A knowledge graph-based shell data management system

Abstract

This invention relates to the field of data management technology, specifically a knowledge graph-based shell casing data management system. The system includes: collecting a raw dataset containing workpiece identification codes, process semantic operators, processing timestamps, tooling entity numbers, and precision deviation feature values; constructing asymmetric node sets using the workpiece identification code and tooling entity number as indices, and generating a workpiece-tooling bipartite graph model by establishing cross-set association edges with multi-dimensional attributes. Based on field semantic comparison logic, the precision deviation feature space is decoupled into an end-face axial deviation sequence and a shell radial deviation sequence; a sliding window operator with a time decay factor is invoked to perform topological cleaning on discrete outliers in the sequences. End-face axial time-varying wear and shell outer diameter time-varying wear features reflecting the physical entity state are extracted from the cleaned sequences. These features are persistently written to a graph database, constructing a knowledge graph structure with spatiotemporal correlation characteristics.

Description

Technical Field

[0001] This invention relates to the field of data management technology, specifically to a knowledge graph-based shell casing data management system. Background Technology

[0002] As manufacturing processes become increasingly data-driven, the amount of heterogeneous data generated on production lines is growing exponentially. However, existing production data organization methods are primarily based on the statistical granularity of "process batches," rather than the individual granularity of "logical entities." The processing parameters and measurement results generated by various control systems are physically stored as scattered, unstructured data silos. The lack of a unified data structure that can dynamically correlate individual workpiece UIDs with tooling fixture numbers and accuracy deviations across physical storage levels prevents the computing system from constructing complete logical links when performing global data retrieval. This fragmentation of the data topology significantly reduces the computational efficiency and accuracy of quality attribution analysis algorithms due to the lack of necessary associative feature inputs.

[0003] In existing data modeling logic, the state attributes of fixture entities are treated as single scalars, ignoring the heteroscedasticity of the physical feature loss caused by machining tasks. The force components generated by bottom edge machining and shell outer diameter machining are not decoupled at the data level, causing logically independent axial and radial accuracy evolution features to collapse and be mapped to the same feature dimension. This excessive coupling of the feature space prevents the data processing system from performing multi-dimensional analysis of the differentiated health of fixture entities, leading to serious rejection errors or missed detection risks in data evaluation models when handling edge conditions (such as excessive wear in a single direction). Due to the opaque nature of the data structure-level correlations, scheduling algorithms cannot call upon the multi-dimensional accuracy margins of each fixture entity in real time when handling resource allocation tasks. This not only leads to the loss of value of data resources in the decision-making process but also prevents the production line from achieving refined predictive maintenance of the remaining value of fixtures through data-driven methods, limiting the global optimization capabilities of data-driven manufacturing systems.

[0004] To achieve a unified data structure that associates the number of a single workpiece, tooling fixture, and accuracy deviation value, and to distinguish between the accuracy deviations caused by two types of machining operations—flattening the bottom edge end face and precision turning the outer diameter of the shell—a shell shell data management system based on a knowledge graph is proposed. Summary of the Invention

[0005] This invention aims to provide a knowledge graph-based shell casing data management system. By constructing a bipartite graph data structure with workpiece identification code and tooling fixture number as dual-node sets, and combining a force decoupling data classification mechanism for two processing scenarios (bottom edge end face and shell outer diameter) with a sliding window wear feature extraction algorithm with time decay function, it achieves independent quantitative modeling and graph-based associated storage of the two-dimensional time-varying wear degree of the fixture, thereby supporting fine scheduling decisions of tooling fixtures based on wear state differences.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A knowledge graph-based shell casing data management system includes: Data Acquisition Module: Acquires the workpiece identification code, process actions, processing timestamp, and tooling fixture numbers passed through the workpiece; and measures and inspects the target surface after the workpiece is processed to obtain the accuracy deviation value. Bipartite graph modeling module: Establish cross-set edges with the workpiece identification code as the left node set and the tooling fixture number as the right node set, and combine the process action, the processing timestamp and the accuracy deviation value as the attributes of the edges to obtain the workpiece-fixture bipartite graph; Fixture wear calculation module: For each tooling fixture number corresponding to the right node, extract the precision deviation value on the connected edge and sort it according to the processing timestamp. Apply a sliding window with time decay function to perform weighted statistical calculation on the sorted precision deviation value, separate random processing error and gradual wear characteristics, and obtain the time-varying wear of the fixture. Graph database storage module: Writes the workpiece-fixture bipartite graph into the graph data storage, and stores the time-varying wear degree of the fixture as the dynamic attribute of the corresponding right node to form an associated knowledge graph.

[0007] Preferably, the acquisition module is specifically used for: The data acquisition process monitors the sequential flow of the workpiece among multiple tooling fixtures, extracting the process actions, processing timestamps, and corresponding tooling fixture numbers. The process actions include axial and radial process actions. Additionally, tool change records for each tooling fixture station are collected. After the workpiece completes processing at any tooling fixture station, the dimensional data of the target surface is collected, and the difference is calculated based on the pre-stored nominal dimensions to obtain the accuracy deviation value. Using two adjacent tool change records as segment boundaries, the accuracy deviation value is divided into the corresponding tool usage cycle data segments.

[0008] Preferably, the bipartite graph modeling module is specifically used for: All workpiece identification codes and tooling fixture numbers are deduplicated to obtain a left node set and a right node set. Based on the actual workstation flow matching relationship in the physical workshop, cross-set edges are established between the left node set and the right node set. The process actions, processing timestamps and accuracy deviation values ​​recorded under the same processing operation are added to the attributes of the cross-set edges to obtain the workpiece-fixture bipartite graph.

[0009] Preferably, the fixture wear calculation module includes: Extract all cross-set edges that are connected to each right node in the set of right nodes one by one, read the process action, the processing timestamp and the precision deviation value in the edge attributes, compare each field one by one, and put the data records whose process action matches the axial process action into the first storage queue, and put the data records whose process action matches the radial process action into the second storage queue. Within the tool usage cycle data segment, the accuracy deviation values ​​in the first storage queue and the second storage queue are arranged in ascending order according to the machining timestamp, so that the accuracy deviation values ​​generated by the axial machining operation and the radial machining operation are stored independently at the data level, respectively obtaining the bottom edge end face deviation data sequence and the housing radial deviation data sequence.

[0010] Preferably, the fixture wear calculation module further includes: The following steps are performed on the bottom edge end face deviation data sequence and the shell radial deviation data sequence: a sliding calculation interval is set on the deviation data sequence, and data points within the interval are extracted as the current calculation window data; a dynamic weight factor set is assigned to each data point in the current calculation window data through a time decay function, with the weight increasing as it is closer to the current time; a weighted average deviation value and a weighted variance value are calculated based on the dynamic weight factor set; data points whose difference from the weighted average deviation value exceeds the tolerance threshold set based on the weighted variance value are identified as discrete abrupt values ​​and are removed to obtain a smooth and effective dataset; After removing the weights of discrete abrupt change points from the dynamic weight factor set corresponding to the retained data points in the smoothed effective dataset, normalize and recalculate to obtain a corrected weight sequence; multiply and accumulate the values ​​in the smoothed effective dataset with the corrected weight sequence point by point to obtain the wear trend feature value; use the wear trend feature value obtained from the bottom edge end face deviation data sequence as the time-varying wear degree of the bottom edge end face, and use the wear trend feature value obtained from the shell radial deviation data sequence as the time-varying wear degree of the shell outer diameter.

[0011] Preferably, the graph database storage module is specifically used for: For each tooling fixture number corresponding to the right node, a time-series attribute linked list is established with the processing batch time as the index dimension. The time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are received and loaded into the corresponding time-series attribute linked list in the form of incremental snapshots to complete the traceability and storage of the fixture wear data trajectory. A non-blocking writing mechanism is adopted, and the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are treated as two independent time series dimensions and asynchronously appended to the preset dynamic attribute slot of the right node; the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are respectively used as the accuracy influence weights under the corresponding processing scenarios and backfilled into the corresponding attribute fields of the process action matching in the cross-set connection, thus forming the associated knowledge graph.

[0012] Preferably, the system further includes a production management decision module, specifically used for: The time-series attribute linked list of each right node in the associated knowledge graph is accessed periodically to read the latest time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell. The deterioration ratio of the two relative to the preset single wear threshold is calculated respectively, and the advantageous processing dimension corresponding to the right node is determined based on the lower deterioration ratio. Obtain the current process action requirements of the workpiece to be transferred. When the process action requirement is the axial process action, filter the set of fixtures whose time-varying wear degree of the bottom edge end face has not yet reached the single wear threshold. When the process action requirement is the radial process action, filter the set of fixtures whose time-varying wear degree of the outer diameter of the housing has not yet reached the single wear threshold. Based on the aforementioned advantageous processing dimensions, a path allocation is performed. For a single fixture that exceeds the single wear threshold in only one of the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell, the remaining available processing scenarios are determined. Based on the determination result, a tooling and fixture scheduling instruction is generated to preferentially allocate the single fixture that exceeds the threshold to the process scenario corresponding to the wear degree that does not exceed the threshold, and output to the production management terminal.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing left and right node sets using workpiece identification codes and tooling fixture numbers respectively, and using process actions and accuracy deviation values ​​as connection attributes, a bipartite graph data structure is created. This allows the many-to-many mapping relationship between a single workpiece and its complete processing history to be queried and reconstructed in the graph database. Compared to the existing technology's batch-level distributed recording method, when any workpiece or fixture exhibits a quality anomaly, the system can directly locate all historical processing records along the connection attributes. This transforms the reconstruction of the quality traceability path from manual comparison to graph query operation, eliminating the risk of traceability link breaks due to human oversight.

[0014] 2. By employing a data domain decoupling mechanism based on process action field comparison, the precision deviation values ​​generated by the bottom edge end face machining action and the housing outer diameter precision machining action are respectively collected into independent time series. After removing discrete and abrupt values ​​through a sliding window with a time decay function, the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the housing outer diameter are output independently. Compared with the existing technology that treats the fixture as a whole for unified wear assessment, this invention achieves independent quantification of the wear degree of the axial positioning surface and the radial positioning surface of the same fixture, avoiding premature overall scrapping of the fixture due to the inability to distinguish the wear differences in the two dimensions.

[0015] 3. The attributed edges constructed by the bipartite graph modeling module provide a multidimensional data foundation for the fixture wear calculation module, which can be accurately retrieved by field. The two-dimensional time-varying wear degree extracted by the fixture wear calculation module is written into the associated knowledge graph in the form of time-series attribute linked lists and accuracy influence weights by the graph database storage module, so that the graph has both wear trajectory tracing and accuracy association query capabilities. On this basis, the production management decision module calculates the two-dimensional deterioration ratio, filters the set of available fixtures, and generates targeted scheduling instructions, transforming the wear-accuracy association relationship stored in the graph into an executable production decision. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a knowledge graph-based shell casing data management system according to the present invention. Figure 2 This is a schematic diagram of the structure of a knowledge graph-based shell casing data management system according to the present invention. Figure 3 This is a schematic flowchart of the wear calculation and decision-making process of the present invention. Detailed Implementation

[0017] 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.

[0018] Please see Figures 1 to 3 This invention provides a knowledge graph-based shell casing data management system, specifically referring to... Figure 1 Step-by-step flowchart Figure 2 Structural diagram and Figure 3 The flowchart for wear degree calculation and decision-making is shown below, and the technical solution is as follows: Data Acquisition Module: Acquires the workpiece identification code, process actions, processing timestamp, and tooling fixture numbers passed through the workpiece; and measures and inspects the target surface after the workpiece is processed to obtain the accuracy deviation value. Bipartite graph modeling module: Establish cross-set edges with the workpiece identification code as the left node set and the tooling fixture number as the right node set, and combine the process action, the processing timestamp and the accuracy deviation value as the attributes of the edges to obtain the workpiece-fixture bipartite graph; Fixture wear calculation module: For each tooling fixture number corresponding to the right node, extract the precision deviation value on the connected edge and sort it according to the processing timestamp. Apply a sliding window with time decay function to perform weighted statistical calculation on the sorted precision deviation value, separate random processing error and gradual wear characteristics, and obtain the time-varying wear of the fixture. Graph database storage module: Writes the workpiece-fixture bipartite graph into the graph data storage, and stores the time-varying wear degree of the fixture as the dynamic attribute of the corresponding right node to form an associated knowledge graph.

[0019] Example 1: This example takes the multi-source heterogeneous data generated by the flow of workpieces between multiple sets of tooling fixtures during the finishing process as the processing object. Through the collaborative operation of the acquisition module, bipartite graph modeling module, fixture wear calculation module, graph database storage module and production management decision module, a workpiece-fixture bipartite graph with workpiece identification code and tooling fixture number as the two-node set is constructed. The accuracy deviation data generated by the two types of processing scenarios, bottom edge end face and shell outer diameter, are decoupled into independent time series. The two-dimensional time-varying wear degree is extracted by time decay sliding window and normalized weight correction. The associated knowledge graph is used as a carrier to support the refined scheduling decision of the remaining use value of the fixture.

[0020] First, the acquisition module is specifically used for: reading the workpiece identification code of the workpiece through an identification reading device deployed before the workpiece enters the finishing process; monitoring the data acquisition process of the workpiece flowing sequentially between multiple sets of tooling fixtures, and extracting the process actions, the processing timestamp, and the corresponding tooling fixture number from the data acquisition interface of each tooling fixture station; the process actions include axial process actions and radial process actions; acquiring the tool change record of each tooling fixture station through a dimensional data acquisition device deployed on each tooling fixture station, the tool change record being automatically generated and written by the data acquisition interface each time a tool change is completed; after the workpiece completes the processing operation at any tooling fixture station, acquiring the dimensional data of the detection target surface through the dimensional data acquisition device, and calculating the difference based on the nominal size pre-stored in the system database to obtain the accuracy deviation value; dividing the accuracy deviation value into the corresponding tool usage cycle data segment using two adjacent tool change records as the segment boundary.

[0021] Specifically, an identification reading device is deployed at the entrance station of the shell casing finishing process. When the workpiece arrival signal is triggered, the identification reading device automatically scans the engraved code on the bottom flange of the workpiece and writes the reading result as the workpiece identification code into the workpiece flow record table of the system database. Each workpiece corresponds to a unique record, and all data collected at each subsequent station are associated and stored with the workpiece identification code as the primary key.

[0022] Each tooling and fixture station's data acquisition interface is connected to the lathe control system via an industrial Ethernet. After the workpiece completes the current station's machining action and the spindle stops, the data acquisition interface automatically reads the currently executed process action type field, the machining timestamp generated by the system clock, and the tooling fixture number corresponding to the station. These three fields are written into the flow details table of the system database in the form of structured records, and each record is associated with the workpiece identification code. The value of the process action type field is predefined by the process route configuration file. Axial process actions correspond to bottom edge end face turning operations, and radial process actions correspond to housing outer diameter precision turning operations. The field value is uniformly updated by process personnel in the system management interface when the process route changes.

[0023] The dimensional data acquisition device is deployed on the side of the machine tool spindle at each tooling fixture station. After each tool change operation, the device is triggered by the operator pressing the confirmation button to write a signal. Upon receiving the signal, the system database automatically generates a tool change event record in the tool change record table, which includes the tooling fixture number and the current timestamp. If the confirmation button signal is not triggered within a preset time window, the system automatically writes an abnormal mark record in the tool change record table and pushes a manual supplementary record reminder to the production management terminal. After the operator manually confirms the tool change time on the management interface, the abnormal mark is replaced with a valid tool change record, thereby ensuring the boundary integrity of the tool usage cycle data segment.

[0024] Data was collected during the stable machining stage after the machine tool's thermal deformation reached thermal equilibrium. It was approximately assumed that the contribution of thermal deformation was relatively stable within a single tool's service life, and the extracted wear trend characteristic value was dominated by the fixture's progressive wear.

[0025] After the workpiece completes the current station machining operation and the spindle stops, the dimensional data acquisition device automatically extends its probe to perform contact measurement on the target surface and sends the acquired measured dimensional data to the data processing unit. The data processing unit reads the nominal dimension of the corresponding part from the system database. The nominal dimension is entered into the process parameter table of the system database by the process personnel according to the process drawings during system initialization and is updated synchronously when the process changes. The data processing unit subtracts the nominal dimension from the measured dimension to obtain the accuracy deviation value. The accuracy deviation value, along with the workpiece identification code, tooling fixture number, and machining timestamp, is written into the flow details table.

[0026] The division of tool usage cycle data segments is based on the tool change records in the system database. During system initialization, the system startup timestamp is automatically used as the starting boundary of the first tool usage cycle, and the timestamp of the first tool change event record is used as the ending boundary of the first cycle and the starting boundary of the second cycle, and so on. The machining timestamp of each accuracy deviation value record in the flow details table is assigned to the corresponding tool usage cycle data segment based on the time interval it falls into. The assignment result is written to the flow details table with the cycle number field, so that the subsequent fixture wear calculation module can extract data by segment.

[0027] By binding the triggering times of identifier reading, process action acquisition, tool change record writing, and accuracy deviation calculation to the workpiece arrival signal, spindle stop signal, and tool change confirmation signal, respectively, it is ensured that the timing of various acquired data strictly corresponds to the actual machining behavior of the workpiece. At the same time, the accuracy deviation baseline jump introduced by tool change is isolated by using the tool usage cycle data segment as the segment unit, providing structured input data with clear source and clear boundary for subsequent two-dimensional wear feature extraction.

[0028] Furthermore, the bipartite graph modeling module is specifically used to: remove duplicates from all workpiece identification codes and tooling fixture numbers to obtain a left node set and a right node set; establish cross-set edges between the left node set and the right node set based on the actual workstation flow matching relationship in the physical workshop; and attach the process actions, processing timestamps, and accuracy deviation values ​​recorded under the same processing operation to the attributes of the cross-set edges to obtain the workpiece-fixture bipartite graph.

[0029] Specifically, the bipartite graph modeling module is triggered by a system timed task after each production batch ends. It reads all workpiece identification code fields of the current batch from the workpiece flow record table in the system database. The input is a string-type code list, which for example contains 120 records from "WP-20240801-001" to "WP-20240801-120". The module performs a deduplication operation based on a hash table on the above list, merging the same codes into a single entry. The output is a set of left nodes without duplicates, which is written to the node table of the graph database. The deduplication judgment rule is an exact match of the complete string of field values. Workpiece identification codes of different production batches are naturally distinguished by the different batch date fields, so there will be no cross-batch erroneous merging. If duplicate codes appear within the same batch, a conflict log is recorded and a manual verification reminder is triggered.

[0030] Furthermore, the module reads all online tooling and fixture number fields from the equipment ledger table in the system database. The input is a string-type encoding list, which for example contains 7 records from "FX-01" to "FX-07". After performing deduplication using a hash table, the output is a set of right nodes and written to the node table of the graph database. The deduplication rules for the right node set are the same as those for the left node set, with the exact matching of the complete string of the encoding field as the criterion.

[0031] Specifically, the establishment of cross-set edges is based on the workstation arrival time sequence records stored in the workpiece flow record table. Using the workpiece identifier code as the primary key, the system queries which tooling fixture numbers correspond to the workstations where the workpiece generated processing timestamp records. For each valid pairing relationship between the workpiece identifier code and the tooling fixture number found, a cross-set edge is established between the corresponding workpiece node in the left node set and the corresponding fixture node in the right node set. The edge is written to the edge table of the graph database using the concatenated string of the workpiece identifier code and the processing fixture number as the unique edge identifier. This scheme uses the actual arrival time sequence records as the matching key instead of a preset process route table template. The reason is that the actual flow sequence of the workpiece among the 7 sets of tooling fixture workstations may be inconsistent with the process route table due to scheduling changes. Using the actual records as the standard can ensure that the bipartite graph truly reflects the clamping relationships that occur in the physical workshop.

[0032] Furthermore, for workpieces that have not passed through all N workstations, the module establishes a corresponding number of connections based on the actual number of workstations that generated records. It is allowed for the number of connections for a single workpiece to be less than N, and the number of connections less than N is not considered abnormal. After completing the connection establishment, the module counts the number of connections for each workpiece. If the number of connections for a workpiece is 0, it means that the workpiece has no processing records in the flow record table. The system marks the workpiece identification code as a data missing state and writes it to the abnormal log, and does not establish any connections for it. When subsequent calculation modules read data, they automatically skip the left nodes marked as data missing states.

[0033] Specifically, the appending operation of the connection attributes uses the edge identifier as the primary key. It queries the workpiece flow record table for all flow details that match the combination of the workpiece identifier and the tooling fixture number. Each detail record contains three attribute fields: process action type, processing timestamp, and accuracy deviation value. The module serializes the above three fields in key-value pair format and appends them to the attribute fields of the corresponding cross-set connection, writing them into the edge attribute table of the graph database. If the same workpiece has multiple processing records at the same tooling fixture station (such as in a rework scenario), each processing record is appended as an independent attribute entry to the attribute list of the same connection in sequence. The attribute list is arranged in ascending order of processing timestamp to ensure that the subsequent modules do not extract the data in chronological order.

[0034] When the same workpiece has multiple clamping and processing records at the same tooling fixture station, the bipartite graph modeling module determines whether the process action field values ​​of two adjacent records are the same. If the process actions are the same (rework scenario), a rework mark field is added to the attribute of the connection to record the number of reworks. This allows the fixture wear calculation module to perform separate grouping or weighted / reduced processing of rework records based on the rework mark field when extracting the deviation data sequence.

[0035] By establishing cross-set connections using actual arrival time records as matching keys, allowing fewer than N connections, and marking abnormal workpieces with zero connections, the bipartite graph not only fully reflects the real flow relationship in the physical workshop but also has the ability to automatically identify and isolate missing data. This provides a clear and well-defined multi-dimensional attribute graph structure input for the subsequent fixture wear calculation module.

[0036] Furthermore, the fixture wear calculation module includes: extracting all cross-set edges connected to each right node in the right node set one by one, reading the process action, the processing timestamp, and the accuracy deviation value in the edge attributes, comparing each field, and classifying the data records whose process actions match the axial process action into the first storage queue, and classifying the data records whose process actions match the radial process action into the second storage queue; within the tool usage cycle data segment, arranging the accuracy deviation values ​​in the first storage queue and the second storage queue in ascending order according to the processing timestamp, so that the accuracy deviation values ​​generated by the axial machining operation and the radial machining operation are stored independently at the data level, respectively obtaining the bottom edge end face deviation data sequence and the housing radial deviation data sequence.

[0037] Specifically, the fixture wear calculation module is triggered by a system timed task after each production batch of data is written. It reads the set of right nodes from the node table of the graph database, taking the seven tooling fixture number nodes from "FX-01" to "FX-07" as input, and performs subsequent processing on each right node in sequence. The module first queries all cross-set edges in the graph database edge table with the current right node number as the endpoint. If the query result is an empty set, it means that the fixture has not undertaken any workpiece processing tasks in the current batch. The module writes an empty record containing the fixture number and the current batch timestamp in the exception log table, skips the subsequent calculation of the right node, and directly processes the next right node without generating any deviation data sequence output.

[0038] Furthermore, for right nodes whose query results are not empty, the module reads the edge attributes of all cross-set edges connected to the right node in batches from the edge attribute table. Each edge attribute record contains three fields: process action type, processing timestamp, and precision deviation value. For example, fixture number "FX-03" is associated with 86 edge attribute records in a certain batch. All records are loaded into memory in list form as the input dataset for field comparison.

[0039] Specifically, the module performs a complete string exact match on the process action type field of each record in the input dataset. Records with a field value of axial process action identifier are written to the first storage queue, and records with a field value of radial process action identifier are written to the second storage queue. When the process action type field of a record is an empty string, a non-predefined identifier, or garbled characters due to acquisition anomalies, the record is not assigned to any queue, but is written to the abnormal record temporary storage table with the current fixture number and the processing timestamp of the record attached. The system also pushes a field anomaly reminder to the production management terminal. The process personnel can check the record in the management interface and manually specify the process action type before re-triggering the re-assignment operation. For example, out of 86 records, 52 matching axial process actions are assigned to the first storage queue, 31 matching radial process actions are assigned to the second storage queue, and 3 records with abnormal field values ​​are written to the temporary storage table.

[0040] Furthermore, the module groups the records in the first and second storage queues according to the tool usage cycle data segment number. Each data segment number corresponds to an independent calculation group. When the same fixture spans multiple tool usage cycle data segments, the records in each data segment participate independently in the sorting and subsequent wear calculation within that segment. The accuracy deviation values ​​between different data segments are not merged. The reason for using segmented independent calculation instead of merging calculation is that merging data across tool change cycles will mix the baseline jump of accuracy deviation introduced by tool change into the sequence, causing the subsequent sliding window to fail to correctly identify the progressive wear trend of the fixture.

[0041] Specifically, within each tool usage cycle data segment, the module sorts all precision deviation value records in the first storage queue in ascending order according to the machining timestamp field, outputting a list of floating-point values ​​sorted in ascending order of timestamps. Physically, this represents the time sequence of bottom edge end face dimension deviations for each workpiece during axial operations performed by the fixture within that tool usage cycle. This data is written to the deviation sequence table in the system database and marked as the bottom edge end face deviation data sequence. The same ascending sorting operation is performed on all precision deviation value records in the same segment of the second storage queue, outputting a list of floating-point values ​​of the same type. Physically, this represents the time sequence of outer diameter deviations for each workpiece during radial operations performed by the fixture within that tool usage cycle. This data is written to the deviation sequence table and marked as the shell radial deviation data sequence. Both sequences are stored using the fixture number, tool usage cycle data segment number, and sequence type identifier as a joint primary key, allowing for querying and retrieval via the time-varying sliding window unit.

[0042] By decoupling the mixed precision deviation data into two independent time series using the process action type field as the split key, isolating tool change baseline jumps using the tool usage cycle data segment as the group boundary, and setting clear handling paths for abnormal field values ​​and unloaded fixtures, it is ensured that the bottom edge end face deviation data series and the shell radial deviation data series have complete engineering feasibility in three dimensions: data source, time sequence boundary, and abnormal isolation.

[0043] Furthermore, the fixture wear calculation module further includes: performing the following operations on the bottom edge end face deviation data sequence and the housing radial deviation data sequence respectively: setting a sliding calculation interval on the deviation data sequence, and extracting data points within the interval as the current calculation window data; assigning a dynamic weight factor set to each data point in the current calculation window data, with the weight increasing as it is closer to the current time, through a time decay function; calculating a weighted average deviation value and a weighted variance value based on the dynamic weight factor set, and identifying data points whose difference from the weighted average deviation value exceeds a tolerance threshold set based on the weighted variance value. The discrete abrupt change values ​​are identified and removed to obtain a smoothed effective dataset. The dynamic weight factor set corresponding to the remaining data points in the smoothed effective dataset is then normalized and recalculated after removing the weights of the discrete abrupt change points to obtain a corrected weight sequence. The values ​​in the smoothed effective dataset are multiplied point-by-point by the corrected weight sequence and accumulated to obtain wear trend characteristic values. The wear trend characteristic values ​​obtained from the bottom edge end face deviation data sequence are used as the time-varying wear degree of the bottom edge end face, and the wear trend characteristic values ​​obtained from the shell radial deviation data sequence are used as the time-varying wear degree of the shell outer diameter.

[0044] Specifically, in this implementation method, before the shell casing precision machining production line is put into operation, the machine tool is required to run continuously under no-load conditions for no less than 40 minutes until the spindle and bed temperatures reach a stable thermal equilibrium. The data acquisition module only collects the dimensional data of the target surface during the normal production stage after the thermal equilibrium has stabilized. Within a single tool life cycle, the contribution of machine tool thermal deformation to the accuracy deviation value is approximately regarded as a relatively stable systematic bias. This bias does not show a monotonically accumulating trend in the time dimension and is fundamentally different in its variation characteristics from the slow drift caused by the gradual wear of the fixture positioning surface. Therefore, the wear trend characteristic value extracted by this scheme is dominated by the gradual wear of the fixture, and the residual influence of machine tool thermal deformation is isolated by the segment boundary of the tool life cycle.

[0045] Further, the module reads the bottom edge face deviation data sequence corresponding to the specified fixture number and the specified tool usage cycle data segment from the deviation sequence table in the system database. The input format is a list of floating-point values ​​arranged in ascending order of machining timestamps. For example, the bottom edge face deviation data sequence of fixture number "FX-03" in the second tool usage cycle contains 52 data points. The module reads the preset sliding calculation interval window length parameter and minimum effective number of points threshold parameter from the system configuration table. For example, the window length is set to 20 data points and the minimum effective number of points threshold is set to 8 data points. The module reads the data in sequence... The latest data point at the end of the column is used as the right endpoint to truncate 20 consecutive data points to be loaded into memory as the current calculation window data. When the total number of points in the sequence is less than 20, the actual number of points is used as the window length. If the actual number of points in the current calculation window data after truncation is lower than the minimum effective number of points threshold of 8, the module writes a "points insufficient" status record containing the fixture number, data segment number and current timestamp into the calculation status table, suspends the wear calculation of the fixture and the data segment, and the scheduled task will re-truncate the operation after the next batch of data is written, until the actual number of points in the window is not less than 8 before the subsequent calculation is performed.

[0046] Specifically, for each data point in the current calculation window, the processing timestamp corresponding to that point is read from the flow details table. The time difference between that timestamp and the current calculation trigger time is calculated. The unit of the time difference is hours, and the smaller the difference, the closer the data point is to the current time. The value of the time decay function is equal to the exponent of the base of the natural logarithm, and the decay rate parameter with a negative exponent is multiplied by the time difference corresponding to the i-th data point. The time difference is the difference between the processing timestamp corresponding to the data point and the current calculation trigger time, and the decay rate parameter is configured to range from 0.01 to 0.05. The module assigns a weight factor to each data point based on the time difference. The weight factor is assigned according to the rule that the smaller the difference, the greater the weight. For example, the time difference of 20 data points in the window ranges from 2 hours to 96 hours, and the corresponding weight factor ranges from 0.95 to 0.12. All 20 weight factors constitute a dynamic weight factor set and are written into memory. The physical meaning of this set is that the contribution of recent processing records to the current fixture wear state is higher than that of historical records.

[0047] Furthermore, the module performs a weighted average operation on the current calculation window data using a dynamic set of weighting factors, and outputs a weighted average deviation value, which physically represents the central estimate of the fixture wear baseline trend within the current window. Simultaneously, the module performs a weighted variance operation on the current calculation window data using the same set of dynamic weighting factors, and outputs a weighted variance value, which physically represents the weighted dispersion of the precision deviation value around the weighted average deviation value within the window. Both calculation results are floating-point values ​​and are temporarily stored in memory for subsequent culling operations.

[0048] Specifically, the module calculates the absolute difference between each data point in the current calculation window and the weighted average deviation value, and compares it with the tolerance threshold set based on the weighted variance value. The tolerance threshold is obtained by multiplying the square root of the weighted variance value by a preset multiplier coefficient in the system configuration table. The multiplier coefficient k ranges from 2.0 to 3.5, preferably 2.5. The number of samples n based on the data points within the current calculation window is selected: when n≤30, k is 2.5-3.5; when n>30, k is 2.0-2.5. The multiplier can be dynamically determined based on the Grubbs test or Dixon test criteria. Data points whose absolute difference exceeds the tolerance threshold are determined to be discrete mutation values ​​caused by random processing errors and removed from the current calculation window data. The remaining data points constitute a smoothed effective dataset written to memory. For example, 3 out of 20 points are determined to be discrete mutation values ​​and removed, resulting in a smoothed effective dataset containing 17 data points. If the number of points in the smoothed effective dataset is 0 after the removal operation, the module writes a "valid data empty" status record to the calculation status table, terminating the current calculation. The system waits for the next batch of new data points to be added to the sequence before re-triggering the sliding window truncation and removal operation, without outputting any wear trend characteristic values.

[0049] Furthermore, the module retrieves the original weight factors corresponding to the 17 retained data points in the smoothed effective dataset from the dynamic weight factor set, removes the original weight factors of the 3 points that are judged to be discrete abrupt changes, performs normalization recalculation on the remaining 17 weight factors, so that the sum of the 17 corrected weight factors equals 1, and outputs the corrected weight sequence written to memory; the physical meaning of the corrected weight sequence is to redistribute the contribution weights of the recent effective data after removing the outlier points, and eliminate the residual interference of the removed jump points.

[0050] Specifically, the module multiplies the floating-point values ​​of 17 data points in the smoothed effective dataset point by point with the corresponding 17 correction weight factors in the correction weight sequence and then sums them up, outputting a single floating-point value, i.e., the wear trend feature value, which is written into the wear degree calculation result table in the system database. When the wear trend feature value obtained from the bottom edge deviation data sequence through the above full process is written into the result table, the label type field is "bottom edge end face time-varying wear degree". When the wear trend feature value obtained from the shell radial deviation data sequence through the same process is written into the result table, the label type field is "shell outer diameter time-varying wear degree". Both wear degree records are stored with the fixture number and the tool usage cycle data segment number as a joint primary key for the image database storage module to read.

[0051] To address the lag issue of the time decay function when handling discontinuous physical damage, a dual-track parallel calculation logic is introduced into the fixture wear calculation module. Specifically, before performing time decay weighted statistics, the module first performs a sudden damage hard-limiting verification on the data captured in the current calculation window. A sudden damage alarm threshold is set (the sudden damage alarm threshold is significantly larger than the tolerance threshold set based on weighted variance, and is exemplarily set to 1.5 times the single wear threshold). When the precision deviation value of the latest data point added in the current calculation window exceeds the sudden damage alarm threshold, the module triggers an interruption takeover mechanism: immediately stops the time decay weighting and smoothing removal operation of the current window; marks the abnormal data point as a "suspected sudden damage event" (such as collision, damage to the positioning reference caused by iron filings), and writes it to the system abnormal log; skips the normalization calculation of the smoothed effective dataset, forcibly updates the current time-varying wear output value of this dimension to the extreme value of the precision deviation of the sudden damage point, and adds a "sudden damage lock" label to the right node status field corresponding to the associated knowledge graph, directly triggering the production management decision module to output the highest priority offline maintenance instruction, ensuring that sudden physical damage is not masked by the time decay filtering algorithm.

[0052] By setting a minimum effective number of points threshold and a dual exit condition of an empty smooth effective dataset, strengthening the trend contribution of recent data with a time decay weight factor, and performing normalized correction weight calculation after removing discrete mutation points, the independent and reliable extraction of the asymptotic loss trend of the fixture bottom edge face and the outer diameter of the shell is achieved under the premise that random machining error and machine tool thermal deformation are approximately stable.

[0053] Furthermore, the graph database storage module is specifically used for: establishing a time-series attribute linked list with the processing batch time as the index dimension for the right node corresponding to each tooling fixture number; receiving the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell, and loading them into the corresponding time-series attribute linked list in the form of incremental snapshots to complete the traceability storage of the fixture wear data trajectory; adopting a non-blocking write mechanism, asynchronously appending the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell as two independent time series dimensions into the preset dynamic attribute slots of the right node; and using the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell as the accuracy influence weights under the corresponding processing scenarios, backfilling them into the corresponding attribute fields of the process action matching in the cross-set connection, thus forming the associated knowledge graph.

[0054] Specifically, during the system initialization phase, the graph database storage module pre-builds a time-series attribute linked list data structure for each tooling fixture number node in the right node set. This structure is mounted on the corresponding right node in the form of graph database node attribute fields. The linked list uses the processing batch timestamp string as the index key and the floating-point value pair of the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell calculated in the current time as the index value. Initially, the linked list is empty. The system configuration table presets the maximum number of storage entries for the time-series attribute linked list, for example, the maximum limit is set to 500 entries. When the number of entries in the linked list reaches the limit, the system automatically archives the earliest 100 historical entries to the historical wear degree archive table in the system database and removes them from the linked list, keeping the number of entries in the linked list always below the limit and avoiding a decrease in query performance due to the unlimited growth of graph database node attribute fields.

[0055] Furthermore, a push-triggered mechanism is adopted instead of a timed polling mechanism. Each time the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the housing are successfully written into the wear degree calculation result table of the system database, a notification message containing the fixture number, the tool usage cycle data segment number and the calculation result is immediately pushed to the graph database storage module through the internal message queue of the system. The graph database storage module subscribes to the message queue. After receiving the push notification, it immediately reads the time-varying wear degree values ​​of the bottom edge end face and the time-varying wear degree values ​​of the outer diameter of the housing carried in the message body, constructs a new linked list entry with the timestamp of the current processing batch as the index key, and writes it to the time-series attribute linked list of the corresponding right node in an incremental append manner. The physical meaning of the linked list entry written this time is a snapshot of the wear state of the fixture in the current processing batch. For example, after 20 consecutive batches of production, the time-series attribute linked list of fixture number "FX-03" contains 20 entries, and each entry corresponds to the bottom edge end face time-varying wear degree and outer diameter time-varying wear degree value pair of one batch.

[0056] Specifically, the graph database storage module employs a non-blocking write mechanism to asynchronously append the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell as two independent time series dimensions to the preset dynamic attribute slots of the right nodes. These dynamic attribute slots are pre-allocated to each right node during system initialization. The time-varying wear degree slots of the bottom edge end face and the outer diameter of the shell are independent and do not interfere with each other. The asynchronous write operation is executed in an independent thread, and the main thread does not block while waiting for the write result. The write result is returned through a callback function. When the asynchronous write operation fails due to graph database connection timeout or... When a write conflict returns a failure status, the callback function pushes the failed write task along with the original value and timestamp into the local retry queue. The retry queue re-initiates the write operation at intervals of 2 seconds, 4 seconds, and 8 seconds according to the exponential backoff strategy, with a maximum of 5 retries. If the write operation still fails after 5 retries, the module marks the write task as "persistence failed" and writes it to the system exception log. At the same time, it pushes a data synchronization exception alarm to the production management terminal. After the operation and maintenance personnel manually check the graph database connection status, they trigger the supplementary entry operation to ensure that the graph data is not permanently lost due to a single write failure.

[0057] Furthermore, the backfilling operation for the accuracy impact weight is triggered by the same message queue after the incremental snapshot of the time-series attribute linked list is written. The module queries the graph database edge table for all cross-set edges with the current fixture number as the right endpoint node and the process action field value in the edge attribute being the axial process action identifier. The time-varying wear value of the bottom edge end face calculated in this batch is updated to the accuracy impact weight attribute field of the above-mentioned edges by overwriting. The module simultaneously queries all cross-set edges with the process action field value being the radial process action identifier and overwrites the time-varying wear value of the outer diameter of the shell into the accuracy impact weight attribute field of the corresponding edge. After the backfilling operation is completed, the cross-set edges in the graph database with the workpiece node as the left endpoint and the fixture node as the right endpoint simultaneously carry the process action, processing timestamp and accuracy deviation value attributes generated by the acquisition layer and the accuracy impact weight attribute generated by the calculation layer. The two types of attributes together constitute a complete association knowledge graph edge structure, which can be retrieved and read by the production management decision module by combining the fixture number and process action type conditions through the graph query language.

[0058] To ensure that wear data is written to the graph immediately after calculation, an exponential backoff retry mechanism guarantees the eventual consistency of non-blocking asynchronous writing, and the storage scale of the graph database is controlled by the upper limit of the number of linked list entries and the historical archiving mechanism, so that the associated knowledge graph maintains a traceable wear trajectory and a real-time and effective data management status for the impact of edge accuracy on weights during continuous production.

[0059] Furthermore, considering the cumulative monotonicity of physical wear on tooling fixtures across tool cycles, this embodiment further establishes an inter-cycle wear baseline transfer mechanism between adjacent tool usage cycle data segments. Since tool replacement primarily eliminates the wear deviation of the tool itself, while the wear on the fixture positioning surface is not recovered before and after tool replacement, the initial wear trend characteristic value should not start from zero when calculating the (k+1)th tool usage cycle data segment. Instead, the final time-varying wear degree recorded at the end of the kth tool usage cycle (i.e., the previous cycle) in the associated knowledge graph is extracted as the wear compensation baseline for the (k+1)th cycle. Specifically, when performing weighted cumulative calculation on the smoothed effective dataset of the (k+1)th cycle, the wear compensation baseline is superimposed. This results in the time-varying wear degree calculated segmented by tool life being stitched together on the macro-time axis into a globally progressive wear curve that exhibits a step-like increase and conforms to the physical dissipation law, avoiding the situation where the actual wear amount of the fixture is periodically reset to zero due to tool replacement segmentation.

[0060] Correspondingly, in the graph database storage module, when the above-mentioned "suspected sudden damage event" is captured, not only is the time-series attribute linked list updated, but also an "abnormal event node" is instantiated. The abnormal event node is then linked with the left node (workpiece identification code) and the right node (tooling fixture number) that caused the deviation. This allows for the precise topological solidification of the source of sudden damage in the graph structure, providing an immutable data traceability link for subsequent tripartite responsibility determination (blank defect leading to tool collision vs. fixture fatigue fracture).

[0061] Furthermore, the system also includes a production management decision module, specifically used for: periodically accessing the time-series attribute linked list of each right node in the associated knowledge graph, reading the latest time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell, calculating the deterioration ratio of the two relative to the preset single wear threshold, and determining the advantageous processing dimension corresponding to the right node based on the lower deterioration ratio; obtaining the current process action requirements of the workpiece to be transferred, and when the process action requirement is the axial process action, filtering the set of fixtures whose time-varying wear degree of the bottom edge end face has not yet reached the single wear threshold; when the process action requirement is the radial process action, filtering the set of fixtures whose time-varying wear degree of the outer diameter of the shell has not yet reached the single wear threshold; performing path allocation based on the advantageous processing dimension, determining the remaining available processing scenarios for fixtures whose time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell exceeds the single wear threshold; generating tooling and fixture scheduling instructions based on the determination results, prioritizing the allocation of the above fixtures to the process scenarios corresponding to their wear degrees that have not exceeded the standard, and outputting them to the production management terminal.

[0062] Specifically, during the initialization phase before the system is put into production, the single-item wear threshold is manually configured by process engineers in the system management interface according to the shell casing accuracy standard and written into the system configuration table. The time-varying wear threshold of the bottom edge end face and the time-varying wear threshold of the shell outer diameter are stored as two independent configuration items. For example, the single-item wear threshold of the bottom edge end face is set to 0.025 mm and the single-item wear threshold of the shell outer diameter is set to 0.018 mm. When the process standard changes due to product model change or accuracy level change, the process engineers can perform update operations on the above threshold parameters in the management interface. The system will synchronously write the update timestamp and the threshold values ​​before and after the modification into the configuration change log table to ensure that the source of the threshold is traceable.

[0063] Furthermore, the production management decision module is triggered periodically according to the polling interval preset in the system configuration table. For example, the polling interval is set to execute once every 10 minutes. Each time it is triggered, the module reads the list of all right node numbers from the graph database node table, and accesses its time-series attribute linked list in turn using each right node number as the query key. It extracts the entry with the largest index key in the linked list (i.e., the latest processing batch time), and reads two floating-point values: the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell. The module divides the time-varying wear degree of the bottom edge end face by the single wear threshold of the bottom edge end face to obtain the bottom edge end face deterioration ratio, and divides the time-varying wear degree of the outer diameter of the shell by the single wear threshold of the outer diameter of the shell to obtain the outer diameter deterioration ratio of the shell. Both deterioration ratios are floating-point numbers greater than 0. The closer the value is to 1, the closer the wear state of that dimension is to the threshold boundary. For example, the bottom edge end face deterioration ratio of the fixture "FX-03" is 0.72, and the outer diameter deterioration ratio of the shell is 0.89.

[0064] Specifically, the module compares the two degradation ratio values ​​of the same right node and determines the dimension with the lower degradation ratio as the dominant processing dimension of the fixture. For example, the degradation ratio of the bottom edge end face of "FX-03" is 0.72, which is lower than the degradation ratio of the outer diameter of the shell is 0.89. Therefore, its dominant processing dimension is determined to be the axial processing scenario. When the two degradation ratio values ​​are exactly equal, the module prioritizes selecting the dominant dimension with the smaller number according to the ascending order of the tooling fixture number string. If multiple fixtures in the same batch have the same degradation ratio and the same number, the module further compares the processing batch timestamp of the latest entry in their time sequence attribute chain. The fixture with the earlier last use time is given priority to obtain the dominant dimension mark to ensure that the determination result is unique in the case of a tie. The above dominant processing dimension determination result is written to the fixture status cache table of the system database with the fixture number as the primary key for subsequent filtering and scheduling operations to read.

[0065] Furthermore, the current process action requirements of the workpiece to be transferred are sent from the production scheduling system to the production management decision module through the system's internal service interface. The interface message body contains the workpiece identification code and the current process action type field. The process action type field value is either an axial process action identifier or a radial process action identifier. After receiving the interface message, the module filters the fixture set that meets the conditions from the fixture status cache table based on the process action type field value. When the process action type is an axial process action, all fixture numbers that have not reached the bottom edge end face single wear threshold when the time-varying wear degree is of the bottom edge end face are filtered. When the process action type is a radial process action, all fixture numbers that have not reached the shell outer diameter single wear threshold when the time-varying wear degree is of the shell outer diameter are filtered. The filtering result is a list of fixture numbers that meet the conditions.

[0066] Specifically, when the screening result is an empty set, it indicates that all online fixtures in the dimension corresponding to the current process action have exceeded the standard. The module immediately pushes a "fixture resource shortage" warning message to the production management terminal. The message body includes the process action type, the number of fixtures exceeding the standard, and the current timestamp. At the same time, it suspends the return of this scheduling instruction to the production scheduling system and writes the process action requirement record into the scheduling pending queue. The pending queue is polled by the system at 30-second intervals. When it is detected that any fixture has completed maintenance and returned to online (i.e., the wear degree of the corresponding dimension has been manually reset to below the threshold), the system automatically re-triggers the screening process for the scheduling requirement of the corresponding process type in the pending queue and resumes scheduling.

[0067] Furthermore, when the screening results are not empty, the module further identifies fixtures with only single-item wear exceeding the standard in the list of fixture numbers obtained from the screening. The identification rule is that the deterioration ratio of the bottom edge end face exceeds 1 and the deterioration ratio of the outer diameter of the shell does not exceed 1, or the deterioration ratio of the outer diameter of the shell exceeds 1 and the deterioration ratio of the bottom edge end face does not exceed 1. For fixtures with a bottom edge end face deterioration ratio exceeding the standard but the outer diameter of the shell does not exceed the standard, it is determined that its remaining usable processing scenario is radial operation. For fixtures with a shell outer diameter deterioration ratio exceeding the standard but the bottom edge end face deterioration ratio does not exceed the standard, it is determined that its remaining usable processing scenario is axial operation. Based on the current process operation requirements and the above determination results, the module prioritizes fixtures with only single-item wear exceeding the standard whose remaining usable processing scenarios match the current process requirements and places them at the front of the scheduling candidate list. It generates tooling fixture scheduling instructions, writes them into the scheduling instruction table of the system database, and outputs them to the production management terminal through the service interface. The scheduling instruction message body includes the recommended fixture number, the current dual-dimensional deterioration ratio of the fixture, and the advantageous processing dimension mark.

[0068] Specifically, for fixtures where both the bottom edge end face degradation ratio and the shell outer diameter degradation ratio exceed 1, the module generates a fixture offline maintenance instruction. The instruction includes the fixture number, the current wear values ​​in two dimensions, and the excess ratio. The fixture offline maintenance instruction is written into the status field of the corresponding right node in the associated knowledge graph in the form of node attribute update, and the status mark of the right node is updated from "online" to "pending maintenance". At the same time, an offline maintenance notification is pushed to the production management terminal through the service interface. After the maintenance personnel confirm receipt, the fixture is arranged for offline maintenance. After the fixture completes maintenance and is manually reset by the process personnel in the management interface, the system restores the status field of the corresponding right node to "online" and resets the wear values ​​in the two dimensions to the initial values. The reset operation is synchronously written into the time-series attribute linked list as a new incremental snapshot entry.

[0069] When neither the time-varying wear degree of the bottom edge end face nor the time-varying wear degree of the outer diameter of the housing reaches the corresponding single wear threshold, the production management decision module will determine that the fixture has both axial machining advantage dimension and radial machining advantage dimension, and will arrange such dual-dimensional available fixtures in the scheduling candidate list before fixtures with only single-dimensional availability, with the priority from high to low as follows: dual-dimensional available fixtures > fixtures with single-dimensional excess but matching remaining available machining scenarios > fixtures with single-dimensional excess and mismatched remaining available machining scenarios (not included in the candidates).

[0070] By configuring single-item wear thresholds in the system configuration table and recording change logs, using production scheduling system interface messages as the input source for process requirements, triggering suspension and recovery mechanisms for screening empty sets, generating offline instructions for fixtures with two exceeding limits and writing them into the graph status field, and setting dual decision rules for the average deterioration ratio based on numbering order and last usage time, the system covers the handling paths for all boundary situations in the scheduling decision-making process. This enables the production management decision-making module to output deterministic results under various abnormal conditions, achieving refined data-driven utilization of the remaining usability of fixtures with only single-item wear exceeding limits.

[0071] This invention constructs a workpiece-fixture bipartite graph and decouples the precision deviation data into a bottom edge end face deviation data sequence and a shell radial deviation data sequence according to the processing scenario. The time-varying wear degree of the bottom edge end face and the time-varying wear degree of the shell outer diameter are extracted through a time decay sliding window and normalized correction weight. Then, the two-dimensional wear degree is persisted to the associated knowledge graph using a time-series attribute linked list and dynamic attribute slots. Finally, the production management decision module realizes the targeted scheduling of fixtures with only single-item wear exceeding the standard based on the deterioration ratio, forming a complete closed loop from data acquisition to scheduling decision.

[0072] Example 2: Based on the above examples, this example addresses a more complex engineering problem in the finishing production line where the same tooling fixture handles a small number of workpieces within a single tooling cycle, resulting in sparse deviation data sequence points and insufficient effective points within the sliding window to support reliable wear estimation. The time-varying sliding window unit and progressive wear feature extraction unit of the fixture wear calculation module are refined.

[0073] Specifically, before performing window truncation, the time-varying sliding window unit first queries the system configuration table for the pre-set adaptive window expansion strategy configuration item for sparse sequence scenarios. This configuration item includes three parameters: baseline window length, minimum effective number of points threshold, and maximum allowable expansion multiple. For example, the baseline window length is 20 points, the minimum effective number of points threshold is 8 points, and the maximum allowable expansion multiple is 3 times. When the actual number of points in the current calculated window data after truncation with the baseline window length is lower than the minimum effective number of points threshold, the module does not immediately enter a suspended waiting state, but starts the adaptive expansion process, gradually expanding the window length in the historical direction with the baseline window length as the step size. After each expansion step, the number of points truncated is recounted until the number of points is not lower than the minimum effective number of points threshold or the window length has been cumulatively expanded to the upper limit corresponding to the maximum allowable expansion multiple. If the number of points is still insufficient after expanding to the upper limit, the module determines that the current sequence is in the data accumulation stage, writes the "still insufficient after adaptive expansion" status record in the calculation status table, and pauses the calculation, waiting for subsequent batches of data to be supplemented before re-triggering.

[0074] Furthermore, after the adaptive expansion window successfully captures data from the current calculation window that is not less than the minimum effective number of points threshold, the weighting strategy of the time decay function needs to be modified for the expanded historical span. Since the time span covered by the expanded window is significantly larger than that of the baseline window, if the same decay rate parameter as the baseline window is still used, the weight factor of the historical points will approach 0 and lose statistical significance. Therefore, this module reads the slow decay rate parameter configured separately for the expanded window scenario from the system configuration table and applies the slow decay rate parameter to the weight factor calculation of all data points after adaptive expansion, so that the weight factor of the earliest point after expansion is not less than 0.15 times the weight factor of the latest point. For example, when expanded to 60 points, the weight factor of the latest point is 0.91 and the weight factor of the earliest point is 0.14, ensuring that the historical data still has a quantifiable statistical contribution to the wear trend judgment. The slow decay rate parameter is configured by the process personnel in the system management interface according to the actual tool change frequency and processing cycle of the production line, and is updated synchronously with the production schedule adjustment.

[0075] Specifically, after the discrete mutation points are removed under the combined effect of adaptive expansion window and slow decay rate, the progressive wear feature extraction unit performs normalization correction weight calculation on the expanded smooth effective dataset. Simultaneously, it checks whether the sum of the weight factors of the latest three points in the correction weight sequence is higher than 0.40 times the sum of the weight factors of all remaining points. This verifies that the correction weight sequence still maintains the characteristic of tilting towards recent data. If the ratio is lower than 0.40, it indicates that the slow decay rate parameter setting is too low, resulting in insufficient contribution weight of recent data. The module writes an alarm record of "recent weight too low" in the calculation status table and pushes parameter adjustment suggestions to the production management terminal. At the same time, it continues to complete the calculation of the wear trend feature value with the current correction weight sequence without interrupting the output. Finally, the wear trend feature value obtained by the adaptive expansion window process and the wear trend feature value obtained by the conventional window process are written into the wear degree calculation result table in the same output format. The field structure and type identification of the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are completely consistent with those in Example 1, ensuring that the graph database storage module can directly read without being aware of the window strategy used upstream.

[0076] This embodiment solves the engineering problem of insufficient reference window points causing long-term suspension of wear calculation in sparse machining sequence scenarios by introducing an adaptive window expansion strategy and a slow decay rate parameter correction mechanism for the expansion span. This enables the system to output statistically significant two-dimensional time-varying wear even in low-frequency usage scenarios where the number of workpieces borne by the fixture is small, thus expanding the applicability of the invention under diverse production scheduling and non-uniform load distribution conditions.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based shell casing data management system, characterized in that, include: Data Acquisition Module: Acquires the workpiece identification code, process actions, processing timestamp, and tooling fixture numbers passed through the workpiece; and measures and inspects the target surface after the workpiece is processed to obtain the accuracy deviation value. Bipartite graph modeling module: Establish cross-set edges with the workpiece identification code as the left node set and the tooling fixture number as the right node set, and combine the process action, the processing timestamp and the accuracy deviation value as the attributes of the edges to obtain the workpiece-fixture bipartite graph; Fixture wear calculation module: For each tooling fixture number corresponding to the right node, extract the precision deviation value on the connected edge and sort it according to the processing timestamp. Apply a sliding window with time decay function to perform weighted statistical calculation on the sorted precision deviation value, separate random processing error and gradual wear characteristics, and obtain the time-varying wear of the fixture. Graph database storage module: Writes the workpiece-fixture bipartite graph into the graph data storage, and stores the time-varying wear degree of the fixture as the dynamic attribute of the corresponding right node to form an associated knowledge graph.

2. The knowledge graph-based shell casing data management system according to claim 1, characterized in that, The acquisition module is specifically used for: The data acquisition process monitors the sequential flow of the workpiece among multiple tooling fixtures, extracting the process actions, processing timestamps, and corresponding tooling fixture numbers. The process actions include axial and radial process actions. Additionally, tool change records for each tooling fixture station are collected. After the workpiece completes processing at any tooling fixture station, the dimensional data of the target surface is collected, and the difference is calculated based on the pre-stored nominal dimensions to obtain the accuracy deviation value. Using two adjacent tool change records as segment boundaries, the accuracy deviation value is divided into the corresponding tool usage cycle data segments.

3. The knowledge graph-based shell casing data management system according to claim 1, characterized in that, The bipartite graph modeling module is specifically used for: All workpiece identification codes and tooling fixture numbers are deduplicated to obtain a left node set and a right node set. Based on the actual workstation flow matching relationship in the physical workshop, cross-set edges are established between the left node set and the right node set. The process actions, processing timestamps and accuracy deviation values ​​recorded under the same processing operation are added to the attributes of the cross-set edges to obtain the workpiece-fixture bipartite graph.

4. A knowledge graph-based shell casing data management system according to claim 2, characterized in that, The fixture wear calculation module includes: Extract all cross-set edges that are connected to each right node in the set of right nodes one by one, read the process action, the processing timestamp and the precision deviation value in the edge attributes, compare each field one by one, and put the data records whose process action matches the axial process action into the first storage queue, and put the data records whose process action matches the radial process action into the second storage queue. Within the tool usage cycle data segment, the accuracy deviation values ​​in the first storage queue and the second storage queue are arranged in ascending order according to the machining timestamp, so that the accuracy deviation values ​​generated by the axial machining operation and the radial machining operation are stored independently at the data level, respectively obtaining the bottom edge end face deviation data sequence and the housing radial deviation data sequence.

5. A knowledge graph-based shell casing data management system according to claim 4, characterized in that, The fixture wear calculation module also includes: The following steps are performed on the bottom edge end face deviation data sequence and the shell radial deviation data sequence: a sliding calculation interval is set on the deviation data sequence, and data points within the interval are extracted as the current calculation window data; a dynamic weight factor set is assigned to each data point in the current calculation window data through a time decay function, with the weight increasing as it is closer to the current time; a weighted average deviation value and a weighted variance value are calculated based on the dynamic weight factor set; data points whose difference from the weighted average deviation value exceeds the tolerance threshold set based on the weighted variance value are identified as discrete abrupt values ​​and are removed to obtain a smooth and effective dataset; After removing the weights of discrete abrupt change points from the dynamic weight factor set corresponding to the retained data points in the smoothed effective dataset, normalize and recalculate to obtain a corrected weight sequence; multiply and accumulate the values ​​in the smoothed effective dataset with the corrected weight sequence point by point to obtain the wear trend feature value; use the wear trend feature value obtained from the bottom edge end face deviation data sequence as the bottom edge end face time-varying wear degree, and use the wear trend feature value obtained from the shell radial deviation data sequence as the shell outer diameter time-varying wear degree, and the bottom edge end face time-varying wear degree and the shell outer diameter time-varying wear degree together constitute the fixture time-varying wear degree.

6. The knowledge graph-based shell casing data management system according to claim 1, characterized in that, The graph database storage module is specifically used for: For each tooling fixture number, a time-series attribute linked list is established with the processing batch time as the ordered index dimension. The time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are received and loaded into the corresponding time-series attribute linked list in the form of incremental snapshots to complete the traceability and storage of the fixture wear data trajectory. A non-blocking writing mechanism is adopted, and the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are treated as two independent time series dimensions and appended to the preset dynamic attribute slot of the right node through an asynchronous message queue; the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell are respectively used as the accuracy influence weights under the corresponding processing scenarios and backfilled into the corresponding attribute fields of the process action matching in the cross-set connection, thus forming the associated knowledge graph.

7. A knowledge graph-based shell casing data management system according to claim 6, characterized in that, The system also includes a production management decision module, specifically used for: The time-series attribute linked list of each right node in the associated knowledge graph is accessed periodically to read the latest time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell. The deterioration ratio of the two relative to the preset single wear threshold is calculated respectively, and the advantageous processing dimension corresponding to the right node is determined based on the lower deterioration ratio. Obtain the current process action requirements of the workpiece to be transferred. When the process action requirement is an axial process action, filter the set of fixtures whose time-varying wear degree of the bottom edge end face has not yet reached the single wear threshold. When the process action requirement is a radial process action, filter the set of fixtures whose time-varying wear degree of the outer diameter of the housing has not yet reached the single wear threshold. Based on the aforementioned advantageous processing dimensions, a path allocation is performed. For a single fixture that exceeds the single wear threshold in only one of the time-varying wear degree of the bottom edge end face and the time-varying wear degree of the outer diameter of the shell, the remaining available processing scenarios are determined. Based on the determination result, a tooling and fixture scheduling instruction is generated to preferentially allocate the single fixture that exceeds the threshold to the process scenario corresponding to the wear degree that does not exceed the threshold, and output to the production management terminal.

Images