CN-121722951-B - Numerical control machine tool part machining data storage method and system

CN121722951BCN 121722951 BCN121722951 BCN 121722951BCN-121722951-B

Abstract

The invention discloses a method and a system for storing machining data of a numerical control machine tool part, wherein the method for storing the machining data of the numerical control machine tool part collects and synchronizes multisource data of a machining process in real time, process segment segmentation based on process entropy is carried out on a data stream, multi-mode characteristics are extracted, data value is estimated through a fusion rule and a machine learning classifier, a characteristic data packet is generated according to the multi-mode characteristics, a data gene code is bound for each part, a machining process knowledge graph is built, full-chain data association and accurate tracing are achieved, a sample is generated based on the characteristic data packet, a process optimization model is trained in a cloud by utilizing a graph neural network and Bayesian optimization, and optimization parameters are fed back to a machine tool for execution to form a closed loop.

Inventors

WANG XIAOHUI
WANG XIAOLIN
Yuan Maicui

Assignees

湖南嵘触智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260227

Claims (8)

1. The method for storing the machining data of the parts of the numerical control machine tool is characterized by comprising the following steps of: S1, acquiring and synchronizing multi-source data streams in the processing process of a numerical control machine tool in real time, marking a unified time mark for all data, and storing the data into a buffer area, wherein the multi-source data streams at least comprise internal state data of the numerical control system, PLC signals and external sensor data; S2, carrying out real-time analysis on the data flow of the buffer area, dividing the data flow into a plurality of process sections based on process state change, extracting multi-mode characteristics of each process section, evaluating the value grade of the data, and carrying out compression or summarization on the original data of the process section by adopting a corresponding storage strategy according to the value grade to generate a characteristic data packet containing metadata and characteristic vectors; S3, binding a unique data gene code for each processing part, carrying out association binding on all the characteristic data packets generated in the step S2 and the data gene codes of the corresponding parts, and constructing a traceable data gene chain together with upstream material data and downstream quality detection data of the parts; S4, closed loop learning and optimization feedback, namely generating a process-feature-result triple sample based on the feature data packet and an associated processing result, uploading a high-value sample to a central knowledge base, training a process optimization model by using an aggregated multi-source sample, transmitting the trained optimization model to an edge side, generating optimized process parameter recommendation for a new processing task, and feeding back new data after execution to the central knowledge base to realize iterative updating of the model; the step S2 of dividing the process into a plurality of process segments based on the process state change specifically includes: S21, carrying out initial segmentation based on a preset strong segmentation trigger event, wherein the strong segmentation trigger event comprises G code line number change, cutter replacement completion signal or feed speed return to zero; S22, calculating a process entropy value of a key variable by adopting a sliding window in a continuous processing interval without a strong segmentation trigger event, judging that the processing state is remarkably changed when the variation of the process entropy value exceeds a preset threshold value, and executing soft segmentation; wherein the process entropy value The calculation formula of (2) is as follows: Wherein, the As the total number of joint states for the key variables, Expressed in a time window Internally observed first The state of the union is seeded, For the empirical probability that the state occurs within the window, A base number that is logarithmic; In the step S2, "adopting a corresponding storage policy" specifically includes: if the value grade is 'high', the original high-frequency data and the complete feature set of the process section are saved by adopting a lossless or near-lossless compression algorithm, and the original high-frequency data and the complete feature set are stored in a high-speed storage medium; if the value grade is 'medium', storing the downsampled data and the feature set to a conventional storage medium; if the value grade is 'low', only the statistical characteristics, semantic characteristics and processing result data of the process section are saved, and the original data are discarded.
2. The method for storing machining data of a part of a numerically-controlled machine tool according to claim 1, wherein extracting the multi-modal feature in the step S2 includes extracting a frequency domain feature, specifically: time series signal for vibration or acoustic emission sensor Performing fast Fourier transform to calculate power spectral density : Wherein, the For signal length, from the power spectral density The first Q dominant frequency components and the corresponding amplitude values are extracted as the frequency domain feature vectors of the process section.
3. The method for storing machining data of a part of a numerically controlled machine tool according to claim 1, wherein the step S2 of evaluating the data value level is specifically: inputting the extracted multi-modal features and semantic features analyzed from G codes into a predefined value evaluation classifier, wherein the value evaluation classifier combines a rule base and a lightweight machine learning model, and outputs the value grade of the process section, and the value grade at least comprises high, medium and low levels; the rule base comprises judging rules based on the processing type, the participation shaft and the cutter information; It is realized in particular by a value evaluation classifier whose decision function Expressed as: Wherein, the For the input multimodal feature and semantic feature vector, As a class of value class, For rule matching score based on a preset rule base containing "IF process type= = 'finish milling' AND participating axis contains judgment rule in the form of 'C axis' THEN value level= 'high', The class probabilities output for a lightweight machine learning model, To weigh the coefficients, the weights used to adjust the rule and model outputs.
4. The method for storing machining data of a part of a numerically-controlled machine tool according to claim 1, wherein the step S3 of constructing a traceable data gene chain specifically comprises: S31, constructing a processing process knowledge graph comprising parts, working procedures, process sections, machine tools, cutters and quality detection result entities by taking the data gene codes as main keys; S32, associating and storing each characteristic data packet as the attribute of the corresponding process section entity, and connecting the characteristic data packet with an upstream process, a part entity and a downstream quality detection result entity through a map relation; Wherein, the processing course knowledge graph Expressed as: Wherein, the Is a node set, comprises parts, working procedures, process sections, machine tools, cutters and quality detection result entities, For a collection of edges, the relationship between entities is represented.
5. The method for storing machining data of a part of a numerically-controlled machine tool according to claim 4, further comprising the step of intelligent tracing: When quality abnormality warning aiming at a specific data gene code is received, in the processing process knowledge graph, graph traversal is carried out by taking a corresponding part entity as a starting point, all related process segment entities and corresponding characteristic data packets are positioned, suspicious stages are screened out according to process segment attributes, and the stored original data are called for deep analysis.
6. The method for storing machining data of a numerically-controlled machine tool part according to claim 1, wherein the training process optimization model using the aggregated multisource samples in step S4 is specifically: training by using a graph neural network model with the processing knowledge graph as an organization structure of training data, wherein the input of the graph neural network model is node characteristics and side relations in the graph, the output is prediction for process parameter adjustment, and the training aim is to minimize the loss between predicted characteristics/results and actual characteristics/results; the graph neural network model updates the node representation through a messaging mechanism: for each node in the knowledge-graph In the first place Characterization of layers Calculated by the following formula: Wherein, the Representing nodes Is defined by a set of neighboring nodes of the network, As a function of the aggregation function, In order for the weight matrix to be trainable, Training loss function of model as nonlinear activation function For the combination of mean square error and cross entropy loss: Wherein, the And Respectively predicting the process characteristics and the processing results of the model, And To be a true value of the value, And Is a loss weight coefficient.
7. The method for storing machining data of a part of a numerically controlled machine tool according to claim 6, wherein the step S4 of generating the optimized process parameter recommendation for the new machining task is specifically: matching or constructing a similar sub-graph structure in the central knowledge base according to the part information, the cutter and the machine tool context of the new task, reasoning the sub-graph by utilizing the trained graph neural network model, solving a process parameter combination which enables a predicted machining result to be optimal through an optimization algorithm, and sending the process parameter combination to an edge side as a recommended parameter set; Building query subgraphs from new task contexts Taking the trained graph neural network model as an evaluation function, and searching for an optimal technological parameter combination by adopting a Bayesian optimization algorithm : Wherein, the The input of the graph neural network model is the injection process parameter Post query subgraph Output is a predicted processing result and a stability index, To map the model output as a function of the integrated utility value, Is the expected utility under model prediction uncertainty.
8. A numerically controlled machine tool part processing data storage system for implementing the method of any one of claims 1-7, the system comprising: edge intelligence storage node is disposed in digit control machine tool side, includes: the data acquisition and synchronization module is used for executing the step S1; The feature extraction and compression engine is used for executing the step S2; The local storage and management module is used for storing the characteristic data packet and the associated index; a first communication interface; The central knowledge base and the optimization platform are deployed on a cloud or a server and comprise: the second communication interface is used for communicating with the edge intelligent storage node; The knowledge graph management module is used for executing the functions of constructing and maintaining the data gene chain and the knowledge graph in the processing process in the step S3; The machine learning training platform is used for executing the model training and optimizing functions in the step S4; A global repository; and the edge intelligent storage node, the central knowledge base and the optimization platform are subjected to data interaction through a network to cooperatively finish the storage, management and optimization of the numerical control machine tool part processing data.

Description

Numerical control machine tool part machining data storage method and system Technical Field The invention relates to the technical field of intelligent manufacturing, in particular to a method and a system for storing machining data of a numerical control machine tool part. Background In the prior art, in modern intelligent manufacturing, a numerical control machine tool is core production equipment. The processing process can generate mass data including instructions and state data in the numerical control system, such as G codes, servo tracking errors and I/O signals of a programmable logic controller, such as cutter numbers, cooling liquid states and various external sensor data, such as vibration, acoustic emission, temperature and power, and the data has extremely high value for process optimization, quality tracing, predictive maintenance and production management. At present, the following problems and disadvantages mainly exist in the aspects of numerical control processing data storage and management in the industry: 1. The storage mode is extensive, and has high cost, namely two extreme modes of full-volume storage or result storage are generally adopted, wherein the full-volume storage or the result storage is used for recording all original high-frequency data without distinction, so that the storage space and the network bandwidth are rapidly exhausted, the cost is greatly increased, and the final processing result such as a size report is only saved, and precious processing process information is discarded, so that process analysis and problem tracing are not from mention. 2. The data island is serious, the tracing is difficult, the processing process data, the material data and the quality detection data are generally scattered in different information systems such as MES, ERP, QMS or databases, an effective association mechanism is lacked, when the quality problem occurs to the part, the defect is difficult to be rapidly and accurately positioned to a specific processing procedure, a machine tool state and even technological parameters, and the investigation efficiency is low. 3. The data value is not deeply mined, the stored data is mostly a 'deep sleep' archive, an effective analysis and utilization means is lacked, the process rule cannot be automatically learned from the historical data, the intelligent and self-adaptive optimization of the process parameters cannot be realized, and the further improvement of the processing quality, efficiency and consistency is limited. The invention patent of China with the application number 202211097029.7 discloses a method for acquiring, storing and applying data of a numerical control machine tool, which comprises the steps of installing a vibration sensor, a noise sensor, an alternating current transmitter and an alternating voltage transmitter on the machine tool, acquiring data acquired by various sensors in real time through a serial port communication technology, and displaying real-time data and information on a man-machine interface of an industrial personal computer and a numerical control machine tool controller in real time through software data processing, so that the data is locally stored in a detection and monitoring system database of a background of the industrial personal computer, or an intelligent gateway of the Internet of things is uploaded to a network cloud platform server through Ethernet, so that data cloud storage and cloud calculation are realized, and cloud storage data of the cloud storage data are displayed through software of a PC (personal computer) or a mobile terminal. However, although the method for acquiring, storing and applying the numerical control machine tool data can automatically acquire key process index data in real time, the storage mode is extensive, the cost is high, the data island is serious, the tracing is difficult, and the data value is not deeply mined, so that a novel data storage method and system capable of intelligently judging the data value, realizing full life cycle data association and driving the process to continuously optimize are urgently needed. Disclosure of Invention The invention aims to provide a method and a system for storing machining data of a part of a numerical control machine tool. In order to achieve the above purpose, the technical scheme provided by the invention is as follows: A method for storing machining data of a part of a numerical control machine tool comprises the following steps: S1, collecting and synchronizing multi-source data streams in the machining process of the numerical control machine tool in real time, marking a unified time mark for all data, and storing the data into a buffer area, wherein the multi-source data streams at least comprise internal state data of the numerical control system, PLC signals and external sensor data. S2, carrying out real-time analysis on the data flow of the buffer area, dividing the data flow into a plurality of proce