Injection molding process knowledge graph-based abnormal working condition diagnosis method and device

Abstract

The application provides an abnormal working condition diagnosis method and device based on an injection molding process knowledge graph, which comprises the following steps: deviation comparison is performed based on a shot pressure curve feature point and a standard process window boundary to obtain a process deviation feature vector; semantic matching is performed based on the process deviation feature vector and a process state entity node to obtain a potential abnormal process event; reverse tracing is performed based on the potential abnormal process event and a cause-effect correlation edge to obtain a target abnormal source node; connectivity constraint verification is performed based on the target abnormal source node and a physical topology to obtain a target propagation link; spatial coincidence degree matching is performed based on the target propagation link and the spatial distribution position of a weld mark defect in a mold cavity to obtain a target diagnosis path; and mapping matching is performed based on the target abnormal source node attribute in the target diagnosis path and a disposal strategy of a historical maintenance knowledge base to obtain a diagnosis conclusion of the weld mark abnormal working condition. The application improves the accuracy of root positioning of the weld mark abnormal working condition.

Description

Technical Field

[0001] This invention relates to the field of injection molding quality control technology, and in particular to an abnormal operating condition diagnosis method and device based on injection molding process knowledge graph. Background Technology

[0002] In the injection molding process, weld lines are a critical defect affecting the appearance and strength of the product. Existing weld line diagnosis methods mainly include: prediction methods based on offline finite element simulation (CAE), which predict the location of weld lines by simulating melt flow; and monitoring methods based on statistical process control (SPC), which issue alarms by monitoring whether the real-time values ​​of key parameters such as injection pressure and holding pressure exceed preset upper and lower limits.

[0003] However, CAE-based simulation methods are computationally intensive and time-consuming, applicable only to the mold design stage. They cannot handle real-time operational drift caused by raw material batch fluctuations, machine hydraulic system aging, or environmental temperature changes during production, leading to a disconnect between the simulation model and actual production conditions. While SPC-based monitoring methods can provide real-time alarms, they are essentially "single-point threshold judgments," ignoring the strong coupling between injection molding process parameters and the spatiotemporal propagation characteristics of defects. For example, when weld line strength is insufficient, SPC may only show normal holding pressure, failing to identify that insufficient shear heat during the filling stage leads to a low melt front temperature, preventing the holding pressure from effectively reaching the weld line area. The lack of analytical capabilities to understand the propagation path of abnormal signals from their source to the defect point makes it difficult for technicians to distinguish between improper machine parameter settings, mold runner blockage, and material property fluctuations, often resulting in blind trial and error based on experience. Therefore, when facing weld line anomalies, the lack of logical deduction capabilities regarding the coupling relationships of process parameters and the propagation path of abnormal signals in the injection molding system leads to low accuracy in locating the root cause of weld line anomalies. Summary of the Invention

[0004] This invention provides a method and apparatus for diagnosing abnormal working conditions based on injection molding process knowledge graph, aiming to improve the accuracy of locating the root cause of abnormal working conditions such as weld lines.

[0005] In a first aspect, the present invention provides an abnormal operating condition diagnosis method based on an injection molding process knowledge graph, including: Based on the deviation comparison between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary, a process deviation feature vector is obtained. Then, based on the semantic matching of the process deviation feature vector and the process state entity nodes in the injection process knowledge graph, potential abnormal process events are obtained. Reverse tracing is performed based on the causal relationship edges between the potential abnormal process events and the injection molding process knowledge graph to obtain the target abnormal source node. Then, based on the physical topology of the target abnormal source node and the injection mold runner system, connectivity constraint verification is performed to obtain the target propagation link. Based on the spatial overlap between the target propagation path and the spatial distribution of weld line defects in the mold cavity, the target diagnostic path is obtained. Based on the mapping and matching of the target anomaly source node attributes in the target diagnostic path with the handling strategies of the historical maintenance knowledge base, a diagnostic conclusion for the abnormal working condition of the weld line is obtained.

[0006] Secondly, the present invention also provides an abnormal operating condition diagnosis device based on an injection molding process knowledge graph, for implementing the abnormal operating condition diagnosis method based on an injection molding process knowledge graph as described in the first aspect; the abnormal operating condition diagnosis device based on an injection molding process knowledge graph includes: The abnormal process event analysis module is used to compare the deviation between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary to obtain the process deviation feature vector, and to perform semantic matching between the process deviation feature vector and the process state entity nodes in the injection process knowledge graph to obtain potential abnormal process events. The anomaly propagation analysis module is used to perform reverse tracing based on the causal relationship edges between the potential abnormal process events and the injection molding process knowledge graph to obtain the target anomaly source node, and to perform connectivity constraint verification based on the physical topology of the target anomaly source node and the injection mold runner system to obtain the target propagation link. The diagnostic path localization module is used to perform spatial overlap matching between the target propagation link and the weld line defect in the mold cavity to obtain the target diagnostic path. The abnormal operating condition diagnosis module is used to map and match the attributes of the target abnormal source node in the target diagnosis path with the handling strategies of the historical maintenance knowledge base to obtain a diagnostic conclusion on the abnormal operating condition of the weld line.

[0007] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for reading and executing the computer program, thereby realizing the abnormal working condition diagnosis method based on the injection molding process knowledge graph as described above.

[0008] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the abnormal operating condition diagnosis method based on the injection molding process knowledge graph as described above.

[0009] Fifthly, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned abnormal working condition diagnosis method based on injection molding process knowledge graph.

[0010] The abnormal operating condition diagnosis method based on injection molding process knowledge graph provided in this invention extracts process deviation feature vectors by comparing the deviations of feature points on the injection pressure curve with the standard process window boundary. Combined with the injection molding process knowledge graph, semantic matching is performed to accurately capture real-time process deviations caused by raw material batch fluctuations, equipment aging, and environmental temperature changes during production. This overcomes the limitations of traditional CAE simulation methods, which are only applicable to the mold design stage, cannot adapt to dynamic drift in production conditions, and are disconnected from actual production. It also eliminates the one-sidedness of traditional SPC methods that rely on single-parameter threshold judgments, achieving comprehensive feature extraction of multi-dimensional process deviations. By relying on the causal relationship edges of the knowledge graph, potential abnormal process events are initially screened for target abnormal source nodes through reverse tracing. Then, by combining the physical topology of the mold runner system to complete connectivity constraint verification, the true and effective abnormal propagation links are locked. This overcomes the shortcomings of traditional technologies that ignore strong coupling relationships of process parameters and cannot identify abnormal signal propagation paths, solving the problem of being unable to identify hidden weld lines caused by abnormal coupling even when a single parameter is normal. By matching the overlap between the target propagation path and the spatial distribution of weld line defects inside the mold cavity, the target diagnostic path directly corresponding to the defect is selected, and irrelevant abnormal interference paths are eliminated, ensuring the targetedness and effectiveness of anomaly tracing. The attributes of the abnormal source nodes in the diagnostic path are mapped and matched with the historical maintenance knowledge base to complete the handling strategy, outputting accurate weld line anomaly diagnostic conclusions. Therefore, it can completely reproduce the entire propagation process of abnormal signals from the source to the weld line defect point in the injection molding system, effectively distinguishing different types of anomaly root causes such as machine parameter settings, mold runner status, and material properties, avoiding the drawbacks of blind trial and error, and improving the accuracy of locating the root cause of weld line anomaly conditions. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the abnormal operating condition diagnosis method based on injection molding process knowledge graph provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the abnormal working condition diagnosis device based on injection molding process knowledge graph provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] Optionally, see Figure 1 , Figure 1 This is a flowchart illustrating the abnormal operating condition diagnosis method based on injection molding process knowledge graph provided by the present invention. In this embodiment of the invention, the execution subject of the abnormal operating condition diagnosis method based on injection molding process knowledge graph is the abnormal operating condition diagnosis device.

[0016] Abnormal operating condition diagnosis methods based on injection molding process knowledge graphs include: Step 10: Based on the deviation comparison between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary, the process deviation feature vector is obtained. Then, based on the semantic matching of the process deviation feature vector and the process state entity nodes in the injection process knowledge graph, potential abnormal process events are obtained.

[0017] Optionally, an injection molding process knowledge graph refers to a structured knowledge base organized in a graph structure, describing various entities and their interrelationships in the injection molding process. An injection molding process knowledge graph consists of three parts: entity nodes, causal relationship edges, and attribute fields. Entity nodes are the basic building blocks of the injection molding process knowledge graph. Each entity node corresponds to a physical object or process state in the injection molding system. Physical objects include injection system execution units and mold runner components. Injection system execution units include nozzle heating coils, injection cylinder speed control valves, and proportional flow valves; mold runner components include nozzles, main runners, branch runners, and gates. Process states include low melt front temperature and slow injection speed. Each entity node carries multiple attribute fields to describe the specific characteristics of the entity. Attribute fields include equipment instance status fields, control loop affiliation fields, and physical component codes. Causal relationship edges are directed edges connecting two entity nodes, representing the causal transmission relationship from one entity node to another. Each causal edge is labeled with a physical domain coupling attribute, a temporal constraint interval, and an action direction attribute. The physical domain coupling attribute characterizes the physical domain in which the coupling occurs between the starting and ending nodes connected by the causal edge. The temporal constraint interval characterizes the time interval from the initial impact of the causal effect represented by the causal edge to the manifestation of the result at the ending node. The action direction attribute characterizes the direction of transmission between cause and effect on the causal edge. Attribute fields are structured data attached to entity nodes or causal edges, used to record the specific parameters and operational status information of the entity or edge. The injection molding process knowledge graph integrates knowledge from multiple physical domains, such as equipment, processes, and molds, into a reasonable graph structure through the association of entity nodes, causal edges, and attribute fields. This structure is used by the abnormal condition diagnosis device for semantic matching, reverse tracing, and path analysis during abnormal diagnosis.

[0018] Optionally, the injection pressure curve for the current injection cycle can be obtained. The injection pressure curve is a continuous curve showing the change in injection pressure over time, collected in real-time by the injection system of the injection molding machine within a complete injection cycle. Characteristic points on the injection pressure curve refer to discrete points on the curve that have process interpretation significance, including the injection start point, injection peak point, and injection holding pressure transition point. The standard process window boundaries refer to the upper and lower limits of normal values ​​corresponding to each characteristic point of the injection pressure curve, determined through prior process verification.

[0019] The actual values ​​of each injection pressure curve feature point within the current cycle are compared with the upper and lower limits of the corresponding standard process window boundary to obtain the deviation direction and degree of deviation of each feature point relative to the boundary. The deviation directions and degrees of deviation of all feature points are then uniformly arranged and combined into a process deviation feature vector. Subsequently, the deviation descriptions contained in the process deviation feature vector are semantically matched with the process state entity nodes stored in the injection molding process knowledge graph.

[0020] Process state entity nodes refer to entity nodes in the injection molding process knowledge graph used to represent various process states during injection molding. Each process state entity node carries a label representing the semantics of that process state. Semantic matching refers to comparing the similarity between the deviation description and the semantic labels of each process state entity node, and selecting the process state represented by the process state entity node with the highest similarity as the potential abnormal process event.

[0021] In one embodiment, the product manufactured is an automotive door inner panel, made of acrylonitrile butadiene styrene copolymer. The mold is a two-cavity mold, with each cavity fed by two side gates. After collecting the injection pressure curve for this cycle, two feature points of the injection pressure curve are extracted: the injection peak point and the injection holding pressure transition point. Assuming the standard process window corresponding to the injection peak point has an upper limit of 90 MPa and a lower limit of 80 MPa, the actual injection peak point for this cycle is 72 MPa, which is 8 MPa below the lower limit. The standard process window corresponding to the injection holding pressure transition point has an upper limit of 60 MPa and a lower limit of 50 MPa, and the actual injection holding pressure transition pressure for this cycle is 46 MPa, which is 4 MPa below the lower limit. Based on this, the process deviation feature vector is obtained as low injection peak point and low injection holding pressure transition pressure. This process deviation feature vector is semantically matched with the process state entity nodes in the injection molding process knowledge graph. A process state entity node indicating low melt front temperature is matched, which is considered a potential abnormal process event.

[0022] Step 20: Perform reverse tracing based on the causal relationship edges between potential abnormal process events and the injection molding process knowledge graph to obtain the target abnormal source node, and perform connectivity constraint verification based on the physical topology of the target abnormal source node and the injection mold runner system to obtain the target propagation link.

[0023] Optionally, starting with a potential abnormal process event, reverse tracing is performed along the causal relationship edges in the injection molding process knowledge graph. A causal relationship edge refers to a directed edge in the injection molding process knowledge graph that connects two entity nodes and represents the causal transmission relationship between them. Reverse tracing refers to starting from the potential abnormal process event as a result, tracing back layer by layer along the causal relationship edges to its cause, and finally locating the target abnormal source node corresponding to the injection molding system execution unit that caused the abnormality, as described in steps 201 to 204.

[0024] The target anomaly source node is mapped onto the physical topology of the injection mold runner system, and connectivity constraint verification is performed. The physical topology refers to a directed graph model established according to the actual geometric connections of the injection mold runner system, recording the physical connectivity relationships of each runner component from the injection nozzle to the cavity. Connectivity constraint verification verifies whether a continuous physical path conforming to the actual melt flow direction exists from the target anomaly source node to the cavity. After connectivity constraint verification, a target propagation link conforming to the physical path of melt flow is obtained, as detailed in steps 205 to 208.

[0025] Step 30: Based on the spatial overlap of the target propagation link and the weld line defect in the mold cavity, the target diagnosis path is obtained.

[0026] Optionally, the cavity area covered by the target propagation link is spatially matched with the actual spatial distribution location of the weld line defect in the mold cavity. The spatial distribution location of the weld line defect in the mold cavity refers to the actual location of the weld line defect detected by the injection molded product surface defect detection system in the cavity coordinate system during this cycle. Spatial overlap matching refers to geometrically comparing the location of the target propagation link that theoretically may induce weld lines in the cavity with the actual detected weld line defect location to determine the degree of spatial overlap between the two. After matching, the portion of the propagation link that overlaps with the actual weld line defect location is retained as the target diagnostic path associated with the weld line formation area, as described in steps 301 to 304.

[0027] Step 40: Based on the mapping and matching of the target anomaly source node attributes in the target diagnostic path with the handling strategies of the historical maintenance knowledge base, a diagnostic conclusion is obtained.

[0028] Optionally, attribute information of the target anomaly source node is extracted from the target diagnostic path, and the attribute information is mapped and matched with the handling strategies in the historical maintenance knowledge base.

[0029] The historical maintenance knowledge base is a pre-accumulated and maintained database that stores the fault phenomena, cause analyses, and proven handling strategies for each anomaly source node in history. The handling strategy refers to the recommended process adjustment or equipment maintenance measures for a specific anomaly source node.

[0030] Mapping matching refers to comparing the similarity between the attributes of the current target anomaly source node and the anomaly source attributes of each fault entry in the historical maintenance knowledge base, selecting the historical fault entry with the highest similarity, and retrieving its corresponding handling strategy as the diagnostic conclusion for the current weld line anomaly condition.

[0031] In one embodiment, if the target anomaly source is a low current output from the proportional flow valve, a historical fault entry matching this attribute is retrieved from the historical maintenance knowledge base. Assume the retrieved historical fault entry states: When the actual output current of the proportional flow valve is lower than the set current, the injection speed will be slower, the shear force on the melt in the flow channel will be weakened, the shear heat will be insufficient, the melt front temperature will drop, and the two melt fronts will not fuse properly when they meet at the centerline of the cavity due to the low temperature, resulting in a weak weld line at the weld line location. The corresponding handling strategy is to check the electrical connections and coil condition of the proportional flow valve, verify the deviation between the actual output current and the set current, and clean or replace the proportional flow valve if necessary. This handling strategy is output as the diagnostic conclusion for this weld line anomaly.

[0032] This invention extracts process deviation feature vectors by comparing the deviations of feature points on the injection pressure curve with the standard process window boundary. Combined with a knowledge graph of injection molding processes, semantic matching is performed to accurately capture real-time process deviations caused by raw material batch fluctuations, equipment aging, and changes in ambient temperature during production. This overcomes the limitations of traditional CAE simulation methods, which are only applicable to the mold design stage, cannot adapt to dynamic drift in production conditions, and are disconnected from actual production. It also eliminates the one-sidedness of traditional SPC methods that rely on single-parameter threshold judgments, achieving comprehensive feature extraction of multi-dimensional process deviations. By relying on the causal relationships of the knowledge graph, potential abnormal process events are traced back to initially screen out target abnormal source nodes. Then, by combining the physical topology of the mold runner system, connectivity constraint verification is performed to lock in the real and effective abnormal propagation links. This overcomes the shortcomings of traditional technologies that ignore strong coupling relationships of process parameters and cannot identify abnormal signal propagation paths, solving the problem of being unable to identify hidden weld lines caused by abnormal coupling even when a single parameter is normal. By matching the overlap between the target propagation path and the spatial distribution of weld line defects inside the mold cavity, the target diagnostic path directly corresponding to the defect is selected, and irrelevant abnormal interference paths are eliminated, ensuring the targetedness and effectiveness of anomaly tracing. The attributes of the abnormal source nodes in the diagnostic path are mapped and matched with the historical maintenance knowledge base to complete the handling strategy, outputting accurate weld line anomaly diagnostic conclusions. Therefore, it can completely reproduce the entire propagation process of abnormal signals from the source to the weld line defect point in the injection molding system, effectively distinguishing different types of anomaly root causes such as machine parameter settings, mold runner status, and material properties, avoiding the drawbacks of blind trial and error, and improving the accuracy of locating the root cause of weld line anomaly conditions.

[0033] Optionally, the processes of steps 201 to 204 include: Step 201: Based on the compatibility matching between the physical effect type of potential abnormal process events and the physical domain coupling attribute of the causal association edge in the injection molding process knowledge graph, extract the multi-physical domain coupling edge that allows the physical effect type as the input end of the physical domain coupling attribute to obtain the first causal candidate edge.

[0034] Optionally, the physical effect type corresponding to the potential abnormal process event can be determined. The physical effect type refers to the category of effect in terms of the physical mechanism to which the potential abnormal process event belongs, such as thermal effect, flow effect, pressure effect, shear effect, etc. Each causal relationship edge in the injection molding process knowledge graph is labeled with a physical domain coupling attribute. The physical domain coupling attribute is used to characterize the physical domain in which the coupling occurs between the starting node and the ending node connected by the causal relationship edge, such as shear domain coupling, thermal domain coupling, pressure domain coupling, etc.

[0035] The physical effect type of potential abnormal process events is matched with the physical domain coupling attribute of each causal edge one by one for compatibility. Only those causal edges whose physical domain coupling attribute allows the physical effect type to be used as the input are retained, thus obtaining the first causal candidate edge.

[0036] In this context, "allowing" refers to the direction of coupling action represented by the causal relationship edge being able to receive the physical effect type as the result of the action.

[0037] In one embodiment, the aforementioned potential abnormal process event is a low melt tip temperature, and its physical effect type is thermal effect. The causal relationships in the injection molding process knowledge graph are traversed, and causal relationships whose physical domain coupling attributes allow thermal effects as input are selected. Assume that after matching, three candidate edges are initially selected: the first connects a low nozzle temperature setting with a low melt temperature, and its physical domain coupling attribute is thermal coupling. The second connects a slow injection speed with insufficient melt shear heat, and its physical domain coupling attribute is shear coupling. The third connects a high mold cooling water temperature with a high melt surface temperature, and its physical domain coupling attribute is thermal coupling. The input corresponding to the thermal coupling attribute of the third edge is a high melt surface temperature, which is opposite to the direction of the current thermal effect of a low melt tip temperature; therefore, this physical effect type is not allowed as an input and is discarded. The first and second causal relationships are retained as the first causal candidate edges.

[0038] Step 202: Based on the device instance status field of the starting node connected by each first causal candidate edge, perform an existence comparison with the device activation mapping table of the current injection molding production cycle, retain the causal association edges that are activated and occupied in the current cycle as indicated by the device instance status field, and obtain the second causal candidate edges.

[0039] Optionally, for each first causal candidate edge, the device instance status field recorded on the starting node connected to that causal relationship edge is read. The device instance status field refers to the actual operating status record of the injection molding system execution unit represented by the starting node in the current injection molding production cycle, including states such as activated and occupied, activated but not occupied, and inactive. The device activation mapping table is a pre-established lookup table that records the actual activation and occupancy status of each injection molding system execution unit in the current injection molding production cycle.

[0040] The device instance status field of the starting node of each first causal candidate edge is compared with the device activation mapping table for existence. Only causal edges that are indeed activated and occupied by the injection molding system execution unit corresponding to the starting node in the current injection molding production cycle are retained, while causal edges that are not activated or occupied by the execution unit corresponding to the starting node are removed. The second causal candidate edge is obtained after filtering.

[0041] In one embodiment, for the aforementioned two first causal candidate edges, the device instance status field of their starting nodes is read. The starting node of the first causal edge is "nozzle temperature setting too low," and the corresponding injection molding system execution unit is the nozzle heating coil. Its device instance status field records that it is activated and occupied in the current cycle. The starting node of the second causal edge is "injection speed too slow," and the corresponding injection molding system execution unit is the injection cylinder speed control valve. Its device instance status field records that it is activated and occupied in the current cycle. A query of the device activation mapping table for the current injection molding production cycle shows that both the nozzle heating coil and the injection cylinder speed control valve are activated and occupied in this cycle. After existence comparison of both causal edges, the two second causal candidate edges are still these two.

[0042] Step 203: Based on the temporal constraint interval corresponding to each second causal candidate edge and the occurrence time of the potential abnormal process event, perform temporal inclusion judgment, retain the causal association edges whose occurrence time is within the temporal constraint interval, and obtain the third causal candidate edge.

[0043] Optionally, for each second causal candidate edge, the time-series constraint interval marked on the causal relationship edge is read. The time-series constraint interval refers to the time interval corresponding to the causal effect represented by the causal relationship edge from the starting node to the ending node, expressed as the start and end times relative to the start time of the injection molding cycle. The occurrence time of the potential abnormal process event is compared with the time-series constraint interval of each second causal candidate edge for time-series inclusion judgment. Only causal relationship edges whose occurrence time is within their time-series constraint interval are retained, and causal relationship edges whose occurrence time falls outside the time-series constraint interval are eliminated to obtain the third causal candidate edge.

[0044] In one embodiment, the aforementioned potential abnormal process event is a low melt tip temperature, which occurs 1.2 seconds after the start of the current injection cycle. For the two second causal candidate edges: the first causal correlation edge, where the nozzle temperature setting is too low to result in a low melt temperature, has a timing constraint interval from 0.1 seconds to 2 seconds relative to the start of the injection cycle. The second causal correlation edge, where the injection speed is too slow to result in insufficient melt shear heat, has a timing constraint interval from 0.5 seconds to 1.5 seconds relative to the start of the injection cycle.

[0045] The potential abnormal process event occurs at 1.2 seconds, which falls within both the first and second temporal constraint intervals. Therefore, both causal edges pass the temporal inclusion judgment, and the resulting third causal candidate edges are still these two.

[0046] Step 204: Based on the control loop attribution field of the starting node connected by the third causal candidate edge, perform loop consistency matching with the main control loop identifier of the potential abnormal process event to obtain the target abnormal source node.

[0047] Optionally, for each third causal candidate edge, the control loop attribution field recorded on the starting node connected to the causal relationship edge is read. The control loop attribution field refers to the control loop identifier to which the injection system execution unit represented by the starting node belongs in the injection control system, such as temperature control loop, injection speed control loop, holding pressure control loop, etc. The main control loop identifier of the potential abnormal process event refers to the main control loop identifier involved in causing the potential abnormal process event. The control loop attribution field of the starting node of each third causal candidate edge is matched with the main control loop identifier of the potential abnormal process event for loop consistency matching. Causal relationship edges with the same control loop attribution field and main control loop identifier or with a preset coupling relationship are retained, and mismatched causal relationship edges are eliminated. The starting node connected to the causal relationship edge that passes the matching is the target abnormal source node, as described in steps 2041 to 2044.

[0048] This invention employs a four-layer progressive screening process—physical domain compatibility matching, device instance state existence comparison, timing constraint interval inclusion judgment, and control loop consistency matching—to gradually narrow the vast number of causal relationships in the injection molding process knowledge graph. The first layer eliminates causal edges with mismatched coupling domains based on physical mechanisms. The second layer eliminates causal edges corresponding to devices not activated in the current cycle. The third layer eliminates causal edges whose occurrence time is outside the causal action time window. The fourth layer eliminates causal edges unrelated to the main control loop. While ensuring the physical rationality of causal relationships, this approach traces potential abnormal process events back to the target abnormal source node that is actually active in the current cycle and consistent with the main control loop. This effectively eliminates interference candidates such as those with mismatched physical mechanisms, non-operating devices, inconsistent time windows, and unrelated control loops, improving the accuracy of abnormal source node location.

[0049] Optionally, the processes of steps 2041 to 2044 include: Step 2041: Based on the control loop attribution field of the starting node connected by the third causal candidate edge, perform loop consistency matching with the main control loop identifier of the potential abnormal process event, retain the causal association edges that have the same control loop attribution field and main control loop identifier or have a preset coupling relationship, and obtain the fourth causal candidate edge.

[0050] Optionally, for each third causal candidate edge, the control loop attribution field of its starting node is read, and this field is matched with the main control loop identifier of the potential abnormal process event for loop consistency matching. A preset coupling relationship refers to a signal linkage or power sharing relationship between two control loops in the injection molding control system. For example, when a proportional flow valve is simultaneously regulated by an injection speed control loop and a back pressure control loop, a preset coupling relationship exists between them. The matching rule in this embodiment is: if the control loop attribution field is exactly the same as the main control loop identifier, or if there is a preset coupling relationship between the control loop corresponding to the control loop attribution field and the control loop corresponding to the main control loop identifier, then the causal edge is retained; otherwise, it is discarded. The retained causal edges are the fourth causal candidate edges.

[0051] Continuing with the above embodiment, among the two third causal candidate edges, the control loop whose nozzle temperature setting is too low leads to a low melt temperature belongs to the temperature control loop, and the control loop whose injection speed is too slow leads to insufficient melt shear heat belongs to the injection speed control loop. The main control loop for potential abnormal process events is identified as the injection speed control loop. Through loop consistency matching, the temperature control loop and the injection speed control loop are neither identical nor have a pre-defined coupling relationship, so they are eliminated. The injection speed control loop has the same identifier as the main control loop and is retained. Therefore, the fourth causal candidate edge obtained is: injection speed too slow leads to insufficient melt shear heat.

[0052] Step 2042: Based on the direction attribute of each fourth causal candidate edge and the reverse transmission rule corresponding to the signal flow direction of the injection control system, perform direction consistency verification, retain the causal correlation edges whose direction attribute conforms to the direction of the result effect pointing to the cause excitation, and obtain the fifth causal candidate edge.

[0053] Optionally, for each fourth causal candidate edge, the action direction attribute marked on the causal association edge is read. The action direction attribute is used to characterize the transmission direction between the cause and the effect on the causal association edge, that is, from the cause end to the effect end. In this embodiment of the invention, the reverse transmission rule corresponding to the signal flow of the injection molding control system means that in the reverse tracing process, the causal association edge used should show a direction from the effect end to the cause excitation end, which is opposite to the direction of the signal flow from the controller to the actuator during normal production. The action direction attribute of each fourth causal candidate edge is checked for consistency with the reverse transmission rule, and only causal association edges whose action direction attribute conforms to the direction from the effect end to the cause excitation are retained, while causal association edges whose direction does not conform are eliminated. After the check, the fifth causal candidate edge is obtained.

[0054] In one embodiment, the directional attribute of the aforementioned fourth causal candidate edge, "slow injection speed leading to insufficient melt shear heat," is from "slow injection speed" to "insufficient melt shear heat." In a reverse tracing context, the potential abnormal process event of low melt front temperature is a result effect, requiring tracing back to the cause, "slow injection speed." This causal relationship edge points from cause to result in the forward physical direction, and from result to cause in the reverse direction, conforming to the reverse propagation rule. Therefore, this causal relationship edge passes the direction consistency check, and the resulting fifth causal candidate edge is this one.

[0055] Step 2043: Construct a reverse causal tree with potential abnormal process events as the endpoint based on each fifth causal candidate edge. Traverse upwards along the reverse causal tree until a boundary node with independent self-checking function or independent control interface is encountered. The boundary node at which the traversal stops is taken as the candidate abnormal source node.

[0056] Optionally, a reverse causal tree is constructed based on each fifth causal candidate edge, with the potential abnormal process event as the endpoint. The reverse causal tree is a tree structure that expands upwards layer by layer along the causal association edges, with the potential abnormal process event as the root node. Specifically, starting from the potential abnormal process event, the process traverses backwards along the fifth causal candidate edge to its starting node. Then, using this starting node as the new endpoint, other causal association edges terminating at this starting node are searched in the injection molding process knowledge graph, and the process continues to expand upwards in reverse. This upward traversal is continued layer by layer until a boundary node with an independent self-checking function or an independent control interface is encountered. An independent self-checking function means that the device unit represented by this node has self-detection and self-diagnosis capabilities, and can determine its own normal operation without relying on other nodes. An independent control interface means that the device unit represented by this node has an independent control input port, which can directly receive control commands issued by the injection molding controller. The boundary node where the traversal stops is designated as the candidate abnormality source node.

[0057] In one embodiment, based on the aforementioned fifth causal candidate edge "slow injection speed leading to insufficient melt shear heat," a reverse causal tree is constructed with the potential abnormal process event "low melt front temperature" as the endpoint. The first layer of reverse traversal leads to "slow injection speed." Continuing with "slow injection speed" as the new endpoint, the causal relationship edge terminating at this endpoint is searched in the injection molding process knowledge graph, finding that "insufficient opening of the injection cylinder speed control valve" leads to "slow injection speed." The reverse traversal continues until "insufficient opening of the injection cylinder speed control valve." The search continues again, finding that "low current output of the proportional flow valve" leads to "insufficient opening of the injection cylinder speed control valve." The proportional flow valve has an independent control interface and can directly receive the analog current signal output by the injection molding controller; it is a boundary node, so the traversal stops. Therefore, the candidate abnormal source node is "low current output of the proportional flow valve."

[0058] Step 2044: Based on the direct causal incoming edges of potential abnormal process events in the injection molding process knowledge graph, perform causal connectivity verification on each candidate abnormal source node, retain the candidate abnormal source nodes that can be directly traced to the potential abnormal process events through the fifth causal candidate edge, and remove the candidate abnormal source nodes that have not established a causal connection path to obtain the target abnormal source node.

[0059] Optionally, for each candidate anomaly source node, the direct causal incoming edges of potential abnormal process events are searched in the injection molding process knowledge graph. Direct causal incoming edges refer to causal association edges that directly point to the node containing the potential abnormal process event. It is verified whether the candidate anomaly source node can be directly traced back to the potential abnormal process event through the reverse causal tree formed by the fifth causal candidate edge, i.e., whether a continuous causal connection path exists between them formed by the fifth causal candidate edge. Candidate anomaly source nodes that can be directly traced back to the potential abnormal process event through the fifth causal candidate edge are retained, while candidate anomaly source nodes without established causal connection paths are eliminated, ultimately yielding the target anomaly source node.

[0060] In one embodiment, the aforementioned candidate anomaly source node is a low proportional flow valve current output. The direct causal input edge for the potential abnormal process event of low melt tip temperature is searched in the injection molding process knowledge graph. This direct causal input edge is confirmed to be from insufficient melt shear heat to low melt tip temperature. The causal path is verified along the reverse causal tree formed by the fifth candidate causal edge: low proportional flow valve current output points to insufficient injection cylinder speed control valve opening; insufficient injection cylinder speed control valve opening points to slow injection speed; slow injection speed points to insufficient melt shear heat; insufficient melt shear heat points to low melt tip temperature, forming a continuous causal connection path. The candidate anomaly source node passes the causal connectivity verification and is thus identified as the target anomaly source node.

[0061] In this embodiment of the invention, on the fourth causal candidate edge obtained through loop consistency screening, direction consistency verification ensures that the direction of the causal relationship edge used is consistent with the reverse tracing logic, avoiding tracing to the wrong node due to misuse of edge direction. By constructing a reverse causal tree with the potential abnormal process event as the endpoint, the tracing process extends upwards layer by layer along the causal relationship until it encounters a boundary node with independent self-testing function or independent control interface, so that the tracing stops at the equipment unit that can be independently detected and maintained. Causal connectivity verification eliminates pseudo-candidate nodes that appear in the tree structure but have not established a continuous causal path with the potential abnormal process event. The final target abnormal source node is both the physical root cause of the abnormality and a specific execution unit with independent detection and maintenance conditions, improving the accuracy of locating the root cause of weld line abnormal conditions.

[0062] Optionally, steps 205 to 208 include: Step 205: Based on the mapping relationship between the physical component code corresponding to the target anomaly source node and the component code corresponding to the physical topology, the topology starting node directly connected to the physical component code is located to obtain the starting flow node.

[0063] Optionally, the physical component code corresponding to the target anomaly source node is obtained. The physical component code is a unique identifier assigned to each injection system execution unit in the injection molding process knowledge graph, corresponding to the component code in the physical topology. The component code mapping relationship corresponding to the physical topology refers to the correspondence between each flow channel component and its code in the physical topology diagram of the injection mold runner system. The physical component code of the target anomaly source node is matched with this mapping relationship, and the topology starting node directly connected to this physical component code is located in the physical topology as the starting flow node.

[0064] In one embodiment, the aforementioned target anomaly source node is a low current output of the proportional flow valve, and its corresponding physical component code is the injection cylinder in the injection system. By querying the component code mapping relationship of the physical topology, the topology starting node directly connected to the injection cylinder is located as the nozzle, which is the starting flow node.

[0065] Step 206: Based on the directed flow channel connection relationship between the starting flow node and the physical topology, perform a forward traversal to extract the continuous flow channel element sequence extending from the starting flow node along the injection direction, and obtain candidate paths.

[0066] Optionally, starting from the initial flow node, a forward traversal is performed along the pre-defined directed flow channel connections in the physical topology. The directed flow channel connections refer to the directed edges connecting various flow channel elements, set according to the actual flow direction of the melt, pointing from upstream to downstream of the melt. During the forward traversal, the flow channel elements are extended along the injection direction, connecting them in series according to the melt flow sequence to form a continuous flow channel element sequence. Each continuous flow channel element sequence reachable from the initial flow node along the injection direction constitutes a candidate path.

[0067] In one embodiment, starting from the aforementioned initial flow node nozzle, the flow is traversed forward along the directed flow channel connection relationship. After the melt exits the nozzle, it enters the main flow channel. The end of the main flow channel connects to the first and second branch flow channels. The end of the first branch flow channel connects to the first gate, and the end of the second branch flow channel connects to the second gate. The traversal yields two candidate paths: the first path is from the nozzle through the main flow channel to the first branch flow channel and then to the first gate; the second path is from the nozzle through the main flow channel to the second branch flow channel and then to the second gate.

[0068] Step 207: Based on the end node attributes of the candidate paths, compare them with the gate type dictionary corresponding to the physical topology, retain the candidate paths whose end node attributes are injection points, and obtain the active injection paths.

[0069] Optionally, for each candidate path, the attributes of its end node are read and compared with a pre-defined gate type dictionary in the physical topology. The gate type dictionary is a lookup table recording various runner element types and their attribute names, where "gate point" refers to the attribute of a gate element that directly leads to the mold cavity. Only candidate paths with the end node attribute of "gate point" are retained, while candidate paths with the end node attribute of "gate point" are discarded. The retained candidate paths are the active gate paths.

[0070] In one embodiment, the terminal nodes of the two candidate paths are the first gate and the second gate, respectively. A query of the gate type dictionary shows that the terminal node attribute of both the first gate and the second gate is "gluing point". Both candidate paths are compared and retained as the active gluing path.

[0071] Step 208: Spatial association is performed between the coordinates of the end gate of each active injection path and the cavity gate belonging area corresponding to the physical topology to obtain the target propagation link.

[0072] Optionally, for each active injection path, the coordinates of its final gate are obtained and spatially correlated with the pre-defined cavity gate assignment area in the physical topology. The cavity gate assignment area refers to the spatial range pre-divided in the physical topology that records the area to which each gate is responsible for filling in the cavity. Through spatial correlation, the filling area covered by each active injection path in the cavity is determined, and all active injection paths affected by the target anomaly source node and their covered cavity areas are combined to form the target propagation link, as described in steps 2081 to 2084.

[0073] This invention maps target anomaly source nodes from the knowledge graph to the physical topology of the injection mold runner system. It locates the initial flow node through the mapping relationship between physical component codes and element codes, obtains a continuous runner element sequence by forward traversing the directed runner connection relationship, and then filters out active injection paths whose endpoints are injection points through gate type dictionary comparison. Finally, it determines the cavity filling sub-region covered by each active injection path by spatially associating the coordinates of the end gate with the cavity gate's belonging region. This invention strictly constrains the impact range of the anomaly to the actual physical path of melt flow, eliminating pseudo-propagation paths in the knowledge graph that are logically reachable but physically disconnected. This ensures that the subsequently obtained target propagation links conform to the actual flow law of melt from the nozzle to the cavity in injection molding, thereby improving the accuracy of root cause localization of weld line anomalies.

[0074] Optionally, the process of steps 2081 to 2084 includes: Step 2081: Based on the spatial association between the end gate coordinates of each active gluing path and the cavity gate belonging area corresponding to the physical topology, determine the cavity filling sub-region dominated by the active gluing path.

[0075] Optionally, for each active injection path, based on the spatial relationship between its end gate coordinates and the area to which the cavity gate belongs, the dominant filling sub-region in the cavity for that active injection path is determined. The filling sub-region refers to the local area of ​​the cavity covered by the melt injected by that gate under standard operating conditions, and its boundary is naturally defined by the flow front of the melt in the cavity.

[0076] In one embodiment, the aforementioned first gate is responsible for filling the left half of the cavity, and its dominant filling sub-region is the left half of the cavity. The second gate is responsible for filling the right half of the cavity, and its dominant filling sub-region is the right half of the cavity.

[0077] Step 2082: Extract the common interface line segments between adjacent sub-regions based on the geometric boundary of the cavity filling sub-region, and perform inclusion judgment based on the cavity entity geometry corresponding to each common interface line segment and the physical topology, retain the common interface line segments located inside the cavity entity, and obtain the weld line segment candidate.

[0078] Optionally, the geometric boundaries of each filled sub-region are extracted, and the common interface segments between two adjacent filled sub-regions are identified. A common interface segment refers to the overlapping line segments on the geometric boundaries of two adjacent filled sub-regions. Subsequently, each common interface segment is compared with the geometry of the cavity entity in the physical topology for inclusion determination. Only common interface segments completely located inside the cavity entity are retained as candidate weld line segments. Common interface segments falling outside the cavity entity are discarded.

[0079] In one embodiment, the geometric boundaries of the aforementioned left and right half-filling sub-regions meet at the middle of the cavity, and the common interface segment is a line segment distributed along the longitudinal centerline of the cavity. This line segment is included in the geometry of the cavity entity; if the line segment is located inside the cavity entity and corresponds to the root of the reinforcing rib of the product, it is retained as a candidate weld line segment.

[0080] Step 2083: Based on the path spatial position between the candidate weld line segment and the active glue injection path, associate each candidate weld line segment with the adjacent active glue injection paths on both sides.

[0081] Optionally, each weld line candidate segment is spatially associated with an active injection path, and the adjacent active injection paths on both sides of each weld line candidate segment are marked. Specifically, based on the position of the weld line candidate segment in the cavity, it is determined which two active injection paths will approach the segment from both sides. In one embodiment, the aforementioned weld line candidate segment is located on the longitudinal centerline of the cavity, the adjacent active injection path on its left is the injection path corresponding to the first gate, and the adjacent active injection path on its right is the injection path corresponding to the second gate. Therefore, the adjacent active injection paths on both sides of the weld line candidate segment are the first injection path and the second injection path, respectively.

[0082] Step 2084: Based on the spatial geometry of candidate weld line segments that belong to active glue injection paths on at least one side, construct potential weld line spatial links affected by the target abnormal source node, and obtain target propagation links that conform to the physical path of melt flow.

[0083] Optionally, candidate weld line segments with at least one adjacent path belonging to an active injection path are selected, and their spatial geometry is combined to construct a potential weld line spatial link affected by the target anomaly source node. A potential weld line spatial link refers to a spatial distribution path formed by connecting these candidate weld line segments, representing the potential for anomalies to induce weld lines within the mold cavity. This spatial link is the target propagation link conforming to the physical path of melt flow. In one embodiment, the first and second injection paths adjacent to both sides of the aforementioned candidate weld line segment are both active injection paths, satisfying the condition that at least one adjacent path belongs to an active injection path. The spatial geometry of the candidate weld line segment is combined with the flow channel spatial positions of the two active injection paths to construct a complete spatial link extending from the nozzle through the main channel, two branch channels, and two gates to the centerline of the mold cavity, serving as the target propagation link.

[0084] This invention, through cavity filling sub-region division, extraction of common interface segments between adjacent sub-regions, association of active injection paths on both sides of weld line candidate segments, and spatial geometric combination of weld line candidate segments that meet the conditions, accurately depicts the spatial location of the weld line formed by the convergence of two melt fronts and its corresponding flow channel source at the cavity geometry level. This ensures that the target propagation link not only includes the mechanical flow channel path from the abnormal source to the gate, but also extends to the spatial link of the weld line formed by the convergence of two melt fronts within the cavity. It clarifies the specific cavity location and injection paths on both sides that may produce weld lines due to the influence of the target abnormal source node, thereby improving the accuracy of root cause localization of weld line abnormal conditions.

[0085] Optionally, the processes of steps 301 to 304 include: Step 301: Based on the matching relationship between the injection system execution unit identifier corresponding to the target propagation link and the gate coordinate mapping relationship in the preset injection mold three-dimensional geometric model database, extract at least one target gate spatial coordinate that is physically connected to the target propagation link.

[0086] Optionally, the injection system execution unit identifier corresponding to the target propagation link is obtained. The injection system execution unit identifier is a unique number for each execution unit in the injection system. A preset injection mold 3D geometric model database is queried, which stores the mapping relationship between the coordinates of each gate and the injection system execution unit identifier. By matching, the 3D spatial coordinates of at least one gate physically connected to the target propagation link are extracted as the target gate spatial coordinates.

[0087] In one embodiment, the injection system execution unit corresponding to the aforementioned target propagation link is the injection system, identified as injection unit 1. The gate coordinate mapping relationship in the injection mold's three-dimensional geometric model database is queried; the gates physically connected to injection unit 1 are the first gate and the second gate. The three-dimensional spatial coordinates of the first gate are extracted, located at the upper left of the cavity, and the three-dimensional spatial coordinates of the second gate are located at the upper right of the cavity.

[0088] Step 302: Based on the spatial coordinates of the target gate and the preset standard working condition melt filling benchmark database, perform an index query to obtain the local flow characteristic data corresponding to the spatial coordinates of the target gate. The local flow characteristic data includes the theoretical filling time of the full cavity mesh nodes and the identifier of the dominant flow source.

[0089] Optionally, based on the spatial coordinates of the target gate, a preset standard operating condition melt filling benchmark database is queried. This database stores melt filling benchmark data for each gate, obtained through simulation or experimental calibration under standard process parameters. By querying the index, local flow characteristic data corresponding to the spatial coordinates of each target gate is obtained. This local flow characteristic data includes the theoretical filling time and dominant flow source identifier for all cavity grid nodes. The theoretical filling time refers to the theoretical time when the melt arrives at each cavity grid node under standard operating conditions. The dominant flow source identifier indicates which gate or flow path the melt at each cavity grid node primarily originates from.

[0090] In one embodiment, the spatial coordinates of the first and second gates are used as indexes to query the standard melt filling reference database. Local flow characteristic data is obtained: the theoretical filling time of the grid nodes in the left half of the cavity is earlier, and the dominant flow source is identified as the first gate. The theoretical filling time of the grid nodes in the right half of the cavity is later, and the dominant flow source is identified as the second gate. The theoretical filling time of the grid nodes near the cavity centerline is between the two, and the dominant flow source identification includes both the first and second gates.

[0091] Step 303: Traverse all cavity mesh nodes within the mold cavity, and select all cavity mesh nodes with two or more dominant flow source identifiers and a difference between the maximum theoretical filling time and the minimum theoretical filling time less than a preset welding time threshold to obtain potential weld line candidate points.

[0092] Optionally, all cavity mesh nodes within the mold cavity are traversed, and a dual screening condition is applied. The first condition is that the number of dominant flow source markers at the mesh node is greater than or equal to two, meaning that at least two melt fronts from different gates converge at this node. The second condition is that the difference between the maximum and minimum theoretical filling time at the node is less than a preset weld time threshold. The preset weld time threshold refers to the maximum allowable time difference for two melt fronts to fuse and form a qualified weld line. If this time difference is exceeded, the melt front temperature drops too much, preventing proper fusion and resulting in a defective weld line. Cavity mesh nodes that simultaneously meet both conditions are retained as potential weld line candidate points.

[0093] In one embodiment, after traversing all grid nodes in the cavity, the grid nodes inside the left and right halves of the cavity each have only one dominant flow source identifier, which does not meet the first condition and is therefore excluded. The grid nodes near the cavity centerline have dominant flow source identifiers that include both the first and second gates, thus satisfying the first condition. The difference between the maximum and minimum theoretical filling times at these nodes is calculated. Assuming this difference is 0.3 seconds and the preset weld time threshold is 0.5 seconds, 0.3 seconds is less than 0.5 seconds, satisfying the second condition. Therefore, these grid nodes near the cavity centerline are retained as potential weld line candidate points.

[0094] Step 304: Based on the three-dimensional spatial coordinates of potential weld line candidate points, perform spatial overlap matching with the camera calibration parameters of the injection molded product surface defect detection system to obtain the target diagnostic path.

[0095] Optionally, the three-dimensional spatial coordinates of potential weld line candidate points are spatially matched with the camera calibration parameters of the injection molded product surface defect detection system. Camera calibration parameters refer to pre-calibrated parameters describing the camera's imaging geometry, including intrinsic and extrinsic parameters. Intrinsic parameters describe the camera's optical imaging characteristics, while extrinsic parameters describe the camera's position and orientation in the mold coordinate system. Spatial matching involves projecting candidate points in three-dimensional space onto the camera's imaging plane and comparing them with the actual detected weld line defect location to determine if they overlap. After matching, a target diagnostic path associated with the actual weld line defect location is obtained, as detailed in steps 3041 to 3044.

[0096] This invention employs gate coordinate extraction, standard working condition melt filling benchmark database index query, full cavity mesh node dual-condition screening, and spatial overlap matching with camera calibration parameters to precisely align theoretically possible weld line candidate locations in the knowledge graph with the actual visually detected weld line defect locations. The first screening condition ensures that two or more melt streams converge at the candidate point, and the second screening condition ensures that the time difference between the two melt fronts is within the fusion range. Combined with the judgment of the projection overlap of three-dimensional coordinates to the imaging plane, the true diagnostic path that coincides with the actual defect location is selected from the theoretical candidates, eliminating theoretically possible but not actually formed weld line candidates. This improves the matching accuracy between the diagnostic path and the actual defect, thereby improving the accuracy of root cause localization of abnormal weld line conditions.

[0097] Optionally, the processes of steps 3041 to 3044 include: Step 3041: Based on the three-dimensional spatial coordinates of the potential weld line candidate points and the camera calibration parameters, perform coordinate projection transformation to convert the three-dimensional spatial coordinates of the potential weld line candidate points into two-dimensional image plane coordinates, and obtain the projection candidate region.

[0098] Optionally, based on the three-dimensional spatial coordinates of potential weld line candidate points, coordinate projection transformation is performed using camera calibration parameters. Coordinate projection transformation refers to the process of mapping points in three-dimensional space to a two-dimensional image plane according to the camera imaging geometry model. Through coordinate projection transformation, the three-dimensional coordinates of each potential weld line candidate point are converted into two-dimensional image plane coordinates, and all the transformed two-dimensional coordinate points form the projection candidate region.

[0099] In one embodiment, the aforementioned potential weld line candidate points are located near the centerline of the cavity, and their three-dimensional spatial coordinates are known. Using camera intrinsic and extrinsic parameters, the three-dimensional coordinates of each candidate point are projected onto the camera imaging plane to obtain the corresponding two-dimensional image plane coordinates. These two-dimensional coordinate points form a projection candidate region distributed along the longitudinal centerline of the product on the image plane.

[0100] Step 3042: Based on the two-dimensional image plane coordinates of the projection candidate region and the defect pixels in the weld line defect segmentation mask image, perform geometric inclusion judgment, retain the projection candidate points located inside the defect pixels, and remove the projection candidate points located outside the defect pixels to obtain the target weld line associated points.

[0101] Optionally, the two-dimensional image plane coordinates of the projection candidate region are geometrically included with the defect pixels in the weld line defect segmentation mask image. The weld line defect segmentation mask image refers to the image obtained by segmenting the product image captured in the current cycle by the product surface defect detection system, and the image in binary form annotating the positions of weld line defect pixels. The geometric inclusion judgment refers to determining whether the two-dimensional coordinates of each projection candidate point fall within the region enclosed by the defect pixels. Projection candidate points located inside the defect pixels are retained as target weld line association points. Projection candidate points located outside the defect pixels are discarded.

[0102] In one embodiment, the aforementioned projection candidate regions are distributed along the longitudinal centerline of the product on the image plane. In this phase of the weld line defect segmentation mask image, the weld line defect pixels are located near the root of the reinforcing rib of the product and are distributed in a longitudinal strip shape. The projection candidate points are geometrically included with the defect pixels one by one, assuming that most projection candidate points fall inside the defect pixel region, and a few fall outside. The projection candidate points that fall inside are retained as the target weld line association points.

[0103] Step 3043: Based on at least two dominant flow source identifiers corresponding to each target weld line association point in the local flow characteristic data, determine the target flow path identifier combination that leads to the formation of the weld line.

[0104] Optionally, for each target weld line associated point, the dominant flow source identifier corresponding to it in the local flow characteristic data is queried. Each target weld line associated point has at least two dominant flow source identifiers, which represent the two flow paths from which the two melt streams converging at that point originate. Combining these at least two dominant flow source identifiers yields the target flow path identifier combination that leads to the formation of the weld line at that point. In one embodiment, the aforementioned target weld line associated point is located near the root of the reinforcing rib of the product. Querying the local flow characteristic data, the dominant flow source identifiers at this location are the first gate and the second gate. Therefore, the target flow path identifier combination that leads to the formation of the weld line at this point is the flow path corresponding to the first gate and the flow path corresponding to the second gate.

[0105] Step 3044: Based on the intersection of the target flow path identifier combination and the controlled flow channel branch identifier in the target propagation link, the target overlapping path identifier is obtained. Based on the target overlapping path identifier, a subgraph search is performed in the directed graph structure of the target propagation link to extract the minimum connected subgraph containing the target overlapping path identifier and its upstream causal related nodes, thus obtaining the target diagnostic path.

[0106] Optionally, the target flow path identifier combination is intersected with the controlled flow channel branch identifiers in the target propagation link to obtain the target overlapping path identifiers. The controlled flow channel branch identifiers refer to the identifiers of each flow channel branch in the target propagation link. The result of the intersection operation is the identifiers commonly included in both the target flow path identifier combination and the controlled flow channel branch identifiers. Subsequently, a subgraph search is performed in the directed graph structure of the target propagation link to extract the minimum connected subgraph containing the target overlapping path identifiers and their upstream causal nodes. The minimum connected subgraph is the minimum directed subgraph formed by tracing back upwards along the causal edges from the target overlapping path identifier as the endpoint to the upstreammost node. This minimum connected subgraph is the target diagnostic path.

[0107] In one embodiment, the aforementioned target flow path identifiers are combined into a first gate flow path and a second gate flow path. The controlled flow channel branch identifiers in the target propagation link include a first branch channel, a second branch channel, etc. The intersection of these two is performed, and the identifiers commonly included are the first gate corresponding branch and the second gate corresponding branch, i.e., the target overlapping path identifiers. In the directed graph structure of the target propagation link, a reverse subgraph search is performed with these two overlapping path identifiers as endpoints to extract the minimum connected subgraph from the proportional flow valve through the injection cylinder, nozzle, main flow channel, two branch channels, two gates to the weld line connection point, which serves as the target diagnostic path.

[0108] This invention utilizes a projection transformation from three-dimensional spatial coordinates to a two-dimensional image plane to accurately map theoretical candidate points within the cavity onto the actual imaging plane. By determining the geometric inclusion of defect pixels in the weld line defect segmentation mask image, only candidate points falling within the actual defect area are retained. The flow path combination leading to the weld line is determined by querying at least two dominant flow source identifiers at each associated point. This path combination is then intersected with the controlled flow channel branch identifiers in the target propagation link, and a minimum connected subgraph is searched in reverse. This ultimately yields a minimum causal subgraph that spatially coincides with the actual weld line defect location and causally leads directly from the target anomaly source node to the defect point. This achieves complete closed-loop localization from the defect location to the anomaly root cause, avoiding misjudging the theoretical weld line location as the actual defect source, thereby improving the accuracy of weld line anomaly root cause localization.

[0109] Furthermore, the abnormal operating condition diagnosis device based on injection molding process knowledge graph provided by the present invention will be described below. The abnormal operating condition diagnosis device based on injection molding process knowledge graph described below can be referred to in correspondence with the abnormal operating condition diagnosis method based on injection molding process knowledge graph described above.

[0110] Optionally, refer to Figure 2 , Figure 2This is a schematic diagram of the abnormal operating condition diagnosis device based on injection molding process knowledge graph provided by the present invention. The abnormal operating condition diagnosis device based on injection molding process knowledge graph includes: The abnormal process event analysis module 210 is used to compare the deviation between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary to obtain the process deviation feature vector, and to perform semantic matching between the process deviation feature vector and the process state entity nodes in the injection process knowledge graph to obtain potential abnormal process events.

[0111] The anomaly propagation analysis module 220 is used to perform reverse tracing based on the causal relationship edges between potential abnormal process events and the injection molding process knowledge graph to obtain the target anomaly source node, and to perform connectivity constraint verification based on the physical topology of the target anomaly source node and the injection mold flow channel system to obtain the target propagation link.

[0112] The diagnostic path localization module 230 is used to perform spatial overlap matching based on the spatial distribution position of the target propagation link and the weld line defect in the mold cavity to obtain the target diagnostic path.

[0113] The abnormal working condition diagnosis module 240 is used to map and match the attributes of the target abnormal source node in the target diagnosis path with the handling strategies of the historical maintenance knowledge base to obtain the diagnostic conclusion of the abnormal working condition of the weld line.

[0114] This invention extracts process deviation feature vectors by comparing the deviations of feature points on the injection pressure curve with the standard process window boundary. Combined with a knowledge graph of injection molding processes, semantic matching is performed to accurately capture real-time process deviations caused by raw material batch fluctuations, equipment aging, and changes in ambient temperature during production. This overcomes the limitations of traditional CAE simulation methods, which are only applicable to the mold design stage, cannot adapt to dynamic drift in production conditions, and are disconnected from actual production. It also eliminates the one-sidedness of traditional SPC methods that rely on single-parameter threshold judgments, achieving comprehensive feature extraction of multi-dimensional process deviations. By relying on the causal relationships of the knowledge graph, potential abnormal process events are traced back to initially screen out target abnormal source nodes. Then, by combining the physical topology of the mold runner system, connectivity constraint verification is performed to lock in the real and effective abnormal propagation links. This overcomes the shortcomings of traditional technologies that ignore strong coupling relationships of process parameters and cannot identify abnormal signal propagation paths, solving the problem of being unable to identify hidden weld lines caused by abnormal coupling even when a single parameter is normal. By matching the overlap between the target propagation path and the spatial distribution of weld line defects inside the mold cavity, the target diagnostic path directly corresponding to the defect is selected, and irrelevant abnormal interference paths are eliminated, ensuring the targetedness and effectiveness of anomaly tracing. The attributes of the abnormal source nodes in the diagnostic path are mapped and matched with the historical maintenance knowledge base to complete the handling strategy, outputting accurate weld line anomaly diagnostic conclusions. Therefore, it can completely reproduce the entire propagation process of abnormal signals from the source to the weld line defect point in the injection molding system, effectively distinguishing different types of anomaly root causes such as machine parameter settings, mold runner status, and material properties, avoiding the drawbacks of blind trial and error, and improving the accuracy of locating the root cause of weld line anomaly conditions.

[0115] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 40.

[0116] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 40.

[0117] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the abnormal working condition diagnosis method based on the injection molding process knowledge graph provided by the above methods, which includes steps 10 to 40.

[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An abnormal operating condition diagnosis method based on injection molding process knowledge graph, characterized in that, include: Based on the deviation comparison between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary, a process deviation feature vector is obtained. Then, based on the semantic matching of the process deviation feature vector and the process state entity nodes in the injection process knowledge graph, potential abnormal process events are obtained. Reverse tracing is performed based on the causal relationship edges between the potential abnormal process events and the injection molding process knowledge graph to obtain the target abnormal source node. Then, based on the physical topology of the target abnormal source node and the injection mold runner system, connectivity constraint verification is performed to obtain the target propagation link. Based on the spatial overlap between the target propagation path and the spatial distribution of weld line defects in the mold cavity, the target diagnostic path is obtained. Based on the mapping and matching of the target anomaly source node attributes in the target diagnostic path with the handling strategies of the historical maintenance knowledge base, a diagnostic conclusion for the abnormal working condition of the weld line is obtained.

2. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 1, characterized in that, The steps involved in obtaining the target diagnostic path through spatial overlap matching include: Based on the matching relationship between the injection system execution unit identifier corresponding to the target propagation link and the gate coordinate mapping relationship in the preset injection mold three-dimensional geometric model database, at least one target gate spatial coordinate physically connected to the target propagation link is extracted. Based on the target gate spatial coordinates and the preset standard working condition melt filling benchmark database, an index query is performed to obtain the local flow characteristic data corresponding to the target gate spatial coordinates; the local flow characteristic data includes the theoretical filling time of the full cavity grid nodes and the identifier of the dominant flow source; Traverse all cavity mesh nodes within the mold cavity, and filter out all cavity mesh nodes with two or more dominant flow source identifiers and a difference between the maximum theoretical filling time and the minimum theoretical filling time less than a preset weld time threshold to obtain potential weld line candidate points; Based on the three-dimensional spatial coordinates of the potential weld line candidate points and the camera calibration parameters of the injection molded product surface defect detection system, the spatial overlap is matched to obtain the target diagnostic path.

3. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 2, characterized in that, The target diagnostic path is obtained by matching the spatial overlap between the three-dimensional spatial coordinates of the potential weld line candidate points and the camera calibration parameters of the injection molded product surface defect detection system, including: Based on the three-dimensional spatial coordinates of the potential weld line candidate points and the camera calibration parameters, coordinate projection transformation is performed to convert the three-dimensional spatial coordinates of the potential weld line candidate points into two-dimensional image plane coordinates, thereby obtaining the projection candidate region. Based on the two-dimensional image plane coordinates of the projection candidate region and the defect pixels in the weld line defect segmentation mask image, a geometric inclusion judgment is performed. Projection candidate points located inside the defect pixels are retained, and projection candidate points located outside the defect pixels are removed to obtain the target weld line associated points. Based on at least two dominant flow source identifiers corresponding to each target weld line association point in the local flow characteristic data, determine the target flow path identifier combination that leads to the formation of the weld line. The target overlapping path identifier is obtained by intersecting the target flow path identifier combination with the controlled flow channel branch identifier in the target propagation link. Then, a subgraph search is performed in the directed graph structure of the target propagation link based on the target overlapping path identifier to extract the minimum connected subgraph containing the target overlapping path identifier and its upstream causal related nodes, thus obtaining the target diagnostic path.

4. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 1, characterized in that, The steps for obtaining the target propagation link through connectivity constraint verification include: Based on the mapping relationship between the physical component code corresponding to the target anomaly source node and the component code corresponding to the physical topology, the topology starting node directly connected to the physical component code is located to obtain the starting flow node; Based on the directed flow channel connection relationship between the starting flow node and the physical topology, a forward traversal is performed to extract a continuous flow channel element sequence extending from the starting flow node along the injection direction, thereby obtaining candidate paths. Based on the comparison between the end node attributes of each candidate path and the gate type dictionary corresponding to the physical topology, candidate paths with end node attributes of injection point are retained to obtain active injection paths. The target propagation link is obtained by spatially associating the coordinates of the end gate of each active injection path with the cavity gate belonging area corresponding to the physical topology.

5. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 4, characterized in that, The target propagation link is obtained by spatially associating the coordinates of the end gate of each active injection path with the cavity gate belonging region corresponding to the physical topology, including: Based on the spatial association between the end gate coordinates of each active gluing path and the cavity gate belonging area corresponding to the physical topology, the cavity filling sub-region dominated by each active gluing path is determined. Based on the geometric boundary of the cavity filling sub-region, common interface line segments between adjacent sub-regions are extracted, and inclusion judgment is performed based on the cavity entity geometry corresponding to each common interface line segment and the physical topology. Common interface line segments located inside the cavity entity are retained to obtain weld line candidate segments. The path spatial position between the candidate weld line segment and the active glue injection path is associated to associate each candidate weld line segment with the adjacent active glue injection paths on both sides. Based on the spatial geometry of candidate weld line segments that belong to active glue injection paths on at least one side, a potential weld line spatial link affected by the target abnormal source node is constructed to obtain a target propagation link that conforms to the physical path of melt flow.

6. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 1, characterized in that, The steps involved in reverse tracing to obtain the target anomaly source node include: Based on the compatibility matching between the physical effect type of the potential abnormal process event and the physical domain coupling attribute of the causal association edge in the injection molding process knowledge graph, the multi-physical domain coupling edge that allows the physical effect type as the input end of the physical domain coupling attribute is extracted to obtain the first causal candidate edge. Based on the existence comparison between the device instance status field of the starting node connected by each first causal candidate edge and the device activation mapping table of the current injection molding production cycle, the causal association edges that are indicated by the device instance status field as being activated and occupied in the current cycle are retained to obtain the second causal candidate edges. Based on the temporal constraint interval corresponding to each second causal candidate edge and the occurrence time of the potential abnormal process event, a temporal inclusion judgment is performed, and causal related edges whose occurrence time is within the temporal constraint interval are retained to obtain the third causal candidate edge. The target anomaly source node is obtained by performing loop consistency matching between the control loop attribution field of the starting node connected by each third causal candidate edge and the main control loop identifier of the potential abnormal process event.

7. The abnormal operating condition diagnosis method based on injection molding process knowledge graph according to claim 6, characterized in that, The control loop attribution field of the starting node connected by each third causal candidate edge is matched with the main control loop identifier of the potential abnormal process event to obtain the target abnormal source node, including: Based on the control loop attribution field of the starting node connected by each third causal candidate edge, loop consistency matching is performed with the main control loop identifier of the potential abnormal process event. Causal association edges with the same control loop attribution field and main control loop identifier or with a preset coupling relationship are retained to obtain the fourth causal candidate edge. Based on the reverse transmission rule corresponding to the action direction attribute of each fourth causal candidate edge and the signal flow direction of the injection control system, the direction consistency is checked, and the causal correlation edge whose action direction attribute conforms to the direction of the result effect to the direction of the cause excitation is retained to obtain the fifth causal candidate edge. Based on each fifth causal candidate edge, construct a reverse causal tree with the potential abnormal process event as the endpoint. Traverse upwards along the reverse causal tree until a boundary node with independent self-checking function or independent control interface is encountered. The boundary node at the point where the traversal stops is taken as the candidate abnormal source node. Based on the direct causal incoming edges of the potential abnormal process events in the injection molding process knowledge graph, causal connectivity is verified for each candidate abnormal source node. Candidate abnormal source nodes that can be directly traced back to the potential abnormal process events through the fifth causal candidate edge are retained, and candidate abnormal source nodes that have not established causal connection paths are eliminated to obtain the target abnormal source node.

8. An abnormal operating condition diagnosis device based on injection molding process knowledge graph, characterized in that, The device is used to implement the abnormal operating condition diagnosis method based on injection molding process knowledge graph as described in any one of claims 1 to 7; the abnormal operating condition diagnosis device based on injection molding process knowledge graph includes: The abnormal process event analysis module is used to compare the deviation between the feature points of the injection pressure curve during the injection cycle and the standard process window boundary to obtain the process deviation feature vector, and to perform semantic matching between the process deviation feature vector and the process state entity nodes in the injection process knowledge graph to obtain potential abnormal process events. The anomaly propagation analysis module is used to perform reverse tracing based on the causal relationship edges between the potential abnormal process events and the injection molding process knowledge graph to obtain the target anomaly source node, and to perform connectivity constraint verification based on the physical topology of the target anomaly source node and the injection mold runner system to obtain the target propagation link. The diagnostic path localization module is used to perform spatial overlap matching between the target propagation link and the weld line defect in the mold cavity to obtain the target diagnostic path. The abnormal operating condition diagnosis module is used to map and match the attributes of the target abnormal source node in the target diagnosis path with the handling strategies of the historical maintenance knowledge base to obtain a diagnostic conclusion on the abnormal operating condition of the weld line.

9. An electronic device, comprising: Memory, used to store computer programs; A processor for reading and executing the computer program, characterized in that, when the processor executes the computer program, it implements the abnormal working condition diagnosis method based on injection molding process knowledge graph as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the abnormal operating condition diagnosis method based on injection molding process knowledge graph as described in any one of claims 1 to 7.

Images