A bearing quality detection system based on artificial intelligence

Abstract

The invention discloses a bearing quality inspection system based on artificial intelligence, which specifically relates to the field of quality inspection, and comprises a data acquisition module, a data preprocessing module, a sound characteristic value calculation module, a bearing image acquisition and processing module, a feature extraction module, and a bearing quality qualified value calculation module; the invention collects the vibration signal of the bearing through a speed sensor, and performs denoising processing on the vibration data by adopting a minimum mean square error filtering algorithm, and then calculates the sound characteristic value from the denoised vibration data; at the same time, a scanner is used to perform 360-degree image acquisition on the bearing surface, and the image quality is improved by mean filtering denoising and grayscale linear transformation; on this basis, a convolutional neural network deep learning algorithm is used to extract key features, and the roundness of the bearing surface is calculated according to the extracted key features; finally, the quality qualified value of the bearing is calculated by comprehensively combining the sound characteristic value and the roundness of the bearing surface, and compared with a preset standard, and an instruction is output.

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection, and more specifically, to a bearing quality detection system based on artificial intelligence. Background Art

[0002] Bearings are important parts of rotating machinery. They play a vital role in the fields of electricity, chemical metallurgy, military industry, aerospace, transportation, and even furniture and household appliances in daily life. With the continuous development of industrial manufacturing technology, bearings are key components in mechanical equipment, and their quality is directly related to the operating efficiency and safety of the entire equipment.

[0003] In recent years, the rapid development of artificial intelligence technology has provided new solutions for bearing quality inspection. By introducing artificial intelligence algorithms and advanced sensor technology, accurate collection and analysis of bearing vibration signals and surface images can be achieved, thereby accurately judging the quality status of the bearing.

[0004] Traditional bearing quality inspection still has some shortcomings in actual use. For example, traditional bearing quality inspection mainly relies on the experience of workers, which has high subjectivity and misjudgment rate; and manual inspection is slow and cannot meet the needs of mass production. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a bearing quality detection system based on artificial intelligence to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Data acquisition module: used to collect the vibration signal of the bearing using a speed sensor, and transmit the collected vibration data to the data preprocessing module;

[0008] Data preprocessing module: used to perform denoising on the collected vibration signal and transmit the processed vibration signal to the sound eigenvalue calculation module;

[0009] Sound characteristic value calculation module: used to calculate the sound characteristic value from the processed vibration signal, and transmit the calculated sound characteristic value to the bearing quality qualified value calculation module;

[0010] Bearing image acquisition and processing module: used to acquire bearing surface images using a scanner, process the acquired bearing surface images, and transmit the bearing surface images to the feature extraction module;

[0011] Feature extraction module: used to extract key features from the bearing surface image using a deep learning algorithm, calculate the extracted key features to obtain the bearing surface roundness value, and transmit the calculated bearing surface roundness to the bearing quality qualified value calculation module;

[0012] Bearing quality qualified value calculation module: used to calculate the sound characteristic value and the roundness of the bearing surface to obtain the bearing quality qualified value, and to judge the calculated bearing quality qualified value with the preset bearing quality qualified value.

[0013] Preferably, in the data acquisition module, the speed sensor is a speed sensor, the speed sensor is mounted on the bearing housing using a fixing device, and the output end of the speed sensor is connected to the data acquisition system computer, and then the sampling rate is set. , assuming the highest frequency of the vibration signal is According to the Nyquist theorem, the minimum sampling rate required is ; Start the data acquisition equipment, start recording vibration signals, monitor signal quality in real time and collect vibration data.

[0014] Preferably, in the data preprocessing module, minimum mean square error filtering is performed on the collected vibration signal to remove noise.

[0015] Preferably, in the sound characteristic value calculation module, the processed vibration signal is calculated to obtain the sound characteristic value, and the sound characteristic value calculation method is specifically as follows:

[0016] , where S represents the sound feature value, Expressed as the vibration frequency, Expressed as a sound wave value, Expressed as sound index value;

[0017] The calculated sound characteristic values ​​are transmitted to the bearing quality qualified value calculation module.

[0018] Preferably, in the bearing image acquisition module, the scanner is fixed on the bracket, the focal length and exposure parameters are adjusted, the bearing is illuminated by an annular light source, and then the scanner is used to scan the bearing image 360 ​​degrees;

[0019] The scanned bearing image is subjected to mean filtering to remove noise, and then the image after mean filtering and denoising is subjected to grayscale linear transformation to enhance the image.

[0020] Preferably, in the feature extraction module, the deep learning algorithm selects a convolutional neural network algorithm, and extracts key features through the neural network algorithm, and the key features include bearing surface roughness, bearing diameter, and bearing width; the calculation method for calculating the bearing surface roundness value by calculating the key features is specifically as follows:

[0021] , where Z represents the roundness of the bearing surface, R represents the bearing surface roughness, W represents the bearing diameter, and V represents the bearing width.

[0022] Preferably, in the bearing quality qualified value calculation module, the calculation method of the bearing quality qualified value is specifically:

[0023] , where Q represents the bearing quality qualification value, S represents the sound characteristic value, and Z represents the bearing surface roundness value. Expressed as the weight of the sound feature value, Expressed as the weight of the rounded value of the bearing surface;

[0024] The calculated bearing quality qualified value is compared with the preset bearing quality qualified value; if the calculated bearing quality qualified value is greater than or equal to the preset bearing quality qualified value, a bearing quality qualified instruction is output, and the bearing is judged to be qualified and can be used; if the calculated bearing quality qualified value is less than the preset bearing quality qualified value, a bearing quality unqualified instruction is output, and the bearing is judged to be unqualified and destroyed by professionals.

[0025] Technical effects and advantages of the present invention:

[0026] The present invention collects bearing vibration data by using a speed sensor and transmits the data to a data preprocessing module for denoising; then, the denoised vibration signal is calculated to obtain a sound characteristic value, and at the same time, a scanner is used to collect and process the bearing surface image, and key features are extracted from the image through a deep learning algorithm to calculate the roundness of the bearing surface. Finally, the sound characteristic value and the roundness of the bearing surface are calculated to obtain a bearing quality qualification value, which is compared with the preset bearing quality qualification value and outputs a corresponding instruction; the present invention improves the accuracy and efficiency of bearing quality detection by introducing artificial intelligence algorithms and advanced sensor technology, and meets the needs of mass production. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of system module connection of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] See also Figure 1As shown, the present invention provides a bearing quality detection system based on artificial intelligence, which includes a data acquisition module, a data preprocessing module, a sound characteristic value calculation module, a bearing image acquisition and processing module, a feature extraction module, and a bearing quality qualified value calculation module.

[0030] The data acquisition module is used to collect vibration data of the bearing using a speed sensor, and transmit the collected vibration data to the data preprocessing module;

[0031] In the data acquisition module, the speed sensor is a speed sensor, which is mounted on the bearing housing using a fixing device, and the output end of the speed sensor is connected to the data acquisition system computer, and then the sampling rate is set. , assuming the highest frequency of the vibration signal is According to the Nyquist theorem, the minimum sampling rate required is ; Start the data acquisition equipment, start recording the vibration signal, monitor the signal quality in real time and collect the vibration signal; the collected vibration data includes vibration frequency, sound wave, and vibration signal.

[0032] The data preprocessing module is used to perform denoising on the collected vibration signal and transmit the processed vibration signal to the sound characteristic value calculation module;

[0033] In the data preprocessing module, the collected vibration signal is subjected to minimum mean square error filtering for denoising. The denoising method is specifically as follows:

[0034] The input signal consists of vibration signal and noise, so the calculation method of the input signal is as follows:

[0035] , where A represents the input signal, B represents the vibration signal, and C represents the noise;

[0036] Extract a containing The calculation method of the input signal window is as follows:

[0037] , where D represents the input signal window, It is represented as the signal value at the ath time point, and T is represented as the transpose, which converts the row vector in the brackets into a column vector;

[0038] Select the order of the filter, set the order of the filter to E, and calculate the output signal of the filter. The specific method for calculating the output signal of the filter is:

[0039] ,in, It represents the output signal of the filter, D represents the input signal window, T represents the transpose, Represented as the weight vector of the filter;

[0040] Calculate the error between the expected signal and the filter output. The error calculation method is as follows:

[0041] , where b represents the error, is represented as the expected signal, Represented as the output signal of the filter;

[0042] The method for obtaining the desired signal can directly measure the signal that is not contaminated by noise as the desired signal; for example, in a laboratory environment, the conditions are controlled to reduce the influence of noise;

[0043] Use the LMS algorithm to update the weights. The calculation method for updating the weights is as follows:

[0044] ,in, Represented as the weight vector of the filter, Represents the weight before updating, It is represented as learning rate, b is represented as error, and D is represented as input signal window;

[0045] Repeat the above calculation process for each new input signal sample a until all data points are processed. The final filter output signal is is the denoised signal; visualize the denoised signal and compare it with the original signal to verify the denoising effect; if the initial effect is not ideal, adjust the learning rate according to the result The filter order is E and the denoising process is rerun.

[0046] The sound characteristic value calculation module is used to calculate the sound characteristic value from the processed vibration signal, and transmit the calculated sound characteristic value to the bearing quality qualified value calculation module;

[0047] In the sound characteristic value calculation module, the processed vibration signal is calculated to obtain the sound characteristic value. The specific method for calculating the sound characteristic value is:

[0048] , where S represents the sound feature value, Expressed as the vibration frequency, Expressed as a sound wave value, Expressed as sound index value;

[0049] The specific method for calculating the sound index value is as follows:

[0050] ,in, Expressed as a sound index value, Expressed as the peak value of the vibration signal, Expressed as root mean square value, Expressed as mean;

[0051] The calculation method of the root mean square value is as follows:

[0052] ,in, It is represented by the root mean square value, H is the number of sampling points, Represented as a vibration signal, ;

[0053] The specific method for calculating the mean is:

[0054] ,in, is represented by the mean, H is represented by the number of sampling points, Represented as a vibration signal, ;

[0055] The calculation method of the vibration signal peak value is as follows:

[0056] After dividing the vibration signal into g segments, find n peaks , ;

[0057] , It is expressed as the peak value of the vibration signal, g is expressed as the number of segments of the vibration signal, Represented as the jth peak;

[0058] The calculated sound characteristic values ​​are transmitted to the bearing quality qualified value calculation module.

[0059] The bearing image acquisition module is used to acquire a bearing surface image using a scanner, process the acquired bearing surface image, and transmit the bearing surface image to the feature extraction module;

[0060] In the bearing image acquisition module, the scanner is fixed on the bracket, and the focus and exposure parameters are adjusted. The bearing is illuminated with a ring light source, and the scanner is used to scan the bearing image 360 ​​degrees;

[0061] The scanned bearing image is subjected to mean filtering to remove noise. The specific method of mean filtering to remove noise is as follows:

[0062] A1: Filter window set to The size of

[0063] A2: For each pixel in the image, define a filter window centered on the pixel; this window will cover the pixel and its surrounding pixels;

[0064] A3: Calculate the average value of pixels in the filter window. The specific method for calculating the average value of pixels in the filter window is:

[0065] ,in, is represented as the new pixel average, u and v are represented as the offset relative to the center pixel, and p and q are represented as a The filter window is represented by k, which is the value in the vertical direction of the image. Represented as the horizontal value in the image;

[0066] A4: Replace the original pixel value at the center of the window with the calculated average value;

[0067] A5: Repeat steps 2 to 4 for each pixel in the image until the entire image is processed to obtain an image denoised using mean filtering.

[0068] The image after mean filtering and denoising is subjected to grayscale linear transformation to enhance the image. The specific method of grayscale linear transformation to enhance the image is:

[0069] B1: Determine the minimum grayscale value of the original image and the maximum gray value , minimum gray value and the maximum gray value It is obtained by traversing all pixels of the image;

[0070] B2: Select a target grayscale range, which is the minimum grayscale value of the transformed image. and the maximum gray value ; For example, for an 8-bit image, the range is 0 to 255;

[0071] B3: The specific method for calculating the gray value after transformation is:

[0072] ,in, Represented as the original image in The gray value at It is represented as the gray value after transformation, m is represented as the scaling value, and r is represented as the offset;

[0073] The calculation method of the scaling value is as follows:

[0074] , where m is the scaling factor, Represented as the maximum gray value of the transformed image, Represents the minimum gray value of the transformed image, Represented as the maximum gray value of the original image, Represented as the minimum gray value of the original image;

[0075] The scaling value controls the degree of grayscale scaling; if , the range of grayscale values ​​will be expanded, and expansion means increasing the contrast; if , the range of grayscale values ​​will be compressed, and compression means reducing the contrast; if 1, the gray value remains unchanged and is only affected by r;

[0076] , where r is the adjustment coefficient, Represents the minimum gray value of the transformed image, It represents the minimum gray value of the original image, and m represents the scaling value;

[0077] Used to adjust the overall brightness of the grayscale value. When it increases all gray values, it makes the image brighter; When it is turned on, it will reduce all gray values ​​and make the image darker;

[0078] B4: Apply the above transformation to each pixel in the image, and the grayscale processing is completed.

[0079] The feature extraction module is used to extract key features from the bearing surface image using a deep learning algorithm, and calculate the extracted key features to obtain the bearing surface roundness value, and transmit the calculated bearing surface roundness to the bearing quality qualified value calculation module;

[0080] In the feature extraction module, the deep learning algorithm selects the convolutional neural network algorithm, and extracts key features through the neural network algorithm. The key features include the surface roughness of the bearing, the bearing diameter, and the bearing width. The calculation method for calculating the roundness value of the bearing surface by calculating the key features is specifically as follows:

[0081] , where Z represents the roundness of the bearing surface, R represents the bearing surface roughness, W represents the bearing diameter, and V represents the bearing width;

[0082] The calculated bearing surface roundness value is transmitted to the bearing quality qualified value calculation module.

[0083] The bearing quality qualified value calculation module is used to calculate the sound characteristic value and the roundness of the bearing surface to obtain the bearing quality qualified value, and to judge the calculated bearing quality qualified value with the preset bearing quality qualified value;

[0084] In the bearing quality qualified value calculation module, the calculation method of the bearing quality qualified value is specifically as follows:

[0085] , where Q represents the bearing quality qualification value, S represents the sound characteristic value, and Z represents the bearing surface roundness value. Expressed as the weight of the sound feature value, Expressed as the weight of the rounded value of the bearing surface;

[0086] The calculated bearing quality qualified value is compared with the preset bearing quality qualified value; if the calculated bearing quality qualified value is greater than or equal to the preset bearing quality qualified value, a bearing quality qualified instruction is output, and the bearing is judged to be qualified and can be used; if the calculated bearing quality qualified value is less than the preset bearing quality qualified value, a bearing quality unqualified instruction is output, and the bearing is judged to be unqualified and destroyed by professionals.

[0087] The thresholds set in the present invention are all set by those skilled in the art and are not limited here.

[0088] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A bearing quality detection system based on artificial intelligence, characterized in that: include: Data acquisition module: used to collect the vibration signal of the bearing using a speed sensor, and transmit the collected vibration data to the data preprocessing module; Data preprocessing module: used to perform denoising on the collected vibration signal and transmit the processed vibration signal to the sound eigenvalue calculation module; Sound characteristic value calculation module: used to calculate the sound characteristic value from the processed vibration signal, and transmit the calculated sound characteristic value to the bearing quality qualified value calculation module; In the sound characteristic value calculation module, the processed vibration signal is calculated to obtain the sound characteristic value. The specific method for calculating the sound characteristic value is: , where S represents the sound feature value, Expressed as the vibration frequency, Expressed as a sound wave value, Expressed as sound index value; The specific method for calculating the sound index value is as follows: ,in, Expressed as a sound index value, Expressed as the peak value of the vibration signal, Expressed as root mean square value, Expressed as mean; Bearing image acquisition and processing module: used to acquire bearing surface images using a scanner, process the acquired bearing surface images, and transmit the bearing surface images to the feature extraction module; Feature extraction module: used to extract key features from the bearing surface image using a deep learning algorithm, calculate the extracted key features to obtain the bearing surface roundness value, and transmit the calculated bearing surface roundness to the bearing quality qualified value calculation module; In the feature extraction module, the deep learning algorithm selects the convolutional neural network algorithm, and extracts key features through the neural network algorithm. The key features include the surface roughness of the bearing, the bearing diameter, and the bearing width. The calculation method for calculating the roundness value of the bearing surface by calculating the key features is specifically as follows: , where Z represents the roundness of the bearing surface, R represents the bearing surface roughness, W represents the bearing diameter, and V represents the bearing width; Bearing quality qualified value calculation module: used to calculate the sound characteristic value and the roundness of the bearing surface to obtain the bearing quality qualified value, and to judge the calculated bearing quality qualified value with the preset bearing quality qualified value; In the bearing quality qualified value calculation module, the calculation method of the bearing quality qualified value is specifically as follows: , where Q represents the bearing quality qualification value, S represents the sound characteristic value, and Z represents the bearing surface roundness value. Expressed as the weight of the sound feature value, Expressed as a weight for the rounded value of the bearing surface.

2. The bearing quality detection system based on artificial intelligence according to claim 1 is characterized in that: In the data acquisition module, the speed sensor is mounted on the bearing housing using a fixing device, and the output end of the speed sensor is connected to the data acquisition system computer, and then the sampling rate is set. , assuming the highest frequency of the vibration signal is According to the Nyquist theorem, the minimum sampling rate required is ; Start the data acquisition equipment, start recording vibration signals, monitor signal quality in real time and collect vibration data.

3. The bearing quality detection system based on artificial intelligence according to claim 1 is characterized in that: In the data preprocessing module, the collected vibration signal is subjected to minimum mean square error filtering for denoising. The denoising method is specifically as follows: The input signal consists of vibration signal and noise, so the calculation method of the input signal is as follows: , where A represents the input signal, B represents the vibration signal, and C represents the noise; Extract a containing The calculation method of the input signal window is as follows: , where D represents the input signal window, It is represented as the signal value at the ath time point, and T is represented as the transpose, which converts the row vector in the brackets into a column vector; Select the order of the filter, set the order of the filter to E, and calculate the output signal of the filter. The specific method for calculating the output signal of the filter is: ,in, It represents the output signal of the filter, D represents the input signal window, T represents the transpose, Represented as the weight vector of the filter; Calculate the error between the expected signal and the filter output. The error calculation method is as follows: , where b represents the error, is represented as the expected signal, Represented as the output signal of the filter; Use the LMS algorithm to update the weights. The calculation method for updating the weights is as follows: ,in, Represented as the weight vector of the filter, Represents the weight before updating, It is represented as learning rate, b is represented as error, and D is represented as input signal window; Repeat the above continuous calculation process for each new input signal sample a until all data points are processed; the final filter output signal is is the denoised signal.

4. The bearing quality detection system based on artificial intelligence according to claim 1 is characterized in that: In the bearing image acquisition module, the scanner is fixed on the bracket, the focal length and exposure parameters are adjusted, the bearing is illuminated with a ring light source, and then the scanner is used to scan the bearing image 360 ​​degrees; The scanned bearing image is subjected to mean filtering to remove noise, and then the image after mean filtering and denoising is subjected to grayscale linear transformation to enhance the image.

5. The bearing quality detection system based on artificial intelligence according to claim 1 is characterized in that: The specific method of image enhancement by grayscale linear transformation is: B1: Determine the minimum grayscale value of the original image and the maximum gray value , minimum gray value and the maximum gray value It is obtained by traversing all pixels of the image; B2: Select a target grayscale range, which is the minimum grayscale value of the transformed image. and the maximum gray value ; B3: The specific method for calculating the gray value after transformation is: ,in, Represented as the original image in The gray value at It is represented as the gray value after transformation, m is represented as the scaling value, and r is represented as the offset; The calculation method of the scaling value is as follows: , where m is the scaling factor, Represented as the maximum gray value of the transformed image, Represents the minimum gray value of the transformed image, Represented as the maximum gray value of the original image, Represented as the minimum gray value of the original image; The scaling value controls the degree of scaling of the grayscale value; if , the range of grayscale values ​​will be expanded, and expansion means increasing the contrast; if , the range of grayscale values ​​will be compressed, and compression means reducing the contrast; if 1, the gray value remains unchanged and is only affected by r; , where r is the adjustment coefficient, Represents the minimum gray value of the transformed image, It represents the minimum gray value of the original image, and m represents the scaling value; Used to adjust the overall brightness of the grayscale value. When it increases all gray values, it makes the image brighter; When it is turned on, it will reduce all gray values ​​and make the image darker; B4: Apply the above transformation to each pixel in the image, and the grayscale processing is completed.

Images