CN-120800242-B - Fine blanking equipment monitoring method and system based on fiber bragg grating sensing and machine learning

CN120800242BCN 120800242 BCN120800242 BCN 120800242BCN-120800242-B

Abstract

The invention provides a fine blanking equipment monitoring method and a fine blanking equipment monitoring system based on fiber bragg grating sensing and machine learning, and relates to the field of fine blanking equipment monitoring; the method comprises the steps of preprocessing strain data to remove noise and nonlinear trend in the strain data, splitting the strain data into single-stroke data from multi-stroke data based on minimum point detection, dividing each single-stroke data into data of four stages of fast forward, detection, fine blanking and fast backward, storing the data into an array, calculating time domain characteristic parameters of data of each stage in the array, setting classification labels for the four stages respectively, constructing a training data set, training a support vector machine classifier based on the training data set to obtain a stage identification model, constructing real-time data monitoring data of a visual interface display stage identification model, and performing fault early warning.

Inventors

WEI WENTING
ZHANG SHUZHAO
HUA LIN
LIU YANXIONG

Assignees

武汉理工大学

Dates

Publication Date: 20260508
Application Date: 20250812

Claims (8)

1. The fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning is characterized by comprising the following steps of: a fiber bragg grating sensor is arranged on a sliding block of a fine blanking machine, and strain data of the sliding block are collected in real time; preprocessing the strain data specifically comprises the following steps: Selecting db4 wavelet basis function in Daubechies wavelet family as decomposition basis function to carry out multi-scale wavelet decomposition on the strain data D1 to obtain a wavelet coefficient set containing noise information; Dividing the wavelet coefficient set into a plurality of analysis windows on average, and estimating a noise standard deviation by a median absolute deviation method; Calculating a window data stability coefficient, dynamically adjusting a weighting coefficient based on the window data stability coefficient to increase the weight of stable data in a window, and calculating the weighted noise standard deviation by using unstable data to occupy smaller weight; calculating Donoho a general threshold value based on the weighted noise standard deviation, further performing soft threshold processing on the wavelet coefficient, setting 0 as noise below the threshold value, smoothing above the threshold value, and reconstructing the processed wavelet coefficient to obtain de-noised strain data D2; Setting the window size of the strain data D2 to be 3 stroke data lengths, forming a trend item array by sliding the local minimum value in each window of the index, and removing trend items in the strain data D2 to obtain strain data D3 with nonlinear trends removed so as to remove noise and nonlinear trends in the strain data; dividing the strain data into single-stroke data from multi-stroke data based on minimum value point detection, dividing each single-stroke data into data of four stages of fast forward, detection, fine blanking and fast backward, and storing the data into an array; Calculating time domain characteristic parameters of data of each stage in the array, and respectively setting classification labels for the four stages to construct a training data set; Training a support vector machine classifier based on the training data set to obtain a stage identification model; and building a visual interface to display real-time data monitoring data of the stage identification model and carrying out fault early warning on a fast forward system, a detection system, a main blanking system and a fast backward system.
2. The fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning of claim 1, wherein the step of splitting the strain data from multi-stroke data into single-stroke data based on minimum point detection specifically comprises: and setting the minimum interval between two minimum value points in the strain data, ensuring that the complete stroke can be split, indexing the minimum value points, and taking the data between two adjacent minimum value points as single-stroke data.
3. The method for monitoring fine blanking equipment based on fiber bragg grating sensing and machine learning as claimed in claim 1, wherein the step of dividing each single-stroke data into four stages of fast forward, detection, fine blanking and fast backward comprises the following steps: dividing each single-stroke data into data of a fast forward stage, a detection stage, a fine blanking stage and a fast backward stage according to strain change characteristics; The fast forward stage is a data segment with the strain increasing from zero to a first maximum point, the detection stage is a data segment with the strain decreasing from the first maximum point to a first minimum point, the fine blanking stage is a data segment with the strain increasing from the first minimum point to a global maximum point, and the fast reverse stage is a data segment with the strain decreasing from the global maximum point to zero.
4. The fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning of claim 1, wherein the time domain characteristic parameters include at least five of maximum value, minimum value, average value, standard deviation, kurtosis, skewness, impact factor, margin factor.
5. The fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning as claimed in claim 1, wherein the step of fault pre-warning specifically comprises: setting time thresholds for four stages of fast forward, detection, fine blanking and fast backward; when the time of the fast forward stage exceeds the preset proportion of the time threshold value, judging that the fast forward system is in fault; When the time of the detection stage exceeds the preset proportion of the time threshold value, judging that the detection system is faulty or the surface of the die is excessively wasted; when the time of the fine blanking stage exceeds the preset proportion of the time threshold value, judging that the main blanking system is in fault; and when the time of the fast-backing stage exceeds the preset proportion of the time threshold value, judging that the fast-backing system fails.
6. Fine blanking equipment monitoring system based on fiber bragg grating sensing and machine learning, which is characterized by comprising: The acquisition module is used for arranging a fiber bragg grating sensor on a sliding block of the fine blanking machine and acquiring strain data of the sliding block in real time; The preprocessing module is used for preprocessing the strain data, and specifically comprises the following steps: Selecting db4 wavelet basis function in Daubechies wavelet family as decomposition basis function to carry out multi-scale wavelet decomposition on the strain data D1 to obtain a wavelet coefficient set containing noise information; Dividing the wavelet coefficient set into a plurality of analysis windows on average, and estimating a noise standard deviation by a median absolute deviation method; Calculating a window data stability coefficient, dynamically adjusting a weighting coefficient based on the window data stability coefficient to increase the weight of stable data in a window, and calculating the weighted noise standard deviation by using unstable data to occupy smaller weight; calculating Donoho a general threshold value based on the weighted noise standard deviation, further performing soft threshold processing on the wavelet coefficient, setting 0 as noise below the threshold value, smoothing above the threshold value, and reconstructing the processed wavelet coefficient to obtain de-noised strain data D2; Setting the window size of the strain data D2 to be 3 stroke data lengths, forming a trend item array by sliding the local minimum value in each window of the index, and removing trend items in the strain data D2 to obtain strain data D3 with nonlinear trends removed so as to remove noise and nonlinear trends in the strain data; The phase division module is used for dividing the strain data into single-stroke data from multi-stroke data based on minimum value point detection, dividing each single-stroke data into data of four phases of fast forward, detection, fine blanking and fast backward, and storing the data into an array; The data set construction module is used for calculating time domain characteristic parameters of data of each stage in the array, and respectively setting classification labels for the four stages to construct a training data set; the model training module is used for training a support vector machine classifier based on the training data set to obtain a fine blanking stage identification model; and the visualization module is used for creating a visual graphical interface, visually displaying the data acquired in real time and the identification result of the fine blanking stage, and carrying out fault early warning on the fast forward system, the detection system, the main blanking system and the fast backward system.
7. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning as claimed in any one of claims 1 to 5.
8. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of the fine blanking apparatus monitoring method based on fiber bragg grating sensing and machine learning as claimed in any one of claims 1 to 5.

Description

Fine blanking equipment monitoring method and system based on fiber bragg grating sensing and machine learning Technical Field The invention relates to the technical field of fine blanking equipment monitoring, in particular to a fine blanking equipment monitoring method and system based on fiber bragg grating sensing and machine learning. Background The fine blanking press is special processing equipment for executing a fine blanking process, is widely used for manufacturing precision parts such as automobiles and electronics, monitors the running state of the fine blanking press in real time and gives fault early warning, and can effectively avoid loss caused by serious faults. The fine blanking process generally comprises four stages of fast forward, detection, fine blanking and fast backward, and the slide strain data can reflect the operation stage of the fine blanking machine. In addition, the data of four stages of the fine blanking process can reflect the running state of a fine blanking machine, and the real-time detection of the fine blanking process stage is important for intelligent development of fine blanking equipment. However, the traditional fine blanking process stage identification method mainly depends on manual experience or a pressure sensor, and the method is low in efficiency and insufficient in precision, and is difficult to meet the intelligent detection requirement. Disclosure of Invention The invention aims to provide a fine blanking equipment monitoring method and system based on fiber bragg grating sensing and machine learning, which are used for solving the problems of low precision and poor instantaneity of the traditional fine blanking process stage identification method in the background art. The fine blanking equipment monitoring method based on fiber bragg grating sensing and machine learning comprises the following steps of arranging a fiber bragg grating sensor on a sliding block of a fine blanking machine, collecting strain data of the sliding block in real time, preprocessing the strain data to remove noise and nonlinear trends in the strain data, splitting the strain data into single-stroke data from multi-stroke data based on minimum point detection, dividing each single-stroke data into data of four stages of fast forward, detection, fine blanking and fast backward, storing the data into an array, calculating time domain feature parameters of the data of each stage in the array, setting classification labels for the four stages respectively, constructing a training data set, training a support vector machine classifier based on the training data set, obtaining a stage identification model, constructing real-time data monitoring data of a visual interface display stage identification model, and carrying out fault early warning. The method comprises the steps of selecting db4 wavelet basis functions in Daubechies wavelet families as decomposition basis functions, carrying out multi-scale wavelet decomposition on the strain data D1 to obtain a wavelet coefficient set containing noise information, dividing the wavelet coefficient set into a plurality of analysis windows in an average mode, estimating a noise standard deviation through a median absolute deviation method, dynamically adjusting weighting coefficients based on window data stability coefficients to increase the weight of stable data in the windows, further calculating weighted noise standard deviation, calculating Donoho universal threshold values based on the weighted noise standard deviation, further carrying out soft threshold value processing on the wavelet coefficients, setting the wavelet coefficient as noise below 0, smoothing the wavelet coefficient below the threshold value and carrying out smoothing on the wavelet coefficient below the threshold value, and reconstructing the processed wavelet coefficient to obtain denoised strain data D2. Optionally, the step of preprocessing the strain data specifically further includes setting the window size of the strain data D2 to be 3 stroke data lengths, forming a trend term array by sliding the local minimum values in each window of the index, and removing the trend term in the strain data D2 to obtain strain data D3 with nonlinear trend removed. Optionally, the step of splitting the strain data into single-stroke data from multi-stroke data based on minimum value point detection specifically comprises setting a minimum interval of two minimum value points in the strain data, ensuring that a complete stroke can be split, indexing the minimum value points, and taking data between two adjacent minimum value points as single-stroke data. Optionally, the step of dividing each single-stroke data into four stages of fast forward, detection, fine blanking and fast backward specifically includes dividing each single-stroke data into a fast forward stage, a detection stage, a fine blanking stage and a fast backward stage according to strain change characteristics,