Automatic production line fault diagnosis method based on multi-source heterogeneous data

Abstract

The application relates to the technical field of data processing, in particular to an automatic production line fault diagnosis method based on multi-source heterogeneous data. The automatic production line is divided into multiple local diagnosis modules according to equipment control ranges, and each local diagnosis module corresponds to a local Petri net; under the current normal transition, the actual local identifier of each library is compared with the reference local identifier before and after the event anchor point, and the residual increment is counted; the state correction vector is formed by the residual increment, and the boundary influence vector is counted; the candidate abnormal transition event is matched with the downstream candidate abnormal transition event, the boundary matching value is determined according to the consistent degree of the residual increment direction, the attribution score is determined in combination with the time difference; and finally, the diagnosis result is output according to the boundary influence vector, the attribution score and the state conflict identifier. The application can effectively improve the accuracy of fault source positioning and enhance the adaptability to unknown fault scenes.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to an automated production line fault diagnosis method based on multi-source heterogeneous data. Background Technology

[0002] Automated production lines typically include CNC machine tools, robots, automated guided vehicles (AGVs), conveying equipment, buffer stations, fixtures, vision inspection equipment, and controllers. Workpieces sequentially complete loading, processing, inspection, transfer, and assembly between various workstations. During workpiece flow, the equipment exhibits not only sequential actions but also synchronous, concurrent, resource-exclusive, and cross-workstation handover relationships. To ensure stable production line operation, the control system typically collects heterogeneous data from multiple sources, including PLC status, robot and AGV control status, vision inspection results, RFID or barcode identification results, current, pressure, torque, speed, work orders, and process recipes.

[0003] Existing fault diagnosis methods for automated production lines often struggle to accurately attribute data from multiple heterogeneous sources, such as vision, PLC, equipment controllers, RFID, and sensors, to the same production action when processing data with different sampling periods, transmission delays, and observed objects. This makes it difficult to establish a consistent correlation between local anomalies and the corresponding workpiece flow, resource occupancy, and module handover processes, thus making it difficult to identify the source of the fault in a timely manner. Summary of the Invention

[0004] To address the technical problem in existing technologies that fail to integrate the relationships between various processes on multi-process production lines and output accurate diagnostic results, the present invention aims to provide an automated production line fault diagnosis method based on multi-source heterogeneous data. The specific technical solution adopted is as follows: This invention proposes an automated production line fault diagnosis method based on multi-source heterogeneous data, the method comprising: The automated production line is divided into multiple local diagnostic modules according to the equipment control range, and each local diagnostic module corresponds to a local Petri net; the standard action in each local Petri net is defined as the normal transition corresponding to the local Petri net; under the current normal transition, all transition events are counted. For each transition event, determine the conflict state of the location identifier and generate a state conflict identifier; before and after the event anchor point occurs, compare the actual local identifier and the reference local identifier of each location, and calculate the residual increment before and after the event anchor point occurs; if there is a residual increment, the corresponding transition event is taken as a candidate abnormal transition event and a state correction vector is calculated; the state correction vector is composed of residual increments, and the identifiers corresponding to the elements have common correlations; based on the state correction vector, calculate the boundary influence vector corresponding to the boundary location between the current normal transition and the downstream normal transition. The candidate abnormal transition events of the current normal transition are matched with the candidate abnormal transition events of each downstream normal transition. The boundary matching value is determined based on the consistency of the residual increment direction between the matched events. The attribution score of the candidate abnormal transition event of the current normal transition is determined based on the boundary matching value and the time difference between the matched events. The diagnostic result is output based on the boundary influence vector, the attribution score and the state conflict identifier.

[0005] Furthermore, the method for classifying the transition events includes: For each normal transition, when the identifier of the preceding repository meets the execution conditions, the various startup data fields corresponding to the normal transition are counted, and the time corresponding to the earliest executed startup data field is taken as the anchor time of the transition event. Starting from the anchor point, if no normal downstream transition occurs within the preset maximum duration, the period from the anchor point to the end of the maximum duration is considered the range of transition events; if a normal downstream transition occurs, the period from the anchor point to the moment when the normal downstream transition occurs is considered the range of transition events.

[0006] Furthermore, the method for obtaining the residual increment includes: For a transition event, within the same repository, the actual local identifier before the event anchor point is subtracted from the reference local identifier before the event anchor point to obtain the residual before the anchor point; the actual local identifier after the event anchor point is subtracted from the reference local identifier after the event anchor point to obtain the residual after the anchor point; the residual increment is obtained by subtracting the residual before the anchor point from the residual after the anchor point.

[0007] Furthermore, the statistical method for the state correction vector includes: The residual increments corresponding to the preceding, subsequent, or directly related resource repositories that belong to the current normal changes are statistically analyzed to obtain the first set; For any two residual increments in the first set, if the state of the same place corresponding to the two is confirmed as opposite by at least two different data sources within the same transition observation window, then the two residual increments are determined to have an observation conflict, and the residual increments involved are split into different conflict subsets according to different observation branches; within each conflict subset, residual increments with the same workpiece identifier, carrier identifier, fixture identifier or shared resource identifier are grouped into the same residual association group; For residual increments that do not have observation conflicts, residual increments with the same workpiece identifier, carrier identifier, fixture identifier, or shared resource identifier are grouped into the same residual association group. All residual increments in each residual association group constitute a state correction vector.

[0008] Furthermore, the method for obtaining the boundary influence vector includes: There is a module boundary library extracted matrix between the local Petri nets of normal transitions between adjacent upstream and downstream. The element corresponding to the boundary library in the module boundary library extracted matrix has a value of 1, and the other elements have a value of 0. Based on the module boundary library extracted matrix, the residual increment value corresponding to the boundary library is obtained by querying the state correction vector and a boundary influence vector is constructed.

[0009] Furthermore, the method for obtaining the boundary matching value includes: For any set of matching events, the boundary influence vector of the candidate abnormal transition event of the current normal transition is used as the first comparison vector, and the boundary influence vector of the candidate abnormal transition event of the downstream normal transition is used as the second comparison vector. Obtain the components in the first comparison vector and the second comparison vector that correspond to the same workpiece identifier, shared resource identifier, or handover identifier; if the non-zero component in the first comparison vector has the same state change direction as the corresponding component in the second comparison vector, then set the boundary matching value of the matching event group to 1; if the non-zero component in the first comparison vector has the same state change direction as the corresponding component in the second comparison vector, and there is no corresponding component with the opposite direction, then set the boundary matching value of the matching event group to 0.5; if there is a corresponding component in the second comparison vector with the opposite direction to the non-zero component in the first comparison vector, then set the boundary matching value of the matching event group to 0.

[0010] Furthermore, the method for obtaining the attribution score includes: Under each matching event, the ratio of the time difference to the maximum duration of the current normal transition is used as the time score, and the initial score of the matching event is obtained based on the boundary matching value and the time score. The initial scores of all matching events for the candidate abnormal transition event of the current normal transition are accumulated as the attribution score.

[0011] Further, the step of outputting diagnostic results based on the boundary influence vector, attribution score, and state conflict identifier includes: If the boundary influence vector of the candidate abnormal transition event of the current normal transition is a vector of all zeros, then the matching process is not executed, and the local execution abnormal result is output. If a candidate abnormal transition event of the current normal transition has a boundary influence vector, the candidate abnormal transition event is stored in a buffer and awaits the matching process. After candidate abnormal transition events of different normal transitions are found in the buffer, the matching process is performed. If no matching event is found within a preset matching time and a state conflict indicator exists, an observation anomaly result is fed back. If no matching event is found and no state conflict indicator exists, a local execution anomaly is fed back. If a matching event is found, the attribution score of each candidate abnormal transition event is obtained. The fault source event is selected based on the attribution score, and the corresponding diagnostic result is fed back based on the identification result of the fault source event.

[0012] Furthermore, the step of filtering out fault source events based on the attribution score includes: Among all candidate abnormal transition events in the statistical buffer, the two candidate abnormal transition events with the highest attribution scores are selected. If the difference in attribution scores between the two candidate abnormal transition events with the highest attribution scores is greater than a preset difference threshold, then the candidate abnormal transition event with the highest attribution score is selected as the fault source event. If the difference is less than or equal to the preset difference threshold, it is determined whether the two candidate abnormal transition events share a common identifier. If any kind of shared identifier exists, the candidate abnormal transition event with the highest attribution score is taken as the fault source event; if no shared identifier exists, the two candidate abnormal transition events are taken as fault source events respectively.

[0013] Furthermore, the step of feeding back the corresponding diagnostic results based on the identification results of the fault source event includes: All matching events corresponding to the fault source event are taken as matching events to be analyzed; If, among all the events to be analyzed and matched, the fault source event has downstream candidate abnormal transition events, and all the downstream candidate abnormal transition events are located in the same adjacent local diagnostic module, and each matching event is sequentially associated through the same workpiece identifier or handover identifier, then the fault source event is determined to be an upstream propagation abnormality. If, among all the events to be analyzed and matched, the downstream candidate abnormal transition events matched by the fault source event are distributed in at least two different local diagnostic modules, and each matched event points to the same shared resource identifier, then the fault source event is determined to be a shared resource abnormality. In other cases, the fault source event is marked as a cross-module anomaly pending confirmation, and all downstream candidate anomaly transition event information matched with it is output.

[0014] The present invention has the following beneficial effects: This invention divides the automated production line into multiple local diagnostic modules according to the equipment control range and establishes corresponding local Petri nets, so that the diagnostic range of each normal transition is limited to the corresponding physical or logical boundary, avoiding cross-module data mixing and improving the accuracy of data attribution.

[0015] By generating state conflict identifiers for each transition event and comparing the actual local identifiers with the reference local identifiers before and after the event anchor point to calculate the residual increment, it is possible to align multi-source data with different sampling periods and different transmission delays to the same production action on the time axis, thus solving the problem of the difficulty in accurately attributing multi-source data.

[0016] When residual increments exist, this invention uses the corresponding transition events as candidate abnormal transition events and calculates state correction vectors, achieving adaptive expansion for unknown faults and process switching scenarios without pre-enumerating all fault types. Based on the state correction vector, the boundary influence vector is calculated, quantifying the impact of anomalies on module boundaries and shared resources, enabling local anomalies to establish a consistent association with workpiece flow, resource occupation, and module handover processes.

[0017] Furthermore, candidate abnormal transition events are matched with downstream normal transition candidate abnormal transition events. Boundary matching values ​​are determined based on the consistency of residual increment directions, and attribution scores are calculated in conjunction with time differences, thereby identifying the most likely fault source among multiple candidate anomalies. Finally, diagnostic results are output based on boundary influence vectors, attribution scores, and state conflict identifiers, effectively improving the accuracy of fault source localization and enhancing the production line's adaptive diagnostic capabilities in scenarios of unknown faults and process switching. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 The flowchart illustrates an automated production line fault diagnosis method based on multi-source heterogeneous data, as provided in one embodiment of the present invention. Detailed Implementation

[0020] Please see Figure 1 The diagram illustrates a flowchart of an automated production line fault diagnosis method based on multi-source heterogeneous data, according to an embodiment of the present invention. The method includes: Step S1: Divide the automated production line into multiple local diagnostic modules according to the equipment control range, with each local diagnostic module corresponding to a local Petri net; define the standard action in each local Petri net as the normal transition corresponding to the local Petri net; under the current normal transition, count all transition events.

[0021] In Petri net theory, a classic Petri net consists of places, transitions, and directed arcs connecting them. Places represent the states of the system, transitions represent changes or actions between states, and the distribution of tokens in the places constitutes the system's marking. This theory has been widely used in the modeling and analysis of discrete event dynamic systems and is a well-known technology in this field. To achieve effective correlation analysis in the process flow, this invention, based on classic Petri net theory, constructs a local Petri net and a local diagnostic module for the multi-source heterogeneous data diagnostic scenario of automated production lines in the following manner: (1) Determine the equipment control range under the current process mode.

[0022] Specifically, the current work order identifier, product model, and process recipe version can be obtained from the MES (Manufacturing Execution System). The current process mode is then determined based on the pre-defined correspondence within the MES. The same equipment may correspond to different workpiece flow paths and resource call methods under different product models or process recipes. Therefore, it is necessary to first determine the valid workpiece status, resource status, and normal transition set based on the current process mode. The normal transition refers to the standard execution actions defined by the PLC sequential control program, robot program, or equipment scheduling configuration in this production line.

[0023] (2) Construct a corresponding local Petri net for each local diagnostic module.

[0024] After determining the current process mode, based on the production line layout diagram, PLC sequential control program, robot program, automated guided vehicle (AGV) scheduling configuration, and equipment I / O point table, the automated production line is divided into multiple local diagnostic modules according to the equipment control range; each local diagnostic module corresponds to a local Petri net. Specifically: 1. The status of the workpiece at each workstation, AGV, buffer position, and input position of the next module is set as the workpiece status library; 2. The idle or occupied status of shared resources (such as AGVs, fixtures, buffer positions, etc.) is set as the resource status library; 3. Standard execution actions such as picking up, placing, processing, inspection, transfer, receiving, and releasing are set as normal transitions; 4. The triggering condition of each normal transition is determined by the token distribution in its preceding library, that is, the normal transition is only ready to execute when the current state meets the preceding condition.

[0025] (3) Establish boundary associations for interactions between different local Petri nets.

[0026] To compare the same shared resource or handover location in different local Petri nets, a unique resource identifier is assigned to the shared resource, and a unique handover identifier is assigned to the output, handover, and input bits between local diagnostic modules. The locations in the local Petri net corresponding to the resource identifier or handover identifier are recorded as boundary locations, and a module boundary location extraction matrix is ​​constructed based on the index of the boundary location in the current local Petri net. This module boundary location extraction matrix is ​​used to extract token components corresponding to the shared resource status and workpiece handover status from the current local identifier according to the location index of the boundary location in the local Petri net, serving as the data interface for subsequent cross-module attribution analysis.

[0027] (4) Establish evidence mapping relationships between multi-source heterogeneous data and states and transitions in local Petri nets.

[0028] To enable multi-source heterogeneous data to be interpreted as evidence of workpiece status, resource status, or normal transitions, a mapping relationship is established between data sources, status evidence, and normal transitions: 1. Workpiece position is confirmed by vision, RFID, barcode, photoelectric, or position sensors; 2. Resource status is confirmed by resource controllers, buffer position sensors, and resource lock records; 3. PLC control outputs and task assignments serve as candidate observations for initiation; 4. Current, pressure, torque, speed, and vibration serve as evidence of execution response.

[0029] The construction process of the aforementioned local Petri nets is a well-known technique in the art, and specific details will not be elaborated further. In practical implementation scenarios, when switching process modes, resetting production lines, or starting a new work order, a reference local identifier is established based on the PLC reset status, the actual status of the equipment controller, the buffer bit occupancy sensor, and the initial confirmation results of the workpiece. This identifier is used to maintain the reference baseline for the current state under normal circumstances. Subsequently, the reference local identifier is only updated based on the normal transition if there are no confirmed local state residuals, state conflicts, or unconfirmed critical states in the current normal transition during the previous diagnostic cycle. For example: When a new work order begins, the PLC reset status confirms no faults, the buffer occupancy sensor shows the current buffer is empty, and the initial workpiece confirmation result confirms the workpiece is located at the output position of the processing module, waiting for AGV transfer. At this time, the established reference local identifiers are: Location A (workpiece at output position): 1; Location B (AGV in idle state): 1; Location C (AGV carrying workpiece): 0. Here, 1 indicates true, and 0 indicates false. The final reference local identifier m = [1, 1, 0]^T, where ^T indicates transpose. This reference local identifier indicates that under normal circumstances, the workpiece should be at the output position, the AGV should be idle, and the AGV should not be carrying the workpiece. When the AGV successfully completes the picking action and confirms that the workpiece has left the output position and the AGV is carrying the workpiece, the system confirms that the current normal transition (AGV picking) has no local state residuals, no state conflicts, and that the critical state has been confirmed. At this time, based on the local state change vector corresponding to this normal transition = [-1, -1, +1]^T (workpiece leaves output position, AGV is no longer idle, AGV carries workpiece), the reference local identifier is updated to: m + =m+t=[1,1,0]^T+[-1,-1,+1]^T=[0,0,1]^T. If, after the AGV completes its picking-up action, the sensor feedback indicates the workpiece is still in the output position (actual state 1), the AGV is in an occupied state (actual state 0), but the AGV is not carrying a workpiece (actual state 0), meaning the actual local identifier is [1,0,0]^T. This indicates a residual between the actual local identifier and the reference local identifier for this event. Therefore, the reference local identifier should not be updated to allow it to enter the next normal transition.

[0030] Thus, each local diagnostic module contains a set of local Petri nets that can describe the workpiece status, resource status, and standard production actions in this area, as well as corresponding multi-source heterogeneous data mapping relationships and reference local identifiers, forming the basic data environment for subsequent residual calculation and fault diagnosis.

[0031] It should be noted that, in order for multi-source heterogeneous data to be interpreted as evidence of workpiece status, resource status, or normal transitions, a mapping relationship between data sources, status evidence, and normal transitions is established. Workpiece position is confirmed by vision, RFID, barcode, photoelectric, or position sensors; resource status is confirmed by resource controllers, buffer position sensors, and resource lock records; PLC control outputs and task issuance serve as candidate observations for initiation; current, pressure, torque, speed, and vibration serve as evidence of execution response.

[0032] The specific collection cycle can be set according to the equipment response speed. For example, the status of PLC and equipment controller can be collected in 50 milliseconds, and the status of AGV can be reported in 100 milliseconds. When the confidence level of visual recognition is not less than 0.90, it is considered valid evidence. When RFID reads the same workpiece mark twice in a row within 100 milliseconds, it is considered valid evidence.

[0033] As can be seen from the above description of constructing a local Petri net, the "normal transition" described in the embodiments of the present invention is actually a type, which can be regarded as a certain execution action in the process flow of an automated production line. In actual implementation scenarios, under the current normal transition, there will be multiple events, such as multiple workpieces simultaneously or about to perform the same action (such as the AGV transfer and carrying action in the above example). Therefore, in the actual scenario, in the current production line process, it is necessary to determine all transition events under the current normal transition. In subsequent steps, each transition event is analyzed as a separate individual to determine whether it is a source of failure.

[0034] Preferably, in some implementations of the embodiments of the present invention, considering that normal transitions are a type of process flow, they have a fixed maximum duration set in the process flow. In specific implementations, the historical execution records of the last 50 times with the same state under the same process mode and the same equipment configuration can be called, and 1.2 times the average value can be taken as the upper limit of the maximum duration. When the historical records are insufficient, the MES cycle time setting value or the PLC sequential control timeout setting value can be used. For example, the historical average value of the transfer action is 8 seconds, and the upper limit of the maximum duration is 9.6 seconds. Therefore, when dividing transition events, for each event, after determining the event start point, i.e., the anchor point time, the range of each transition event can be determined according to the maximum duration of the event, specifically including: For each normal transition, when the identifier of the preceding repository meets the execution conditions, the various startup data fields corresponding to the normal transition are statistically analyzed, and the time corresponding to the earliest executed startup data field is taken as the anchor time of the transition event.

[0035] The startup data fields include the rising edge of the PLC control output, the switching of the robot program execution state, the switching of the automated guided vehicle task execution state, the switching of the machine tool processing startup state, and the first change in the shared resource occupancy state. The observation records of the state switching that occur in the above startup data fields within the current acquisition period are recorded as the transition startup candidate observation records of the normal transition. Then, these transition startup candidate observation records are counted, and the time corresponding to the earliest executed startup data field is taken as the anchor time of the transition event.

[0036] It should be noted that the aforementioned state setting of "when the identifier of the pre-position warehouse meets the execution conditions" indicates that event anchor points are only established for normal transitions that are currently likely to be executed. For example, when the automated guided vehicle is in an occupied state, it means that the identifier of the pre-position warehouse does not meet the execution conditions. Even if the processing module PLC outputs a transfer request again, the transfer request will not be directly attributed to the new automated guided vehicle receiving the workpiece transition, that is, no anchor point moment is generated. This may be because an anomaly occurred in the upstream transition event (in this case, according to the diagnostic process of the present invention, after the upstream transition event is executed, a local execution anomaly diagnosis result will be fed back, indicating that the event has generated an anomaly such as a jam in its own workflow, resulting in no downstream execution action. Therefore, the current normal transition event should not exist as a downstream).

[0037] Starting from the anchor point, if no normal downstream transition occurs within the preset maximum duration, it indicates that the current normal transition may have timed out due to anomalies or other factors. In this case, the period from the anchor point to the end of the maximum duration is defined as the range of transition events. This setting informs the diagnostic system that if a timeout anomaly occurs, it will no longer wait for the occurrence of downstream events and will directly provide diagnostic feedback. If a normal downstream transition occurs, it indicates that the transition can be properly defined, and the period from the anchor point to the time when the normal downstream transition occurs is defined as the range of transition events.

[0038] After defining the scope of the transition events, the data such as the pre-state evidence, initiation candidate observations, execution response evidence, post-state evidence, and boundary resource evidence within that scope need to be recorded synchronously as transition evidence for analysis and processing in subsequent steps.

[0039] Step S2: For each transition event, determine the conflict state of the place identifier and generate a state conflict identifier; before and after the event anchor point occurs, compare the actual local identifier and the reference local identifier of each place, and calculate the residual increment before and after the event anchor point occurs; if there is a residual increment, the corresponding transition event is taken as a candidate abnormal transition event and a state correction vector is calculated; the state correction vector is composed of residual increments, and the identifiers corresponding to the elements have a common correlation; based on the state correction vector, calculate the boundary influence vector corresponding to the boundary place between the downstream normal transitions of the current normal transition.

[0040] After dividing each transition event, it is determined that the transition event has been completed in the actual scenario, and it is necessary to retrieve the transition evidence for analysis. First, this embodiment of the invention determines whether there are conflicting states in the evidence. Because in automated production lines, the status of the same storage location is usually observed by multiple types of data sources. For example, whether a workpiece is located at a certain workstation can be confirmed by various data sources such as visual recognition results, RFID reading results, photoelectric sensor signals, and position sensor signals. However, different data sources have different sampling periods (e.g., PLC status is collected in 50ms cycles, while visual recognition is processed in frames), different transmission delays (e.g., wired sensors are at the millisecond level, while visual data transmitted via Ethernet may have a delay of tens of milliseconds), and different validity conditions (e.g., visual recognition is not accepted when the confidence level is below 0.90). These factors lead to different judgments from different data sources regarding the same storage location status within the same transition event. For example, a photoelectric sensor may falsely report "workpiece in place" due to reflections on the workpiece surface, while the visual recognition system correctly judges "workpiece not in place," thus creating a conflict. This conflict state is crucial data for subsequent fault diagnosis; therefore, a state conflict identifier can be generated if a conflict state occurs. In a specific implementation of this invention, if a conflict state exists, the state conflict identifier can be set to 1; otherwise, it can be set to 0. This identifier can be used in subsequent steps to distinguish between two situations: a genuine physical anomaly and a contradiction within the data source itself.

[0041] Next, for each transition event, before and after the event anchor point occurs, the actual local identifiers and reference local identifiers of each location are compared, and the residual increment before and after the event anchor point is calculated. The above comparison process includes two steps: before the event anchor point and after the event anchor point. Specifically: Before the event anchor point occurs, that is, at the instant before the current normal transition begins execution, the actual local identifier at this moment of reconstruction is compared with the reference local identifier to obtain the state residual before the anchor point. The significance of this comparison is to identify whether there are abnormal states in the system before the current normal transition starts: due to the sequential dependencies between the actions of the production line, the preceding state of the current transition may have deviated in the upstream links. If there are abnormalities before the action starts, these abnormalities should be attributed to the upstream links and should not be included in the execution result of the current transition.

[0042] After the event anchor point occurs, once the current normal transition is completed or terminated, the actual local identifier at this point is compared with the reference local identifier to obtain the post-anchor state residual. The reference local identifier represents the state that the current normal transition should reach under normal conditions. The significance of this comparison is to assess whether the system state has changed as expected after the current normal transition is completed: if the actual state matches the expected state, it indicates that the current transition is executing normally; if there is a deviation, it indicates that an anomaly occurred during the execution of the current transition.

[0043] Further statistical analysis was conducted to determine the residual increment before and after the event anchor point occurred. This residual increment represents the new state deviation generated during the current normal transition execution, that is, the state abnormality caused by the current action itself after removing the abnormalities that existed before the action started.

[0044] Taking the AGV picking action as an example, we focus on two states: the workpiece is located at the output position (warehouse A) and the AGV is carrying the workpiece (warehouse B). If the reference local identifier before the anchor point is [1,0]^T (workpiece is at the output position, AGV is not carrying the workpiece), and the actual local identifier matches it, then the residual before the anchor point is [0,0]^T, indicating that there is no abnormality before the action starts; if the output position is actually confirmed to be empty, then the residual before the anchor point is [-1,0]^T, indicating that there is a workpiece missing abnormality in the upstream link. After the action is completed, if the actual state is [0,1]^T (workpiece has been moved out, AGV has been carried), which matches the expected local identifier [0,1]^T, then the residual after the anchor point is [0,0]^T, and the residual increment is also [0,0]^T, indicating that the current transition is executed normally, and no candidate abnormal transition is constructed. If the actual state is [1,0]^T (the workpiece is still in the output position and the AGV does not carry the workpiece), then the residual after the anchor point is [+1,-1]^T, and the residual increment is also [+1,-1]^T, indicating that the current transition is abnormal: where the positive number (+1) indicates that the workpiece is extraly retained in the output position, and the negative number (-1) indicates that the AGV carrying the workpiece is not formed as expected. The two together constitute the state correction vector of the candidate abnormal transition.

[0045] If there is an upstream workpiece missing anomaly before the anchor point (residual of [-1,0]^T), and the actual state after the action is completed is [0,0]^T (AGV failed to carry the workpiece), then the residual after the anchor point is [0,-1]^T (workpiece missing state maintained, carrying state missing), and the residual increment is [+1,-1]^T. This indicates that the current transition has worsened the single upstream missing anomaly into a combined anomaly of "workpiece missing and carrying not formed" during the execution process. This residual increment will be recorded as the state correction vector of the candidate anomaly transition for subsequent attribution analysis.

[0046] Therefore, if the final residual increment is all zeros, it indicates that no new state deviation was generated during the current normal transition execution, and no candidate abnormal transition event is constructed. If a residual increment exists, the current transition event is a candidate abnormal transition event, and a state correction vector needs to be calculated. This state correction vector is composed of residual increments. Considering that the places corresponding to the residual increments may not be related to each other, and that the current candidate abnormal transition event may contain different and unrelated vehicle identifiers, the identifiers corresponding to each element in the state correction vector should have a common correlation.

[0047] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the residual increment includes: For a transition event, within the same repository, the actual local identifier before the event anchor point is subtracted from the reference local identifier before the event anchor point to obtain the residual before the anchor point; the actual local identifier after the event anchor point is subtracted from the reference local identifier after the event anchor point to obtain the residual after the anchor point; the residual increment is obtained by subtracting the residual before the anchor point from the residual after the anchor point.

[0048] Preferably, in some implementations of the embodiments of the present invention, the statistical method for the state correction vector includes: The residual increments corresponding to the preceding, subsequent, or directly related resource repositories that belong to the current normal changes are statistically analyzed to obtain the first set; For any two residual increments in the first set, if the state of the same place corresponding to the two is confirmed as opposite by at least two different data sources within the same transition observation window, then the two residual increments are determined to have an observation conflict, and the residual increments involved are split into different conflict subsets according to different observation branches; within each conflict subset, residual increments with the same workpiece identifier, carrier identifier, fixture identifier or shared resource identifier are grouped into the same residual association group; For residual increments that do not have observation conflicts, residual increments with the same workpiece identifier, carrier identifier, fixture identifier, or shared resource identifier are grouped into the same residual association group. All residual increments in each residual association group constitute a state correction vector.

[0049] Taking the normal transition of the machining module from the machining position to the output position as an example, this illustrates how the residual increment is divided into different residual association groups, thus forming the state correction vector: Assuming that after the transition event is completed, the following residual increments can be calculated: Workpiece at the processing station: residual increment is +1, associated workpiece identifier WK-001, this library belongs to the preceding library of the current normal transition; Workpiece at the output station: residual increment is -1, associated workpiece identifier WK-001, associated handover identifier J01, this library belongs to the following library of the current normal transition; Fixture holding status: residual increment is -1, associated resource identifier F01, this library belongs to the directly associated resource library of the current normal transition; Workpiece at the next station: residual increment is -1, associated workpiece identifier WK-002, this library belongs to the following library of the current normal transition; Conveyor belt status: residual increment is +1, associated resource identifier C01, this library does not belong to the preceding, following, or directly associated resource library of the current normal transition.

[0050] The residual increments "Workpiece at processing position +1" and "Workpiece at output position -1" are associated with the same workpiece identifier WK-001, and respectively correspond to the preceding and following storage locations of the current normal transition. There is a direct causal relationship between them, and they are grouped into the same residual association group. The fixture corresponding to the residual increment "Fixture holding status -1" is a shared resource that needs to be called during the execution of the current normal transition. This fixture belongs to the directly associated resource location of the current normal transition and shares the same normal transition as the aforementioned two residual increments, and is grouped into the same residual association group. The residual increment "Workpiece at next station -1" is associated with workpiece identifier WK-002. Unlike the workpiece identifiers associated with the aforementioned residual increments, this workpiece is not the workpiece object processed by the current normal transition; it belongs to another independent event and is not grouped into the aforementioned residual association group. A new residual association group is established separately. The conveyor belt corresponding to the residual increment "conveyor belt status +1" is not the preceding, following, or directly associated resource warehouse of the current normal change. This warehouse has no direct causal relationship with the current normal change and is not part of the residual association group, nor is it used to construct the state correction vector. If the residual increment of "workpiece at output position" has an observation conflict within the same observation window, i.e., visual recognition confirms that there is a workpiece at the output position (residual +1), while the photoelectric sensor confirms that the output position is empty (residual -1), then groups are formed according to different observation branches: In the branch based on the visual confirmation result, the residual increment is +1, and it forms the first residual association group with other residual increments that have the same workpiece identification and normal change; In the branch based on the photoelectric sensor confirmation result, the residual increment is -1, and it forms the second residual association group with other residual increments that have the same workpiece identification and normal change; The two branches independently generate candidate abnormal changes and enter the subsequent matching stage. Ultimately, all residual increments in each residual association group constitute a state correction vector, which is used for the construction of subsequent candidate anomaly transitions and cross-module attribution analysis.

[0051] In step S1, to compare the same shared resource or handover position in different local diagnostic modules, a unique resource identifier is assigned to the shared resource, and a unique handover identifier is assigned to the output, handover, and input bits between local diagnostic modules. The locations in the local Petri net corresponding to the resource identifier or handover identifier are recorded as boundary locations. A boundary location represents the interaction interface between a local diagnostic module and an external module for workpiece handover or shared resource invocation. Specifically, boundary locations include the following two categories: 1. Boundary locations corresponding to handover identifiers, such as the output bit of this module, the handover bit shared with adjacent modules, and the input bit of adjacent modules, used to characterize the flow status of workpieces between different local diagnostic modules; 2. Boundary locations corresponding to resource identifiers, such as AGV occupancy status, fixture clamping status, and buffer bit occupancy status, used to characterize the idle or occupied status of shared resources shared by multiple local diagnostic modules.

[0052] This invention further extracts a boundary influence vector. The physical meaning of this boundary influence vector is to compress the impact of candidate abnormal transitions on the system state from all libraries to a dimension that only includes module boundaries, thereby quantifying the degree and direction of the impact of the abnormality on adjacent modules or shared resources. For example, when a component of the boundary influence vector is positive, it indicates that the state corresponding to the boundary library is additionally retained or appears relative to the normal expectation. For example, a workpiece's output position, which should be empty, is abnormally retained, or a shared resource, which should be released, is abnormally occupied. When a component of the boundary influence vector is negative, it indicates that the state corresponding to the boundary library has not formed or is abnormally missing relative to the normal expectation. For example, a workpiece's input position, which should be reached, is not detected, or a shared resource, which should be occupied, remains idle. When a component of the boundary influence vector is zero, it indicates that the candidate abnormal transition has no impact on the corresponding boundary library, and the state of the boundary library is consistent with the expectation.

[0053] In the subsequent diagnostic process, the boundary influence vector is the core data interface for cross-module attribution analysis: if the boundary influence vector is all zero, it means that the candidate anomaly only affects the internal state of this local diagnostic module and belongs to the local execution anomaly. The diagnostic result can be output directly without waiting for data from other modules. If the boundary influence vector has a non-zero component, it means that the anomaly has affected the module boundary or shared resources. The candidate anomaly transition and its boundary influence vector need to be uploaded to the cache of the coordination module and wait to be matched with the candidate anomaly transition of the adjacent modules to further determine whether the anomaly is propagating upstream, downstream, or a shared resource anomaly.

[0054] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the boundary influence vector includes: There is a module boundary library extracted matrix between the local Petri nets of normal transitions between adjacent upstream and downstream. The element corresponding to the boundary library in the module boundary library extracted matrix has a value of 1, and the other elements have a value of 0. Based on the module boundary library extracted matrix, the residual increment value corresponding to the boundary library is obtained by querying the state correction vector and a boundary influence vector is constructed.

[0055] Step S3: Match the candidate abnormal transition events of the current normal transition with the candidate abnormal transition events of each downstream normal transition. Determine the boundary matching value based on the consistency of the residual increment direction between the matched events. Determine the attribution score of the candidate abnormal transition events of the current normal transition based on the boundary matching value and the time difference between the matched events. Output the diagnostic results based on the boundary influence vector, attribution score, and state conflict identifier.

[0056] Local diagnostic modules can only observe the status of workpieces and resources within their own scope. Taking the example of a downstream workpiece not being received, the anomaly could be caused by the local receiving mechanism, upstream release actions, shared automated guided vehicles, shared buffer bits, or observation equipment. Therefore, the boundary impact of candidate anomaly transitions should be compared based on the same resource identifier and handover identifier. That is, it is necessary to analyze whether a candidate anomaly transition event will affect the normal transition downstream, thus leading to the formation of a candidate anomaly transition in the normal downstream transition, based on the upstream and downstream relationships between various execution actions on the automated production line.

[0057] The specific diagnostic system of this invention includes a coordination module, which is a cross-module attribution calculation unit deployed in a production line server, industrial computer, or edge computing gateway. This module receives candidate abnormal transition summary information uploaded by each local diagnostic module and groups them into attribution groups based on the same workpiece identifier, shared resource identifier, or handover identifier. The coordination module receives the candidate abnormal transition identifier, workpiece identifier, shared resource identifier or handover identifier, event anchor time, boundary influence vector, boundary state residual vector, state conflict identifier, and supporting evidence source identifier of the candidate abnormal transition event of the current normal transition and stores them in a cache. When the boundary influence vector of the candidate abnormal transition event is all 0, clearly indicating that it will not have downstream influence, the diagnostic result can be directly output. When boundary influence vector information exists, it is necessary to wait in the cache for the generation of downstream normal transition candidate abnormal transition events and output the diagnostic result based on the generation result.

[0058] The candidate abnormal transition identifier is used to uniquely identify a candidate abnormal transition generated by a local diagnostic module during a normal transition execution. The candidate abnormal transition identifier can be generated by combining the local diagnostic module number, the current normal transition number, the event anchor time, and the residual association group number.

[0059] The supporting evidence source identifier is used to record the data source type and fields on which candidate abnormal transitions, residual increments, or state conflicts are based. Specifically, when the local diagnosis module generates transition initiation candidate observations, local state residuals, or state conflict identifiers, the data source type, device number, data field name, and observation time involved in the determination are combined into the supporting evidence source identifier.

[0060] Within the coordination module's buffer, candidate abnormal transition events of the current normal transition can be matched with candidate abnormal transition events of each downstream normal transition. The boundary matching value is determined based on the consistency of the residual increment direction between the matched events. If an upstream candidate abnormality causes a missing state at a certain boundary place (e.g., a workpiece not pushed to the output position), this missing state will propagate downstream, causing downstream candidate abnormalities to exhibit the same missing state at the same junction position (e.g., a workpiece not detected at the downstream input position). In this case, the state change directions of the two at the boundary are consistent, indicating that the upstream abnormality is the cause of the downstream abnormality, and the two belong to the same causal chain. Conversely, if an upstream abnormality causes an redundant state at a certain boundary place (e.g., a workpiece is repeatedly pushed in), but the downstream does not exhibit a corresponding redundant state or shows the opposite state, the directions of the two are inconsistent, indicating that there is no direct causal relationship between them. By analyzing the consistency of directions, the causal relationship between candidate abnormalities can be effectively identified, avoiding the incorrect attribution of unrelated abnormal events.

[0061] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the boundary matching value includes: For any set of matching events, the boundary influence vector of the candidate abnormal transition event of the current normal transition is used as the first comparison vector, and the boundary influence vector of the candidate abnormal transition event of the downstream normal transition is used as the second comparison vector. Obtain the components in the first comparison vector and the second comparison vector corresponding to the same workpiece identifier, shared resource identifier, or handover identifier. If the non-zero component in the first comparison vector has the same direction of state change as the corresponding component in the second comparison vector, then set the boundary matching value of the matching event group to 1, indicating that the current candidate anomaly event can completely explain the state deviation of the downstream candidate anomaly event, and the two constitute a strong causal relationship. If the non-zero component in the first comparison vector has a partially consistent direction of state change with the corresponding component in the second comparison vector, and there is no corresponding component with the opposite direction, then set the boundary matching value of the matching event group to 0.5, indicating that the current candidate anomaly event can partially explain the downstream state deviation, but there are still deviation components that cannot be explained. If there is a corresponding component in the second comparison vector with the opposite direction of the non-zero component in the first comparison vector, it indicates that there is a contradiction between the current candidate anomaly and the downstream candidate anomaly, and there is no causal relationship between the two. This matching event should be excluded when calculating the attribution score in the subsequent process, and then set the boundary matching value of the matching event group to 0.

[0062] Taking AGV transfer as an example, the candidate abnormal transition boundary influence vector of the current module (processing module) is [-1] (workpiece not pushed to the output position, missing from the boundary library). The candidate abnormal transition boundary influence vector of the downstream module (detection module) is [-1] (workpiece not detected at the input position, missing from the boundary library). Both correspond to the same handover identifier and have completely consistent directions (both are missing). The boundary matching value is 1, indicating that the upstream workpiece not being pushed out is the direct cause of the downstream input position having no workpiece. If the downstream candidate abnormal boundary influence vector is [0] (input position is normal) or [+1] (input position is abnormally occupied), the directions are inconsistent, the boundary matching value is 0, and the causal relationship between the two is excluded.

[0063] By further combining the boundary matching value and the time difference between the matching events, the attribution score of the candidate abnormal transition event for the current normal transition can be determined. A larger time difference indicates that the current candidate abnormal transition event occurred earlier than downstream events, making it more likely to be the source. Simultaneously, if multiple matching events exist, and each matching event has a large boundary matching value, it indicates that the impact of the current candidate abnormal transition event is more severe. Therefore, a higher attribution score indicates that the current candidate abnormal transition event is more likely to be the source of the fault. Finally, an accurate diagnostic result can be output by combining the boundary influence vector, attribution score, and the state conflict identifier.

[0064] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the attribution score includes: For each matching event, the ratio of the time difference to the maximum duration of the current normal transition is used as the time score. Since the time difference is the time difference between the event anchor points of two events, using the maximum duration as the denominator allows for normalization of the time difference, as the numerator, being a related upstream and downstream process, will only be less than or equal to the maximum duration. The initial score for the matching event is obtained based on the boundary matching value and the time score. In a specific implementation of this invention, the time score is added to a positive integer 1 to obtain a scaling factor between 1 and 2. This scaling factor is then multiplied by the boundary matching value as an amplification weight, achieving a gain based on the time difference information to obtain the initial score for the matching event.

[0065] The initial scores of all matching events for the candidate abnormal transition event of the current normal transition are accumulated as the attribution score.

[0066] Preferably, in some implementations of the present invention, the diagnostic result is output based on the boundary influence vector, the attribution score, and the state conflict identifier. This process is essentially a multi-layered decision-making process, specifically including: The first layer of judgment determines whether the candidate abnormal transitions of the current normal transition will have an impact on the downstream: If the boundary influence vector of the candidate abnormal transition event of the current normal transition is a vector of all zeros, it means that the current candidate abnormal transition event will not affect the downstream normal transition, but only affect the internal state of this local diagnostic module. For example, if the workpiece slips during the gripping process of the robot arm, but the workpiece does not block the output bit or affect the shared resources, and does not affect the workpiece flow of other modules, then the matching process will not be executed, and the local execution abnormal result will be output.

[0067] The second layer of judgment, assuming that the current candidate abnormal transition event has a boundary influence vector, further analyzes whether there are downstream candidate abnormal transition events in the buffer for it to participate in matching: If a candidate abnormal transition event for the current normal transition has a boundary influence vector, it indicates that the current candidate abnormal transition event may affect downstream normal transitions or shared resources. For example, if the push rod is stuck, causing the workpiece to remain at the output position (abnormal boundary storage state), it is necessary to determine whether it has a related impact on downstream modules. In this case, the candidate abnormal transition event is stored in the cache area and waits for the matching process to be executed.

[0068] After candidate abnormal transition events with different normal transitions exist in the buffer, a matching process is performed. If no matching event exists within the preset matching time, that is, no downstream candidate abnormal transition event has the same workpiece identifier, shared resource identifier, or handover identifier as the current candidate abnormal transition event, it means that the boundary influence of the current candidate abnormal transition event has failed to trigger the downstream module to generate the corresponding candidate abnormality. At this time, if the candidate abnormal transition event has a state conflict identifier (i.e., the state conflict identifier is 1), it means that the sensor data inside the current module is contradictory. For example, visual recognition confirms that the workpiece is in the output position, but the photoelectric sensor confirms that the output position is empty, and then the observation abnormality result is fed back. If there is no state conflict identifier (i.e., the state conflict identifier is 0), it means that the physical state is indeed abnormal, but the downstream is not affected. For example, the push rod is stuck, causing the workpiece to remain in this station, but the downstream module happens to not start for other reasons, and then a local execution abnormality is fed back. In the specific implementation of this embodiment, the preset matching time can be set to the maximum duration of the downstream normal transition, that is, it essentially waits for the execution of the downstream normal transition, and directly receives the information of the downstream candidate abnormal transition event after the execution is completed or interrupted.

[0069] The third layer of judgment involves, in the case of matching events, statistically analyzing the matching results of each candidate abnormal transition event, quantifying the attribution score, filtering out the fault source events, and providing further diagnostic result feedback: If a matching event exists, the attribution score of each candidate abnormal transition event is obtained. The fault source event is selected based on the attribution score, and the corresponding diagnostic result is fed back based on the identification result of the fault source event.

[0070] Furthermore, the specific events for identifying fault sources based on the attribution scores include: Among all candidate anomaly transition events in the buffer, the two candidate anomaly transition events with the highest attribution scores are analyzed. If the difference in attribution scores between these two events exceeds a preset threshold, it indicates that the highest-scoring candidate anomaly transition event is significantly better than the second-highest-scoring event in explaining downstream anomalies. The difference is significant, and there is no possibility of confusion. In this case, the candidate anomaly transition event with the highest attribution score is taken as the fault source event. It should be noted that because the matching in this embodiment is mandatory with downstream candidate anomaly events, there are no further downstream normal transitions within the buffer. That is, there are two types of normal transitions within the buffer: the current normal transition and downstream normal transitions. Therefore, only candidate anomaly transition events generated by the current normal transition can have their attribution scores calculated, thus explaining downstream anomalies.

[0071] If the difference in attribution scores between the two candidate anomalous transition events with the highest attribution scores is less than or equal to a preset difference threshold, it indicates that the two candidate anomalous transition events are similar in their ability to explain downstream anomalies, and it is impossible to distinguish between primary and secondary events simply by their scores. At this point, it is further determined whether the two candidate abnormal transition events share a common identifier (i.e., whether they have the same workpiece identifier, shared resource identifier, or handover identifier): If any kind of shared identifier exists, it means that although the two candidate abnormal transition events have different sources, they point to the same workpiece, the same shared resource, or the same handover location, and are actually two manifestations of the same root cause event. For example, the "workpiece not ejected" of the processing module and the "workpiece not received" of the detection module both point to the same workpiece identifier and the same handover identifier. In this case, the candidate abnormal transition event with the highest attribution score is taken as the fault source event, and the second highest event is output as an accompanying phenomenon. If no shared identifier exists, it means that the two candidate abnormal transition events point to completely different physical objects and are two independent concurrent events. For example, there is no shared identifier between the processing module push rod jamming (associated with workpiece WK-001) and the detection module camera contamination (associated with camera ID-Cam05), and there is no causal relationship between the two. In this case, the two candidate abnormal transition events are each taken as a fault source event, and the multi-fault source diagnosis result is output. In the specific implementation of this embodiment, the difference threshold can be set to 0.1.

[0072] Furthermore, in this embodiment of the invention, the corresponding diagnostic result is fed back based on the identification result of the fault source event, including: All matching events corresponding to the fault source event are taken as the matching events to be analyzed. A matching event refers to a pair of related events formed between the fault source event and downstream candidate anomaly transition events through the same workpiece identifier, shared resource identifier, or handover identifier. Each matching event records the causal relationship between the two candidate anomalies, the corresponding time difference, boundary matching value, and other information. After determining the matching events to be analyzed, they are classified according to their distribution characteristics and shared identifier types as follows: If, among all the events to be analyzed and matched, the fault source event has downstream candidate abnormal transition events, and all of these downstream candidate abnormal transition events are located in the same adjacent local diagnostic module, and each matching event is sequentially associated through the same workpiece identifier or handover identifier, it indicates that the abnormal impact of the fault source event is transmitted downstream along a single workpiece path or a single handover position. That is, the state deviation generated by this module is captured by the only adjacent module. For example, if the workpiece is stuck at the output position due to the push rod jamming in the processing module, and this output position happens to be the only input source of the detection module, the detection module generates a candidate abnormality of "input position missing" because the precondition is not met. The two events are sequentially associated through the same workpiece identifier and handover identifier. Then, the fault source event is determined to be an upstream propagation abnormality, indicating that the fault is propagated unidirectionally from the current module to the adjacent downstream module.

[0073] If, among all the events to be analyzed and matched, the downstream candidate anomaly transition events matched by the fault source event are distributed across at least two different local diagnostic modules, and each matched event points to the same shared resource identifier, it indicates that the abnormal impact of the fault source event does not propagate along a single workpiece path, but rather affects multiple modules simultaneously through a shared resource (such as the same AGV, the same robot, or the same cache position). For example, if an AGV malfunctions during transport, causing the processing module to report "AGV pickup timeout" due to the AGV's failure to pick up the goods in time, the detection module to report "AGV unloading timeout" due to the AGV's failure to deliver the goods in time, and the cache module to report "AGV transport timeout" due to the AGV's failure to transport the goods in time, and the candidate anomalies of the three modules all point to the same shared resource identifier AGV_A03, then the fault source event is determined to be a shared resource anomaly, indicating that the fault originates from equipment or resources shared by multiple modules.

[0074] If the matching pattern of the fault source event in all the events to be analyzed does not meet any of the above conditions, for example, the matched downstream candidate anomalies are distributed in multiple different modules but the matched events do not share the same resource identifier, nor do they form a single sequential association through the same workpiece or handover identifier, it means that the current anomaly propagation mode cannot be classified into the typical single propagation or shared resource fault category, and belongs to the atypical anomaly situation. At this time, the fault source event is marked as a cross-module anomaly to be confirmed, and all the downstream candidate anomaly transition event information matched with it is output for operation and maintenance personnel to further analyze and locate the problem in combination with the actual situation on site.

[0075] This invention is now complete.

[0076] In summary, this invention divides the automated production line into multiple local diagnostic modules according to the equipment control range. Each local diagnostic module corresponds to a local Petri net, and the standard actions in the local Petri net are defined as normal transitions. Under the current normal transition, the conflict state identified by the database for each transition event is judged, and a state conflict identifier is generated. Before and after the event anchor point, the actual local identifier of each database is compared with the reference local identifier, and the residual increment is calculated. If a residual increment exists, the corresponding transition event is taken as a candidate abnormal transition event, and a state correction vector is constructed from the residual increment. The boundary influence vector is calculated based on the state correction vector. The candidate abnormal transition events are matched with downstream candidate abnormal transition events, and the boundary matching value is determined according to the consistency of the residual increment direction. The attribution score is determined by combining the time difference. Finally, the diagnostic result is output based on the boundary influence vector, the attribution score, and the state conflict identifier. This invention can effectively improve the accuracy of fault source location and enhance the adaptability to unknown fault scenarios.

[0077] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for fault diagnosis of automated production lines based on multi-source heterogeneous data, characterized in that, The method includes: The automated production line is divided into multiple local diagnostic modules according to the equipment control range, and each local diagnostic module corresponds to a local Petri net; the standard action in each local Petri net is defined as the normal transition corresponding to the local Petri net; under the current normal transition, all transition events are counted. For each transition event, determine the conflict state of the location identifier and generate a state conflict identifier; before and after the event anchor point occurs, compare the actual local identifier and the reference local identifier of each location, and calculate the residual increment before and after the event anchor point occurs; if there is a residual increment, the corresponding transition event is taken as a candidate abnormal transition event and a state correction vector is calculated; the state correction vector is composed of residual increments, and the identifiers corresponding to the elements have common correlations; based on the state correction vector, calculate the boundary influence vector corresponding to the boundary location between the current normal transition and the downstream normal transition. The candidate abnormal transition events of the current normal transition are matched with the candidate abnormal transition events of each downstream normal transition. The boundary matching value is determined based on the consistency of the residual increment direction between the matched events. The attribution score of the candidate abnormal transition event of the current normal transition is determined based on the boundary matching value and the time difference between the matched events. The diagnostic result is output based on the boundary influence vector, the attribution score and the state conflict identifier.

2. The automatic production line fault diagnosis method based on multi-source heterogeneous data according to claim 1, characterized in that, The methods for classifying transition events include: For each normal transition, when the identifier of the preceding repository meets the execution conditions, the various startup data fields corresponding to the normal transition are counted, and the time corresponding to the earliest executed startup data field is taken as the anchor time of the transition event. Starting from the anchor point, if no normal downstream transition occurs within the preset maximum duration, the period from the anchor point to the end of the maximum duration is considered the range of transition events; if a normal downstream transition occurs, the period from the anchor point to the moment when the normal downstream transition occurs is considered the range of transition events.

3. The method for fault diagnosis of an automated production line based on multi-source heterogeneous data according to claim 1, characterized in that, The method for obtaining the residual increment includes: For a transition event, within the same repository, the actual local identifier before the event anchor point is subtracted from the reference local identifier before the event anchor point to obtain the residual before the anchor point; the actual local identifier after the event anchor point is subtracted from the reference local identifier after the event anchor point to obtain the residual after the anchor point; the residual increment is obtained by subtracting the residual before the anchor point from the residual after the anchor point.

4. The automatic production line fault diagnosis method based on multi-source heterogeneous data according to claim 1, characterized in that, The statistical methods for the state correction vector include: The residual increments corresponding to the preceding, subsequent, or directly related resource repositories that belong to the current normal changes are statistically analyzed to obtain the first set; For any two residual increments in the first set, if the state of the same place corresponding to the two is confirmed as opposite by at least two different data sources within the same transition observation window, then the two residual increments are determined to have an observation conflict, and the residual increments involved are split into different conflict subsets according to different observation branches; within each conflict subset, residual increments with the same workpiece identifier, carrier identifier, fixture identifier or shared resource identifier are grouped into the same residual association group; For residual increments that do not have observation conflicts, residual increments with the same workpiece identifier, carrier identifier, fixture identifier, or shared resource identifier are grouped into the same residual association group. All residual increments in each residual association group constitute a state correction vector.

5. The method for fault diagnosis of an automated production line based on multi-source heterogeneous data according to claim 1, characterized in that, The method for obtaining the boundary influence vector includes: There is a module boundary library extracted matrix between the local Petri nets of normal transitions between adjacent upstream and downstream. The element corresponding to the boundary library in the module boundary library extracted matrix has a value of 1, and the other elements have a value of 0. Based on the module boundary library extracted matrix, the residual increment value corresponding to the boundary library is obtained by querying the state correction vector and a boundary influence vector is constructed.

6. The method for fault diagnosis of an automated production line based on multi-source heterogeneous data according to claim 1, characterized in that, The method for obtaining the boundary matching value includes: For any set of matching events, the boundary influence vector of the candidate abnormal transition event of the current normal transition is used as the first comparison vector, and the boundary influence vector of the candidate abnormal transition event of the downstream normal transition is used as the second comparison vector. Obtain the components in the first comparison vector and the second comparison vector that correspond to the same workpiece identifier, shared resource identifier, or handover identifier; if the non-zero component in the first comparison vector has the same state change direction as the corresponding component in the second comparison vector, then set the boundary matching value of the matching event group to 1; if the non-zero component in the first comparison vector has the same state change direction as the corresponding component in the second comparison vector, and there is no corresponding component with the opposite direction, then set the boundary matching value of the matching event group to 0.5; if there is a corresponding component in the second comparison vector with the opposite direction to the non-zero component in the first comparison vector, then set the boundary matching value of the matching event group to 0.

7. The method for automatic production line fault diagnosis based on multi-source heterogeneous data according to claim 1, characterized in that, The method for obtaining the attribution score includes: Under each matching event, the ratio of the time difference to the maximum duration of the current normal transition is used as the time score, and the initial score of the matching event is obtained based on the boundary matching value and the time score. The initial scores of all matching events for the candidate abnormal transition event of the current normal transition are accumulated as the attribution score.

8. The method for automatic production line fault diagnosis based on multi-source heterogeneous data according to claim 1, characterized in that, The step of outputting diagnostic results based on the boundary influence vector, attribution score, and state conflict identifier includes: If the boundary influence vector of the candidate abnormal transition event of the current normal transition is a vector of all zeros, then the matching process is not executed, and the local execution abnormal result is output. If a candidate abnormal transition event of the current normal transition has a boundary influence vector, the candidate abnormal transition event is stored in a buffer and awaits the matching process. After candidate abnormal transition events of different normal transitions are found in the buffer, the matching process is performed. If no matching event is found within a preset matching time and a state conflict indicator exists, an observation anomaly result is fed back. If no matching event is found and no state conflict indicator exists, a local execution anomaly is fed back. If a matching event is found, the attribution score of each candidate abnormal transition event is obtained. The fault source event is selected based on the attribution score, and the corresponding diagnostic result is fed back based on the identification result of the fault source event.

9. The method for fault diagnosis of an automated production line based on multi-source heterogeneous data according to claim 8, characterized in that, The step of filtering out fault source events based on the attribution score includes: Among all candidate abnormal transition events in the statistical buffer, the two candidate abnormal transition events with the highest attribution scores are selected. If the difference in attribution scores between the two candidate abnormal transition events with the highest attribution scores is greater than a preset difference threshold, then the candidate abnormal transition event with the highest attribution score is selected as the fault source event. If the difference is less than or equal to the preset difference threshold, it is determined whether the two candidate abnormal transition events share a common identifier. If any kind of shared identifier exists, the candidate abnormal transition event with the highest attribution score is taken as the fault source event; if no shared identifier exists, the two candidate abnormal transition events are taken as fault source events respectively.

10. The method for automatic production line fault diagnosis based on multi-source heterogeneous data according to claim 8, characterized in that, The step of feeding back corresponding diagnostic results based on the identification results of the fault source event includes: All matching events corresponding to the fault source event are taken as matching events to be analyzed; If, among all the events to be analyzed and matched, the fault source event has downstream candidate abnormal transition events, and all the downstream candidate abnormal transition events are located in the same adjacent local diagnostic module, and each matching event is sequentially associated through the same workpiece identifier or handover identifier, then the fault source event is determined to be an upstream propagation abnormality. If, among all the events to be analyzed and matched, the downstream candidate abnormal transition events matched by the fault source event are distributed in at least two different local diagnostic modules, and each matched event points to the same shared resource identifier, then the fault source event is determined to be a shared resource abnormality. In other cases, the fault source event is marked as a cross-module anomaly pending confirmation, and all downstream candidate anomaly transition event information matched with it is output.

Images