CN-121785238-B - Computer numerical control system data desensitizing method based on machine learning

CN121785238BCN 121785238 BCN121785238 BCN 121785238BCN-121785238-B

Abstract

The invention relates to the technical field of data processing, in particular to a computer numerical control system data desensitizing method based on machine learning. The method comprises the steps of taking a real-time cutting load sequence of a numerical control system based on a main shaft position, eliminating the influence of rotation speed fluctuation, carrying out linear fitting by taking a locally stored standard cutting load sequence as an independent variable and taking the real-time cutting load sequence as an independent variable, decoupling to obtain a fitting slope representing a hardness coefficient of a material and a residual sequence representing a wear characteristic component, carrying out normalization processing on the characteristic component by using the hardness coefficient of the material and calculating energy density to generate a wear energy accumulated value, uploading the wear energy accumulated value, the hardness coefficient of the material and a fault occurrence position to a cloud storage, and carrying out fault early warning. The embodiment of the invention can realize data desensitization on the premise of not uploading the geometric original data of the workpiece, and effectively solves the technical problems of false alarm of cutter wear monitoring and geometric privacy leakage of the workpiece caused by material hardness fluctuation.

Inventors

DOU XIAOMU
WANG GANG

Assignees

南京高商机电科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260306

Claims (10)

1. A machine learning-based computer numerical control system data desensitization method, the method comprising: Acquiring a real-time cutting load sequence of the numerical control system based on the position of the main shaft; Taking elements in a standard cutting load sequence stored locally as independent variables, and taking elements in the real-time cutting load sequence as dependent variables, performing linear fitting to obtain a slope and a residual sequence of a fitting result, wherein the slope is taken as a material hardness coefficient, and the residual sequence is taken as a wear characteristic component; And carrying out cloud storage by taking the wear energy accumulation value, the material hardness coefficient and the fault occurrence position as storage data, positioning the fault occurrence position based on the energy density, and feeding back an early warning signal according to the wear energy accumulation value and the material hardness coefficient.
2. The machine learning based computer numerical control system data desensitization method according to claim 1, wherein said standard cutting load sequence acquisition method comprises: and under a template construction mode of the numerical control system, counting cutting load sequences generated by a preset number of complete processing cycles as a basic sequence, and averaging the basic sequence after alignment to obtain the standard cutting load sequence.
3. The machine learning based computer numerical control system data desensitization method according to claim 1, wherein said linear fitting process is preceded by an alignment process of standard cutting load sequences with real-time cutting load sequences, said alignment process comprising: searching reference elements in a standard cutting load sequence and a real-time cutting load sequence respectively, wherein the reference elements are elements with loads larger than a preset contact threshold value for the first time and larger than the previous element; Constructing an offset interval by taking the index offset as a center, and carrying out resampling interpolation reconstruction on the real-time cutting load sequence based on each candidate offset in the offset interval to obtain a reconstruction sequence; and obtaining a correlation coefficient between the reconstruction sequence and the standard cutting load sequence, and selecting the reconstruction sequence with the maximum correlation coefficient as an alignment result of the real-time cutting load sequence and the standard cutting load sequence.
4. A machine learning based computer numerical control system data desensitization method according to claim 3, wherein said correlation coefficient is pearson correlation coefficient.
5. The machine learning based computer numerical control system data desensitization method according to claim 1, wherein in the fitting result, further comprising: If the material hardness of the fitting result is in the preset dangerous interval, a severe fault alarm signal is fed back, the residual sequence acquisition process is interrupted, and the material hardness is stored in the cloud.
6. The machine learning based computer numerical control system data desensitization method according to claim 1, wherein said energy density acquisition method comprises: and for each abrasion characteristic in the abrasion characteristic components, mapping the abrasion characteristic after eliminating the basic noise threshold by using a material hardness coefficient to obtain the energy density, wherein the energy density is larger than 0.
7. The machine learning based computer numerical control system data desensitizing method according to claim 1, wherein said fault occurrence location locating method comprises: and selecting the spindle position corresponding to the maximum energy density as the fault occurrence position.
8. The machine learning based computer numerical control system data desensitizing method according to claim 1, wherein said feeding back early warning signals according to said wear energy accumulation value and material hardness coefficient comprises: Counting a machining cycle corresponding to the real-time cutting load sequence, and counting a wear energy accumulation value sequence formed by continuous repeated machining cycles before the machining cycle and a material hardness coefficient sequence; If the real-time material hardness coefficient of the real-time cutting load sequence exceeds a preset safety range, judging whether the real-time material hardness coefficient generates step change according to the material hardness coefficient sequence, and if so, feeding back a material batch hardness abnormal signal; If the real-time material hardness coefficient does not generate step change, performing linear fitting on the abrasion energy accumulation value sequence to obtain an abrasion increase slope, and according to the abrasion increase slope and the difference between the real-time abrasion energy accumulation value of the real-time cutting load sequence and a preset failure threshold value, obtaining the expected life; and obtaining a difference value of the abrasion energy accumulation value between the real-time processing cycle and the previous processing cycle, and if the difference value of the abrasion energy accumulation value is larger than a preset tipping threshold value, feeding back and blocking the subsequent cutting signals.
9. The machine learning based computer numerical control system data desensitizing method according to claim 8, wherein said determining whether a real-time material hardness factor produces a step change according to said material hardness factor sequence, comprises: if the difference between the real-time material hardness coefficient and the preset stable hardness threshold is larger than the preset difference threshold and the standard deviation of the material hardness coefficient sequence is smaller than the preset standard deviation threshold, judging that step change is generated.
10. The machine learning based computer numerical control system data desensitization method according to claim 8, wherein said life expectancy acquisition method comprises: and taking the difference value of the failure threshold value and the real-time abrasion energy accumulation value as a molecule, taking the abrasion increase slope as a denominator, and rounding down the obtained ratio to obtain the life expectancy.

Description

Computer numerical control system data desensitizing method based on machine learning Technical Field The invention relates to the technical field of data processing, in particular to a computer numerical control system data desensitizing method based on machine learning. Background In the field of modern precision manufacturing, particularly aerospace and high-end consumer electronics processing, spindle load data of Computer Numerical Control (CNC) systems is considered as a core index for process monitoring. With the development of the industrial Internet of things (IIoT), the collected high-frequency load data is uploaded to the cloud for big data analysis, and the method has become a mainstream trend for realizing full life cycle management and predictive maintenance of the cutter. However, this process faces serious data security challenges. Since the variation characteristics of the load waveform in the cutting process in the time domain or the angle domain are highly correlated with the geometric profile of the workpiece, the original load data substantially implies CAD design drawing information of the workpiece. For manufacturers involving core confidential or high value added products, there is a great risk of reverse engineering to directly upload raw data containing waveform details, and there is a great need for a data desensitization technique that can effectively strip geometrically sensitive information on the edge side. However, the existing data processing method is difficult to achieve desensitization and meanwhile is difficult to achieve monitoring robustness, and particularly poor in performance when dealing with batch fluctuation of blank materials. In mass production, there is often a difference in hardness of the casting blanks in batches. An increase in material hardness results in an integral linear amplification of the cutting force amplitude, while tool wear is more manifested as non-linear distortion of the cutting force. Conventional monitoring means are generally based on fixed thresholds or simple statistical features (e.g., mean, effective values) and cannot effectively distinguish between "overall increase in cutting force due to hardening of the material" and "change in mechanical features due to tool wear". This coupling phenomenon often results in a large number of false alarms generated by the monitoring system when changing material batches, or is forced to relax the threshold in order to be compatible with hardness fluctuations, thus missing early weak wear. Disclosure of Invention In order to solve the technical problems that the existing data processing method cannot take monitoring robustness into account and is difficult to distinguish the increase cause of cutting force while realizing the desensitization of cutting data of a numerical control system, the invention aims to provide a machine learning-based computer numerical control system data desensitization method, which adopts the following specific technical scheme: the invention provides a computer numerical control system data desensitization method based on machine learning, which comprises the following steps: Acquiring a real-time cutting load sequence of the numerical control system based on the position of the main shaft; Taking elements in a standard cutting load sequence stored locally as independent variables, and taking elements in the real-time cutting load sequence as dependent variables, performing linear fitting to obtain a slope and a residual sequence of a fitting result, wherein the slope is taken as a material hardness coefficient, and the residual sequence is taken as a wear characteristic component; And carrying out cloud storage by taking the wear energy accumulation value, the material hardness coefficient and the fault occurrence position as storage data, positioning the fault occurrence position based on the energy density, and feeding back an early warning signal according to the wear energy accumulation value and the material hardness coefficient. Further, the method for acquiring the standard cutting load sequence comprises the following steps: and under a template construction mode of the numerical control system, counting cutting load sequences generated by a preset number of complete processing cycles as a basic sequence, and averaging the basic sequence after alignment to obtain the standard cutting load sequence. Further, the linear fitting process is preceded by an alignment process of a standard cutting load sequence and a real-time cutting load sequence, the alignment process comprising: searching reference elements in a standard cutting load sequence and a real-time cutting load sequence respectively, wherein the reference elements are elements with loads larger than a preset contact threshold value for the first time and larger than the previous element; Constructing an offset interval by taking the index offset as a center, and carrying out resampling interpolation reco