Dial plate processing optimization system based on big data

Abstract

The invention discloses a dial plate processing optimization system based on big data, and relates to the technical field of dial plate processing, data is collected, anti-fact sample data is generated, a causal atlas structure including processing process nodes and an edge weight function model is constructed, processing path performance is evaluated based on an abstract objective function, and the processing process is optimized. According to the method, the causal atlas is constructed based on multi-source data such as tool wear degree data and spindle vibration amplitude data, and path deduction and parameter optimization of a target state are realized in combination with an anti-fact sample; self-adaptive control over the machining state is achieved through a structured deviation matrix, a causal relationship and a weight function model are dynamically corrected through a feedback learning mechanism, and data-driven optimization and defect prevention in the machining process are achieved.

Description

Technical Field

[0001] The present invention relates to the field of dial processing technology, and in particular to a dial processing optimization system based on big data. Background Art

[0002] Current dial machining quality optimization relies on empirically driven, single-factor control models, making it difficult to address the complex causal relationships between multi-parameter coupling, state evolution, and micro-defect formation. Existing systems lack a unified model for the combined effects of factors such as tool wear, spindle vibration, and environmental fluctuations. This often prevents early warning of defect generation paths and lacks feedback control capabilities. Machining path optimization often relies on static rules and empirical templates, making it impossible to reverse-engineer the target quality state and incapable of real-time adjustment of machining strategies to process fluctuations.

[0003] Currently, a Chinese invention patent application numbered CN202311586173.1 discloses an instrument and its processing system and method. The instrument includes a housing, wherein a plurality of rubber blocks are arranged in an annular shape on the inner side of the housing, and slots are provided on the plurality of rubber blocks. The plurality of rubber blocks are inserted into the outer periphery of the dial through the slots, and the dial is arranged in the housing. An instrument processing system includes a clamping ring, wherein two clamping rings are arranged in front and back, and a plurality of positioning grooves are provided in an annular shape on the opposite surfaces of the two clamping rings. The two clamping rings are clamped on the plurality of rubber blocks, and the plurality of rubber blocks are respectively located in the plurality of positioning grooves. An instrument processing method includes the following steps: S1: clamping two round seats on the front and back sides of the dial respectively, and clamping two clamping rings on the outside of the plurality of rubber blocks; S2: placing the dial and the plurality of rubber blocks into the housing through the two clamping rings; S3: gluing a conical rubber ring to the outer edges of the front and back sides of the housing respectively; S4: covering the outside of the two conical rubber rings with a transparent cover.

[0004] The above technologies make it difficult to build a unified model with a temporal causal structure based on multi-source heterogeneous data, and are unable to achieve reverse path deduction of the target state, adaptive parameter adjustment, and dynamic update and closed-loop optimization of the causal graph. Summary of the Invention

[0005] The technical problem solved by the present invention is that it is difficult for the existing technology to construct a unified model with a temporal causal structure based on multi-source heterogeneous data, and it is impossible to achieve reverse path deduction of the target state, adaptive parameter adjustment, and dynamic update and closed-loop optimization of the causal graph.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: A dial processing optimization system based on big data, including a data modeling module, a cause-effect diagram module, a solution deduction module, an adaptive control module, and a feedback learning module: The data modeling module is used to collect tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, processing image data and process log data, and generate counterfactual sample data; The causal graph module is used to construct a causal graph structure including a processing process node and an edge weight function model; The solution deduction module is used to evaluate the processing path performance based on the abstract objective function; The adaptive control module is used to compare state deviations and generate standardized parameter instructions; The feedback learning module is used to construct a quality event dataset.

[0007] Preferably, the data modeling module includes a multi-source acquisition unit, a vector generation unit and a dual construction unit: The multi-source acquisition unit is used to collect raw data of the entire machining process, and the raw data of the entire machining process includes tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, machining image data and process log data; The vector generation unit is used to convert the original data of the whole processing process into the original vector data of the whole processing process with unified dimension and spatial identification and time identification through a convolutional network; The dual construction unit is used to generate counterfactual sample data based on the original vector data of the entire processing process and the preset physical processing model. The generation process of the counterfactual sample data includes parameter perturbation simulation, environmental variable deduction, reverse reasoning of the target state and construction of counterfactual samples containing actual results and multiple virtual samples. The counterfactual sample data includes parameter perturbation simulation data, environmental variable deduction data, target reverse reasoning path and batch virtual experiment generation samples.

[0008] Preferably, the causal graph module includes a node definition unit, an edge weight learning unit and a structure update unit: The node definition unit is used to define a processing node, wherein the processing node includes an equipment node, a tool node, a material batch node, a processing event node and a state parameter node; The edge weight learning unit is used to calculate the causal strength, directionality and time lag parameters between each node in the processing process based on the original data of the entire processing process and the counterfactual sample data through statistical regression and graph learning algorithms, identify the causal relationship type according to the causal strength, directionality and time lag parameters, and the causal relationship type includes positive and negative relationships. The effect of the causal edge is quantitatively evaluated, and based on the original data of the entire process and the counterfactual sample data, a frequency statistical analysis and Bayesian inference method are used to construct a weight function model, which includes causal strength data, establish a time response mechanism, establish a lag time window model based on the tool wear data and the corresponding counterfactual sample data, generate a verification set in combination with the counterfactual sample data, verify the edge weight distribution of the weight function model, and establish a causal graph. The causal graph is the topological basis of the knowledge graph jointly constructed by the node definition unit and the edge weight learning unit.

[0009] Preferably, the structure updating unit is used to dynamically adjust the cause-effect graph according to tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data and process log data, environmental variable deduction data and state vector collected during the machining process, and the dynamic adjustment includes: When the multi-source acquisition unit identifies an unseen combination of processing events, state parameters, or environmental variables, it automatically adds a corresponding new node to the processing process node and establishes a causal edge link with the existing node; Based on the activity statistics of new nodes in the causal graph and the stability evaluation of weight values ​​in the edge weight function model, nodes and edges that have not been called within the specified time window or whose weights have decayed below the threshold are removed; When it is monitored that the tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data, and process log data collected during the processing process have significant deviations at the statistical distribution level, or when the model parameter reset conditions are triggered, the corresponding subgraph structure in the causal graph is reconstructed.

[0010] Preferably, the solution deduction module includes a target setting unit, a path evaluation unit and a pruning optimization unit: The target setting unit is used to construct an abstract objective function for the solution deduction module based on the historical tool wear data, image detection data, process log data, and processing energy consumption record data collected by the multi-source acquisition unit, combined with the production capacity constraint conditions input by the user, and define an abstract target node corresponding to the abstract objective function in the causal graph. The abstract objective function includes: Determine defect limits corresponding to surface quality requirements based on historical image inspection data; Based on tool wear data and process log data, set the minimum acceptable yield threshold per unit processing time; Combine energy consumption record data and material utilization rate records to set the resource cost evaluation function corresponding to each path; The abstract target node serves as the endpoint constraint for graph path traversal and pruning logic; The path evaluation unit is based on the topological structure constructed by the node definition unit and the edge weight learning unit in the causal graph. With the target node as the end point and the current state node as the starting point, it uses the graph traversal algorithm to enumerate all structurally reachable paths and performs serialized causal inference analysis on each path. During the path evaluation process, the time delay parameters and causal strength values ​​of the edge weight function model are used to accumulate the change sequences of the spindle vibration amplitude, tool wear growth rate, and coolant flow rate fluctuations contained in the path. Map the change sequence to the texture deviation distribution, reflective consistency scoring model, and processing time and energy consumption two-dimensional function corresponding to the image detection data, and calculate the degree of achievement of the corresponding objective function; The performance results of all paths are sorted to form a path priority candidate list.

[0011] Preferably, the pruning optimization unit receives the path candidate list output by the path evaluation unit, compares the path nodes one by one to see whether there are spindle vibration mutation nodes, coolant flow rate sudden drop nodes or status nodes associated with image detection abnormality labels marked as abnormal nodes in the causal graph, and deletes the corresponding paths; Stability criteria are applied based on the parameters recorded in the process log data, eliminating paths containing spindle speed or feed rate fluctuations exceeding a set threshold; For the remaining paths, a virtual sample library generated by the counterfactual sample construction unit is called to perform control verification on each process step step by step, eliminate the paths that deviate from the objective function beyond the limit range, and output the optimal path instruction set, which includes the target state vector.

[0012] Preferably, the adaptive control module includes a monitoring fusion unit, a strategy generation unit and a device control unit: The monitoring fusion unit is used to integrate tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, coolant flow rate data and processing image data from the multi-source acquisition unit, and compare them with the optimal path instruction set and change sequence; Based on the set sliding time window, the current state parameter vector is extracted in each cycle and the difference between it and the target state vector is calculated. The vector difference metric is output, and a state deviation scalar indicator based on the difference calculation is generated. The full data between the current state parameter vector and the target state vector is constructed into a structured deviation matrix. The strategy generation unit extracts the historical path and its corresponding response parameters that match the current state parameter vector based on the structured deviation matrix and counterfactual sample data; When the current spindle vibration amplitude data and coolant flow rate data are in an abnormal range, the preset simulation optimization engine is called to compare the historical feasible adjustment paths in the process log data and identify the path parameter combination whose texture consistency score change in the processing image data association score is less than the preset threshold; Based on the quality constraints and energy consumption parameter weights set in the objective function, the candidate paths are scored one by one, the strategic path closest to the target is selected, and standardized parameter instructions are generated. The standardized parameter instructions include the feed rate adjustment value, the spindle speed adjustment ratio, and the coolant flow rate correction coefficient; The equipment control unit receives standardized parameter instructions, maps the standardized parameter instructions into control commands compatible with the PLC control protocol, and sends them to the CNC system through the field equipment driver layer. It collects status feedback data sent back by the multi-source acquisition unit, and determines whether the current tool wear changes, spindle vibration spectrum response and surface image real-time detection results are consistent with the strategy expectations. If they are inconsistent, the local strategy is triggered to re-match, and all execution behaviors and offset feedback results are synchronously written into the data modeling module.

[0013] Preferably, the feedback learning module includes a quality perception unit and a graph correction unit: The quality perception unit is used to obtain the final processed image data and quality inspection report data from the final inspection link, and identify surface microcracks, texture deviation and brightness unevenness features in combination with manual annotation, and output them as current defect events; Through structured analysis, a defect label set is constructed to clarify the defect type, location, severity, and its potential association with spindle vibration amplitude data, tool wear data, and coolant flow rate data. After combining the equipment number, parameter setting, and processing time period information contained in the process log data, a quality event dataset is generated.

[0014] Preferably, the graph correction unit receives the quality event data set and maps it to the path structure in the cause-effect graph, and identifies the defective processing event node path associated with the current defective event according to the processing operation sequence recorded in the process log data; The weight values ​​of the corresponding edges in the defective processing event node path are adjusted to reduce or increase the influence intensity of the causal edge according to the interactive relationship between the defect frequency and severity and the known spindle vibration amplitude data and coolant flow rate data.

[0015] Preferably, the edge weight learning unit further combines the spindle vibration amplitude data, tool wear data, and the operation time series recorded in the process log data to establish a time response mechanism in the event causal path. Based on the sliding time window statistical method, the time correlation relationship between the increase in the spindle vibration amplitude, the change in the tool wear degree, and the change in the surface texture reflected in the image detection data is extracted, and a time lag model between the state parameter node and the defect event node is constructed. Based on the time-weighted factor model, the triggering sequence and response time of the edges in the causal graph are normalized and adjusted, and a multi-input single-output causal mapping structure is established. The spindle vibration amplitude data, coolant flow rate data and tool wear data are uniformly mapped to the state parameter nodes in the graph and connected to the corresponding defect event nodes. Based on the Bayesian causal inference algorithm, the significance and synergy coefficient of the interaction between each state parameter node are quantified, and the state parameter nodes whose significance and synergy coefficient are greater than the preset significance threshold and synergy coefficient threshold are embedded in the causal graph structure in the form of cross nodes, and the joint triggering conditions of the spindle vibration amplitude mutation and the coolant flow rate lower than the preset value are recorded.

[0016] Beneficial effects of the present invention: The present invention constructs a causal graph based on multi-source data such as tool wear data and spindle vibration amplitude data, combines counterfactual samples to achieve path deduction and parameter optimization of the target state, realizes adaptive control of the processing state through a structured deviation matrix, and uses a feedback learning mechanism to dynamically correct the causal relationship and weight function model, thereby realizing data-driven optimization and defect prevention of the processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the basic flow of a dial processing optimization system based on big data provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0019] Example, see Figure 1 , provides a dial processing optimization system based on big data, including data modeling module, cause-effect diagram module, solution deduction module, adaptive control module and feedback learning module: The data modeling module is used to collect tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, processing image data and process log data, and generate counterfactual sample data.

[0020] The causal graph module is used to construct a causal graph structure that includes processing process nodes and edge weight function models.

[0021] The solution deduction module is used to evaluate the performance of the machining path based on the abstract objective function.

[0022] The adaptive control module is used to compare state deviations and generate standardized parameter instructions.

[0023] The feedback learning module is used to construct the quality event dataset.

[0024] The data modeling module includes a multi-source acquisition unit, a vector generation unit, and a dual construction unit: The multi-source acquisition unit is used to collect the original data of the entire machining process, which includes tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, machining image data and process log data.

[0025] The multi-source acquisition unit realizes the full-process collection of tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, processing image data and process log data, providing high-integrity raw data input for subsequent modeling and deduction.

[0026] The vector generation unit is used to convert the original data of the entire processing process into original vector data of the entire processing process with unified dimension and spatial and temporal identification through a convolutional network.

[0027] The vector generation unit converts heterogeneous raw data into raw vector data of the entire processing process with unified dimensions, and introduces spatial and temporal identifiers to achieve comparability and temporal expression of data in the same computing space.

[0028] The dual construction unit is used to generate counterfactual sample data based on the original vector data of the entire processing process and the preset physical processing model. The generation process of counterfactual sample data includes parameter perturbation simulation, environmental variable deduction, reverse reasoning of the target state and construction of counterfactual samples containing actual results and multiple virtual samples. The counterfactual sample data includes parameter perturbation simulation data, environmental variable deduction data, target reverse reasoning path and batch virtual experiment generation samples.

[0029] The dual construction unit performs perturbation deduction and reverse logic modeling on the original vector data based on the physical processing model, generating counterfactual sample data covering multiple possible processing states, thereby improving the system's generalization and optimization capabilities under unknown parameter combinations.

[0030] The data modeling module realizes the structured expression and simulation expansion of multi-source raw data in the entire dial processing process, builds a data foundation with time series identification and causal logic support, and provides unified high-dimensional expression and virtual sample support for causal graph construction, path deduction and strategy optimization.

[0031] The causal graph module includes a node definition unit, an edge weight learning unit, and a structure update unit: The node definition unit is used to define the processing nodes, which include equipment nodes, tool nodes, material batch nodes, processing event nodes and status parameter nodes.

[0032] The node definition unit abstracts the equipment information, tool status, material source, operation events and continuous parameters in the processing process into standardized processing process nodes to construct the basic unit set of the graph.

[0033] The edge weight learning unit is used to calculate the causal strength, directionality and time lag parameters between each node in the processing process based on the original data of the entire processing process and the counterfactual sample data through statistical regression and graph learning algorithms, identify the causal relationship type according to the causal strength, directionality and time lag parameters, and the causal relationship type includes positive and negative relationships, and quantitatively evaluate the effect of the causal edge. Based on the original data of the entire process and the counterfactual sample data, the frequency statistical analysis and Bayesian inference method are used to construct a weight function model, which includes causal strength data, establish a time response mechanism, establish a lag time window model based on the tool wear data and the corresponding counterfactual sample data, generate a verification set in combination with the counterfactual sample data, verify the edge weight distribution of the weight function model, and establish a causal graph. The causal graph is the topological basis of the knowledge graph jointly constructed by the node definition unit and the edge weight learning unit.

[0034] The edge weight learning unit also combines the spindle vibration amplitude data, tool wear data and the operation time series recorded in the process log data to establish a time response mechanism in the event causal path. Based on the sliding time window statistical method, it extracts the time correlation relationship between the increase in spindle vibration amplitude, the change in tool wear degree and the surface texture change reflected in the image detection data, and constructs a time lag model between the state parameter node and the defect event node.

[0035] Based on the time-weighted factor model, the triggering sequence and response time of the edges in the causal graph are normalized and adjusted, and a multi-input single-output causal mapping structure is established. The spindle vibration amplitude data, coolant flow rate data and tool wear data are uniformly mapped to the state parameter nodes in the graph and connected to the corresponding defect event nodes. Based on the Bayesian causal inference algorithm, the significance and synergy coefficient of the interaction between each state parameter node are quantified, and the state parameter nodes whose significance and synergy coefficient are greater than the preset significance threshold and synergy coefficient threshold are embedded in the causal graph structure in the form of cross nodes, and the joint triggering conditions of the spindle vibration amplitude mutation and the coolant flow rate lower than the preset value are recorded.

[0036] Based on the original data of the entire machining process and counterfactual sample data, the edge weight learning unit establishes a causal strength, directionality and time lag model between the nodes of the machining process, generates an edge weight function model, integrates sliding window statistics and Bayesian inference, quantifies the causal effect and synergistic relationship between machining factors, and combines the spindle vibration amplitude data, coolant flow rate data and tool wear data to establish a lagged causal chain between the machining status and defect events, forming a unified multi-input causal path mapping and joint trigger condition description.

[0037] The structure update unit is used to dynamically adjust the cause-effect graph based on the tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data, process log data, environmental variable deduction data, and state vector collected during the machining process. The dynamic adjustment includes: When the multi-source acquisition unit identifies an unseen combination of processing events, state parameters, or environmental variables, it automatically adds a corresponding new node to the processing process node and establishes a causal edge link with the existing nodes.

[0038] Based on the activity statistics of new nodes in the causal graph and the stability evaluation of weight values ​​in the edge weight function model, nodes and edges that have not been called within the specified time window or whose weights have decayed below the threshold are removed.

[0039] When it is monitored that the tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data, and process log data collected during the processing process have significant deviations at the statistical distribution level, or when the model parameter reset conditions are triggered, the corresponding subgraph structure in the causal graph is reconstructed.

[0040] The structure update unit dynamically adds, deletes or reconstructs nodes and causal edges in the graph according to changes in multi-source collected data and deduced data, keeping the causal graph structure consistent with the actual processing status.

[0041] The causal graph module constructs and dynamically maintains a causal graph structure covering the entire dial processing process, realizing automatic identification of causal paths between processing factors, edge weight learning, and self-updating of the graph structure, providing a structured reasoning basis for parameter deduction and defect prediction.

[0042] The scenario deduction module includes a goal setting unit, a path evaluation unit, and a pruning optimization unit: The target setting unit is used to construct an abstract objective function for the solution deduction module based on the historical tool wear data, image detection data, process log data, and processing energy consumption record data collected by the multi-source acquisition unit, combined with the production capacity constraints input by the user, and define the abstract target node corresponding to the abstract objective function in the cause-effect diagram. The abstract objective function includes: Determine the defect limit conditions corresponding to the surface quality requirements based on historical image inspection data.

[0043] Based on tool wear data and process log data, set the minimum acceptable yield threshold per unit processing time.

[0044] Combined with energy consumption record data and material utilization rate records, the resource cost evaluation function corresponding to each path is set.

[0045] The abstract target node serves as the endpoint constraint for graph path traversal and pruning logic.

[0046] The target setting unit uses historical tool wear data, image detection data, process log data and processing energy consumption record data, combined with the production capacity constraints set by the user, to construct an abstract objective function and define target nodes in the graph, providing a clear basis for quality, efficiency and resource constraints for path deduction.

[0047] The path evaluation unit is based on the topological structure constructed by the node definition unit and the edge weight learning unit in the causal graph. It takes the target node as the end point and starts from the current state node. It uses the graph traversal algorithm to enumerate all structurally reachable paths and performs serialized causal inference analysis on each path.

[0048] During the path evaluation process, the change sequence of spindle vibration amplitude, tool wear growth rate and coolant flow rate fluctuation contained in the accumulated path is calculated according to the time delay parameter and causal strength value of the edge weight function model.

[0049] The change sequence is mapped to the texture deviation distribution, reflective consistency scoring model and two-dimensional function of processing time and energy consumption corresponding to the image detection data, and the degree of achievement of the corresponding objective function is calculated.

[0050] The performance results of all paths are sorted to form a path priority candidate list.

[0051] The path evaluation unit relies on the causal graph topology structure, starting from the current state node, enumerates all reachable paths to the abstract target node, calculates the change sequence of spindle vibration amplitude, tool wear growth and coolant flow rate fluctuation in the path through the edge weight function model, and maps it to the image quality and energy consumption model, quantifies the target achievement of each path and generates a priority candidate list.

[0052] The pruning optimization unit receives the path candidate list output by the path evaluation unit, compares the path nodes one by one to see whether there are spindle vibration mutation nodes, coolant flow rate drop nodes or status nodes associated with image detection abnormal labels marked as abnormal nodes in the causal graph, and deletes the corresponding paths.

[0053] A stability criterion is implemented based on the parameters recorded in the process log data, and paths containing spindle speed or feed rate fluctuations exceeding a set threshold are eliminated.

[0054] For the remaining paths, a virtual sample library generated by the counterfactual sample construction unit is called to perform control verification on each process step step by step, eliminate the paths that deviate from the objective function beyond the limit range, and output the optimal path instruction set, which includes the target state vector.

[0055] The pruning optimization unit compares the candidate paths to see whether there are abnormal state nodes or unstable execution parameters in the process logs. It verifies each process step by step based on the counterfactual sample library, eliminates paths that deviate from the objective function, and finally outputs the optimal path instruction set containing the target state vector for subsequent control execution and scheduling.

[0056] The solution deduction module is based on historical processing data and user constraints, combined with the causal graph structure, to realize the path deduction, effect evaluation and pruning screening of the processing objective function, and finally generate the optimal path instruction set that meets the quality, efficiency and resource constraints.

[0057] The adaptive control module includes a monitoring fusion unit, a strategy generation unit, and a device control unit: The monitoring fusion unit is used to integrate tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, coolant flow rate data and processing image data from multi-source acquisition units, and compare them with the optimal path instruction set and change sequence.

[0058] Based on the set sliding time window, the current state parameter vector is extracted in each cycle, and the difference between it and the target state vector is calculated. The vector difference metric is output, and a state deviation scalar indicator based on the difference calculation is generated. The full data between the current state parameter vector and the target state vector is constructed as a structured deviation matrix.

[0059] The monitoring fusion unit integrates tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, coolant flow rate data and processing image data, compares them with the optimal path instruction set, calculates the difference measure between the current state parameter vector and the target state vector, and constructs a structured deviation matrix for subsequent strategy derivation.

[0060] The strategy generation unit extracts the historical path and its corresponding response parameters that match the current state parameter vector based on the structured deviation matrix and counterfactual sample data.

[0061] When the current spindle vibration amplitude data and coolant flow rate data are in an abnormal range, the preset simulation optimization engine is called to compare the historical feasible adjustment paths in the process log data, and identify the path parameter combination in which the texture consistency score change in the processing image data association score is less than the preset threshold.

[0062] According to the quality constraints and energy consumption parameter weights set in the objective function, the candidate paths are scored one by one in the set, the strategic path closest to the target is selected, and standardized parameter instructions are formed. The standardized parameter instructions include the feed rate adjustment value, the spindle speed adjustment ratio and the coolant flow rate correction coefficient.

[0063] The strategy generation unit matches historical paths and adjustment parameters based on the structured deviation matrix and counterfactual sample data, combines the feasible paths in the process log data, scores and selects the optimal strategy path that meets the quality constraints and energy consumption weight requirements, and outputs standardized parameter instructions for real-time adjustment.

[0064] The equipment control unit receives standardized parameter instructions, maps the standardized parameter instructions into control commands compatible with the PLC control protocol, and sends them to the CNC system through the on-site equipment driver layer. It collects status feedback data sent back by the multi-source acquisition unit to determine whether the current tool wear changes, spindle vibration spectrum response and surface image real-time detection results are consistent with the strategy expectations. If they are inconsistent, the local strategy is triggered to re-match, and all execution behaviors and offset feedback results are synchronously written into the data modeling module.

[0065] The equipment control unit converts standardized parameter instructions into PLC-compatible control commands and sends them to the CNC system, monitoring changes in tool wear, spindle vibration spectrum and processing image results in real time. If there is any deviation from the strategy expectation, local strategy rematching is triggered, and feedback is written into the data modeling module to achieve continuous learning.

[0066] The adaptive control module realizes real-time status monitoring, deviation analysis, strategy generation and control instruction execution during the machining process, and builds a closed-loop control mechanism based on the target state vector to ensure that the system maintains machining process stability and target consistency under dynamically changing conditions.

[0067] The feedback learning module includes a quality perception unit and a graph correction unit: The quality perception unit is used to obtain the final processed image data and quality inspection report data from the final inspection link, and combine it with manual annotation to identify surface microcracks, texture deviation and brightness unevenness features, and output them as current defect events.

[0068] Through structured analysis, a defect label set is constructed to clarify the defect type, location, severity, and its potential association with spindle vibration amplitude data, tool wear data, and coolant flow rate data. After combining the equipment number, parameter setting, and processing time period information contained in the process log data, a quality event dataset is generated.

[0069] The quality perception unit collects final processing image data and quality inspection report data, combines manual annotation to identify defect types and characteristics, establishes a defect label set through structured analysis, and generates a quality event data set based on spindle vibration amplitude data, tool wear data, coolant flow rate data and process log data.

[0070] The graph correction unit receives the quality event data set and maps it to the path structure in the cause-effect graph. According to the processing operation sequence recorded in the process log data, the defective processing event node path associated with the current defective event is identified.

[0071] The weight values ​​of the corresponding edges in the defective processing event node path are adjusted to reduce or increase the influence intensity of the causal edge according to the interactive relationship between the defect frequency and severity and the known spindle vibration amplitude data and coolant flow rate data.

[0072] The graph correction unit maps the quality event data set to the corresponding path in the causal graph, identifies the processing nodes related to the defect events, and adjusts the weight of the causal edge between the spindle vibration amplitude data and the coolant flow rate data based on the defect frequency and severity, thereby realizing dynamic optimization and updating of the graph.

[0073] The feedback learning module realizes structured recognition and retrospective modeling of defect information in the final inspection stage, dynamically corrects the edge weights of the processing path in the causal graph based on the quality event data set, constructs a feedback closed-loop mechanism between processing defects and parameter anomalies, and improves the system's self-learning and defect prevention capabilities.

[0074] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A dial processing optimization system based on big data, characterized in that: It includes data modeling module, cause-effect diagram module, scenario deduction module, adaptive control module and feedback learning module: The data modeling module is used to collect tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, processing image data and process log data, and generate counterfactual sample data; The causal graph module is used to construct a causal graph structure including a processing process node and an edge weight function model; The solution deduction module is used to evaluate the processing path performance based on the abstract objective function; The adaptive control module is used to compare state deviations and generate standardized parameter instructions; The feedback learning module is used to construct a quality event dataset.

2. The dial processing optimization system based on big data according to claim 1, characterized in that: The data modeling module includes a multi-source acquisition unit, a vector generation unit and a dual construction unit: The multi-source acquisition unit is used to collect raw data of the entire machining process, and the raw data of the entire machining process includes tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, machining image data and process log data; The vector generation unit is used to convert the original data of the whole processing process into the original vector data of the whole processing process with unified dimension and spatial identification and time identification through a convolutional network; The dual construction unit is used to generate counterfactual sample data based on the original vector data of the entire processing process and the preset physical processing model. The generation process of the counterfactual sample data includes parameter perturbation simulation, environmental variable deduction, reverse reasoning of the target state and construction of counterfactual samples containing actual results and multiple virtual samples. The counterfactual sample data includes parameter perturbation simulation data, environmental variable deduction data, target reverse reasoning path and batch virtual experiment generation samples.

3. The dial processing optimization system based on big data according to claim 2, characterized in that: The causal graph module includes a node definition unit, an edge weight learning unit, and a structure update unit: The node definition unit is used to define a processing node, wherein the processing node includes an equipment node, a tool node, a material batch node, a processing event node and a state parameter node; The edge weight learning unit is used to calculate the causal strength, directionality and time lag parameters between each node in the processing process based on the original data of the entire processing process and the counterfactual sample data through statistical regression and graph learning algorithms, identify the causal relationship type according to the causal strength, directionality and time lag parameters, and the causal relationship type includes positive and negative relationships. The effect of the causal edge is quantitatively evaluated, and based on the original data of the entire process and the counterfactual sample data, a frequency statistical analysis and Bayesian inference method are used to construct a weight function model, which includes causal strength data, establish a time response mechanism, establish a lag time window model based on the tool wear data and the corresponding counterfactual sample data, generate a verification set in combination with the counterfactual sample data, verify the edge weight distribution of the weight function model, and establish a causal graph. The causal graph is the topological basis of the knowledge graph jointly constructed by the node definition unit and the edge weight learning unit.

4. The dial processing optimization system based on big data according to claim 3, characterized in that: The structure updating unit is used to dynamically adjust the cause-effect graph based on tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data, process log data, environmental variable deduction data, and state vector collected during the machining process. The dynamic adjustment includes: When the multi-source acquisition unit identifies an unseen combination of processing events, state parameters, or environmental variables, it automatically adds a corresponding new node to the processing process node and establishes a causal edge link with the existing node; Based on the activity statistics of new nodes in the causal graph and the stability evaluation of weight values ​​in the edge weight function model, nodes and edges that have not been called within the specified time window or whose weights have decayed below the threshold are removed; When it is monitored that the tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, image detection data, and process log data collected during the processing process have significant deviations at the statistical distribution level, or when the model parameter reset conditions are triggered, the corresponding subgraph structure in the causal graph is reconstructed.

5. The dial processing optimization system based on big data according to claim 4, characterized in that: The program deduction module includes a target setting unit, a path evaluation unit and a pruning optimization unit: The target setting unit is used to construct an abstract objective function for the solution deduction module based on the historical tool wear data, image detection data, process log data, and processing energy consumption record data collected by the multi-source acquisition unit, combined with the production capacity constraint conditions input by the user, and define an abstract target node corresponding to the abstract objective function in the causal graph. The abstract objective function includes: Determine defect limits corresponding to surface quality requirements based on historical image inspection data; Based on tool wear data and process log data, set the minimum acceptable yield threshold per unit processing time; Combine energy consumption record data and material utilization rate records to set the resource cost evaluation function corresponding to each path; The abstract target node serves as the endpoint constraint for graph path traversal and pruning logic; The path evaluation unit is based on the topological structure constructed by the node definition unit and the edge weight learning unit in the causal graph. With the target node as the end point and the current state node as the starting point, it uses the graph traversal algorithm to enumerate all structurally reachable paths and performs serialized causal inference analysis on each path. During the path evaluation process, the time delay parameters and causal strength values ​​of the edge weight function model are used to accumulate the change sequences of the spindle vibration amplitude, tool wear growth rate, and coolant flow rate fluctuations contained in the path. Map the change sequence to the texture deviation distribution, reflective consistency scoring model, and processing time and energy consumption two-dimensional function corresponding to the image detection data, and calculate the degree of achievement of the corresponding objective function; The performance results of all paths are sorted to form a path priority candidate list.

6. The dial processing optimization system based on big data according to claim 5, characterized in that: The pruning optimization unit receives the path candidate list output by the path evaluation unit, compares the path nodes one by one to see whether there are spindle vibration mutation nodes, coolant flow rate sudden drop nodes or status nodes associated with image detection abnormality labels marked as abnormal nodes in the causal graph, and deletes the corresponding paths; Stability criteria are applied based on the parameters recorded in the process log data, eliminating paths containing spindle speed or feed rate fluctuations exceeding a set threshold; For the remaining paths, a virtual sample library generated by the counterfactual sample construction unit is called to perform control verification on each process step step by step, eliminate the paths that deviate from the objective function beyond the limit range, and output the optimal path instruction set, which includes the target state vector.

7. The dial processing optimization system based on big data according to claim 6, characterized in that: The adaptive control module includes a monitoring fusion unit, a strategy generation unit and a device control unit: The monitoring fusion unit is used to integrate tool wear data, spindle vibration amplitude data, ambient temperature and humidity data, coolant flow rate data and processing image data from the multi-source acquisition unit, and compare them with the optimal path instruction set and change sequence; Based on the set sliding time window, the current state parameter vector is extracted in each cycle and the difference between it and the target state vector is calculated. The vector difference metric is output, and a state deviation scalar indicator based on the difference calculation is generated. The full data between the current state parameter vector and the target state vector is constructed into a structured deviation matrix. The strategy generation unit extracts the historical path and its corresponding response parameters that match the current state parameter vector based on the structured deviation matrix and counterfactual sample data; When the current spindle vibration amplitude data and coolant flow rate data are in an abnormal range, the preset simulation optimization engine is called to compare the historical feasible adjustment paths in the process log data and identify the path parameter combination whose texture consistency score change in the processing image data association score is less than the preset threshold; Based on the quality constraints and energy consumption parameter weights set in the objective function, the candidate paths are scored one by one, the strategic path closest to the target is selected, and standardized parameter instructions are generated. The standardized parameter instructions include the feed rate adjustment value, the spindle speed adjustment ratio, and the coolant flow rate correction coefficient; The equipment control unit receives standardized parameter instructions, maps the standardized parameter instructions into control commands compatible with the PLC control protocol, and sends them to the CNC system through the field equipment driver layer. It collects status feedback data sent back by the multi-source acquisition unit, and determines whether the current tool wear changes, spindle vibration spectrum response and surface image real-time detection results are consistent with the strategy expectations. If they are inconsistent, the local strategy is triggered to re-match, and all execution behaviors and offset feedback results are synchronously written into the data modeling module.

8. The dial processing optimization system based on big data according to claim 7, characterized in that: The feedback learning module includes a quality perception unit and a graph correction unit: The quality perception unit is used to obtain the final processed image data and quality inspection report data from the final inspection link, and identify surface microcracks, texture deviation and brightness unevenness features in combination with manual annotation, and output them as current defect events; Through structured analysis, a defect label set is constructed to clarify the defect type, location, severity, and its potential association with spindle vibration amplitude data, tool wear data, and coolant flow rate data. After combining the equipment number, parameter setting, and processing time period information contained in the process log data, a quality event dataset is generated.

9. The dial processing optimization system based on big data according to claim 8, characterized in that: The graph correction unit receives the quality event data set and maps it to the path structure in the cause-effect graph. According to the processing operation sequence recorded in the process log data, the defective processing event node path associated with the current defective event is identified. The weight values ​​of the corresponding edges in the defective processing event node path are adjusted to reduce or increase the influence intensity of the causal edge according to the interactive relationship between the defect frequency and severity and the known spindle vibration amplitude data and coolant flow rate data.

10. The dial processing optimization system based on big data according to claim 9, characterized in that: The edge weight learning unit also combines the spindle vibration amplitude data, tool wear data and the operation time series recorded in the process log data to establish a time response mechanism in the event causal path. Based on the sliding time window statistical method, it extracts the time correlation between the increase in spindle vibration amplitude, the change in tool wear degree and the change in surface texture reflected in the image detection data, and constructs a time lag model between the state parameter node and the defect event node. Based on the time-weighted factor model, the triggering sequence and response time of the edges in the causal graph are normalized and adjusted, and a multi-input single-output causal mapping structure is established. The spindle vibration amplitude data, coolant flow rate data and tool wear data are uniformly mapped to the state parameter nodes in the graph and connected to the corresponding defect event nodes. Based on the Bayesian causal inference algorithm, the significance and synergy coefficient of the interaction between each state parameter node are quantified, and the state parameter nodes whose significance and synergy coefficient are greater than the preset significance threshold and synergy coefficient threshold are embedded in the causal graph structure in the form of cross nodes, and the joint triggering conditions of the spindle vibration amplitude mutation and the coolant flow rate lower than the preset value are recorded.

Images