organize the connections from defect nodes to target nodes into defect target edges. According to the casting process path, link splicing is performed on parameter state edge, state defect edge and defect target edge, and the spliced ​​link extending from parameter node to target node under the same casting batch identifier is encapsulated as closure mapping edge. The process evidence set is organized into a sequence of evidence units according to the casting process sequence. Field encoding is performed on the parameter values, measured states and predicted states in the evidence unit sequence. The field encoding results are matched with the execution positions of parameter nodes, state nodes and defect nodes in the closure mapping edge to generate a closure mapping record. A consistency determination is performed on the measured state and the predicted state in the closure mapping record. Evidence propositions are generated according to the consistency determination results. The evidence propositions, closure mapping records and position identifiers of closure mapping edges are combined to generate casting twin focal elements. The occurrence count of evidence propositions and the total number of units corresponding to evidence unit sequences within the same closure mapping edge are counted. The occurrence count and the total number of units are normalized to generate basic probability assignments. The basic probability assignments are written into the casting twin focal elements, and the casting twin focal elements are collected to generate a casting twin focal element set.

[0009] Optionally, step four specifically includes: The process evidence set is organized into an evidence unit sequence according to the casting batch identifier, process identifier, and casting process sequence. The evidence unit sequence is then matched with the position identifier of the closure mapping edge to form a residual calculation unit. Perform state field alignment on the measured state and predicted state in the residual calculation unit, and convert the field-aligned measured state into a measured state vector and the field-aligned predicted state into a predicted state vector. The measured state vector and the predicted state vector are used to calculate the corresponding position difference, generate the state difference vector, and perform amplitude adjustment and direction calibration on the state difference vector to form the state residual corresponding to the same closure mapping edge; Write the state residual, the position identifier of the closure mapping edge, the casting batch identifier, and the process identifier into the same residual entry, and arrange the residual entries according to the casting process sequence to generate twin residual records.

[0010] Optionally, step five specifically includes: Match the residual entries in the twin residual record with the closure mapping edges in the closure bearing structure according to the position identifier of the closure mapping edge, determine the position of the state node in the closure mapping edge, and generate a residual location record. Following the reverse connection order of the closure mapping edges, the residual location record is pointed back from the state node through the defect node to the parameter node, forming a residual rollback link; Write the status residual, casting batch identifier, and process identifier from the residual location record into the residual rollback link, and organize the residual rollback link according to the casting process sequence to generate parameter node conflict records. Based on the amplitude results and direction calibration in the parameter node conflict record, the residual rollback links associated with the same parameter node are subjected to direction consistency merging and amplitude accumulation processing to form the parameter node conflict quantity; The parameter node conflict quantity, residual rollback link, and parameter node are associated and organized according to the position identifier of the closure mapping edge to generate a twin residual conflict rollback front.

[0011] Optionally, step six specifically includes: The closure-bearing structure is grouped according to the position identifier of the closure mapping edge. The parameter nodes, state nodes, defect nodes and target nodes associated with the same position identifier are extracted, and the closure-bearing unit is generated according to the connection order from the parameter node to the target node. Extract the position identifiers of the closure mapping edges carried by the casting twin focal elements from the set of casting twin focal elements, perform matching between the casting twin focal elements and the closure carrying units corresponding to the same position identifier, and generate focal element carrying index records. Based on the focal element carrying index record, the range of focal element union in the Dubois-Prade conflict fusion rule is restricted to the same closure carrying unit, forming a closure carrying constraint; Align the twin residual conflict rollback front edge with the closure bearing unit according to the position identifier of the closure mapping edge, extract the residual rollback link, parameter node and parameter node conflict amount, and generate conflict rollback index record; Based on the conflict rollback index record, the conflict quality transfer path in the Dubois-Prade conflict fusion rule is limited to the residual rollback link, and the conflict quality transfer endpoint is limited to the parameter node in the residual rollback link, thus forming a rollback transfer constraint. Based on the number of parameter node conflicts, the conflict quality configuration of the same residual rewind link is assigned a weight, and the assigned weight is associated with the parameter node to form a parameter adjustment focal element allocation rule. The improved Dubois-Prade algorithm is constructed by incorporating the closure bearer unit, focal element bearer index record, closure bearer constraint, conflict rollback index record, rollback transfer constraint, and parameter adjustment focal element allocation rule into the Dubois-Prade conflict fusion rule.

[0012] Optionally, step seven specifically includes: The set of casting twin coke elements is input into the improved Dubois-Prade algorithm, and the closure bearing unit corresponding to each casting twin coke element is determined according to the coke element bearing index record. Within the same closure carrying unit, the intersection determination of the casting twin coke elements is performed. The casting twin coke element combination with an empty intersection is taken as the conflict coke element combination. The product calculation is performed on the basic probability assignment carried by the conflict coke element combination to generate conflict mass. According to the closure bearing constraint, the conflict mass is confined to the closure bearing unit corresponding to the conflict focal element combination, thus forming the closure constraint conflict mass; The residual rollback link corresponding to the closure constraint conflict quality is determined according to the conflict rollback index record, and the closure constraint conflict quality is transferred to the parameter node in the residual rollback link according to the rollback transfer constraint. Extract the allocation weights corresponding to the parameter nodes according to the parameter adjustment focal element allocation rules, write the closure constraint conflict quality passed to the parameter nodes into the parameter adjustment focal element according to the allocation weights, and generate the parameter adjustment focal element quality. The position identifiers of the parameter adjustment focal element quality, parameter nodes, residual rewind links, and closure mapping edges are associated and organized to generate parameter correction records.

[0013] Optionally, step eight specifically includes: Extract the position identifiers of parameter-adjusted coke quality, parameter nodes, residual rewind links, and closure mapping edges from the parameter correction records, and locate the process parameter combination in the casting twin model according to the parameter nodes; The direction calibration and amplitude results are extracted from the residual rewind link. The direction calibration is converted into the parameter correction direction. The amplitude results are multiplied and converted with the parameter-adjusted focal element quality to generate the parameter correction amplitude. Incremental updates are performed on the parameter values ​​in the process parameter combination according to the parameter correction direction and parameter correction magnitude to generate updated parameter values; Write the updated parameter values ​​into the parameter nodes, and organize the mapping relationship between the parameter nodes and the updated parameter values ​​according to the position identifier of the closure mapping edge to generate the updated process parameter combination; The updated process parameter combinations, parameter correction records, and closure mapping edge position identifiers are linked and organized to output the casting process parameter optimization results.

[0014] A casting process parameter optimization system based on digital twin according to an embodiment of the present invention includes: The process evidence module is used to obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; The twin modeling module is used to create a casting twin model based on the casting process path and form a closed-loop load-bearing structure. The focal element generation module is used to establish closure mapping edges and generate a set of casting twin focal elements carrying basic probability assignments. The residual calculation module is used to calculate the state residuals corresponding to the same closure mapping edge and generate twin residual records; The rollback front module is used to trace backward along the closure mapping edge to the parameter node and generate a twin residual conflict rollback front. The algorithm building module is used to write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to build an improved Dubois-Prade algorithm. The parameter correction module is used to process the set of casting twin coke elements using the improved Dubois-Prade algorithm and generate parameter correction records. The optimized output module is used to update the combination of process parameters in the casting twin model based on the parameter correction records, and output the optimization results of the casting process parameters.

[0015] The beneficial effects of this invention are: This invention generates a process evidence set by sequentially associating casting production records and twin prediction records according to the casting process sequence. This ensures that the actual measured state, predicted state, and parameter values ​​maintain a correspondence within the same process link, reducing misalignment issues in time, process, and state fields between data from different sources. By establishing a closure-bearing structure composed of parameter nodes, state nodes, defect nodes, and target nodes, and forming closure mapping edges within this structure, a traceable mapping link is created between casting process parameters, process states, defect evolution, and quality objectives. This improves the completeness of the casting twin model's representation of the production process and the interpretability of the parameter optimization basis.

[0016] This invention further utilizes twin residual records to construct a twin residual conflict rewind front, and incorporates the closure bearing structure and the twin residual conflict rewind front into the Dubois-Prade conflict fusion rule, forming an improved Dubois-Prade algorithm. This allows multi-source evidence conflicts to no longer be limited to risk assessment, but can be allocated to parameter adjustment focal elements along the residual rewind link, generating parameter correction records. This transforms the deviation between simulation predictions and actual production measurements into an incremental update basis for process parameter combinations, improving the targeting of casting process parameter corrections, the stability of closed-loop adjustment, and the reliability of casting quality optimization results. Attached Figure Description

[0017] 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 casting process parameter optimization method based on digital twin proposed in this invention; Figure 2 This is a schematic diagram of the closure bearing structure of a casting process parameter optimization method based on digital twin proposed in this invention. Detailed Implementation

[0018] 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.

[0019] refer to Figures 1-2 A method for optimizing casting process parameters based on digital twins includes the following steps: Step 1: Obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; Step 2: Establish a casting twin model based on the casting process path, connect parameter nodes to state nodes, connect state nodes to defect nodes, and connect defect nodes to target nodes to form a closed-loop support structure. Step 3: Establish a closure mapping edge from the parameter node to the target node in the closure bearing structure, and map the process evidence set into the closure mapping edge to generate a set of casting twin focal elements carrying basic probability assignments; Step 4: Compare the measured states in the process evidence set with the predicted states in the twin prediction records, calculate the state residuals corresponding to the same closure mapping edge, and generate twin residual records; Step 5: Based on the twin residual record, trace backward along the closure mapping edge to the parameter node to generate the twin residual conflict retrace front; Step 6: Write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to construct the improved Dubois-Prade algorithm; Step 7: Use the improved Dubois-Prade algorithm to process the set of casting twin coke elements, distribute the conflict mass along the twin residual conflict rollback front to the parameter-adjusting coke elements, and generate parameter correction records; Step 8: Update the process parameter combination in the casting twin model based on the parameter correction record, and output the casting process parameter optimization results.

[0020] In this embodiment, step one specifically includes: During the casting production process, the process execution record, process parameter record, and quality inspection record are read according to the process execution sequence. The casting batch identifier, process identifier, and collection time are extracted from the process execution record. The casting batch identifier and process identifier are used as the production attribution field, and the collection time is used as the production time field. Then, the parameter values ​​are extracted from the process parameter record, and the measured status is extracted from the quality inspection record. The production attribution field, production time field, parameter values, and measured status are associated and organized to generate the casting production record. During the casting twin simulation process, the prediction output record corresponding to the casting production process is read, and the casting batch identifier, process identifier, prediction time and prediction status are extracted from the prediction output record. The casting batch identifier and process identifier are used as prediction attribution fields, and the prediction time is used as prediction time field. The prediction attribution field, prediction time field and prediction status are associated and organized to generate twin prediction records. The casting batch identifier and process identifier in the casting production record are compared with the casting batch identifier and process identifier in the twin prediction record. The matching results that pass the consistency comparison are retained. Then, the collection time and prediction time in the matching results are aligned according to the time sequence, so that the casting production record under the same casting batch identifier and the same process identifier and the twin prediction record form a time correspondence relationship, and process time sequence association record is generated. Arrange the process sequence association records according to the casting process sequence. Read the parameter values, measured status and predicted status from each process sequence association record, and write the parameter values, measured status and predicted status into the same evidence unit. So that each evidence unit corresponds to a casting batch identifier, a process identifier and a time correspondence. Collect the evidence units to generate a process evidence set.

[0021] In this embodiment, step two specifically includes: The casting batch identifier, process identifier, and collection time are read from the process evidence set. The casting batch identifier is used as the path attribution field, the process identifier is used as the path node field, and the collection time is used as the path sorting field. The process identifiers under the same casting batch identifier are arranged in order according to the path sorting field to form a casting process path that represents the sequence of casting processes. When establishing a casting twin model based on the casting process path, the casting process path is used as the temporal skeleton of the model, and the process evidence set is used as the data source of the model. This allows the casting twin model to carry parameter values, measured states, and predicted states in the order of the processes. Node generation rules and node connection rules are configured. The node generation rules limit the way the process evidence set is written to the nodes, and the node connection rules limit the directed connection between nodes. According to the node generation rules, the parameter values ​​in the process evidence set are read, and the process identifier is used as the node to generate parameter nodes. The measured state and predicted state in the process evidence set are read, and the process identifier is used as the node to generate state nodes. Then, the process identifier corresponding to the state node is used as the node to generate defect nodes, and the casting batch identifier is used as the node to generate target nodes, so that the parameter nodes, state nodes, defect nodes and target nodes all correspond to the source of the process evidence set. In the casting twin model, directed connections are established according to the node connection rules. Parameter nodes are connected to the state nodes corresponding to the same process identifier, state nodes are connected to the defect nodes corresponding to the same process identifier, and defect nodes are connected to the target nodes corresponding to the same casting batch identifier. The directed connection results are then organized according to the casting process path to form a closed-loop bearing structure.

[0022] In this embodiment, step three specifically includes: Perform direction calibration on the node connection relationship in the closure bearing structure, determine the connection category according to the start and end points of the node connection, organize the connection from parameter node to state node into parameter state edge, organize the connection from state node to defect node into state defect edge, and organize the connection from defect node to target node into defect target edge, and write the corresponding start point identifier, end point identifier and direction identifier for parameter state edge, state defect edge and defect target edge. According to the casting process path, link splicing is performed on parameter state edge, state defect edge and defect target edge. First, the parameter node under the same casting batch identifier is used as the link starting point, and then the state node, defect node and target node under the corresponding process identifier are connected in sequence. The complete splicing link extending from the parameter node to the target node is encapsulated as a closure mapping edge. The process evidence set is organized into a sequence of evidence units according to the casting process sequence. The parameter values, measured states and predicted states in the evidence unit sequence are encoded by field, so that the parameter values ​​correspond to the parameter field codes, and the measured states and predicted states correspond to the state field codes. Then, the field encoding results are matched with the parameter nodes, state nodes and defect nodes in the closure mapping edge to generate a closure mapping record containing the field encoding results and the node position relationship. A consistency determination is performed on the measured state and the predicted state in the closure mapping record. The state values ​​of the measured state and the predicted state are compared to form a consistency determination result. An evidence proposition is generated according to the consistency determination result. The evidence proposition, the closure mapping record and the position identifier of the closure mapping edge are combined to generate a casting twin focal element with path source and state determination meaning. The occurrence frequency of evidence propositions within the same closure mapping edge is counted, and the total number of units corresponding to the evidence unit sequence is counted. The ratio of the occurrence frequency to the total number of units is calculated to form a normalized result. The normalized result is used as the basic probability assignment and written into the casting twin focal element. Then, the casting twin focal elements are collected according to the position identifier of the closure mapping edge to generate the casting twin focal element set.

[0023] In this embodiment, step four specifically includes: The process evidence set is grouped according to the casting batch identifier. Under the same casting batch identifier, the evidence units are arranged according to the process identifier and the casting process sequence to form an evidence unit sequence. Then, the process identifier in the evidence unit sequence is matched with the position identifier of the closure mapping edge so that the evidence units belonging to the same closure mapping edge are grouped into the same calculation group to form a residual calculation unit. The measured and predicted states in the residual calculation unit are aligned with the state fields. First, the field correspondence is established according to the state field name. Then, the field values ​​in the measured and predicted states are arranged according to the field correspondence to make the measured and predicted states have the same field order. Then, the field-aligned measured states are converted into measured state vectors, and the field-aligned predicted states are converted into predicted state vectors. The corresponding position difference is calculated for the measured state vector and the predicted state vector. The measured state vector component is subtracted from the predicted state vector component according to the vector position to generate the state difference vector. The absolute value of each position difference in the state difference vector is sorted to form the amplitude result. The positive and negative states of each position difference are calibrated to form the state residual corresponding to the same closure mapping edge. The state residual is associated with the position identifier of the closure mapping edge, and the casting batch identifier and process identifier are written into the associated record position to form a residual entry. The residual entries are then arranged according to the casting process sequence to ensure that the state residual, the position identifier of the closure mapping edge, the casting batch identifier and the process identifier correspond in the same record, thus generating a twin residual record.

[0024] In this embodiment, step five specifically includes: The residual entries in the twin residual record are classified according to the position identifier of the closure mapping edge. The classified residual entries are compared with the closure mapping edge in the closure bearing structure. The state node is located in the closure mapping edge with the same identifier. The state residual, the position identifier of the closure mapping edge and the position of the state node are written into the same location record to generate the residual location record. Establish a reverse connection sequence in the opposite direction of the node connection direction in the closure mapping edge, so that the state node is connected to the corresponding defect node, the defect node is connected to the corresponding parameter node, and the residual location record is then passed along the reverse connection sequence, so that the state residual is pointed back from the state node through the defect node to the parameter node, forming a residual rollback link. Write the status residual, casting batch identifier, and process identifier from the residual location record into the corresponding node position of the residual rewind link, and arrange the residual rewind links under the same casting batch identifier according to the casting process sequence, so that the status residual and parameter node maintain the path correspondence relationship and generate parameter node conflict record. Extract the residual rollback links associated with the same parameter node from the parameter node conflict record, perform consistency comparison on the direction calibration in the residual rollback links, group the state residuals with consistent direction calibration into the same direction group, and then perform accumulation processing on the amplitude results in the same direction group to form the parameter node conflict quantity. Write the parameter node conflict quantity into the corresponding parameter node, and associate and organize the position identifiers of the parameter node, residual rollback link and closure mapping edge so that each parameter node conflict quantity corresponds to a residual rollback link and a closure mapping edge position identifier, generating a twin residual conflict rollback front.

[0025] In this embodiment, step six specifically includes: The closure-bearing structure is grouped according to the position identifier of the closure mapping edge. Within each group, the parameter node, state node, defect node and target node associated with the same position identifier are extracted and connected in the order of parameter node pointing to state node, state node pointing to defect node, and defect node pointing to target node to generate closure-bearing units that are associated with the closure mapping edge one by one. Extract the position identifiers of the closure mapping edges carried by each casting twin focal element from the set of casting twin focal elements one by one. Perform consistency matching between the extracted position identifiers and the position identifiers of the closure carrying units. Write the matching casting twin focal elements and closure carrying units into the same index entry to generate a focal element carrying index record. Based on the coke element bearing index record, the bearing range of the cast twin coke element within the closure bearing unit is determined. The range of the coke element union generated by the cast twin coke element in the Dubois-Prade conflict fusion rule is limited to the same closure bearing unit. The limited bearing range is written into the rule constraint field to form a closure bearing constraint. The twin residual conflict rollback front is aligned with the closure carrying unit according to the position identifier of the closure mapping edge. The residual rollback link, parameter node and parameter node conflict amount are extracted from the alignment result with consistent position identifiers. The residual rollback link, parameter node and parameter node conflict amount are written into the same rollback index entry to generate a conflict rollback index record. Based on the conflict rollback index record, the transfer path of conflict quality within the closure bearer unit is determined. The conflict quality transfer path in the Dubois-Prade conflict fusion rule is limited to the residual rollback link, and the endpoint of the conflict quality transfer is limited to the parameter node in the residual rollback link, thus forming a rollback transfer constraint. Based on the number of parameter node conflicts, the conflict quality corresponding to the same residual rewind link is weighted and configured. The number of parameter node conflicts is used as the basis for weight calculation, and the obtained allocation weight is written into the allocation field corresponding to the parameter node, so that the allocation weight is associated with the parameter node and a parameter adjustment focal element allocation rule is formed. The Dubois-Prade conflict fusion rule incorporates closure bearer units, focal element bearer index records, closure bearer constraints, conflict rollback index records, rollback transfer constraints, and parameter adjustment focal element allocation rules as rule components. This restricts the focal element union range to the closure bearer units and transfers conflict quality to parameter nodes along the residual rollback link, thus constructing an improved Dubois-Prade algorithm.

[0026] The improvements to the Dubois-Prade conflict fusion rule in this invention are mainly reflected in three aspects: the scope of coke element bearing capacity, the conflict transfer path, and parameter adjustment and allocation. First, the closure bearing structure is organized into closure bearing units according to the positional identifiers of the closure mapping edges. This ensures that the coke element union generated by the casting twin coke elements is no longer within the ordinary open evidence space, but is confined to the process link composed of parameter nodes, state nodes, defect nodes, and target nodes corresponding to the same closure mapping edge, thereby preventing the diffusion of conflict quality between irrelevant nodes. Second, the twin residual conflict rollback leading edge is aligned with the closure bearing unit. The conflict quality transfer path is limited to the residual rollback link through the conflict rollback index record, and the transfer endpoint is limited to the parameter nodes in the residual rollback link. This makes the original method, which was only used for uncertain... The qualitative expression of conflict quality can be traced back along the casting process to the parameter position where the deviation occurred. Furthermore, the allocation weight is configured according to the conflict amount of the parameter node, and the allocation weight is written into the allocation field corresponding to the parameter node. This allows the conflict quality to be allocated to the parameter adjustment focal element according to the residual amplitude and the degree of conflict concentration, forming a rule output oriented towards process parameter correction. The improved algorithm can unify multi-source evidence conflict, twin prediction bias, and parameter node adjustment relationship into the same closure bearing structure, improve the process interpretability of the conflict fusion result, enhance the traceability of the parameter correction direction, reduce the invalid uncertainty aggregation caused by the excessively wide focal element union in the traditional Dubois-Prade rule, and transform the optimization of casting process parameters from simple state judgment to executable closed-loop parameter update.

[0027] In this embodiment, step seven specifically includes: The set of casting twin focal elements is input into the improved Dubois-Prade algorithm. According to the position identifier of the closure mapping edge in the focal element carrying index record, the casting twin focal elements in the set of casting twin focal elements are matched item by item. Casting twin focal elements with the same position identifier are grouped into the same closure carrying unit, so that the casting twin focal elements and the closure carrying unit are established to correspond to each other. Within the same closure carrying unit, the intersection determination of the casting twin focal elements is performed. The evidence propositions corresponding to the casting twin focal elements participating in the determination are calculated for intersection. The casting twin focal element combination with empty intersection is determined as the conflict focal element combination. Then, the basic probability assignment carried by each casting twin focal element in the conflict focal element combination is read. The basic probability assignment values ​​are multiplied to generate the conflict quality. The conflict quality is limited in scope according to the closure bearing constraint. The conflict quality is written into the closure bearing unit where the conflict focal element combination is located, and the position identifier of the conflict quality and the closure bearing unit are kept consistent to form the closure constraint conflict quality. Match the closure constraint conflict quality with the position identifier of the closure mapping edge according to the conflict rollback index record, determine the residual rollback link associated with the closure constraint conflict quality, and then pass the closure constraint conflict quality along the residual rollback link according to the rollback transfer constraint, and pass the closure constraint conflict quality to the parameter node in the residual rollback link. According to the parameter adjustment focal element allocation rule, the allocation weight corresponding to the parameter node is extracted. The closure constraint conflict quality passed to the parameter node is multiplied with the allocation weight to calculate the allocation quality corresponding to the parameter node. The allocation quality is then written into the parameter adjustment focal element to generate the parameter adjustment focal element quality. The parameters are linked to the quality of the coke element and the parameter node. The position identifiers of the residual rewind link and the closure mapping edge are written into the same linked record. The linked records are arranged according to the casting process sequence to generate the parameter correction record.

[0028] In this embodiment, step eight specifically includes: Extract the position identifiers of parameter adjustment coke quality, parameter nodes, residual rewind links and closure mapping edges from the parameter correction records. Match the parameter nodes with the parameter nodes in the casting twin model, and locate the parameter values ​​corresponding to the parameter nodes based on the matching results to form the positioning results of the process parameter combination. The direction calibration and amplitude results are extracted from the residual rewind link. The direction calibration is matched with the adjustment direction of the parameter value to form the parameter correction direction. The amplitude result is then multiplied with the parameter-adjusted focal quality, and the multiplication result is used as the parameter correction amplitude corresponding to the parameter node. The direction of parameter increase or decrease is determined according to the parameter correction direction, the amount of change of parameter value is determined according to the parameter correction magnitude, and incremental updates are performed on the parameter values ​​in the process parameter combination to generate updated parameter values. Write the updated parameter values ​​into the parameter nodes, and organize the parameter nodes and updated parameter values ​​according to the position identifier of the closure mapping edge, so that the parameter nodes under the same closure mapping edge position identifier and the updated parameter values ​​maintain a mapping relationship, and generate the updated process parameter combination. The updated process parameter combinations, parameter correction records, and closure mapping edge position identifiers are linked and organized. The linked and organized records are arranged according to the casting process sequence, and the casting process parameter optimization results are output.

[0029] A casting process parameter optimization system based on digital twins, comprising: The process evidence module is used to obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; The twin modeling module is used to create a casting twin model based on the casting process path and form a closed-loop load-bearing structure. The focal element generation module is used to establish closure mapping edges and generate a set of casting twin focal elements carrying basic probability assignments. The residual calculation module is used to calculate the state residuals corresponding to the same closure mapping edge and generate twin residual records; The rollback front module is used to trace backward along the closure mapping edge to the parameter node and generate a twin residual conflict rollback front. The algorithm building module is used to write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to build an improved Dubois-Prade algorithm. The parameter correction module is used to process the set of casting twin coke elements using the improved Dubois-Prade algorithm and generate parameter correction records. The optimized output module is used to update the combination of process parameters in the casting twin model based on the parameter correction records, and output the optimization results of the casting process parameters.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to the optimization of process parameters for aluminum alloy shell castings. These castings are highly sensitive to pouring temperature, holding pressure, holding time, and cooling intensity during production. Discrepancies frequently occur between on-site sensor data and simulation prediction data, especially in continuous production batches. Heat accumulation in the mold can lead to inconsistencies between the predicted solidification state and the actual detection state, resulting in recurring shrinkage cavities, porosity, and dimensional deviations. Traditional methods typically involve process engineers manually reviewing pouring records after quality inspection and adjusting parameters based on simulation curves. However, due to the lack of evidence linking production records, prediction records, and quality results within the same process chain, parameter adjustments easily rely on experience, making it difficult to determine which type of process parameter the defect should be traced back to.

[0031] In this embodiment, the system collects the process execution records, process parameter records, and quality inspection records for each batch of castings, and simultaneously receives the prediction output records generated by the casting twin simulation process. The system first performs source matching and temporal alignment between the casting production records and the twin prediction records according to the casting batch identifier, process identifier, acquisition time, and prediction time, forming a process evidence set. Then, it establishes a casting twin model based on the casting process path, enabling parameter nodes, state nodes, defect nodes, and target nodes to form a closure-bearing structure along the process path. For each closure mapping edge extending from the parameter node to the target node, the system maps the process evidence set into the closure mapping edge, generating a casting twin focal element set carrying basic probability assignments. When inconsistencies arise between the measured state and the predicted state, the system calculates the state residual and forms a twin residual record, then points back along the closure mapping edge from the state node through the defect node to the parameter node, forming a twin residual conflict retracement front. The system then writes the closure bearing structure and twin residual conflict rewind front into the Dubois-Prade conflict fusion rule, so that the evidence conflict no longer stays at the level of defect risk judgment, but is allocated to the parameter adjustment focal element along the residual rewind link, and finally generates parameter correction records, which are used to update the process parameter combination in the casting twin model.

[0032] During the verification process, casting data from the same product model, mold structure, and similar raw material batches were compared. Traditional methods use manual experience combined with conventional twin simulation feedback for parameter adjustment. The present invention employs a closed-loop load-bearing structure and an improved Dubois-Prade algorithm for parameter correction. The verification indicators cover four aspects: data alignment, residual tracking, parameter correction, and quality results, as shown in Table 1. Table 1 Comparison of Casting Closed-Loop Optimization Effects

[0033] As shown in Table 1, this invention establishes a stable correspondence between production records and twin prediction records through process evidence sets, increasing the data alignment success rate from 86.4% to 97.8%, indicating a significant improvement in the misalignment problem of data from different sources in the process chain. The average state residual decreased from 0.182 to 0.071, indicating that the casting twin model's fit with the measured state improved after parameter updates. The proportion of conflict evidence that can be rolled back to parameter nodes increased from 42.6% to 91.3%, indicating that the improved Dubois-Prade algorithm can allocate conflict quality along the twin residual conflict rollback front to the parameter adjustment focal element, rather than just providing a defect risk judgment. The single-batch parameter convergence time and average parameter adjustment rounds were significantly reduced, indicating that the parameter correction process is more concentrated. In terms of quality results, the recurrence rate of shrinkage cavities and porosity and the rate of exceeding dimensional deviation limits decreased, and the first-pass yield increased to 96.2%, indicating that this invention can transform prediction deviations and detection anomalies into executable process parameter correction bases, improving the stability and traceability of casting process parameter optimization.

[0034] 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 casting process parameters based on digital twins, characterized in that, Includes the following steps: Step 1: Obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; Step 2: Establish a casting twin model based on the casting process path, connect parameter nodes to state nodes, connect state nodes to defect nodes, and connect defect nodes to target nodes to form a closed-loop support structure. Step 3: Establish a closure mapping edge from the parameter node to the target node in the closure bearing structure, and map the process evidence set into the closure mapping edge to generate a set of casting twin focal elements carrying basic probability assignments; Step 4: Compare the measured states in the process evidence set with the predicted states in the twin prediction records, calculate the state residuals corresponding to the same closure mapping edge, and generate twin residual records; Step 5: Based on the twin residual record, trace backward along the closure mapping edge to the parameter node to generate the twin residual conflict retrace front; Step 6: Write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to construct the improved Dubois-Prade algorithm; Step 7: Use the improved Dubois-Prade algorithm to process the set of casting twin coke elements, distribute the conflict mass along the twin residual conflict rollback front to the parameter-adjusting coke elements, and generate parameter correction records; Step 8: Update the process parameter combination in the casting twin model based on the parameter correction record, and output the casting process parameter optimization results.

2. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step one specifically involves: Collect process execution records, process parameter records, and quality inspection records generated during the casting production process. Extract casting batch identifiers, process identifiers, and collection times from the process execution records, extract parameter values ​​from the process parameter records, and extract measured states from the quality inspection records to generate casting production records. Read the prediction output record generated during the casting twin simulation process, extract the casting batch identifier, process identifier, prediction time, and prediction status from the prediction output record, and generate twin prediction record; The casting production records and twin prediction records are matched based on the casting batch identifier and process identifier. The matching results are aligned according to the acquisition time and prediction time to generate process time sequence association records. Arrange the process sequence records according to the casting process sequence, and write the parameter values, actual measured status and predicted status into the same evidence unit, and collect the evidence units to generate a process evidence set.

3. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step two specifically involves: Extract casting batch identifiers, process identifiers, and collection times from the process evidence set; arrange the process identifiers under the same casting batch identifier according to the collection time to generate casting process paths. A casting twin model is established based on the casting process path. The casting twin model is a directed mapping model that organizes the process evidence set according to the casting process path, and node generation rules and node connection rules are configured. According to the node generation rules, parameter values ​​are extracted from the process evidence set and parameter nodes are generated according to the process identifier. Measured and predicted states are extracted from the process evidence set and state nodes are generated according to the process identifier. Defect nodes are generated according to the process identifier corresponding to the state nodes. Target nodes are generated according to the casting batch identifier. According to the node connection rules, the parameter nodes are connected to the status nodes corresponding to the same process identifier, the status nodes are connected to the defect nodes corresponding to the same process identifier, and the defect nodes are connected to the target nodes corresponding to the same casting batch identifier, forming a closed-loop bearing structure.

4. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step three specifically involves: Perform direction calibration on the node connection relationships in the closure bearing structure, organize the connections from parameter nodes to state nodes into parameter state edges, organize the connections from state nodes to defect nodes into state defect edges, and organize the connections from defect nodes to target nodes into defect target edges. According to the casting process path, link splicing is performed on parameter state edge, state defect edge and defect target edge, and the spliced ​​link extending from parameter node to target node under the same casting batch identifier is encapsulated as closure mapping edge. The process evidence set is organized into a sequence of evidence units according to the casting process sequence. Field encoding is performed on the parameter values, measured states and predicted states in the evidence unit sequence. The field encoding results are matched with the execution positions of parameter nodes, state nodes and defect nodes in the closure mapping edge to generate a closure mapping record. A consistency determination is performed on the measured state and the predicted state in the closure mapping record. Evidence propositions are generated according to the consistency determination results. The evidence propositions, closure mapping records and position identifiers of closure mapping edges are combined to generate casting twin focal elements. The occurrence count of evidence propositions and the total number of units corresponding to evidence unit sequences within the same closure mapping edge are counted. The occurrence count and the total number of units are normalized to generate basic probability assignments. The basic probability assignments are written into the casting twin focal elements, and the casting twin focal elements are collected to generate a casting twin focal element set.

5. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step four specifically involves: The process evidence set is organized into an evidence unit sequence according to the casting batch identifier, process identifier, and casting process sequence. The evidence unit sequence is then matched with the position identifier of the closure mapping edge to form a residual calculation unit. Perform state field alignment on the measured state and predicted state in the residual calculation unit, and convert the field-aligned measured state into a measured state vector and the field-aligned predicted state into a predicted state vector. The measured state vector and the predicted state vector are used to calculate the corresponding position difference, generate the state difference vector, and perform amplitude adjustment and direction calibration on the state difference vector to form the state residual corresponding to the same closure mapping edge; Write the state residual, the position identifier of the closure mapping edge, the casting batch identifier, and the process identifier into the same residual entry, and arrange the residual entries according to the casting process sequence to generate twin residual records.

6. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step five specifically involves: Match the residual entries in the twin residual record with the closure mapping edges in the closure bearing structure according to the position identifier of the closure mapping edge, determine the position of the state node in the closure mapping edge, and generate a residual location record. Following the reverse connection order of the closure mapping edges, the residual location record is pointed back from the state node through the defect node to the parameter node, forming a residual rollback link; Write the status residual, casting batch identifier, and process identifier from the residual location record into the residual rollback link, and organize the residual rollback link according to the casting process sequence to generate parameter node conflict records. Based on the amplitude results and direction calibration in the parameter node conflict record, the residual rollback links associated with the same parameter node are subjected to direction consistency merging and amplitude accumulation processing to form the parameter node conflict quantity; The parameter node conflict quantity, residual rollback link, and parameter node are associated and organized according to the position identifier of the closure mapping edge to generate a twin residual conflict rollback front.

7. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step six specifically involves: The closure-bearing structure is grouped according to the position identifier of the closure mapping edge. The parameter nodes, state nodes, defect nodes and target nodes associated with the same position identifier are extracted, and the closure-bearing unit is generated according to the connection order from the parameter node to the target node. Extract the position identifiers of the closure mapping edges carried by the casting twin focal elements from the set of casting twin focal elements, perform matching between the casting twin focal elements and the closure carrying units corresponding to the same position identifier, and generate focal element carrying index records. Based on the focal element carrying index record, the range of focal element union in the Dubois-Prade conflict fusion rule is restricted to the same closure carrying unit, forming a closure carrying constraint; Align the twin residual conflict rollback front edge with the closure bearing unit according to the position identifier of the closure mapping edge, extract the residual rollback link, parameter node and parameter node conflict amount, and generate conflict rollback index record; Based on the conflict rollback index record, the conflict quality transfer path in the Dubois-Prade conflict fusion rule is limited to the residual rollback link, and the conflict quality transfer endpoint is limited to the parameter node in the residual rollback link, thus forming a rollback transfer constraint. Based on the number of parameter node conflicts, the conflict quality configuration of the same residual rewind link is assigned a weight, and the assigned weight is associated with the parameter node to form a parameter adjustment focal element allocation rule. The improved Dubois-Prade algorithm is constructed by incorporating the closure bearer unit, focal element bearer index record, closure bearer constraint, conflict rollback index record, rollback transfer constraint, and parameter adjustment focal element allocation rule into the Dubois-Prade conflict fusion rule.

8. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step seven specifically involves: The set of casting twin coke elements is input into the improved Dubois-Prade algorithm, and the closure bearing unit corresponding to each casting twin coke element is determined according to the coke element bearing index record. Within the same closure carrying unit, the intersection determination of the casting twin coke elements is performed. The casting twin coke element combination with an empty intersection is taken as the conflict coke element combination. The product calculation is performed on the basic probability assignment carried by the conflict coke element combination to generate conflict mass. According to the closure bearing constraint, the conflict mass is confined to the closure bearing unit corresponding to the conflict focal element combination, thus forming the closure constraint conflict mass; The residual rollback link corresponding to the closure constraint conflict quality is determined according to the conflict rollback index record, and the closure constraint conflict quality is transferred to the parameter node in the residual rollback link according to the rollback transfer constraint. Extract the allocation weights corresponding to the parameter nodes according to the parameter adjustment focal element allocation rules, write the closure constraint conflict quality passed to the parameter nodes into the parameter adjustment focal element according to the allocation weights, and generate the parameter adjustment focal element quality. The position identifiers of the parameter adjustment focal element quality, parameter nodes, residual rewind links, and closure mapping edges are associated and organized to generate parameter correction records.

9. The method for optimizing casting process parameters based on digital twins according to claim 1, characterized in that, Step eight specifically involves: Extract the position identifiers of parameter-adjusted coke quality, parameter nodes, residual rewind links, and closure mapping edges from the parameter correction records, and locate the process parameter combination in the casting twin model according to the parameter nodes; The direction calibration and amplitude results are extracted from the residual rewind link. The direction calibration is converted into the parameter correction direction. The amplitude results are multiplied and converted with the parameter-adjusted focal element quality to generate the parameter correction amplitude. Incremental updates are performed on the parameter values ​​in the process parameter combination according to the parameter correction direction and parameter correction magnitude to generate updated parameter values; Write the updated parameter values ​​into the parameter nodes, and organize the mapping relationship between the parameter nodes and the updated parameter values ​​according to the position identifier of the closure mapping edge to generate the updated process parameter combination; The updated process parameter combinations, parameter correction records, and closure mapping edge position identifiers are linked and organized to output the casting process parameter optimization results.

10. A casting process parameter optimization system based on digital twins, comprising executing the casting process parameter optimization method based on digital twins as described in any one of claims 1 to 9, characterized in that, include: The process evidence module is used to obtain casting production records and twin prediction records, and generate a process evidence set by associating them according to the casting process time sequence; The twin modeling module is used to create a casting twin model based on the casting process path and form a closed-loop load-bearing structure. The focal element generation module is used to establish closure mapping edges and generate a set of casting twin focal elements carrying basic probability assignments. The residual calculation module is used to calculate the state residuals corresponding to the same closure mapping edge and generate twin residual records; The rollback front module is used to trace backward along the closure mapping edge to the parameter node and generate a twin residual conflict rollback front. The algorithm building module is used to write the conflicting rollback front of the closure bearing structure and the twin residual into the Dubois-Prade conflict fusion rule to build an improved Dubois-Prade algorithm. The parameter correction module is used to process the set of casting twin coke elements using the improved Dubois-Prade algorithm and generate parameter correction records. The optimized output module is used to update the combination of process parameters in the casting twin model based on the parameter correction records, and output the optimization results of the casting process parameters.

Images