A knowledge graph-driven manufacturing process optimization system

Abstract

The present invention relates to the field of process optimization technology, specifically a knowledge graph-driven manufacturing process optimization system, which includes a graph node generation module, a dependency relationship construction module, a semantic entity matching module, a process path screening module, and a trend-driven early warning module. The present invention realizes the recognition of semantic implicit relationships through sequence alignment and numerical difference comparison, introduces quantitative screening between processing dimensions and tolerance levels in path construction, ensures that the path connectivity structure is optimized while meeting dimensional accuracy constraints, and combines the time series trend splitting and continuous fluctuation consistency judgment mechanism of equipment operating status to complete the dynamic marking of process deviation trends and the generation of early warning annotation sets at the node path level, realizing closed-loop processing from knowledge modeling, path screening to trend prediction, and effectively improving the adaptability of process paths, parameter configuration accuracy, and the advance of abnormality identification.

Description

Technical Field

[0001] The present invention relates to the field of process optimization technology, and in particular to a knowledge graph-driven manufacturing process optimization system. Background Art

[0002] The field of process optimization technology focuses on improving efficiency, product quality, and resource utilization in the manufacturing process, specifically including aspects such as manufacturing parameter configuration, process path adjustment, process collaborative scheduling, and equipment operation optimization. Technologies in this field typically combine industrial engineering, control theory, and algorithm modeling to achieve dynamic optimization and control of key indicators such as energy consumption, time cost, and quality fluctuations in the manufacturing process. In the context of informatization and intelligence, process optimization is widely introducing specific technical solutions such as multi-objective optimization algorithms, model predictive control, data-driven Bayesian optimization, and reinforcement learning to address process constraints, nonlinear coupling, and uncertain disturbances in highly complex manufacturing scenarios.

[0003] The knowledge graph-driven manufacturing process optimization system is a process optimization solution based on semantic associations and structured knowledge representation. It aims to achieve explicit modeling and reasoning of manufacturing knowledge by constructing a knowledge graph that includes multiple entities and relationships, such as process parameters, equipment performance, product structure, and quality feedback. The system is designed to support manufacturing companies in making rapid and intelligent decision recommendations in scenarios such as process planning, anomaly diagnosis, and parameter adjustment. By mining semantic paths and entity relationships in the knowledge graph, it improves the accuracy and adaptability of process solution selection.

[0004] When constructing manufacturing process knowledge graphs, traditional optimization systems lack the ability to recognize the differences between semantically heterogeneous operational entities. Path construction is based only on static structural reasoning, and no dynamic constraint mechanism for adapting to machining dimensional tolerances has been formed. This leads to the accumulation of dimensional errors in path selection due to inaccurate matching of control parameter ranges. In anomaly prediction, they often rely on single-point parameter threshold judgments, making it difficult to accurately identify potential anomalies caused by continuous fluctuation trends during the operation phase. In typical cases, for example, when the spindle vibration value does not exceed the limit but continues to rise, it is still difficult to capture it. This reduces the timeliness of process anomaly response and the foresight of decision-making, and limits the ability to actively intervene in manufacturing process disturbances in complex scenarios. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a knowledge graph-driven manufacturing process optimization system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a knowledge graph-driven manufacturing process optimization system, the system comprising:

[0007] The graph node generation module obtains the process documents, assembly flow charts and equipment control item descriptions in the CNC machining line, constructs three types of knowledge graph entity nodes, establishes the entity basic network structure, and generates a process graph node set;

[0008] The dependency relationship construction module maps and groups the cutting parameter ranges and equipment linkage tags based on the process map node set to form a dependency linkage item table, and embeds the association paths in the map structure in an edge connection manner to generate a map dependency structure network;

[0009] The semantic entity matching module calls the graph dependency structure network to obtain device operation entities with semantically heterogeneous expressions in the graph, performs sequence alignment judgment and numerical offset calculation in sequence, sets graph interaction connections based on the co-occurrence frequency in the structure network, and generates an entity semantic fusion graph;

[0010] The process path screening module is based on the entity semantic fusion graph, calls the control parameter edge connecting each process node in the path, statistically outputs the difference between the processing dimension value and the corresponding upstream and downstream nodes, establishes the path connectivity structure between nodes and marks the path optimization level, and generates a process path optimization graph.

[0011] The present invention has the following improvements: the process graph node set includes an operation action node set, a component structure node set, a control parameter node set, a node-to-node pointing relationship set and a graph initial connection structure; the graph dependency structure network includes a step number sequence, an equipment linkage mark set, a control parameter edge set, a parameter dependency connection path set and a node-to-node linkage structure; the entity semantic fusion graph includes a semantic alignment node mapping, an offset difference annotation node, a parameter co-occurrence frequency record, a semantically heterogeneous node group and a fused path topology structure; the process path optimization graph includes a path node connectivity structure, an error elimination path list, a preferred level mark group, a processing target matching index and an optional path set.

[0012] The present invention is improved in that the graph node generation module includes:

[0013] The document parsing submodule obtains process documents, assembly flow charts, and equipment control item descriptions from the CNC machining line. It extracts the operating steps recorded in the process documents, the component numbers shown in the assembly flow charts, and each type of equipment control field in the equipment control item descriptions. After splitting the three types of text items into semantic paragraphs, the frequency of keyword occurrence is counted and matched against the operation vocabulary, component dictionary, and control parameter standard items. The word frequency distribution data for operation terms, structure terms, and control terms is established, generating a word frequency distribution ratio matrix.

[0014] The structure recognition submodule determines the content dependency relationships among the operation-type, structure-type, and control-type terms based on the term frequency distribution ratio matrix, sets the term groups whose dependency frequencies are higher than the structure matching benchmark value as candidate node groups, performs entity label normalization processing on the terms in the candidate node groups and maps them to standard term codes to obtain a set of terms with unique entity identifiers, and generates a standard term entity comparison set;

[0015] The node construction submodule calls the standard term entity reference set, sets three types of graph nodes according to the term categories based on the standard term codes of the operation class, structure class and control class, configures directed connection relationships between the three types of nodes according to the term dependency path, establishes a structured node network with complete directed edges, and generates a process graph node set.

[0016] The present invention is improved in that the dependency relationship construction module includes:

[0017] The reference extraction submodule extracts the corresponding numbers in the operation action node and the control parameter node based on the process map node set, obtains the processing instruction sequence number in the operation node and the equipment control code associated with the parameter node, and screens the node pairs with a number reference frequency greater than twice as the direct reference relationship unit by comparing the frequency of simultaneous occurrence of the numbers in the task flow list, and establishes the operation parameter reference frequency value;

[0018] The parameter combination submodule calls each set of operation step numbers and control parameter numbers based on the reference frequency value of the operation parameters, extracts the cutting depth, feed rate, and spindle speed of the corresponding process link, classifies and labels them based on the equipment type label, and reconstructs the operation step numbers in chronological order using the equipment number as the primary key. The parameter items are uniformly grouped into a fixed field set to establish a combination sequence mapping result;

[0019] The path connection submodule sets a directed connection path for the dependency relationship between nodes based on the result of the combined sequence mapping and the operation action node and control parameter node corresponding to each grouping field in the graph, writes the path structure into the node edge set, and marks the path source device number to generate a graph dependency structure network.

[0020] The present invention is improved in that the semantic entity matching module includes:

[0021] The structure extraction submodule calls the graph dependency structure network, extracts the corresponding application device number, operation sequence number and parameter value field according to the parameter item corresponding to the control parameter group in each node, groups the device numbers in the parameter group, arranges the operation sequence fields in the group in ascending order by number, maps the parameter value fields according to the node path number to generate a distribution set, and establishes a structure distribution relationship value;

[0022] The difference judgment submodule selects a set of parameter groups with consistent sequential numbers under each device number based on the structural distribution relationship value, extracts the parameter value fields in the corresponding nodes, calculates the absolute value of the difference between the parameter values under the same number and the overall distribution standard deviation, calculates the relative difference offset, and extracts the structural common nodes and path numbers based on the parameter groups whose relative difference offset values are lower than the offset difference threshold to generate an offset alignment node group;

[0023] The semantic fusion submodule aligns the node groups according to the offset, counts the frequency values of the node paths in the graph structure, assigns interaction identifiers to the path groups whose frequency values are greater than the co-occurrence threshold, updates the node-pointing connection edge set in the entity graph, and establishes an entity semantic fusion graph.

[0024] The present invention is improved in that the process path screening module includes:

[0025] The path extraction submodule extracts the path structure under each group of nodes pointing to the same processing target based on the entity semantic fusion graph, records the process node numbers and connection control parameter edge information in the path, establishes a pairing relationship between the path number and the control parameter edge, and generates the process path structure quantity;

[0026] The error judgment submodule extracts the processing dimension value output by each process node and the processing dimension value of the corresponding upstream and downstream nodes based on the process path structure, calculates the absolute difference between the nodes and compares the difference with the dimensional error critical value in the IT tolerance standard grade, calculates the processing deviation, summarizes the error marks for the path nodes with processing deviation greater than zero, screens out the corresponding path structure, obtains the compliant path number sequence, and establishes processing deviation elimination information;

[0027] Based on the processing deviation elimination information, the grade marking submodule re-counts the average difference between the node pairs in each path according to the compliant path number sequence, divides the path into multi-level range segments according to the proportional distribution between the statistical results and the minimum tolerance value of the tolerance grade, and adds a grade identification field to each path to establish a process path optimization map.

[0028] The present invention is improved in that the system further comprises:

[0029] The trend-driven early warning module calls the process path optimization map, collects the time series values of the real-time control parameters in operation, compares the consistency of the fluctuation direction within each cycle segment, and counts the number of continuous cycles. When there is a continuous fluctuation direction of the node and the number of cycles exceeds the standard value, the corresponding edge in the map is marked as a warning edge, and a map abnormal trend annotation set is generated;

[0030] The graph abnormal trend annotation set specifically includes periodic consistency marker edges, continuous offset path indexes, equipment abnormal fluctuation labels, warning path connection identifiers, and dynamic trend annotation mapping tables.

[0031] The present invention is improved in that the trend-driven early warning module includes:

[0032] The time series acquisition submodule calls the process path optimization map, extracts the periodic variation sequence of the spindle vibration value, the workpiece temperature rise value, and the spindle speed value during the processing stage according to the time series value of the operation control parameters of the equipment associated with the path node, and segments the parameter time series to establish a processing stage time series set;

[0033] The fluctuation identification submodule calls the processing stage time series set, identifies the fluctuation direction of three parameters: spindle vibration value, workpiece temperature rise value, and spindle speed value in each period, determines the positive and negative relationship between the numerical differences between adjacent data points in each period, and performs quantitative statistics on the sections with the same direction, calculates the fluctuation trend intensity, marks the continuous trend change period segments according to the fluctuation trend intensity value, and establishes the trend continuous intensity value;

[0034] Based on the trend continuity strength value, the anomaly marking submodule determines whether there is a situation where the continuous periodic trend strength on the edge corresponding to the path node exceeds the standard value of the control parameter fluctuation, marks the edges that meet the conditions in the path graph as warning edges, and establishes a graph abnormal trend annotation set.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are:

[0036] In the present invention, by extracting semantic entities from process documents, assembly processes and equipment control items and performing structured construction, the multi-dimensional connections between processing operations, component structures and control parameters can be covered, and a dependent linkage network can be established based on the reference relationship and parameter mapping between nodes. The heterogeneous expressions of equipment operation entities are integrated in the semantic comparison process, and semantic implicit relationship identification is achieved through sequence alignment and numerical difference comparison. Quantitative screening between processing dimensions and tolerance grades is introduced in path construction to ensure that the path connectivity structure is optimized while meeting dimensional accuracy constraints. Combined with the time series trend splitting and continuous fluctuation consistency judgment mechanism of the equipment operating status, it is possible to complete the dynamic marking of process deviation trends and the generation of warning annotation sets at the node path level, realizing closed-loop processing from knowledge modeling, path screening to trend prediction, and effectively improving the adaptability of process paths, parameter configuration accuracy and advance abnormality identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system flow chart of the present invention;

[0038] Figure 2 This is a flow chart of the graph node generation module of the present invention;

[0039] Figure 3 A flow chart of the dependency building module of the present invention;

[0040] Figure 4 This is a flowchart of the semantic entity matching module of the present invention;

[0041] Figure 5 This is a flow chart of the process path screening module of the present invention;

[0042] Figure 6 This is a flow chart of the trend-driven early warning module of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0045] See also Figure 1 The present invention provides a technical solution: a knowledge graph-driven manufacturing process optimization system, the system comprising:

[0046] The graph node generation module obtains the process documents, assembly flow charts and equipment control item descriptions in the CNC machining line, extracts the operation items in the process documents, the component nodes in the assembly flow charts and the control parameter fields in the equipment control item descriptions, and constructs three types of knowledge graph entity nodes respectively. It sets the corresponding pointing relationships between the operation action nodes, component structure nodes and control parameter nodes, establishes the entity basic network structure, and generates the process graph node set;

[0047] The dependency relationship construction module extracts the direct reference relationship between the operation action node and the control parameter node based on the process graph node set, classifies and combines the operation step numbers, cutting parameter ranges and equipment linkage tags involved, restores the operation step numbers to a sequence, maps and groups the cutting parameter ranges and equipment linkage tags to form a dependency linkage item table, and embeds the association paths in the graph structure in an edge-connected manner to generate a graph dependency structure network.

[0048] Cutting parameter range is widely used in the manufacturing industry to indicate the operating boundary parameters of spindle speed, feed rate, cutting depth, etc.

[0049] The semantic entity matching module calls the graph dependency structure network. Based on each node and path, it identifies the distribution structure of the control parameter group within the node. It then establishes a multidimensional comparison matrix based on the application devices, adjustment order, and numerical differences of the parameters. It obtains device operation entities with semantically heterogeneous expressions in the graph, performs sequence alignment judgment and numerical offset calculation in sequence, and selects operation nodes with implicit associations based on the offset difference threshold. It then sets graph interaction connections based on the co-occurrence frequency in the structure network to generate an entity semantic fusion graph.

[0050] The process path screening module extracts all optional path structures under the same processing target based on the entity semantic fusion graph, calls the control parameter edge connecting each process node in the path, statistically outputs the difference between the processing dimension value and the corresponding upstream and downstream nodes, and compares the difference with the critical value in the IT tolerance standard level one by one, eliminating paths with excessive errors, establishing the path connectivity structure between nodes, marking the path optimization level, and generating a process path optimization graph;

[0051] The trend-driven early warning module calls the process path optimization map and collects the time series values of the real-time control parameters in operation according to the equipment associated with the path nodes. It then splits the time series of the spindle vibration value, workpiece temperature rise value, and spindle speed value into corresponding periodic segments according to the processing stage. It compares the consistency of the fluctuation direction within each periodic segment and counts the number of continuous periods. When there is a continuous fluctuation direction of the same node and the number of periods exceeds the standard value, the corresponding edge in the map is marked as a warning edge, and a map abnormal trend annotation set is generated.

[0052] The process graph node set includes the operation action node set, the component structure node set, the control parameter node set, the node-to-node pointing relationship set and the graph initial connection structure. The graph dependency structure network includes the step number sequence, the equipment linkage mark set, the control parameter edge set, the parameter dependency connection path set and the node-to-node linkage structure. The entity semantic fusion graph includes the semantic alignment node mapping, the offset difference annotation node, the parameter co-occurrence frequency record, the semantic heterogeneous node group and the fusion path topology structure. The process path optimization graph includes the path node connectivity structure, the error elimination path list, the optimization level mark group, the processing target matching index and the optional path set. The graph abnormal trend annotation set specifically includes the periodic consistency mark edge, the continuous offset path index, the equipment abnormal fluctuation label, the warning path connection identifier and the dynamic trend annotation mapping table.

[0053] See also Figure 2 , the graph node generation module includes:

[0054] The document parsing submodule obtains process documents, assembly flow charts, and equipment control item descriptions from the CNC machining line. It extracts the operating steps recorded in the process documents, the component numbers shown in the assembly flow charts, and each type of equipment control field in the equipment control item descriptions. After splitting the three types of text items into semantic paragraphs, the frequency of keyword occurrence is counted and matched against the operation vocabulary, component dictionary, and control parameter standard items. The word frequency distribution data for operation terms, structure terms, and control terms is established, generating a word frequency distribution ratio matrix.

[0055] Obtain the process documents, assembly flow charts and equipment control item descriptions in the CNC machining line, extract the operation step content, part number and control field name in turn, take the five process documents recorded in the machining line in 2023 as an example, each process document contains 15 to 30 operation instructions, each instruction corresponds to the sequence code and action keyword of the machining link, divide the process document into paragraphs, extract the three most frequently occurring keywords in each paragraph to form an operation word frequency statistical set, for the numbering structure marked in the assembly flow chart, extract the numbering items, position identification and their adjacency relationship information in the figure, and count the frequency of repeated numbering to form a part number word frequency statistical set, further analyze the content of the equipment control item description, extract the "speed", "feed", "temperature" and other keywords. The control fields with "degree" as the first word are matched with the standard control parameter set, the spelling similarity of the field name and the degree of root association are compared, the corresponding control keywords are extracted, the frequency of each control word is counted and its unit and default value range are recorded. For example, the spindle speed is in "rpm" and the common range is 500 to 12000rpm, forming a control parameter word frequency statistical set. The above three types of text are divided into text paragraphs and matched with keywords to form three types of word frequency statistical data. The word frequency item groups of operation class, structure class and control class are constructed respectively. The word frequency normalization method is used to standardize the word frequency to the interval [0,1], and the word frequency distribution ratio matrix is constructed. Each unit in the matrix represents the relative frequency ratio of a certain type of word in the current text source, see Table 1;

[0056] Table 1 Word frequency ratio matrix of process document samples

[0057]

[0058] As shown in Table 1, the frequency of control terms is relatively low in some documents, and the coverage of this type of terms in the corresponding NC document content is relatively narrow. Based on the actual ratio matrix data, the word frequency distribution ratio matrix is obtained.

[0059] The structure recognition submodule determines the content dependency relationships between operational, structural, and control terms based on the word frequency distribution ratio matrix. Term groups with dependency frequencies higher than the structural matching benchmark value are set as candidate node groups. Term items in the candidate node groups are normalized by entity labels and mapped to standard term codes to obtain a set of terms with unique entity identifiers, generating a standard term entity comparison set.

[0060] Based on the word frequency distribution ratio matrix, the content dependency relationship between operation, structure and control terms is judged. For the dependency relationship judgment between terms, the frequency of co-occurrence of terms in the same paragraph is used for cumulative statistics. For example, if the "clamping" operation term and the "workpiece A5" structure term co-occur 7 times in 10 documents, the co-occurrence frequency of the two is 70%. The structural matching benchmark value is set to 60%. According to this benchmark, the term pairs with a co-occurrence frequency exceeding this value are formed into candidate node groups; the candidate terms are then mapped to standard item codes, such as "spindle speed ” is mapped to the equipment control standard field “SPINDLE_SPEED”, and the equipment standard field set defined in GB / T3811-2008 is used as the mapping dictionary to construct a standardized label list; the word combination [“feeding” + “conveyor B7”] with a co-occurrence frequency of 78% and the combination [“cooling” + “cooling unit X3”] with a co-occurrence frequency of 83% are confirmed to meet the set benchmark values, and unique entity codes are constructed for them. Mapping and matching are performed against the standard structural part numbering system to complete the entity label normalization processing and obtain the standard word entity reference set.

[0061] The node construction submodule calls the standard term entity reference set, sets three types of graph nodes according to the term categories based on the standard term codes of operation, structure, and control classes, configures directed connections between the three types of nodes according to the term dependency paths, establishes a structured node network with complete directed edges, and generates a process graph node set.

[0062] The constructed standard term entity comparison set is called to compare the standard term codes of the operation class, structure class and control class. The node "NODE_ACT_CLAMP" is constructed with the "clamping" operation, the node "NODE_COMP_A5" is constructed with the "workpiece A5" structure, and the node "NODE_CTRL_SSPEED" is constructed with the "SPINDLE_SPEED" control field. The connection direction is determined according to the co-occurrence relationship between the terms. If the operation class appears before the structure class, the starting point of the directed edge is set to point to the structure class node; further, an edge is set between the operation class and the control class. The edge weight depends on the term co-occurrence frequency and the encoding matching score of the standard term. If the co-occurrence frequency of "clamping" and "SPINDLE_SPEED" is 65% and the matching score is 90, the edge weight is calculated as To ensure the validity of the edge, the weight threshold is set to 0.5, and the following formula is used to determine whether to retain the edge relationship:

[0063] ;

[0064] in, Represents the normalized frequency value of the operation-related terms in the paragraph. Represents the normalized frequency value of the control term, represents the co-occurrence frequency of the two. Represents the semantic similarity score of the two in the standard item encoding. If the calculated result value exceeds 0.5, a connection edge relationship is established, and finally a process map node set is constructed. The benefit of the formula is that by comprehensively considering the word frequency difference and semantic similarity, the map connection relationship can more accurately reflect the actual dependency relationship in the control logic. Among them, Representative The normalized frequency value of the operation-type terms in the paragraph, Representative The normalized frequency value of the control class terms, Representative The co-occurrence frequency of word pairs, Representative The semantic similarity score of the word pair in the standard encoding matching dictionary is [0, 1]. Based on the word frequency of "clamping" in the sample document being 0.48 and the word frequency of "spindle speed" being 0.22, the co-occurrence frequency of the two is 0.65, and the semantic matching score is 0.9, the formula is substituted into the calculation:

[0065] ;

[0066] Since 0.2015 is lower than the edge setting threshold of 0.5, the edge is not established. For the combination of "feeding" and "cooling unit X3" with word frequencies of 0.45 and 0.40, if the co-occurrence frequency is 0.88 and the semantic score is 0.92, the calculation is:

[0067] ;The result shows that if the weight is too low, no edge connection will be set. Only when the weight is greater than 0.5, the combination will be embedded in the process map node network.

[0068] See also Figure 3 , the dependency building blocks include:

[0069] The reference extraction submodule extracts the corresponding numbers in the operation action node and the control parameter node based on the process map node set, obtains the processing instruction sequence number in the operation node and the equipment control code associated with the parameter node, and compares the frequency of simultaneous occurrence of the numbers in the task flow list. It selects node pairs with a number reference frequency greater than twice as direct reference relationship units and establishes the operation parameter reference frequency value.

[0070] When the reference extraction submodule extracts the numbers of operation action nodes and control parameter nodes based on the process map node set, it first calls all nodes marked as operation classes from the node set and reads the processing instruction serial numbers stored in each node in turn. The numbering format is set to four-digit Arabic numerals (such as "OPC1" to "OPC120"). Then, the control parameter nodes with the same serial numbers are extracted. The serial numbers in the control parameter nodes are the equipment control field codes, using the three-segment labeling rule "EQ-XY-Z", where "XY" is the equipment type code and "Z" is the control parameter serial number. For example, "EQ-MC-7" represents the control parameter number 7 of the CNC milling machine. Then, a task number index table is constructed from the task flow list. The number of times each pair of operation instruction numbers and control parameter numbers co-occurs in the task flow sequence is counted. For example, "OPC12" and "EQ-MC-7" appear simultaneously in task numbers "T-008", "T-017", and "T-021". Because their co-occurrence number is 3, which exceeds the set threshold of 2, they are filtered as reference relationship node pairs. The threshold here is set based on the reference frequency distribution of standard production line tasks. By counting 200 historical task samples, the average co-occurrence number of operation-parameter pairs is 1.68. The integer value corresponding to the 70% percentile is taken as the screening threshold, so the threshold is set to 2 to ensure that only core pairs of repeated calls are retained. The node pairs that meet the conditions are included in the reference relationship set, and their co-occurrence frequency is recorded as the reference frequency value. If a group of operation parameter node pairs, such as "OPC12-EQ-MC-7", has an occurrence frequency of 3, then its corresponding reference frequency value is recorded as 3, which constitutes the basis for subsequent parameter combination judgment. The operation parameter reference frequency value includes the operation instruction number, control parameter number, and reference frequency.

[0071] The parameter combination submodule calls each set of operation step numbers and control parameter numbers based on the reference frequency value of the operation parameters, extracts the cutting depth, feed rate, and spindle speed of the corresponding process link, classifies and labels them based on the equipment type label, and reconstructs the operation step numbers in chronological order using the equipment number as the primary key. The parameter items are uniformly grouped into a fixed field set to establish the combination sequence mapping result;

[0072] Based on the reference frequency value set output by the previous submodule, the parameter combination submodule extracts data for each pair of operation step number and control parameter number in turn. First, the task execution log is called with the operation step number as the index to obtain the processing parameter items set in the corresponding process link, where the cutting depth is recorded in the unit "mm", the feed speed is recorded in "mm / min", and the spindle speed is recorded in "rpm". Then, the device information associated with the control parameter number is read to identify its device type label, such as "MC" for CNC milling machine, "DR" for drilling unit, etc. After packaging and marking the three parameters and equipment types, a multi-field mapping matrix is established. For the node pair "OPC12-EQ-MC-7" in task number "T-017," the query results show a cutting depth of 0.5 mm, a feed rate of 450 mm / min, a spindle speed of 1200 rpm, and an equipment type tag of "MC." Based on this data, the operation step numbers are sorted by task time, and their execution order is reconstructed as "OPC11 → OPC12 → OPC15," forming a mapping sequence. The three parameter fields are then grouped, and a fixed field set is set to [cutting depth, feed rate, spindle speed, equipment tag]. With the equipment number "MC-001" as the primary key, a correspondence is established between the sequence and the parameter set to generate a combined sequence mapping result. The following table lists sample combinations:

[0073] Table 2 Example of process task parameter combinations

[0074]

[0075] As shown in Table 2, in different step combinations, the parameter values are fine-tuned according to the device type and task load, forming a grouped data structure with timing constraints. The combined sequence mapping result includes the device number, operation step sequence, and grouped parameter field value.

[0076] Based on the combined sequence mapping results, the path connection submodule calls the operation action node and control parameter node corresponding to each grouped field in the graph, sets a directed connection path for the dependency relationship between nodes, writes the path structure into the node edge set, and annotates the path source device number to generate a graph dependency structure network;

[0077] Based on the combined sequence mapping results, the path connection submodule sequentially reads the operation action nodes and control parameter nodes corresponding to each device number, searches the graph structure for corresponding node instances, and checks whether there are parameter dependencies or instruction call logic between them. By comparing the "call order" field with the "parameter dependency index" in the node attribute table, a directed path relationship is established if the call order is less than the dependency index and the parameter fields are nested. For device number "MC-001," for example, its operation sequence is "OPC11 → OPC12 → OPC15," and its corresponding control parameter nodes are "EQ-MC-5," "EQ-MC-7," and "EQ-MC-9." The path direction is recorded in the node edge connection as "OPC11 → EQ-MC-5 → OPC12 → EQ-MC-7 → OPC15 → EQ-MC-9," and the source device of the path field is marked as "MC-001." In the path connection logic, if a path direction violates the device dependency rule, such as calling the "spindle speed" before setting the "feed rate," the path setting is canceled. Finally, the structured path is written into the node edge set. The benchmark value for judging the establishment of path connection is parameter dependency matching rate ≥ 85%. This ratio is calculated by the field structure similarity in the standard device control link. 12 groups of standard reference paths are set, and the average matching rate is 87.3%. Therefore, 85% is used as the benchmark value for valid path construction. In the example, the path "OPC11→EQ-MC-5→OPC12→EQ-MC-7" has a matching rate of 91%, which meets the set standard. The graph dependency structure network specifically includes node path direction groups, device number mapping tags, and control relationship edge weights.

[0078] See also Figure 4 , the semantic entity matching module includes:

[0079] The structure extraction submodule calls the graph dependency structure network, extracts the corresponding application device number, operation sequence number and parameter value field according to the parameter items corresponding to the control parameter group in each node, groups the device numbers in the parameter group, arranges the operation sequence fields in the group in ascending order by number, maps the parameter value fields to the node path number to generate a distribution set, and establishes the structure distribution relationship value;

[0080] When the structure extraction submodule calls the graph dependency structure network, it first obtains the parameter item information contained in the control parameter group in each node, and parses each control parameter group in turn to obtain its corresponding device number field, operation sequence number field and parameter value field. The specific calling form is to read the "Dev_ID", "Step_Seq" and "Param_Value" columns in the structure node data table. The device number such as "EQ-101", "EQ-202" is expressed in a unified format, and the operation sequence number is a set of ascending sequences such as "001", "002", "003", etc. After pre-normalization, the parameter value fields are represented as floating-point numbers in the 0-1 range. Sample data is shown in Table 4. All control parameter groups are then grouped by device number. For example, if the device "EQ-101" has five parameter sets, each appearing in operation steps 001 to 005, the groups are labeled "EQ-101-G1" to "EQ-101-G5." Within each group, the operation sequence numbers are sorted in ascending order to ensure consistency in the operation sequence. Using the node path number field as an index, the parameter value fields are mapped sequentially onto the path sequence to form a distribution vector set. To clarify the distribution relationship, the aggregation trend of the parameter vector sequence along the path number is calculated within each device group to obtain the parameter structure distribution relationship value. This value is generated by calculating the sequence similarity and value distribution consistency of each parameter set along the path. The value range is 0-1, with higher values indicating greater structural stability. For example, for the device "EQ-101-G2", its parameter value sequence [0.55, 0.63, 0.61] corresponds to the path sequence [ND-1, ND-3, ND-5], showing a consistent distribution structure in the path node graph, and the structural distribution relationship value is 0.89. The structural distribution relationship value specifically includes the path number sequence, the parameter normalized value sequence, and the sequence distribution trend indicator.

[0081] The difference judgment submodule selects the parameter group set with the same sequence number under each device number according to the structural distribution relationship value, extracts the parameter value field in the corresponding node, and calculates the absolute value of the difference between the parameter values under the same number and the overall distribution standard deviation using the formula:

[0082] ;

[0083] The relative difference offset is obtained by operation, and according to the parameter group whose relative difference offset value is lower than the offset difference threshold, the structural common nodes and path numbers are extracted to generate the offset alignment node group;

[0084] in, Indicates the Device The normalized value of the control parameter, Indicates the corresponding The normalized value of the control parameter, Indicates the The average value of the normalized values of all parameters in the device parameter group, represents the average value of the normalized values of all parameters in the reference device parameter group, Indicates the total number of control parameters involved in the calculation for each group, It represents the relative difference deviation between two equipment operating parameter groups, which is a measure of the overall deviation degree of the value distribution structure of the two groups of parameters;

[0085] The difference judgment submodule extracts parameter groups with the same operation sequence number set under different device numbers based on the aforementioned structural distribution relationship value, and obtains group pairs with consistent sequence numbers under multiple devices through structural index matching. For example, "EQ-101-G2" and "EQ-202-G2" both contain operation steps 001 to 003. Under the premise of ensuring the consistency of the number of parameters, the parameter value fields under the same number in the two groups are extracted respectively and the corresponding matrix is constructed. The two groups of data are set as and For example, the value of "EQ-101-G2" is [0.55, 0.63, 0.61], and the value of "EQ-202-G2" is [0.52, 0.64, 0.59]. The relative difference deviation is calculated using the following formula:

[0086] ;

[0087] in, Indicates the Device The normalized value of the control parameter, which is derived from the original control parameter (such as spindle speed, feed speed, etc.) after normalization transformation of the maximum and minimum values; Indicates the reference device The item corresponds to the normalized value of the control parameter; It represents the arithmetic mean of the normalized values of all control parameters of the device involved in the comparison, and is calculated as follows: ; It represents the average value of the normalized value of the control parameter at the same position in the reference device; Indicates the total number of control parameters in each group involved in the comparison; Indicates the relative difference between two device parameter groups.

[0088] The structural logic of the above formula is:

[0089] Molecular part: , represents the absolute value difference between each corresponding parameter item, and sums them up to reflect the total offset between the overall parameters;

[0090] Denominator: , which represents the square root of the sum of the squares of the internal deviations from the mean of the two groups of parameters, that is, the combined amount of standard deviation, which is used to measure the degree of fluctuation of the parameter value itself;

[0091] Overall logic: normalize the absolute difference and the joint standard deviation. If the means of the two groups of parameters are close, the fluctuation is small and the difference is large, then The value is large, otherwise it is small. This structure avoids the situation where the comparison fails due to excessive fluctuations in the original parameter distribution, making the comparison robust and widely adaptable.

[0092] In the example, set the "EQ-101-G2" parameter group to ,average value , "EQ-202-G2" parameter group is ,average value , bring each value into the formula to perform the operation:

[0093] ;

[0094] ;

[0095] ;

[0096] The offset difference threshold is set to 0.6. The setting basis is: the median of the S value calculated from the 50 control parameter pairs in the typical equipment comparison data set is 0.59, which is selected as the acceptable upper limit and rounded up to 0.6. Based on this, the S value is compared with the threshold. , the offset differences between these parameter groups are judged to be acceptable, and then the common structural nodes and path numbers can be extracted to establish offset-aligned node groups. This result shows that the device has high consistency in the control parameters executed under the same operation steps, and has mappable dependency features in the structure, which can be used for subsequent structural fusion and semantic binding processing.

[0097] The semantic fusion submodule aligns node groups based on offsets, counts the frequency of node paths in the graph structure, assigns interaction identifiers to path groups with frequency values greater than the co-occurrence threshold, updates the node-pointing edge set in the entity graph, and establishes an entity semantic fusion graph.

[0098] The semantic fusion submodule, based on offset-aligned node groups, traverses all path instances in the graph structure, counting the frequency of occurrence of each path number within these node groups. A path statistics table records the number of times each path is referenced in the graph structure, and then selects path groups with a co-occurrence frequency greater than a co-occurrence threshold. The co-occurrence threshold is set to ≥8 path occurrences. Based on the summary statistics within the graph task dataset, the average occurrence of same-node path groups is 7.4, with an upper integer value of 8. For example, if the path groups "P-01, P-02, and P-03" appear 12, 9, and 11 times, respectively, meeting the threshold, they are assigned the interaction identifier "SIG-P." The corresponding path-pointing edge set is updated in the graph database, and the field "Semantic_Bind" with its value "SIG-P" is added, indicating that the path group forms a semantic fusion relationship, ultimately forming an entity semantic fusion graph. In this graph structure, each fused path corresponds to a set of offset-aligned nodes, annotated with the corresponding fusion label and co-occurring device group number. The entity semantic fusion graph includes a fused path list, a set of semantic interaction identifiers, and a fused device number relationship table.

[0099] See also Figure 5 , the process path screening module includes:

[0100] The path extraction submodule is based on the entity semantic fusion graph. It extracts the path structure under each group of nodes pointing to the same processing target, records the process node numbers and connection control parameter edge information in the path, establishes a pairing relationship between the path number and the control parameter edge, and generates the process path structure quantity.

[0101] After obtaining the node group pointing to the same processing target in the entity semantic fusion graph, first extract the node set aggregated under the same target based on the unique identifier of each node in the graph, read the processing step number contained in each node path, and construct a path structure index. Further, based on the annotation information of the connecting edges, obtain the control parameter edges contained in each path segment one by one, extract the equipment action instruction number and parameter field corresponding to the control parameter edge, and establish a one-to-one pairing array of path numbers and control edge items. For example, for the path P-101 corresponding to the node group G1, its process node sequence is N1→N2→N3, and the control edges are C1, C2, and C3 in sequence. The corresponding pairing relationship is established as {P-101:[(N1,C1),(N2,C2),(N3,C3)]}, thereby completing the generation of a multivariate mapping structure set of path structure and control parameter edge items as the basis for the composition of the process path structure quantity.

[0102] The error judgment submodule extracts the processing dimension value output by each process node and the processing dimension value of the corresponding upstream and downstream nodes according to the process path structure, calculates the absolute difference between the nodes, and compares the difference with the dimensional error critical value in the IT tolerance standard grade using the formula:

[0103] ;

[0104] Obtain machining deviation through calculation, summarize error marks for path nodes with machining deviation greater than zero, screen out corresponding path structures, obtain compliant path number sequences, and establish machining deviation elimination information;

[0105] in, Indicates the The upstream node of each path node outputs the normalized value of the processing size. Indicates the The downstream node of the path node inputs the normalized value of the processing size, Indicates the The absolute value of the measured error of each path node, Indicates the The IT tolerance critical value corresponding to each path node, Indicates the total number of nodes involved in the judgment in the path, It represents the overall machining deviation of the path, which is used to measure the cumulative machining error level between all nodes on the path;

[0106] According to the control parameter edges involved in each path in the process path structure, the processing dimension fields in each path node are extracted, and a processing dimension mapping set is constructed, calling the upstream dimension fields output by each process node The input size field received by its immediate downstream nodes , subtract the two and take the absolute value , to obtain the actual processing deviation value of each node, and extract the absolute value of the measured error of each node , and its corresponding IT tolerance critical value , the error ratio term Perform square operation and sum, combined with the number of nodes in each path Calculate the total deviation expression , the specific calculation formula is as follows:

[0107] ;

[0108] in: For the The upstream node of each path node outputs the normalized value of the processing size, which is expressed as the target size provided by the previous device or node after the processing is completed;

[0109] For the The normalized value of the processing size of the downstream node input of each path node is the required specification of the input size of the next level process;

[0110] For the The actual processing deviation value of each path node is collected by the online dimension measurement system and is the offset between the dimension after processing and the target dimension.

[0111] For the The IT level critical value corresponding to each node is taken from the error tolerance limit corresponding to the standard size segment;

[0112] is the total number of nodes involved in error judgment in the path;

[0113] To obtain the average deviation, we divide it by the total number of nodes to make the results comparable.

[0114] As shown in Table 3, four paths numbered P-101 to P-104 are selected, and their upstream and downstream dimensions, measured errors, and IT tolerance values are obtained respectively. Then, the above formulas are substituted in turn to calculate the dimension difference and error term of each path node, and the mean is obtained. , the sum of squared error terms is 1.20, the square root is 1.0954, and the final deviation is .

[0115] Table 3 Process path processing error data table:

[0116]

[0117] This result shows that the final deviation If it is less than zero, it means that the actual machining errors of all nodes on the current path are within the tolerance range. The path is a compliant path and is retained.

[0118] The benefit of the formula is that, through the composite calculation of the machining dimension difference and the standard error tolerance value, it can comprehensively measure the overall error distribution of all nodes in the path structure. It not only reflects the average error level, but also effectively weakens the influence of local extreme errors, thereby improving the robustness and accuracy of path elimination judgment, and avoiding the situation where the quality of the path is misjudged due to single-point error.

[0119] The grade marking submodule recalculates the average difference between node pairs in each path based on the processing deviation elimination information and the compliant path number sequence. Based on the proportional distribution between the statistical results and the minimum tolerance value of the tolerance grade, the path is divided into multiple grade range segments, and a grade identification field is added to each path to establish a process path optimization map.

[0120] After extracting all the retained path number sequences from the obtained processing deviation elimination amount, the size differences formed by adjacent node pairs in each path are re-counted, and the average difference of each path is calculated. For example, the size differences between the three nodes of path P-101 are 0.10mm, 0.15mm, and 0.05mm respectively, and the average difference is mm, and compare the value with the minimum tolerance value in the tolerance grade. For example, if the minimum tolerance value is 0.02mm, the current path is in a proportional interval of 5 times the tolerance range. When dividing, the classification basis is set as follows: a tolerance multiple less than 2 is defined as a first-level path, 2 to 4 times is a second-level path, and more than 4 times is a third-level path. Therefore, P-101 is classified into the third-level path segment, and a grade field Grade=3 is added to it. At the same time, a grade annotation label is added to the atlas structure for subsequent calling of the process path optimization atlas.

[0121] See also Figure 6 , the trend-driven early warning module includes:

[0122] The time series acquisition submodule calls the process path optimization map and extracts the periodic variation sequence of the spindle vibration value, workpiece temperature rise value, and spindle speed value during the processing stage based on the time series values of the operation control parameters of the equipment associated with the path nodes. It then segments the parameter time series to establish a processing stage time series set.

[0123] According to the path numbers and node sets recorded in the process path optimization map, the manufacturing equipment identifiers associated with the nodes are extracted one by one, and the operation log of each device or the control parameter data recorded in the industrial control system are called to collect the spindle vibration value, workpiece temperature rise value and spindle speed value during the processing execution phase to form three independent time series record sets. In order to enhance data adaptability, the three types of time series are synchronously associated according to the node timestamps and segmented with the time window as the period boundary. If the timestamps of nodes N1~N3 with path number PT-201 during the processing are 13s, 27s and 41s respectively, the spindle vibration value in each period is extracted with each 14s as the period window. , workpiece temperature rise And spindle speed value , generate an independent sequence segment for each cycle segment, and record its segment number as f, forming a time series set after segmentation structure. If a node is missing data in a cycle segment, it is filled by taking the median of its adjacent time window data to avoid the breakpoint affecting trend identification. Finally, a processing stage time series set based on the one-to-one mapping between path nodes and time windows is formed in the processing stage. The time window length is set to a fixed 14s to match the periodic frequency of the spindle vibration signal in the processing stage.

[0124] The fluctuation identification submodule calls the time series set of the processing stage to identify the fluctuation direction of the three parameters of spindle vibration value, workpiece temperature rise value and spindle speed value in each cycle segment, judges the positive and negative relationship of the numerical difference between adjacent data points in each segment, and counts the number of segments with the same direction using the formula:

[0125] ;

[0126] Obtain the fluctuation trend strength through calculation, mark the continuous trend change period according to the fluctuation trend strength value, and establish the trend continuous strength value;

[0127] in, Indicates the The incremental normalized value of the spindle vibration value within a period, Indicates the Normalized value of measurement time within a period, Indicates the Normalized value of acceleration amplitude within a period, Indicates the The normalized value of the workpiece temperature rise increment within a period, Indicates the normalized benchmark of the initial temperature rise of the workpiece, Indicates the The incremental normalized value of the spindle speed value within a cycle segment, Indicates the normalized value of the spindle reference speed. Indicates the number of periodic segments containing temperature rise and speed trends, Indicates the number of period segments containing oscillation trends, Indicates the trend continuity strength value;

[0128] Call each parameter sequence in the time series set of the processing stage, and compare the positive and negative relationship of the numerical differences of adjacent sampling points in the same period. If the values of two consecutive sampling points are increasing, the fluctuation direction is marked as positive, otherwise it is negative. Count the length of the interval with the same fluctuation direction of the three types of parameters in each period, and extract the incremental normalized value of the spindle vibration value in each period. , the normalization method is to divide the instantaneous change value by the maximum vibration range; extract the measurement time The normalized value of the total sampling time of the periodic segment is normalized by dividing the total segment duration by the maximum segment duration; the acceleration amplitude is extracted The normalized value is the ratio of the maximum vibration amplitude of the current period to the maximum value of the entire data period. Substitute the above three parameters into the vibration trend item Perform periodic summation in , and then take the temperature rise trend item and speed trend items , using their corresponding initial values 、 After normalization, the final trend value is calculated as:

[0129] ;

[0130] In this formula:

[0131] Indicates the The amplitude of the spindle vibration change in each period is normalized to represent the vibration fluctuation trend;

[0132] Indicates the sampling duration of the period, reflecting the duration of vibration;

[0133] Indicates the peak acceleration amplitude in this segment, which is used to measure the vibration intensity;

[0134] Indicates the temperature rise change of the workpiece in this section;

[0135] It is the normalized value of the initial temperature of the workpiece during machining, and is used as the normalized benchmark for temperature rise;

[0136] is the spindle speed change;

[0137] Normalized value of the reference speed when the spindle machining starts;

[0138] Represents the total number of cycle segments containing the main shaft vibration trend;

[0139] Indicates the total number of cycle segments containing temperature rise and speed trend items;

[0140] The calculation logic is: by summing the vibration trend intensity terms in all period segments and subtracting the mean of the temperature rise and speed trend terms, a fluctuation trend intensity measurement value is formed. , used to mark the strength of trend continuity.

[0141] Take the vibration trend sampling data shown in Table 4 as an example for calculation:

[0142] Table 4 Vibration trend period sample data table

[0143]

[0144] Calculate using the data from the first three periods in the table above:

[0145] The cumulative part of the vibration term is:

[0146] ;

[0147] The average part of temperature rise and speed is:

[0148] ;

[0149] Final trend strength value ,The result shows that the trend fluctuation amplitude is low, which belongs to the stable fluctuation trend section.

[0150] The benefit of the formula is that it enhances the global nature of trend identification through the combined evaluation of vibration, temperature rise, and speed parameters. In particular, the expression of the vibration part after compound adjustment of acceleration and sampling time effectively alleviates the risk of misjudgment of abnormal fluctuations caused by differences in measurement cycles, ensuring the relative consistency of trend strength and periodic stability.

[0151] The anomaly annotation submodule determines whether the trend strength on the edge corresponding to the path node exceeds the standard value of the control parameter fluctuation based on the trend continuity strength value, marks the edge that meets the conditions in the path graph as a warning edge, and establishes a graph anomaly trend annotation set;

[0152] According to the trend continuous strength value The calculation results of the path node are used to determine the periodic segments associated with the adjacent edges. Whether the value exceeds the standard fluctuation control parameter range continuously, the control threshold is set to ±0.025. If it is satisfied within three consecutive period segments, or , then a warning mark is placed on the edge in the path structure, and the structure is marked as an abnormal trend area in the processing technology map. For example, the node pair (N2→N3) has three consecutive vibration trend values of 0.031, 0.037, and 0.029 in the path numbered PT-203, all of which are higher than the standard value. Then its connecting edge is marked as a warning edge and enters the map abnormal trend annotation set. This set is retained as the basis for subsequent path optimization elimination or regulation. The trend strength annotation of all edges is displayed in an interval range with the format of "0.029±0.004", indicating that the continuous trend of the edge under a specific path shows concentrated fluctuations.

[0153] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A knowledge graph-driven manufacturing process optimization system, characterized in that: The system comprises: The graph node generation module obtains the process documents, assembly flow charts and equipment control item descriptions in the CNC machining line, constructs three types of knowledge graph entity nodes, establishes the entity basic network structure, and generates a process graph node set; The dependency relationship construction module maps and groups the cutting parameter ranges and equipment linkage tags based on the process map node set to form a dependency linkage item table, and embeds the association paths in the map structure in an edge connection manner to generate a map dependency structure network; The semantic entity matching module calls the graph dependency structure network to obtain device operation entities with semantically heterogeneous expressions in the graph, performs sequence alignment judgment and numerical offset calculation in sequence, sets graph interaction connections based on the co-occurrence frequency in the structure network, and generates an entity semantic fusion graph; The process path screening module, based on the entity semantic fusion graph, calls the control parameter edge connected to each process node in the path, statistically outputs the difference between the processing dimension value and the corresponding upstream and downstream nodes, establishes the path connectivity structure between the nodes, and marks the path optimization level, thereby generating a process path optimization graph; The process path screening module includes: The path extraction submodule is based on the entity semantic fusion graph and, according to the node groups pointing to the same processing target, extracts the path structure under each group of nodes, records the process node numbers and connection control parameter edge information in the path, establishes a pairing relationship between the path number and the control parameter edge, and generates the process path structure quantity.

2. The knowledge graph-driven manufacturing process optimization system according to claim 1, characterized in that: The process graph node set includes an operation action node set, a component structure node set, a control parameter node set, a node-to-node pointing relationship set and a graph initial connection structure; the graph dependency structure network includes a step number sequence, an equipment linkage mark set, a control parameter edge set, a parameter dependency connection path set and a node-to-node linkage structure; the entity semantic fusion graph includes a semantic alignment node mapping, an offset difference annotation node, a parameter co-occurrence frequency record, a semantically heterogeneous node group and a fused path topology structure; the process path optimization graph includes a path node connectivity structure, an error elimination path list, a preferred level mark group, a processing target matching index and an optional path set.

3. The knowledge graph-driven manufacturing process optimization system according to claim 2, characterized in that: The graph node generation module includes: The document parsing submodule obtains process documents, assembly flow charts, and equipment control item descriptions from the CNC machining line. It extracts the operating steps recorded in the process documents, the component numbers shown in the assembly flow charts, and each type of equipment control field in the equipment control item descriptions. After splitting the three types of text items into semantic paragraphs, the frequency of keyword occurrence is counted and matched against the operation vocabulary, component dictionary, and control parameter standard items. The word frequency distribution data for operation terms, structure terms, and control terms is established, generating a word frequency distribution ratio matrix. The structure recognition submodule determines the content dependency relationships among the operation-type, structure-type, and control-type terms based on the term frequency distribution ratio matrix, sets the term groups whose dependency frequencies are higher than the structure matching benchmark value as candidate node groups, performs entity label normalization processing on the terms in the candidate node groups and maps them to standard term codes to obtain a set of terms with unique entity identifiers, and generates a standard term entity comparison set; The node construction submodule calls the standard term entity reference set, sets three types of graph nodes according to the term categories based on the standard term codes of the operation class, structure class and control class, configures directed connection relationships between the three types of nodes according to the term dependency path, establishes a structured node network with complete directed edges, and generates a process graph node set.

4. The knowledge graph-driven manufacturing process optimization system according to claim 3, characterized in that: The dependency building module includes: The reference extraction submodule extracts the corresponding numbers in the operation action node and the control parameter node based on the process map node set, obtains the processing instruction sequence number in the operation node and the equipment control code associated with the parameter node, and screens the node pairs with a number reference frequency greater than twice as the direct reference relationship unit by comparing the frequency of simultaneous occurrence of the numbers in the task flow list, and establishes the operation parameter reference frequency value; The parameter combination submodule calls each set of operation step numbers and control parameter numbers based on the reference frequency value of the operation parameters, extracts the cutting depth, feed rate, and spindle speed of the corresponding process link, classifies and labels them based on the equipment type label, and reconstructs the operation step numbers in chronological order using the equipment number as the primary key. The parameter items are uniformly grouped into a fixed field set to establish a combination sequence mapping result; The path connection submodule sets a directed connection path for the dependency relationship between nodes based on the result of the combined sequence mapping and the operation action node and control parameter node corresponding to each grouping field in the graph, writes the path structure into the node edge set, and marks the path source device number to generate a graph dependency structure network.

5. The knowledge graph-driven manufacturing process optimization system according to claim 4, characterized in that: The semantic entity matching module includes: The structure extraction submodule calls the graph dependency structure network, extracts the corresponding application device number, operation sequence number and parameter value field according to the parameter item corresponding to the control parameter group in each node, groups the device numbers in the parameter group, arranges the operation sequence fields in the group in ascending order by number, maps the parameter value fields according to the node path number to generate a distribution set, and establishes a structure distribution relationship value; The difference judgment submodule selects a set of parameter groups with consistent sequential numbers under each device number based on the structural distribution relationship value, extracts the parameter value fields in the corresponding nodes, calculates the absolute value of the difference between the parameter values under the same number and the overall distribution standard deviation, calculates the relative difference offset, and extracts the structural common nodes and path numbers based on the parameter groups whose relative difference offset values are lower than the offset difference threshold to generate an offset alignment node group; The semantic fusion submodule aligns the node groups according to the offset, counts the frequency values of the node paths in the graph structure, assigns interaction identifiers to the path groups whose frequency values are greater than the co-occurrence threshold, updates the node-pointing connection edge set in the entity graph, and establishes an entity semantic fusion graph.

6. The knowledge graph-driven manufacturing process optimization system according to claim 5, characterized in that: The process path screening module also includes: The error judgment submodule extracts the processing dimension value output by each process node and the processing dimension value of the corresponding upstream and downstream nodes based on the process path structure, calculates the absolute difference between the nodes and compares the difference with the dimensional error critical value in the IT tolerance standard grade, calculates the processing deviation, summarizes the error marks for the path nodes with processing deviation greater than zero, screens out the corresponding path structure, obtains the compliant path number sequence, and establishes processing deviation elimination information; Based on the processing deviation elimination information, the grade marking submodule re-counts the average difference between the node pairs in each path according to the compliant path number sequence, divides the path into multi-level range segments according to the proportional distribution between the statistical results and the minimum tolerance value of the tolerance grade, and adds a grade identification field to each path to establish a process path optimization map.

7. The knowledge graph-driven manufacturing process optimization system according to claim 6, characterized in that: The system further comprises: The trend-driven early warning module calls the process path optimization map, collects the time series values of the real-time control parameters in operation, compares the consistency of the fluctuation direction within each cycle segment, and counts the number of continuous cycles. When there is a continuous fluctuation direction of the node and the number of cycles exceeds the standard value, the corresponding edge in the map is marked as a warning edge, and a map abnormal trend annotation set is generated; The graph abnormal trend annotation set specifically includes periodic consistency marker edges, continuous offset path indexes, equipment abnormal fluctuation labels, warning path connection identifiers, and dynamic trend annotation mapping tables.

8. The knowledge graph-driven manufacturing process optimization system according to claim 7, characterized in that: The trend-driven early warning module includes: The time series acquisition submodule calls the process path optimization map, extracts the periodic variation sequence of the spindle vibration value, the workpiece temperature rise value, and the spindle speed value during the processing stage according to the time series value of the operation control parameters of the equipment associated with the path node, and segments the parameter time series to establish a processing stage time series set; The fluctuation identification submodule calls the processing stage time series set, identifies the fluctuation direction of three parameters: spindle vibration value, workpiece temperature rise value, and spindle speed value in each period, determines the positive and negative relationship between the numerical differences between adjacent data points in each period, and performs quantitative statistics on the sections with the same direction, calculates the fluctuation trend intensity, marks the continuous trend change period segments according to the fluctuation trend intensity value, and establishes the trend continuous intensity value; Based on the trend continuity strength value, the anomaly marking submodule determines whether there is a situation where the continuous periodic trend strength on the edge corresponding to the path node exceeds the standard value of the control parameter fluctuation, marks the edges that meet the conditions in the path graph as warning edges, and establishes a graph abnormal trend annotation set.

Images