Rubber product quality online detection method and system based on intelligent sensor

Abstract

The invention discloses a rubber product quality online detection method and system based on an intelligent sensor, and relates to the related technical field of rubber products, and the method comprises the steps: connecting a rubber edge removal operation production line of a rubber product production line, and integrating an intelligent sensing unit; detecting the rubber edge removal sample of the rubber product to be detected; performing quality detection feature extraction on the laser-rubber edge contour detection data, the spectrum-rubber edge impurity detection data and the optical-rubber edge surface detection data; and inputting the real-time rubber edge detection characteristics into the rubber edge-product quality evaluation conversion model, and sending an early warning feedback signal to the rubber product production line according to the product quality evaluation index. The technical problems that in the prior art, due to the fact that quality detection of the rubber edge removing procedure of the rubber product lacks real-time accurate monitoring, the rubber edge removing quality is uncontrollable, and the defective rate is high are solved, and the technical effects that online real-time detection of the rubber edge quality of the rubber product is achieved, so that the quality of the rubber product is improved, and the defective rate is reduced are achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of rubber products, and particularly relates to a rubber product quality online detection method and system based on an intelligent sensor. BACKGROUND

[0002] As important industrial materials, rubber products usually include multiple links such as mixing, molding, vulcanization and subsequent processing in the manufacturing process. In the processing stage after vulcanization, the removal of rubber edges is a key process for ensuring the dimensional accuracy and appearance quality of products. However, the traditional quality detection of rubber products is performed on the finished products, and the rubber edge removal operation lacks real-time and accurate quality monitoring means. For example, the detection of the overflow thickness of a cutting piece, the impurity content of a cutting edge and the surface analysis of the cutting edge may lead to incomplete or excessive cutting of the rubber edge, and it is difficult to quantitatively detect and accurately identify small impurities or surface cracks.

[0003] Therefore, in the related art, there is a technical problem that the quality detection of the rubber edge removal process lacks real-time and accurate monitoring, resulting in uncontrollable rubber edge removal quality and high scrap rate. SUMMARY

[0004] The application provides a rubber product quality online detection method and system based on an intelligent sensor, which solves the technical problem that the quality detection of the rubber edge removal process lacks real-time and accurate monitoring in the prior art, resulting in uncontrollable rubber edge removal quality and high scrap rate, and achieves the technical effect of realizing online real-time detection of the rubber edge quality, thereby improving the quality of rubber products and reducing the scrap rate.

[0005] The application provides a rubber product quality online detection method based on an intelligent sensor, which comprises the following steps: connecting a rubber edge removal operation line of a rubber product production line, integrating an intelligent sensing unit on the rubber edge removal operation line, wherein the intelligent sensing unit comprises a 3D laser scanner, a multispectral camera and a surface optical detector; detecting a rubber edge removal sample of a rubber product to be detected by using the intelligent sensing unit, and acquiring laser-rubber edge profile detection data, spectral-rubber edge impurity detection data and optical-rubber edge surface detection data; performing quality detection feature extraction on the laser-rubber edge profile detection data, the spectral-rubber edge impurity detection data and the optical-rubber edge surface detection data, and outputting real-time rubber edge detection features, including rubber edge overflow dispersion, impurity ratio and surface defect density; inputting the real-time rubber edge detection features into a rubber edge-product quality evaluation conversion model, outputting product quality evaluation indexes based on the rubber edge-product quality evaluation conversion model, and sending a warning feedback signal to the rubber product production line according to the product quality evaluation indexes.

[0006] In a possible implementation, the intelligent sensor-based online rubber product quality detection method further performs the following processing: when the rubber edge removal sample of the rubber product to be detected enters the detection area, starting the 3D laser scanner to perform point-by-point scanning on the rubber edge removal sample to obtain rubber edge surface height distribution data; performing rubber edge profile curve conversion on the rubber edge surface height distribution data to obtain laser-rubber edge profile detection data.

[0007] In a possible implementation, the intelligent sensor-based online rubber product quality detection method further performs the following processing: starting the multi-spectrum camera to perform multi-spectrum imaging on the rubber edge removal sample of the rubber product to be detected to obtain multi-spectrum imaging data, wherein the multi-spectrum camera includes a foreign matter spectrum library; identifying the type and proportion of foreign matters in the multi-spectrum imaging data based on the foreign matter spectrum library to obtain spectrum-rubber edge foreign matter detection data.

[0008] In a possible implementation, the intelligent sensor-based online rubber product quality detection method further performs the following processing: collecting good product rubber edge profile detection data; calculating the average thickness and average width of the rubber edge profile based on the good product rubber edge profile detection data, and generating a good product profile curve based on the average thickness and the average width; analyzing the fluctuation range threshold of the good product rubber edge profile detection data based on the good product profile curve; performing quality detection feature extraction on the laser-rubber edge profile detection data based on the good product profile curve and the fluctuation range threshold, and outputting thickness dispersion and width dispersion; and performing weight calculation according to the thickness dispersion and the width dispersion, and outputting rubber edge overflow dispersion.

[0009] In a possible implementation, the intelligent sensor-based online rubber product quality detection method further performs the following processing: traversing the spectrum-rubber edge foreign matter detection data to obtain a total number of foreign matter pixels, and outputting the ratio of the total number of foreign matter pixels to the total number of image pixels as a foreign matter ratio.

[0010] In a possible implementation, the intelligent sensor-based online rubber product quality detection method further performs the following processing: wherein the optical-rubber edge surface detection data includes a surface texture image, performing gray scale processing on the surface texture image to obtain a surface texture gray scale image; pre-training a rubber edge surface defect identification model by known surface defect labeling, performing binary segmentation on the surface texture gray scale image by using the rubber edge surface defect identification model to output a segmentation mask image; identifying a total number of defect pixels according to the segmentation mask image, and outputting a surface defect density based on the percentage of the total number of defect pixels to the total number of pixels of the segmentation mask image.

[0011] In a possible implementation, the rubber product quality online detection method based on the intelligent sensor further performs the following processing: collecting historical rubber edge detection feature samples and corresponding rubber product quality detection results, the rubber product quality detection results including mechanical performance indexes, durability indexes, and aging resistance test indexes; performing correlation analysis on the historical rubber edge detection feature samples and the corresponding rubber product quality detection results by using a correlation analysis algorithm to obtain a correlation coefficient network; defining a conversion relationship of a rubber edge-product quality evaluation conversion model based on the correlation coefficient network to obtain the rubber edge-product quality evaluation conversion model.

[0012] In a possible implementation, the rubber product quality online detection method based on the intelligent sensor further performs the following processing: the rubber edge-product quality evaluation conversion model performs correlation conversion on the real-time rubber edge detection features according to the correlation coefficient network to output mechanical performance conversion indexes, durability conversion indexes, and aging resistance conversion indexes; and outputting product quality evaluation indexes according to the mechanical performance conversion indexes, the durability conversion indexes, and the aging resistance conversion indexes.

[0013] In a possible implementation, the rubber product quality online detection method based on the intelligent sensor further performs the following processing: obtaining product quality evaluation indexes output by the rubber edge-product quality evaluation conversion model and corresponding real rubber product quality detection indexes; obtaining a prediction error based on the product quality evaluation indexes and the real rubber product quality detection indexes; performing incremental learning update on the rubber edge-product quality evaluation conversion model according to the prediction error to output an updated rubber edge-product quality evaluation conversion model.

[0014] The application also provides a rubber product quality online detection system based on intelligent sensors, which comprises: an intelligent sensing unit integrated module, which is used to connect a rubber edge removal operation line of a rubber product production line and integrate an intelligent sensing unit on the rubber edge removal operation line, the intelligent sensing unit comprising a 3D laser scanner, a multispectral camera and a surface optical detector; a detection data acquisition module, which is used to detect rubber edge removal samples of a rubber product to be detected by using the intelligent sensing unit, and acquire laser-rubber edge profile detection data, spectrum-rubber edge impurity detection data and optical-rubber edge surface detection data; a detection feature extraction module, which is used to extract quality detection features from the laser-rubber edge profile detection data, the spectrum-rubber edge impurity detection data and the optical-rubber edge surface detection data, and output real-time rubber edge detection features, including rubber edge overflow dispersion, impurity ratio and surface defect density; and a quality evaluation index output module, which is used to input the real-time rubber edge detection features into a rubber edge-product quality evaluation conversion model, output product quality evaluation indexes based on the rubber edge-product quality evaluation conversion model, and send early warning feedback signals to the rubber product production line according to the product quality evaluation indexes.

[0015] The rubber product quality online detection method and system based on intelligent sensors provided by the application are used to connect a rubber edge removal operation line of a rubber product production line and integrate an intelligent sensing unit; detect rubber edge removal samples of a rubber product to be detected; extract quality detection features from laser-rubber edge profile detection data, spectrum-rubber edge impurity detection data and optical-rubber edge surface detection data; input real-time rubber edge detection features into a rubber edge-product quality evaluation conversion model, and send early warning feedback signals to the rubber product production line according to product quality evaluation indexes. The technical problems of lack of real-time and accurate monitoring of rubber product rubber edge removal process quality, uncontrollable rubber edge removal quality and high scrap rate in the prior art are solved, and the technical effects of realizing online and real-time detection of rubber product rubber edge quality and improving rubber product quality and reducing the scrap rate are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. In the present application, a flowchart is used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. Meanwhile, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 The rubber product quality online detection method based on intelligent sensors provided by the embodiments of the present application is shown in the flowchart.

[0018] Figure 2 A structure schematic diagram of a rubber product quality online detection system based on intelligent sensors is provided for the embodiments of the present application.

[0019] The reference signs are explained as follows: intelligent sensing unit integrated module 10, detection data acquisition module 20, detection feature extraction module 30, and quality evaluation index output module 40. DETAILED DESCRIPTION

[0020] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined object, the specific embodiments, structures, features and effects according to the present application are described in detail below in combination with the drawings and preferred embodiments.

[0021] The embodiments of the present application provide a rubber product quality online detection method based on intelligent sensors, as shown in the method, which comprises the following steps: Figure 1 Step S100, connecting a rubber product production line to a flash removal operation line, integrating an intelligent sensing unit on the flash removal operation line, the intelligent sensing unit comprising a 3D laser scanner, a multispectral camera and a surface optical detector.

[0022] Preferably, the rubber product production line is connected to the flash removal operation line, and the intelligent sensing unit comprising the 3D laser scanner, the multispectral camera and the surface optical detector is integrated on the flash removal operation line, that is, the 3D laser scanner, the multispectral camera and the surface optical detector are installed on the production line rack or gantry through mechanical structures such as mounting brackets, bases or protective covers and are communicatively connected with the production line control system, so that the intelligent sensing unit can receive the workpiece in-place trigger signal of the production line control system and can feed back the detection results and warning signals to the production line control system. After the rubber product completes the flash removal, the intelligent sensing unit is used for online real-time detection, wherein the 3D laser scanner emits a laser line or a laser dot array to the flash area of the rubber product and captures the laser deformation by using a camera to quickly acquire the three-dimensional point cloud data of the surface of the flash area, including X, Y and Z coordinates, to accurately represent the profile shape, height, width and other geometric information after the flash removal; the multispectral camera can capture images at multiple specific narrow-band wavelengths, and based on the different absorption and reflection characteristics of different materials to specific wavelengths, the rubber matrix and non-rubber impurities such as fibers, plastics, metal particles or undispersed additives embedded or attached in the flash area can be distinguished; the surface optical detector refers to a high-resolution industrial camera equipped with a specific illumination light source system, which can acquire high-contrast and high-definition two-dimensional surface texture images of the flash area for detecting microscopic surface defects such as tearing, scratching, unevenness and burrs.

[0023] ​Step S200, using the intelligent sensing unit to detect the rubber product edge removal sample, obtain laser-edge profile detection data, spectrum-edge impurity detection data and optical-edge surface detection data.

[0024] Preferably, when the rubber product completes the edge removal, it is automatically transported or moved to the detection area of the intelligent sensing unit to trigger the detection, i.e. using the 3D laser scanner, the multi-spectral camera and the surface optical detector to detect the rubber product edge removal sample, respectively, to obtain laser-edge profile detection data, spectrum-edge impurity detection data and optical-edge surface detection data, wherein the optical-edge surface detection data is the surface texture image of the rubber product edge removal sample.

[0025] Further, step S200 further comprises step S210, when the rubber product edge removal sample enters the detection area, starting the 3D laser scanner to scan the rubber product edge removal sample point by point to obtain the edge surface height distribution data; step S220, converting the edge surface height distribution data to edge profile curve to obtain the laser-edge profile detection data.

[0026] Preferably, when the rubber product edge removal sample is accurately moved and positioned into the detection area of the 3D laser scanner by the conveying device such as the conveying belt, the robot arm or the rotating table, the control signal of the production line control system is received and the 3D laser scanner is activated to scan the rubber product edge removal sample point by point. Specifically, the 3D laser scanner projects a laser line onto the surface of the rubber product edge area through the internal laser emitter, and the camera captures the deformed laser line due to the fluctuation of the object surface profile, and calculates the three-dimensional space coordinates of each point on the laser line based on the triangulation principle, and then obtains the three-dimensional point cloud representing the surface morphology of the edge and its adjacent area, i.e. the edge surface height distribution data; then the edge surface height distribution data is converted to edge profile curve, i.e. in the cross section perpendicular to the theoretical profile of the product, the three-dimensional point cloud is virtually cut along the edge extension path, all points intersecting the cross section and the point cloud are projected onto the two-dimensional plane, for example, the path length along the edge is taken as the horizontal coordinate, the height is taken as the vertical coordinate, the projected points are connected in order to form a curve representing the actual profile of the edge on the cross section, and finally the laser-edge profile detection data is obtained, containing multiple ordered position-height coordinate pairs, clearly showing the height and shape of the residual edge.

[0027] Further, step S200 further comprises step S230 of starting the multi-spectral camera to perform multi-spectral imaging on the sample of the rubber product to be detected, and acquiring multi-spectral imaging data, wherein the multi-spectral camera comprises an impurity spectrum library; and step S240 of identifying the type and proportion of impurities in the multi-spectral imaging data based on the impurity spectrum library, and acquiring spectrum-rubber edge impurity detection data.

[0028] Preferably, the multi-spectral camera is started to perform multi-spectral imaging on the sample of the rubber product to be detected, that is, the multi-spectral camera is used to capture images of the rubber edge area at multiple specific and discrete narrow wave bands, wherein the multiple wave bands include part of the visible light and non-visible light wave bands such as near-infrared, to obtain the reflection intensity value of each pixel at different wave bands, and then obtain multi-spectral imaging data, wherein the multi-spectral camera comprises an impurity spectrum library, that is, the standard spectral reflection characteristics of various known impurities such as fibers, different types of plastics, metal oxides, insufficiently dispersed carbon black or chemical additives at the same wave band are pre-established and stored; then the type of impurities in the multi-spectral imaging data is identified based on the impurity spectrum library, that is, the spectral characteristics of each pixel are compared with the standard spectral reflection characteristics in the impurity spectrum library, such as using least squares method to calculate the similarity, and when the spectral characteristics of a certain pixel match the standard spectral reflection characteristics of a certain type of impurity in the impurity spectrum library by more than a certain threshold, the pixel is determined to be the impurity; after completing the impurity classification of all pixels in the image, the proportion of the total number of pixels identified as impurities to the total number of pixels in the image region is calculated to determine the proportion of impurities, to realize the spatial area ratio quantification of the content of impurities, and finally output the spectrum-rubber edge impurity detection data, including the list of impurity types, the proportion of each type of impurity, and the position distribution of impurities in the rubber edge area.

[0029] Step S300, quality detection feature extraction is performed on the laser-rubber edge profile detection data, spectrum-rubber edge impurity detection data, and optical-rubber edge surface detection data, and real-time rubber edge detection features including rubber edge overflow dispersion, impurity ratio, and surface defect density are output.

[0030] Preferably, quality detection feature extraction is performed on the laser-rubber edge profile detection data, spectrum-rubber edge impurity detection data, and optical-rubber edge surface detection data respectively, that is, the unevenness and fluctuation degree of the residual profile after the rubber edge is removed are quantified, the spatial coverage area proportion of impurities in the rubber edge detection area is quantified, and the concentration degree of surface microscopic damage such as tearing, scratching, depression, and pitting in the detection area is quantified, and then the rubber edge overflow dispersion, impurity ratio, and surface defect density are output as real-time rubber edge detection features.

[0031] Further, step S300 further comprises step S310, collecting good product rubber edge profile detection data; step S320, calculating the good product rubber edge profile detection data to obtain the average thickness and average width of the rubber edge profile, and generating a good product profile curve based on the average thickness and the average width; step S330, analyzing the fluctuation range threshold of the good product rubber edge profile detection data based on the good product profile curve; step S340, performing quality detection feature extraction on the laser-rubber edge profile detection data based on the good product profile curve and the fluctuation range threshold, and outputting thickness dispersion and width dispersion; and step S350, performing weight calculation according to the thickness dispersion and the width dispersion, and outputting rubber edge overflow dispersion.

[0032] Preferably, a 3D laser scanner is used to scan the rubber edges of a plurality of rubber products known to be good products to obtain good product rubber edge profile detection data, and then the good product rubber edge profile detection data is analyzed and calculated, that is, the average values of the rubber edge profile thickness and width are calculated respectively to obtain the average thickness and average width of the rubber edge profile, and then a good product profile curve representing the ideal state of the rubber product is obtained based on the average thickness and the average width, and each point on the curve represents the ideal value of the rubber edge thickness and width at that position. Then, the fluctuation range threshold of the good product rubber edge profile detection data based on the good product profile curve is analyzed, that is, the deviation of all good product rubber edge profile detection data from the thickness and width at the corresponding position of the good product profile curve is calculated, the acceptable normal fluctuation range is determined by calculating the standard deviation, and then the fluctuation range threshold is determined, for example, the allowed deviation of thickness is ±0.1 mm, and the allowed deviation of width is ±0.2 mm. Then, quality detection feature extraction is performed on the laser-rubber edge profile detection data based on the good product profile curve and the fluctuation range threshold, that is, the laser-rubber edge profile detection data is compared with the good product profile curve point by point, the thickness deviation values of a plurality of sampling points are calculated along the profile line, and statistical analysis is performed to calculate the standard deviation, and the thickness dispersion is outputted, which is used to quantify the fluctuation of the rubber edge in the vertical direction, and the greater the thickness dispersion value, the more uneven the thickness. Similarly, the width deviation values of a plurality of sampling points are calculated and the standard deviation is calculated to obtain the width dispersion, which is used to quantify the fluctuation of the rubber edge in the horizontal direction, and the greater the width dispersion value, the more inconsistent the width of the rubber edge residue. Finally, weight calculation is performed according to the thickness dispersion and the width dispersion, wherein the weight coefficient is determined by historical data analysis and process knowledge, the thickness dispersion and the width dispersion are weighted and combined, and the rubber edge overflow dispersion is outputted, and the smaller the value, the more stable the overall rubber edge profile of the current rubber product, and the better the quality.

[0033] Further, step S300 further comprises step S360, traversing the spectrum-rubber edge impurity detection data to obtain the total number of impurity pixels, and outputting the ratio of the total number of impurity pixels to the total number of image pixels as the impurity ratio.

[0034] Preferably, the spectral-rubber edge impurity detection data is subjected to quality detection feature extraction, wherein each pixel of the spectral-rubber edge impurity detection data contains a category label, the background / rubber matrix pixel is labeled as 0 or represented by black, and different types of impurity pixels are labeled as 1, 2, 3, and the like, respectively. Then, the corresponding category labels are obtained by traversing the spectral-rubber edge impurity detection data, the pixel of the impurity category is added, the total number of impurity pixels is obtained, the ratio of the total number of impurity pixels to the total number of image pixels is calculated, and the impurity ratio is output, which directly represents the area percentage of the impurity region in the entire detection region.

[0035] Further, step S300 further includes step S370, wherein the optical-rubber edge surface detection data includes a surface texture image, the surface texture image is subjected to gray scale processing to obtain a surface texture gray scale image; step S380, a pre-trained rubber edge surface defect recognition model is trained by known surface defects, the rubber edge surface defect recognition model is used to binarize and segment the surface texture gray scale image, and a segmentation mask image is output; step S390, the total number of defect pixels is identified according to the segmentation mask image, and the surface defect density is output based on the percentage value of the total number of defect pixels to the total number of pixels of the segmentation mask image.

[0036] Preferably, the optical-rubber edge surface detection data is subjected to quality detection feature extraction, wherein the optical-rubber edge surface detection data includes a surface texture image, which is used to clearly display the microscopic texture of the rubber edge surface. The surface texture image is subjected to gray scale processing to convert the original color image into a gray scale image, that is, each pixel is converted from a value representing three channels of red, green, and blue into a single gray scale intensity value, a surface texture gray scale image is obtained to reduce the data dimension and the calculation complexity, and eliminate the interference of color information. Then, a pre-trained rubber edge surface defect recognition model is trained by known surface defects. Specifically, the known surface defect labeling means that a large number of pixel regions belonging to defects such as scratches, tears, and pits are labeled in a large number of surface texture images, the labeled surface texture image is established as training sample data, a recognition model is constructed based on a convolutional neural network, the training sample data is used to pre-train the recognition model, the complex features of distinguishing normal surface texture and various defect textures are learned, and then the rubber edge surface defect recognition model is obtained. Then, the surface texture gray scale image is input into the rubber edge surface defect recognition model for binarization and segmentation to obtain a binarization image and further determine a segmentation mask image, wherein the pixel region recognized as a defect is labeled as 1 or white, and the pixel region recognized as a normal surface / background is labeled as 0 or black. Then, the binarized segmentation mask image is traversed to count the number of all pixels labeled as 1 to obtain the total number of defect pixels, and finally the percentage value of the total number of defect pixels to the total number of pixels of the segmentation mask image is calculated to determine the surface defect density, which directly represents the percentage of the surface area occupied by the defect region to the total detection area.

[0037] Step S400, input the real-time flash detection features into a flash-product quality evaluation conversion model, output product quality evaluation indicators based on the flash-product quality evaluation conversion model, and send a warning feedback signal to the rubber product production line according to the product quality evaluation indicators.

[0038] Preferably, the real-time flash detection features including flash overflow dispersion, impurity ratio, and surface defect density are input into a flash-product quality evaluation conversion model, wherein the flash overflow dispersion is strongly correlated with the mechanical property indicator-tensile strength of the rubber product, the impurity ratio is strongly correlated with the durability indicator-fatigue life of the rubber product, and the surface defect density is strongly correlated with the aging resistance indicator-performance retention rate after aging of the rubber product; the flash-product quality evaluation conversion model is an intelligent model based on a correlation coefficient network and trained by historical data, which is used to output quantitative and predictive product quality evaluation indicators, such as predicted tensile strength, predicted fatigue cycle number, and predicted performance retention rate after aging; then the output product quality evaluation indicators are compared with a preset qualified standard range, and different levels of warning feedback signals are sent to the rubber product production line according to the comparison results, such as an adjustment signal is sent when the product quality evaluation indicators are near the critical value, and an audible and visual alarm is sent to the operator interface when the product quality evaluation indicators are significantly over the standard, and a pause signal is sent to the production line control system to request equipment maintenance or process adjustment if multiple consecutive products are unqualified, so as to realize accurate quality detection of the rubber product flash, minimize the waste rate, save costs, and realize intelligent closed-loop management of rubber product production.

[0039] Further, step S400 further includes step S401 of collecting historical flash detection feature samples and corresponding rubber product quality detection results, wherein the rubber product quality detection results include mechanical property indicators, durability indicators, and aging resistance test indicators; step S402 of performing correlation analysis on the historical flash detection feature samples and the corresponding rubber product quality detection results by using a correlation analysis algorithm to obtain a correlation coefficient network; and step S403 of defining a conversion relationship of the flash-product quality evaluation conversion model based on the correlation coefficient network to obtain the flash-product quality evaluation conversion model.

[0040] Preferably, a large number of historical edge detection feature samples are collected from historical production data, including edge overflow dispersion, impurity ratio, surface defect density of each rubber product edge removal sample, while recording the corresponding final rubber product quality detection results of each sample determined by laboratory standard detection, including mechanical property indexes such as tensile strength, elongation at break, hardness, durability indexes such as fatigue life, wear resistance, and aging performance retention rate and other aging resistance test indexes; using correlation analysis algorithms such as Pearson correlation coefficient and Spearman rank correlation coefficient, the correlation between the historical edge detection feature samples and the corresponding rubber product quality detection results is analyzed, including calculating the correlation strength and direction between each historical edge detection feature sample and the rubber product quality detection results, for example, edge overflow dispersion and tensile strength are strongly negatively correlated, impurity ratio and fatigue life are strongly negatively correlated, surface defect density and aging resistance performance are moderately negatively correlated, etc., and then constructing a network diagram according to the correlation analysis results to obtain a correlation coefficient network, clearly showing the correlation coefficient values between all edge detection features and all rubber product quality detection indexes, indicating the influence of each edge detection feature on the final product; then define the conversion relationship of the edge-product quality evaluation conversion model based on the correlation coefficient network, that is, select the edge detection features strongly correlated with the quality detection indexes as input, eliminate weakly correlated features to simplify the conversion model and improve prediction accuracy, and finally obtain the edge-product quality evaluation conversion model for converting real-time edge detection features to determine the predicted value of the final product quality index, thereby realizing forward-looking quality control warning.

[0041] Further, step S400 further comprises step S404 of obtaining product quality evaluation indexes output by the edge-product quality evaluation conversion model and corresponding real product quality detection indexes; step S405, based on the product quality evaluation indexes and the real product quality detection indexes, obtains a prediction error, and updates the edge-product quality evaluation conversion model according to the prediction error to output an updated edge-product quality evaluation conversion model.

[0042] Preferably, the product quality evaluation index output by the flash-product quality evaluation conversion model is obtained, and then a destructive test is performed on the product by using a standard detection method in a laboratory to obtain a real and accurate product quality real detection index. The product quality evaluation index and the product quality real detection index are compared to calculate a deviation value as a prediction error, i.e., a prediction error is calculated for each quality detection index, which is used to quantify the error degree of the current prediction of the flash-product quality evaluation conversion model. Then, the flash-product quality evaluation conversion model is updated by incremental learning according to the prediction error. Specifically, the real-time flash detection features of the product and the product quality real detection index are used as new training samples to train the flash-product quality evaluation conversion model, the internal parameters of which are fine-tuned. Then, the random gradient descent is used to make a small adjustment to the parameters such as the weight and bias of the flash-product quality evaluation conversion model according to the calculated prediction error, so that the prediction value of the flash-product quality evaluation conversion model is closer to the real value. Finally, the updated flash-product quality evaluation conversion model is output, so as to ensure the prediction accuracy of the model and improve the adaptability to the current production state.

[0043] Further, step S400 further includes step S410, in which the flash-product quality evaluation conversion model performs correlation conversion on the real-time flash detection features according to the correlation coefficient network to output a mechanical property conversion index, a durability conversion index, and an aging resistance conversion index; and step S420, in which a product quality evaluation index is output according to the mechanical property conversion index, the durability conversion index, and the aging resistance conversion index.

[0044] Preferably, the real-time flash detection features are input into the flash-product quality evaluation conversion model, and correlation conversion is performed based on the correlation coefficient network, and the feature relationship with the strongest correlation is preferentially used for calculation. Specifically, the predicted tensile strength value is calculated and determined based on the flash overflow dispersion and the mapping relationship between the flash overflow dispersion and the tensile strength value, and the mechanical property conversion index is output. The predicted fatigue cycle number is calculated and determined based on the impurity ratio and the mapping relationship between the impurity ratio and the fatigue life, and the durability conversion index is output. The predicted performance retention rate is calculated and determined based on the surface defect density and the mapping relationship between the surface defect density and the performance retention rate after aging, and the aging resistance conversion index is output. Finally, the mechanical property conversion index, the durability conversion index, and the aging resistance conversion index are compared with respective qualified / unqualified threshold values, and a product quality evaluation index is output according to a preset decision rule. For example, if all the three conversion indexes are within the excellent range, the product is determined to be of excellent quality; if all the three conversion indexes are within the qualified range, the product is determined to be qualified; if any one of the conversion indexes is lower than the qualified threshold value, the product is determined to be unqualified; and if multiple conversion indexes are seriously out of the standard, the product is determined to have serious defects, so as to ensure the scientificity and accuracy of the quality index prediction.

[0045] In the foregoing, reference is made to Figure 1The method for online detection of rubber product quality based on intelligent sensors according to the embodiment of the present application is described in detail. Next, the method for online detection of rubber product quality based on intelligent sensors according to the embodiment of the present application will be described in detail with reference to the accompanying drawings Figure 2 The system for online detection of rubber product quality based on intelligent sensors according to the embodiment of the present application is described.

[0046] The system for online detection of rubber product quality based on intelligent sensors according to the embodiment of the present application is used to solve the technical problem of lack of real-time and accurate monitoring of the quality detection of the rubber product edge removal process in the prior art, which leads to uncontrollable edge removal quality and high rate of defective products. The system achieves online and real-time detection of the quality of the rubber product edge, thereby improving the quality of the rubber product and reducing the rate of defective products. As shown in Figure 2 The system for online detection of rubber product quality based on intelligent sensors includes an intelligent sensing unit integrated module 10, a detection data acquisition module 20, a detection feature extraction module 30, and a quality evaluation index output module 40.

[0047] The intelligent sensing unit integrated module 10 is used to connect the rubber product production line edge removal operation line, and integrate intelligent sensing units on the edge removal operation line. The intelligent sensing units include a 3D laser scanner, a multispectral camera, and a surface optical detector. The detection data acquisition module 20 is used to detect the edge removal sample of the rubber product to be detected by using the intelligent sensing units, and to acquire laser-edge profile detection data, spectral-edge impurity detection data, and optical-edge surface detection data. The detection feature extraction module 30 is used to extract quality detection features from the laser-edge profile detection data, spectral-edge impurity detection data, and optical-edge surface detection data, and to output real-time edge detection features, including edge overflow dispersion, impurity ratio, and surface defect density. The quality evaluation index output module 40 is used to input the real-time edge detection features into an edge-product quality evaluation conversion model, to output product quality evaluation indexes based on the edge-product quality evaluation conversion model, and to send a warning feedback signal to the rubber product production line according to the product quality evaluation indexes.

[0048] Next, the specific configuration of the detection data acquisition module 20 will be described in detail. The detection data acquisition module 20 further includes: when the edge removal sample of the rubber product to be detected enters the detection area, starting the 3D laser scanner to scan the edge removal sample point by point to obtain edge surface height distribution data; and converting the edge surface height distribution data into an edge profile curve to acquire laser-edge profile detection data.

[0049] Next, the specific configuration of the detection data acquisition module 20 will be described in detail. The detection data acquisition module 20 further comprises: starting the multi-spectral camera, performing multi-spectral imaging on the rubber product edge removal sample to be detected, and acquiring multi-spectral imaging data, wherein the multi-spectral camera comprises an impurity spectrum library; identifying the impurity type and impurity proportion in the multi-spectral imaging data based on the impurity spectrum library, and acquiring spectrum-rubber edge impurity detection data.

[0050] Next, the specific configuration of the detection feature extraction module 30 will be described in detail. The detection feature extraction module 30 further comprises: collecting good product edge contour detection data; calculating the average thickness and average width of the edge contour based on the good product edge contour detection data, and generating a good product contour curve based on the average thickness and average width; analyzing the good product edge contour detection data based on the fluctuation range threshold of the good product contour curve; performing quality detection feature extraction on the laser-rubber edge contour detection data based on the good product contour curve and the fluctuation range threshold, and outputting thickness dispersion and width dispersion; and performing weight calculation according to the thickness dispersion and width dispersion, and outputting edge overflow dispersion.

[0051] Next, the specific configuration of the detection feature extraction module 30 will be described in detail. The detection feature extraction module 30 further comprises: traversing the spectrum-rubber edge impurity detection data to obtain the total number of impurity pixels, and outputting the ratio of the total number of impurity pixels to the total number of image pixels as an impurity ratio.

[0052] Next, the specific configuration of the detection feature extraction module 30 will be described in detail. The detection feature extraction module 30 further comprises: wherein the optical-rubber edge surface detection data comprises a surface texture image, the surface texture image is subjected to grayscale processing to obtain a surface texture grayscale image; a pre-trained rubber edge surface defect recognition model is trained by known surface defect labels, the rubber edge surface defect recognition model is used to perform binaryzation segmentation on the surface texture grayscale image, and a segmentation mask image is output; the total number of defect pixels is identified according to the segmentation mask image, and the surface defect density is output based on the percentage value of the total number of defect pixels to the total number of pixels of the segmentation mask image.

[0053] Below, the specific configuration of the quality evaluation index output module 40 will be described in detail. The quality evaluation index output module 40 further comprises: collecting historical rubber edge detection feature samples and corresponding rubber product quality detection results, wherein the rubber product quality detection results include mechanical property indexes, durability indexes, and aging resistance test indexes; performing correlation analysis on the historical rubber edge detection feature samples and the corresponding rubber product quality detection results by using a correlation analysis algorithm to obtain a correlation coefficient network; defining a conversion relationship of a rubber edge-product quality evaluation conversion model based on the correlation coefficient network to obtain the rubber edge-product quality evaluation conversion model.

[0054] Below, the specific configuration of the quality evaluation index output module 40 will be described in detail. The quality evaluation index output module 40 further comprises: the rubber edge-product quality evaluation conversion model performs correlation conversion on the real-time rubber edge detection features according to the correlation coefficient network to output mechanical property conversion indexes, durability conversion indexes, and aging resistance conversion indexes; and outputs product quality evaluation indexes according to the mechanical property conversion indexes, the durability conversion indexes, and the aging resistance conversion indexes.

[0055] Below, the specific configuration of the quality evaluation index output module 40 will be described in detail. The quality evaluation index output module 40 further comprises: obtaining product quality evaluation indexes output by the rubber edge-product quality evaluation conversion model and corresponding real product quality detection indexes; obtaining a prediction error based on the product quality evaluation indexes and the real product quality detection indexes; performing incremental learning update on the rubber edge-product quality evaluation conversion model according to the prediction error to output an updated rubber edge-product quality evaluation conversion model.

[0056] The rubber product quality online detection system based on intelligent sensors provided in the embodiments of the present application can execute the rubber product quality online detection method based on intelligent sensors provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with reference to the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make slight changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the scope of the technical solution of the present application. Any simple modification, equivalent change, and modification of the above embodiments based on the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. A method for online detection of quality of rubber products based on intelligent sensors, characterized in that, The method comprises: connecting the rubber product production line to the edge removal operation line, integrating an intelligent sensing unit on the edge removal operation line, the intelligent sensing unit comprising a 3D laser scanner, a multispectral camera, and a surface optical detector; detecting the edge removal sample of the rubber product to be detected using the intelligent sensing unit, obtaining laser-edge profile detection data, spectral-edge impurity detection data, and optical-edge surface detection data; extracting quality detection features from the laser-edge profile detection data, spectral-edge impurity detection data, and optical-edge surface detection data, and outputting real-time edge detection features, including edge overflow dispersion, impurity ratio, and surface defect density; inputting the real-time edge detection features into an edge-product quality evaluation conversion model, outputting product quality evaluation indicators based on the edge-product quality evaluation conversion model, and sending warning feedback signals to the rubber product production line according to the product quality evaluation indicators.

2. The intelligent sensor based online detection method of rubber product quality as claimed in claim 1 wherein, The method for detecting the edge removal sample of the rubber product to be detected using the intelligent sensing unit to obtain laser-edge profile detection data comprises: when the edge removal sample of the rubber product to be detected enters the detection area, starting the 3D laser scanner to scan the edge removal sample point by point to obtain edge surface height distribution data; converting the edge surface height distribution data to obtain laser-edge profile detection data.

3. The intelligent sensor based online detection method of rubber product quality as claimed in claim 1 wherein, The method for detecting the edge removal sample of the rubber product to be detected using the intelligent sensing unit to obtain spectral-edge impurity detection data comprises: starting the multispectral camera to perform multispectral imaging on the edge removal sample of the rubber product to be detected, and obtaining multispectral imaging data, wherein the multispectral camera comprises an impurity spectrum library; identifying the impurity type and impurity proportion in the multispectral imaging data based on the impurity spectrum library to obtain spectral-edge impurity detection data.

4. The intelligent sensor based online detection method of rubber product quality as claimed in claim 1 wherein, The method for extracting quality detection features from the laser-edge profile detection data comprises: collecting good product edge profile detection data; calculating the good product edge profile detection data to obtain the average thickness and average width of the edge profile, and generating a good profile curve based on the average thickness and average width; analyzing the fluctuation range threshold of the good product edge profile detection data based on the good profile curve; extracting quality detection features from the laser-edge profile detection data based on the good profile curve and the fluctuation range threshold, and outputting thickness dispersion and width dispersion; performing weight calculation according to the thickness dispersion and width dispersion, and outputting edge overflow dispersion.

5. The intelligent sensor based online rubber product quality detection method as claimed in claim 1 wherein, The method for extracting quality detection features from the spectral-edge impurity detection data comprises: traversing the spectral-edge impurity detection data to obtain the total number of impurity pixels, and outputting the ratio of the total number of impurity pixels to the total number of image pixels as the impurity ratio.

6. The intelligent sensor based online rubber product quality detection method as claimed in claim 1 wherein, The method for extracting quality detection features from the optical-edge surface detection data to obtain surface defect density comprises: ​ The optical-rubber edge surface detection data includes a surface texture image, the surface texture image is subjected to gray scale processing, and a surface texture gray scale image is obtained; The pre-trained rubber edge surface defect recognition model is labeled by known surface defects, the rubber edge surface defect recognition model is used for binaryzation segmentation of the surface texture gray scale image, and a segmentation mask image is output; According to the segmentation mask image, the total number of defect pixels is identified, and based on the percentage value of the total number of defect pixels and the total number of pixels of the segmentation mask image, the surface defect density is output.

7. The intelligent sensor based online rubber product quality detection method as claimed in claim 1 wherein, The method for constructing the rubber edge-product quality evaluation conversion model includes: Collecting historical rubber edge detection feature samples and corresponding rubber product quality detection results, the rubber product quality detection results including mechanical performance indicators, durability indicators, and aging resistance test indicators; Correlation analysis is performed on the historical rubber edge detection feature samples and corresponding rubber product quality detection results using a correlation analysis algorithm, and a correlation coefficient network is obtained; Based on the correlation coefficient network, a conversion relationship for training the rubber edge-product quality evaluation conversion model is defined, and the rubber edge-product quality evaluation conversion model is obtained.

8. The smart sensor based online detection method of rubber article quality as claimed in claim 7, wherein, The method for inputting the real-time rubber edge detection feature into the rubber edge-product quality evaluation conversion model and outputting product quality evaluation indicators based on the rubber edge-product quality evaluation conversion model includes: The rubber edge-product quality evaluation conversion model performs correlation conversion on the real-time rubber edge detection feature based on the correlation coefficient network, and outputs mechanical performance conversion indicators, durability conversion indicators, and aging resistance conversion indicators; According to the mechanical performance conversion indicators, durability conversion indicators, and aging resistance conversion indicators, product quality evaluation indicators are output.

9. The smart sensor based online detection method of rubber article quality as claimed in claim 7, wherein, After obtaining the rubber edge-product quality evaluation conversion model, the method further includes: Obtaining product quality evaluation indicators output by the rubber edge-product quality evaluation conversion model and corresponding product quality real detection indicators; Based on the product quality evaluation indicators and the product quality real detection indicators, a prediction error is obtained, the rubber edge-product quality evaluation conversion model is updated by incremental learning according to the prediction error, and an updated rubber edge-product quality evaluation conversion model is output.

10. A system for online quality detection of rubber products based on intelligent sensors, characterized in that, The system is used to implement the online detection method of the rubber product quality based on the intelligent sensor according to any one of claims 1 to 9, and the system includes: An intelligent sensing unit integration module is used to connect a rubber edge removal operation line of a rubber product production line, and an intelligent sensing unit is integrated on the rubber edge removal operation line, the intelligent sensing unit including a 3D laser scanner, a multispectral camera, and a surface optical detector; A detection data acquisition module is used to detect rubber edge removal samples of a rubber product to be detected by using the intelligent sensing unit, and to acquire laser-rubber edge contour detection data, spectral-rubber edge impurity detection data, and optical-rubber edge surface detection data; The detection feature extraction module is configured to perform quality detection feature extraction on the laser-rubber edge profile detection data, the spectrum-rubber edge impurity detection data, and the optical-rubber edge surface detection data, and output real-time rubber edge detection features, including rubber edge overflow dispersion, impurity ratio, and surface defect density. The quality evaluation index output module is configured to input the real-time rubber edge detection features into a rubber edge-product quality evaluation conversion model, output product quality evaluation indexes based on the rubber edge-product quality evaluation conversion model, and send a warning feedback signal to the rubber product production line according to the product quality evaluation indexes.

Images