New energy battery coating thickness control method

Abstract

The invention provides a new energy battery coating thickness control method, which comprises the following steps: presetting a chromatic value range of a coating color, segmenting a stable image into a coating area and a non-coating area, carrying out binarization processing on the segmented coating area to obtain a coating binarization image, according to the segmented coating binarization image, carrying out closed operation on a coating area, filling small holes in the coating area, and removing noisy points outside the coating area through open operation to obtain a coating area image after morphological processing; a motion model of color change is established according to historical focusing data, the color change trend of a next frame of image is predicted, a predicted value is used as a feedforward control signal of a focusing system, and oscillation of the focusing system is suppressed.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for controlling the thickness of a coating on a new energy battery. Background Art

[0002] In the production process of new energy battery coatings, the changes in the frequency and amplitude of coating color changes will cause the quality of the collected images to be unstable, and the contrast and clarity will change with the change of color. This rapidly changing image feature seriously affects the subsequent image processing and analysis. At the same time, the automatic focusing system based on these unstable images is difficult to respond accurately, and the focusing accuracy and speed cannot keep up with the rhythm of color changes, causing the acquired coating surface images to be frequently out of focus and blurred. This not only affects the real-time monitoring of coating quality, but may also cause the coating machine control system to receive erroneous feedback signals, thereby affecting the precise control of coating thickness. In addition, the local overexposure and underexposure problems of the image further aggravate the difficulty of image analysis, which greatly reduces the accuracy of coating edge detection and area calculation. This series of interrelated technical problems seriously restricts the automation level and product quality control capabilities of the new energy battery coating production process. Summary of the invention

[0003] The present invention provides a new energy battery coating thickness control method, which mainly includes:

[0004] Real-time acquisition of new energy battery coating images, obtaining a high-speed continuous image sequence with an image frame frequency greater than the target frequency, extracting unstable images from the continuous image sequence, and using an adaptive histogram equalization algorithm to enhance the contrast of unstable images to obtain a stable image with a contrast variation range reduced to a preset variation range;

[0005] A chromaticity value range of the coating color is preset, and the stable image is segmented into a coating area and a non-coating area. The chromaticity value of the coating area is within the preset chromaticity value range of the coating color, and the segmented coating area is binarized to obtain a coating binary image;

[0006] According to the segmented coating binary image, a closing operation is performed on the coating area to fill the small holes inside the coating area, and an opening operation is performed to remove the noise outside the coating area to obtain a coating area image after morphological processing;

[0007] Identify the local over-exposed and under-exposed areas of the coating area image, suppress the brightness of the local over-exposed area and enhance the brightness of the local under-exposed area according to the brightness distribution of the coating area, obtain a coating image with uniform brightness, transmit the coating image with uniform brightness to a preset focusing system, extract the color feature vector of the coating image, use the color feature vector of the coating image as a feedback control signal of the focusing system, and adjust the focal length of the liquid lens in real time;

[0008] A motion model of color change is established based on historical focusing data to predict the color change trend of the next frame of image. The predicted value is used as a feedforward control signal of the focusing system to suppress the oscillation of the focusing system.

[0009] The coating surface image obtained by the preset focusing system is evaluated for quality, and the gradient variance value of the image is calculated. The gradient variance value is used as a quantitative index of image clarity. If the gradient variance value is greater than the gradient variance threshold, it is judged as a clear image. If the gradient variance value is less than the gradient variance threshold, it is judged as a blurred image and the blurred image is removed.

[0010] Based on the coating surface image with the blurred image removed, the sub-pixel level contour of the coating edge is extracted, and the pixel area of ​​the coating is obtained by calculating the contour area. Combined with the focal length parameters of the preset focusing system, the thickness value of the coating is calculated, and the thickness value is used as the feedback control quantity of the coater to achieve online control of the coating thickness.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a real-time monitoring and control method for the thickness of a new energy battery coating. The method collects coating images at high speed, performs adaptive histogram equalization on unstable images, and segments coating areas based on a preset chromaticity range. Subsequently, morphological processing is performed on the coating area to identify and correct local exposure unevenness, thereby achieving brightness uniformity of the coating image. The present invention uses color feature vectors as feedback signals, combined with feedforward signals predicted by historical data, to adjust the focal length of the liquid lens in real time and suppress focusing oscillations. Image quality is evaluated by calculating gradient variance, sub-pixel-level contours of clear images are extracted, and coating thickness is calculated in combination with focal length parameters. Finally, the present invention uses the thickness value as the feedback control quantity of the coating machine to achieve online precise control of coating thickness, effectively improving the quality and efficiency of new energy battery coating production. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flow chart of a new energy battery coating thickness control method.

[0014] Figure 2 It is a schematic diagram of a new energy battery coating thickness control method of the present invention.

[0015] Figure 3 It is another schematic diagram of a new energy battery coating thickness control method of the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0017] like Figure 1 -3. A new energy battery coating thickness control method in this embodiment may specifically include:

[0018] S101. Real-time acquisition of new energy battery coating images, obtaining a high-speed continuous image sequence with an image frame frequency greater than a target frequency, extracting unstable images from the continuous image sequence, and using an adaptive histogram equalization algorithm to perform contrast enhancement on the unstable images to obtain a stable image with a contrast variation range reduced to a preset variation range.

[0019] Grayscale matrix data is obtained according to the coordinates of the center point of the coating image acquisition area, and the grayscale matrix data is processed by a Gaussian filter. When the mean square error of the grayscale matrix data is less than the mean square error threshold, a filtered image is obtained; a histogram statistical operation is performed on the filtered image, and the histogram data is decomposed by wavelet transform, and the decomposed data is clustered by an adaptive threshold segmentation algorithm. If the inter-class variance obtained by the clustering operation is greater than the preset inter-class variance threshold, it is determined to be the unstable image; an adaptive histogram equalization parameter matrix is ​​constructed for the unstable image, and a block enhancement method is used to adjust the contrast of the unstable image, and a bilinear interpolation algorithm is used to smooth the block boundaries to obtain an equalized image; a contrast range statistics is performed on the equalized image, and a region growing algorithm is used to adjust the pixel points that exceed the preset contrast range, and a median filter is used to smooth the adjusted image to obtain a stable image.

[0020] Exemplarily, the target image area is collected according to the coordinates of the center point of the coating image collection area and the preset sampling window area, the image grayscale matrix data is obtained by a photoelectric sensor, the grayscale matrix data is filtered by a Gaussian filter, and the filtered image is obtained when the mean square error of the grayscale matrix data is less than the preset threshold. The grayscale values ​​in the filtered image are statistically analyzed by histogram, the histogram data is decomposed in multiple scales by wavelet transform, the decomposed data is clustered by an adaptive threshold segmentation algorithm, and the image is determined to be unstable when the inter-class variance obtained by the clustering operation is greater than the preset inter-class variance threshold. The adaptive histogram equalization parameter matrix is ​​constructed according to the grayscale distribution mean and variance of the unstable image, the local contrast of the unstable image is adjusted by a block enhancement method, and the block boundary is smoothed by a bilinear interpolation algorithm to obtain the equalized image. The contrast variation range of the equalized image is statistically calculated, the pixel points beyond the preset contrast range are adjusted by a region growing algorithm, and the adjusted image is smoothed by a median filter to obtain a stable image with stable contrast. Coating image acquisition is usually carried out in real time on the production line. The sampling window area is determined according to the coating width. The window area is set to 200×200 pixels. The sampling position selects the center area of ​​the coating. High-speed acquisition is performed through a sampling frequency of 2000 Hz. The quantization accuracy of the photoelectric sensor is 12 bits, and the grayscale value range is 0 to 4095. Considering the vibration of the production line and the interference of ambient light, a 5×5 Gaussian filter kernel is used to reduce the noise of the image. When the mean square error of the image grayscale matrix is ​​less than 20, the noise reduction effect is judged to be up to standard. The image grayscale histogram reflects the distribution of different grayscale levels in the image. The filtered image is statistically analyzed at 256 grayscale levels. The three-layer wavelet decomposition is used to perform multi-scale analysis on the histogram data. The decomposed data is divided into 4 categories through an adaptive threshold algorithm. If the inter-class variance is greater than 150, it is judged as an unstable image. This judgment method can effectively identify image quality abnormalities caused by uneven coating surface. Adaptive histogram equalization is performed on unstable images. The image is divided into 8×8 sub-blocks, each of which is 25×25 pixels in size. The contrast enhancement parameters are calculated based on the mean and variance of the grayscale distribution of the sub-blocks. The adjacent sub-blocks are smoothly transitioned using bilinear interpolation to avoid block effects. During the equalization process, the slope magnification of each sub-block is limited to no more than 3 times to prevent excessive noise amplification. The contrast variation range is obtained by calculating the grayscale difference between adjacent pixels. The preset contrast range is set to 30 to 120. The pixels outside the range are adjusted using the 8-neighborhood region growth method, and the growth threshold is set to 15. The noise introduced in the adjustment process is eliminated through a 5×5 median filter, and finally an image with stable contrast is obtained. In practical applications, this method has significant effects on the detection of surface defects of metal coatings such as aluminum foil and copper foil, and can accurately identify defects such as scratches, bubbles, and wrinkles on the coating surface.During the coating production process, local reflections will occur on the coating surface, resulting in excessive contrast in the collected image. By limiting the contrast range, the defect characteristics can be highlighted. When applied to different coating materials, the parameters need to be adjusted according to the reflective characteristics of the coating. For example, the preset contrast range of the glass base coating should be set to 20 to 80, while the preset contrast range of the plastic base coating should be set to 40 to 150. For ultra-thin coatings with a coating thickness of less than 10 microns, the sampling frequency needs to be increased to 4000 Hz to ensure image acquisition clarity.

[0021] S102, a chromaticity value range of a preset coating color is set, and the stable image is segmented into a coating area and a non-coating area. The chromaticity value of the coating area is within the preset chromaticity value range of the coating color, and the segmented coating area is binarized to obtain a coating binary image.

[0022] The color space transformation matrix is ​​used to perform color space conversion on the original image, and the image chromaticity value distribution histogram is obtained by color component calculation; according to the pixel points in the chromaticity value distribution histogram that are located in the preset coating color chromaticity value range, the maximum inter-class variance algorithm is used to calculate the coating area segmentation threshold to obtain the initial coating area image; for the initial coating area image, the morphological opening operation is used to correct the boundary of the marked area, and the corrected area is expanded by the neighborhood growing algorithm to obtain the complete coating area image; the grayscale value statistics are performed on the coating area in the complete coating area image, if the grayscale value of the pixel point is greater than the optimal binarization threshold, it is assigned a value of 255, if the grayscale value of the pixel point is less than the optimal binarization threshold, it is assigned a value of 0 to obtain the coating binary image.

[0023] Exemplarily, according to the red, green and blue three-channel data in the original image, the color space transformation matrix is ​​used to convert the image into color space, and the chromaticity value of each pixel of the image is obtained by calculating the color component, and it is determined whether the chromaticity value is within the chromaticity value range of the preset coating color to obtain the chromaticity distribution histogram. Statistics are taken for the pixel points in the chromaticity distribution histogram that are within the chromaticity value range of the preset coating color, and the coating area segmentation threshold is calculated using the maximum inter-class variance algorithm. The area with a chromaticity value greater than the coating area segmentation threshold is marked by the regional marking algorithm to obtain the initial coating area image. According to the regional connectivity characteristics in the initial coating area image, the morphological opening operation with the structural element area as the preset coating area is used to correct the boundary of the marked area, and the neighborhood growing algorithm is used to expand the corrected area to obtain the complete coating area image. The grayscale value statistics of the coating area in the complete coating area image are performed, and the optimal binarization threshold is calculated using the Otsu threshold segmentation algorithm. The pixel points with grayscale values ​​greater than the optimal binarization threshold are assigned 255, and the pixel points less than the optimal binarization threshold are assigned 0, so as to obtain the coating binary image. The coating color is usually represented in the red, green and blue color space, and the preset coating color chromaticity value range sets upper and lower limits on the three components of hue, saturation and brightness, such as the hue range of the blue coating is 200 to 240 degrees, the saturation range is 60% to 90%, and the brightness range is 40% to 75%. The color space transformation uses a standard conversion matrix to convert the red, green and blue three-channel values ​​into chromaticity values, and the chromaticity value of each pixel is obtained by a weighted combination of the three components. The chromaticity distribution histogram reflects the distribution of different chromaticity values ​​in the image, with the horizontal axis being the chromaticity value and the vertical axis being the corresponding number of pixels. In the coating image, the chromaticity values ​​of the coating area are concentrated in the preset range, while the chromaticity values ​​of the non-coating area are more dispersed. The maximum inter-class variance algorithm iteratively calculates the inter-class variance under different thresholds. When the inter-class variance reaches the maximum value, it is the optimal segmentation threshold. This threshold is usually located at the junction of the chromaticity distribution of the coating area and the background area. In practical applications, the coating edge may be blurred or broken. The selection of the structural element directly affects the effect of the morphological opening operation. For a standard coating area with an area of ​​2500 square pixels, a square structural element of 5×5 pixels is selected for opening operation, which can effectively remove edge burrs and small area noise. The neighborhood growth algorithm uses the marked coating area as the seed point. When the difference between the chromaticity value of the neighborhood pixel and the seed point is less than the set threshold, the pixel is classified into the coating area. The grayscale value distribution of the coating area presents a bimodal characteristic, with one peak corresponding to the coating area and the other peak corresponding to the background area. The Otsu threshold segmentation algorithm determines the optimal binarization threshold by maximizing the variance between the two peaks. In coating image processing, the grayscale value of the coating area is generally higher than that of the background area, and the optimal binarization threshold is usually located at the bottom of the valley between the two peaks.When the grayscale value of the background area is between 50 and 100, and the grayscale value of the coating area is between 150 and 200, the optimal binarization threshold is generally between 120 and 140. In coating quality inspection, different coating materials correspond to different chromaticity characteristics. Due to the reflective surface of the metal coating, the chromaticity value range is wider, and the saturation range needs to be set to 40% to 95%. The ceramic coating has uniform color and concentrated chromaticity value distribution, and the saturation range can be set to 70% to 90%. For gradient color coatings, a 30-degree transition interval needs to be reserved within the hue range to ensure that the gradient part is not segmented incorrectly. Coating defects with an area of ​​less than 100 square pixels are easily filtered out by the opening operation. At this time, a small structural element of 3×3 pixels should be used.

[0024] S103, according to the segmented coating binary image, a closing operation is performed on the coating area to fill the small holes inside the coating area, and the noise points outside the coating area are removed by an opening operation to obtain a coating area image after morphological processing.

[0025] With respect to the connected domain features in the coating binary image, an area screening method is used to mark the connected domain features, the total number of pixels in each connected domain is calculated, and a four-neighborhood connectivity analysis method is used to identify non-coating pixels. The ratio of the area to the perimeter of the connected domain formed by the non-coating pixels is calculated to obtain the coating image to be processed; with respect to the holes to be filled in the coating image to be processed, a morphological closing operation is used for processing, and a closed operation coating image is obtained by setting a structural element size larger than an inscribed circle diameter corresponding to a preset hole feature ratio; based on the closed operation coating image, a morphological opening operation is used to smooth the edge of the coating area, and a morphologically processed coating area image is obtained by setting a structural element size smaller than a circumscribed circle diameter corresponding to a preset minimum coating area.

[0026] Exemplarily, according to the connected domain characteristics of the coating area in the coating binary image, the connected domain is marked by an area screening method, the total number of pixels of each connected domain is calculated by regional statistics, and the connected domains with an area smaller than the preset minimum coating area are marked and deleted to obtain the initial coating area binary image. According to the pore characteristics in the initial coating area binary image, the four-neighborhood connectivity analysis method is used to identify the non-coating pixels inside the region, and the area and perimeter ratio of the connected domain formed by the non-coating pixels are calculated. The area with a contrast value smaller than the preset hole feature ratio is marked as a hole to be filled, and the coating image to be processed is obtained. According to the coating image to be processed, the coating area is filled by a morphological closing operation, and the holes to be filled are filled by setting the size of the structural element larger than the diameter of the inscribed circle corresponding to the preset hole feature ratio, and the closed operation coating image is obtained. For the closed operation coating image, the coating area is processed by morphological opening operation. By setting the size of the structural element smaller than the diameter of the circumscribed circle corresponding to the preset minimum coating area, the edge of the coating area is smoothed to obtain the coating area image after the morphological processing. There are usually two types of typical defects in the coating image after binarization, internal holes and edge noise. Connected domain analysis of the coating area is the basis for processing these defects. By statistically analyzing the area characteristics of the connected domain and setting the minimum coating area threshold to 400 square pixels, the small area of ​​false detection caused by imaging noise can be effectively filtered out. In practical applications, the area of ​​the coating area is usually more than 10,000 square pixels, while the area of ​​the false detection area caused by noise is mostly less than 200 square pixels. The four-neighborhood connectivity analysis can accurately identify non-coated pixels inside the coating area, which are often caused by defects such as bubbles and scratches on the coating surface. The characteristic ratio of the hole is defined as the ratio of area to perimeter. When the ratio is less than 0.25, it indicates that the area is slender or irregular in shape and belongs to a defective area that needs to be filled. For circular holes, the ratio is close to 0.35, while for crack-like defects, the ratio is usually less than 0.15. The morphological closing operation is to achieve hole filling through a combination of dilation and erosion operations. The size of the structure element directly affects the filling effect. For a circular hole with an area of ​​100 square pixels, the diameter of its inscribed circle is about 11 pixels. Selecting a 13×13 pixel square structure element for closing operation can completely fill the hole of this size. When there are continuous small bubbles on the coating surface, the hole chain formed by these bubbles may require a larger size structure element. The opening operation is used to smooth the protrusions and burrs on the edge of the coating. The size of the structure element needs to be smaller than the characteristic size corresponding to the minimum coating area. For a circular area with an area of ​​400 square pixels, the diameter of its circumscribed circle is about 23 pixels. Selecting a 7×7 pixel square structure element for opening operation can retain the main shape of the coating while removing edge noise.In practical applications, metal coatings often have jagged edges due to surface reflection, requiring 2 to 3 iterative opening operations. Common types of defects in coating inspection also include coating breakage and coating overlap. Coating breakage can cause a complete coating area to be divided into multiple small areas. At this time, the preset minimum coating area needs to be set to 1 / 4 of the standard coating area to avoid misjudging the broken area as noise. Overlapping coating areas will form strips with higher grayscale values ​​in the image, which are prone to false edges after binarization. At this time, the structural element of the opening operation should be selected to be of a size equivalent to the coating width. For example, if the coating width is 50 pixels, a 9×9 pixel structural element should be selected.

[0027] S104, identifying local over-exposed and under-exposed areas of the coating area image, suppressing the brightness of the local over-exposed area and enhancing the brightness of the local under-exposed area according to the brightness distribution of the coating area, to obtain a coating image with uniform brightness, transmitting the coating image with uniform brightness to a preset focusing system, extracting a color feature vector of the coating image, using the color feature vector of the coating image as a feedback control signal of the focusing system, and adjusting the focal length of the liquid lens in real time.

[0028] According to the brightness histogram distribution of the coating area image, an adaptive segmentation algorithm is used to divide the brightness value into intervals, and the pixel points in the coating area that exceed the preset overexposure threshold and are lower than the preset underexposure threshold are calculated to obtain an abnormal brightness image; for the abnormal brightness image, a gamma transformation function is used to attenuate the overexposed pixels, a block enhancement algorithm is used to compensate for the underexposed pixels, and a bilinear interpolation algorithm is used to smooth the block boundaries to obtain the coating image with uniform brightness; according to the red, green and blue three-channel data of the coating image with uniform brightness, a color space conversion matrix is ​​used to calculate the chromaticity, and the chromaticity mean vector and the chromaticity gradient vector are obtained through a deep feature extraction network to obtain the color feature vector; for the color feature vector, a feedback controller is used to calculate the Euclidean distance with the preset standard feature vector, and a driving voltage is applied to the liquid lens through a proportional integral control algorithm to obtain a focal length adjustment signal.

[0029] Exemplarily, according to the brightness histogram distribution of the coating area image, an adaptive segmentation algorithm is used to divide the image brightness value into intervals, and the pixels whose brightness values ​​in the coating area exceed the preset overexposure threshold are marked as overexposed pixels, and the pixels whose brightness values ​​are lower than the preset underexposure threshold are marked as underexposed pixels, so as to obtain the brightness abnormality image. For the overexposed pixels in the brightness abnormality image, a gamma transformation function is used to attenuate the brightness value, a block enhancement algorithm is used to perform local brightness compensation on the underexposed pixels, and a bilinear interpolation algorithm is used to smooth the block boundaries, so as to obtain the brightness balanced image. According to the red, green and blue three-channel data of the brightness balanced image, a color space conversion matrix is ​​used to calculate the chromaticity of the image, and the chromaticity mean vector, chromaticity standard deviation vector and chromaticity gradient vector of the image are obtained through a deep feature extraction network to obtain the color feature vector. For the color feature vector, a feedback controller is used to calculate the Euclidean distance between the color feature vector and the preset standard feature vector of the focusing system, and a corresponding driving voltage is applied to the liquid lens through a proportional integral control algorithm to obtain the focal length adjustment signal. Overexposure and underexposure of coating images are usually caused by uneven lighting or improper camera exposure parameter settings. Brightness histogram analysis shows that the brightness value distribution of normal coating images is concentrated between 128 and 192. When the brightness value exceeds 230, it is judged as overexposure, and when it is less than 50, it is judged as underexposure. In practical applications, metal coatings are more commonly overexposed locally due to surface reflection, while dark coatings are prone to underexposure in the edge area. Gamma transform is an effective method to deal with local overexposure. By setting the gamma coefficient to 1.8, the overexposed pixels are nonlinearly compressed to map the brightness value to a reasonable range. The block enhancement algorithm uses an 8×8 block size to adaptively enhance the brightness of the underexposed area. The compensation coefficient is dynamically adjusted according to the average brightness value of the local area, and the compensation range is between 1.2 and 2.5 times. The brightness distribution of the processed image is more uniform, which is conducive to subsequent feature extraction. Color space conversion is a key step in image feature extraction. The RGB color space is converted to the HSV space, where the H channel reflects the hue information of the coating, the S channel represents the saturation of the color, and the V channel corresponds to the brightness value. The chromaticity mean vector contains 3 components, which correspond to the average values ​​of the three HSV channels respectively. The chromaticity standard deviation vector reflects the discrete degree of each channel, and the chromaticity gradient vector represents the spatial variation characteristics of the color. In practical applications, the hue mean of the standard coating is between 180 and 220 degrees, and the saturation mean is between 0.6 and 0.8. The focal length adjustment of the liquid lens is achieved by changing the driving voltage, and the voltage and focal length are nonlinearly related. When the Euclidean distance between the color eigenvector and the standard eigenvector is greater than 0.15, the focal length adjustment is started. The proportional coefficient is set to 0.8 and the integral time constant is 0.2 seconds, which ensures fast response while ensuring the stability of the adjustment. For a liquid lens with a focal length of 50 mm, the adjustment range is plus or minus 5 mm, and the driving voltage is between 35 and 65 volts.During the coating inspection process, coatings of different materials have different characteristic parameter requirements. The surface of the ceramic coating is smooth, and the preset overexposure threshold can be increased to 240, and the underexposure threshold is reduced to 40. For matte coatings, the overexposure threshold should be reduced to 220, and the underexposure threshold should be increased to 60 to adapt to its narrower brightness dynamic range. When the production line speed is 2 meters per second, the focusing response time of the liquid lens needs to be controlled within 50 milliseconds, requiring the control algorithm to have high real-time performance. Due to the change in curvature, the edge area of ​​the coating often requires a larger focal length adjustment range. At this time, the integral time constant can be adjusted to 0.3 seconds to improve the adjustment accuracy.

[0030] S105 , establishing a color change motion model based on historical focusing data, predicting the color change trend of the next frame of image, and using the predicted value as a feedforward control signal of the focusing system to suppress oscillation of the focusing system.

[0031] The color feature vector sequence collected by the focusing system is segmented by using a preset window length, and the change speed and change acceleration of the feature vector are calculated by the least square method to obtain the color motion feature vector; for the color motion feature vector, a motion model of the color change is constructed by a recursive neural network, and the feature vector at the next moment is predicted according to the motion model to obtain a motion prediction vector; according to the deviation value between the motion prediction vector and the current color feature vector, a proportional integral differential controller is used to calculate the correction parameter, and the corrected prediction vector is smoothed by the exponential weighted average method to obtain a feedforward compensation vector; for the feedforward compensation vector, high frequency suppression is performed by a low-pass filter, and the amplitude of the feedforward compensation vector is constrained by a nonlinear limiter to obtain a focusing anti-vibration signal.

[0032] Exemplarily, according to the historical color feature vector sequence collected by the focusing system, the feature vector is segmented using a time window with a length of the preset window length, and the change speed and change acceleration of the feature vector in the time window are calculated by the least squares method to obtain the color motion feature vector. For the color motion feature vector, a recursive neural network is used to construct the motion model of the color change, and the motion trajectory parameters of the feature vector are calculated by the motion model. The color feature vector at the next moment is predicted to obtain the motion prediction vector. According to the deviation value between the motion prediction vector and the current color feature vector, a proportional integral differential controller is used to calculate the correction parameter of the motion prediction vector, and the corrected prediction vector is smoothed by the exponential weighted average method to obtain the feedforward compensation vector. For the feedforward compensation vector, a low-pass filter is used to suppress the high frequency of the feedforward compensation vector, and the amplitude of the feedforward compensation vector is constrained by a nonlinear limiter to obtain the focusing anti-vibration signal. The color change in the coating production process shows obvious time series characteristics, and the color change trend can be predicted by analyzing historical data. The choice of time window directly affects the prediction accuracy. When the window length is set to 32 frames, it can retain enough historical information without causing excessive computational burden. The change speed calculated by the least squares method reflects the first-order derivative of the color feature, and the change acceleration reflects the second-order derivative. These two parameters together constitute the motion feature vector. The recursive neural network has the ability to remember historical information and is suitable for processing time series data such as color changes. The number of nodes in the network input layer is 6, corresponding to the position, speed and acceleration information of the color feature. The hidden layer uses 100 neurons, and the number of nodes in the output layer is 3, corresponding to the predicted color feature of the next frame. The training data is collected from normal production conditions and contains 5000 sets of time series samples. The network training adopts the stochastic gradient descent method, the learning rate is set to 0.01, and the training error converges to less than 0.001. There is a deviation between the motion prediction vector and the actual feature vector, which needs to be dynamically corrected by the controller. The proportional coefficient is set to 0.8, the integral time constant is 0.5 seconds, and the differential time constant is 0.1 seconds. The smoothing coefficient in the exponential weighted average method is set to 0.7. A larger smoothing coefficient is conducive to retaining the rapid response characteristics of the predicted value. Under the condition of sudden color change of the coating, the prediction deviation may reach 15%. After correction by the controller, the deviation is reduced to less than 5%. The feedforward compensation vector often contains high-frequency oscillation components. The cutoff frequency of the low-pass filter is set to 20 Hz, which can effectively suppress high-frequency interference. The nonlinear limiter uses a hyperbolic tangent function with a limiting range of plus or minus 1 to ensure that the compensation signal does not cause excessive response of the focusing system. When the production line speed is 2 meters per second, the oscillation amplitude of the focusing system is controlled within 0.2 mm. The change law of the coating color is closely related to the coating material. Due to the reflective surface of the metal coating, the color characteristics change rapidly. The time window length should be shortened to 16 frames to improve the real-time performance of the prediction.For matte coatings, the color changes smoothly and the window length can be extended to 48 frames to improve the smoothness of the prediction. Fluctuations in coating thickness can cause periodic changes in color. Fourier analysis found that the main period is between 0.5 and 2 seconds, and the cutoff frequency of the low-pass filter should cover this frequency range. When the coating formula is switched, the color characteristics will change suddenly. At this time, the exponential average smoothing coefficient should be reduced to 0.5 to speed up the tracking speed of the predicted value.

[0033] S106, performing a quality assessment on the coating surface image acquired by the preset focusing system, calculating the gradient variance value of the image, and using the gradient variance value as a quantitative indicator of image clarity. If the gradient variance value is greater than the gradient variance threshold, the image is judged to be a clear image; if the gradient variance value is less than the gradient variance threshold, the image is judged to be a blurred image, and the blurred image is discarded.

[0034] The Sobel operator is used to calculate the gradients of the coating surface image obtained by the focusing system in the horizontal and vertical directions, and the gradient amplitude matrix is ​​obtained by square and square root operations; the image is divided into a number of sub-regions by a regional blocking method for the gradient amplitude matrix, and the local variance of the gradient amplitude in each sub-region is calculated and normalized to obtain the regional gradient variance value; weighted average fusion is performed according to the spatial distribution of the regional gradient variance value, wherein the weight coefficient is proportional to the sub-region area, and the fused variance value is standardized to obtain the image gradient variance value; if the image gradient variance value is greater than a preset clarity threshold, it is determined to be a clear image and retained, and if the image gradient variance value is less than the preset clarity threshold, it is determined to be a blurred image and eliminated.

[0035] Exemplarily, according to the coating surface image obtained by the preset focusing system, the Sobel operator is used to calculate the gradient of the image in the horizontal and vertical directions, and the gradient amplitude matrix is ​​obtained by square and square root operations. The statistical variance of the gradient amplitude matrix is ​​calculated to obtain the initial gradient variance value. For the initial gradient variance value, the image is divided into several sub-regions by using the regional block method, and the local variance of the gradient amplitude in each sub-region is calculated, and the local variance is normalized to obtain the regional gradient variance value. According to the spatial distribution of the regional gradient variance value, the gradient variance values ​​of each sub-region are fused by the weighted average method, and the variance value after fusion is standardized by setting the weight coefficient to be proportional to the area of ​​the sub-region to obtain the image gradient variance value. For the image gradient variance value, the threshold judgment method is used to evaluate the image quality. If the image gradient variance value is greater than the preset clarity threshold, it is determined to be a clear image and retained. If it is less than the preset clarity threshold, it is determined to be a blurred image and removed. The clarity evaluation of the coating surface image is based on the image gradient features. The Sobel operator calculates the gradients in the horizontal and vertical directions respectively through a 3×3 convolution kernel. For a clear image, the gradient value of the edge part is significantly higher than that of the background area. The typical edge gradient amplitude is between 50 and 150, while the gradient amplitude of the background area is usually less than 20. The initial gradient variance reflects the discrete degree of the gradient distribution of the entire image. The initial gradient variance value of the clear image is usually above 800. Image block processing can more finely evaluate the local clarity. The 640×480 image is divided into 8×6 sub-regions, each with a size of 80×80 pixels. The selection of sub-regions takes into account the spatial distribution characteristics of the coating surface features, which should ensure sufficient local details while avoiding noise interference. In practical applications, the local gradient variance value of the metal coating surface fluctuates between 200 and 500, while the local gradient variance of the blurred image is generally less than 100. The regional weighted average takes into account the importance of sub-regions at different positions. The weight coefficient of the central area of ​​the image is 1.0, which decreases to 0.6 towards the edge. This weight distribution takes into account that the central area of ​​coating detection is usually more important. The normalization process maps the gradient variance value to a range of 0 to 1, which is convenient for comparison with the preset threshold. In actual production, different coating materials correspond to different gradient characteristics, and the gradient variance value of bright coatings is generally higher than that of matte coatings. The selection of the clarity threshold requires a trade-off between the sensitivity and reliability of the detection. When the threshold is set to 0.6, it can effectively distinguish the image blur caused by focal length offset. When the production line speed is 2 meters per second, the camera exposure time is set to 0.5 milliseconds, and the influence of motion blur can be ignored. When there are defects such as scratches and bubbles on the coating surface, these defective areas will produce additional gradient contributions, resulting in an increase in the variance value. Therefore, the influence of these abnormal areas needs to be eliminated before defect detection. In different application scenarios, the performance of gradient features is also different.Since the surface of glass base coating is smooth, the gradient mainly comes from the edge of the coating and surface defects. The variance threshold of a clear image can be reduced to 0.4. For textured coatings, the inherent texture of the surface will produce a lot of gradient information, and the threshold needs to be increased to above 0.8. When the coating thickness is less than 10 microns, the surface reflection will reduce the image contrast. At this time, the threshold needs to be reduced to 0.5 and the image preprocessing intensity needs to be increased. On high-temperature coating production lines, thermal radiation will reduce image quality. It is recommended to increase the threshold to 0.7 and shorten the camera exposure time to 0.3 milliseconds.

[0036] S107. Extract the sub-pixel contour of the coating edge based on the coating surface image after removing the blurred image, calculate the pixel area of ​​the coating by contour area calculation, calculate the thickness value of the coating in combination with the focal length parameters of the preset focusing system, and use the thickness value as the feedback control quantity of the coating machine to achieve online control of the coating thickness.

[0037] A Gaussian template is used to perform a convolution operation on the coating edge, and the grayscale centroid coordinates of the edge point are calculated by sub-pixel interpolation to obtain a sub-pixel edge point sequence; for the sub-pixel edge point sequence, a rectangular integral method is used to calculate the coating contour area, and the interior of the contour is filled by a region growing algorithm, and the coating area value is obtained according to the pixel count of the filled area; according to the coating area value and the focal length value of the focusing system, an optical imaging formula is used to calculate the spatial size of the coating area, and the coating thickness value is obtained by correction through preset calibration parameters; a deviation value is calculated between the coating thickness value and the preset coating target thickness, a proportional integral controller is used to calculate the coating machine control voltage, and the coating thickness is adjusted by the coating machine actuator to obtain a coating control parameter.

[0038] Exemplarily, according to the grayscale distribution of the clear coating surface image, a Gaussian template is used to perform convolution operation on the coating edge, and the grayscale centroid coordinates of the edge point are calculated by sub-pixel interpolation. The centroid coordinate sequence is interpolated by cubic spline to obtain the sub-pixel edge point sequence. For the sub-pixel edge point sequence, the rectangular integral method is used to calculate the area of ​​the coating contour, and the internal area of ​​the contour is filled by the region growing algorithm. The pixel points of the filled area are accurately counted to obtain the coating area value.

[0039]

[0040] , A represents the area of ​​the coating profile, h represents the width of the rectangle, y i Represents the height of the ith rectangle, and n represents the number of rectangles. According to the coating area value and the focal length value of the preset high-speed focusing system, the spatial size of the coating area is calculated using the optical imaging formula, and the spatial size is corrected by the preset calibration parameters to obtain the coating thickness value.

[0041]

[0042] , A represents the coating area, d represents the coating diameter, f 1 Represents the object focal length, f 2 Represents the focal length of the image plane, which is used to calculate the actual imaging area. According to the deviation value between the coating thickness value and the preset coating target thickness, the proportional integral controller is used to calculate the control voltage of the coating machine, and the coating thickness is dynamically adjusted by the coating machine actuator to obtain the coating control parameter. Sub-pixel positioning of the coating edge is the key to achieve high-precision thickness measurement. The Gaussian template is selected to be 5×5 in size, and the standard deviation is set to 1.2 to smooth the edge area. The grayscale center of gravity of the edge point is obtained by weighted average calculation of the grayscale values ​​of the surrounding 25 pixels, and the weight coefficient decays exponentially with the increase of the distance from the center point. In practical applications, the width of the grayscale transition band of the coating edge is usually between 3 and 5 pixels, and sub-pixel interpolation can improve the edge positioning accuracy to 0.1 pixel. The rectangular integration method divides the coating contour into several tiny rectangular strips, each with a width of 1 pixel. The region growing algorithm starts from the seed point inside the contour and gradually expands outward to fill the entire coating area. When the coating area is 10,000 square pixels, the error of the precise counting method is less than 0.1%. For irregularly shaped coatings, non-integer coordinate points on the contour line are accurately located by cubic spline interpolation, and the interpolation node interval is set to 1 pixel. In the optical imaging system, the object distance is set to 300 mm, the lens focal length is 50 mm, and the system magnification is about 0.2 times when the imaging distance is 60 mm. The preset calibration parameters are obtained through the standard thickness block, and the calibration curve adopts a three-segment piecewise linear fit, and the segmentation points are set at 25% and 75% of the coating thickness. The spatial resolution changes with the change of focal length. At the nominal working distance, 1 pixel corresponds to an actual size of 0.01 mm. The control of the coater adopts an incremental proportional integral controller, with a proportional coefficient set to 0.8 and an integral time constant of 2 seconds. The adjustment range of the control voltage is 0 to 10 volts, and the resolution is 0.01 volt. The response time of the coater actuator is about 200 milliseconds, and the control dead zone is set to plus or minus 1% of the target thickness. Under the condition of a production line speed of 2 meters per second, the control accuracy of the coating thickness is better than plus or minus 5 microns. The edge characteristics of the coating are closely related to the material properties. Due to the reflective surface of the metal coating, the edge grayscale gradient is large, and the standard deviation of the Gaussian template should be increased to 1.5. For translucent coatings, the edge transition zone is wide, and the template size needs to be increased to 7×7. When the coating thickness is less than 50 microns, the edge contrast is reduced. At this time, the depth of field of the optical system should be reduced and the imaging distance should be shortened to 55 mm. In a high temperature environment, thermal disturbances will affect the imaging quality. It is recommended to reduce the exposure time to 0.2 milliseconds and increase the image gain compensation to reduce the brightness loss caused by the exposure time.

[0043] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A new energy battery coating thickness control method, characterized in that: The method comprises: Real-time acquisition of new energy battery coating images, obtaining a high-speed continuous image sequence with an image frame frequency greater than the target frequency, extracting unstable images from the continuous image sequence, and using an adaptive histogram equalization algorithm to enhance the contrast of unstable images to obtain a stable image with a contrast variation range reduced to a preset variation range; A chromaticity value range of the coating color is preset, and the stable image is segmented into a coating area and a non-coating area. The chromaticity value of the coating area is within the preset chromaticity value range of the coating color, and the segmented coating area is binarized to obtain a coating binary image; According to the segmented coating binary image, a closing operation is performed on the coating area to fill the small holes inside the coating area, and an opening operation is performed to remove the noise outside the coating area to obtain a coating area image after morphological processing; Identify the local over-exposed and under-exposed areas of the coating area image, suppress the brightness of the local over-exposed area and enhance the brightness of the local under-exposed area according to the brightness distribution of the coating area, obtain a coating image with uniform brightness, transmit the coating image with uniform brightness to a preset focusing system, extract the color feature vector of the coating image, use the color feature vector of the coating image as a feedback control signal of the focusing system, and adjust the focal length of the liquid lens in real time; A motion model of color change is established based on historical focusing data to predict the color change trend of the next frame of image. The predicted value is used as a feedforward control signal of the focusing system to suppress the oscillation of the focusing system. The coating surface image obtained by the preset focusing system is evaluated for quality, and the gradient variance value of the image is calculated. The gradient variance value is used as a quantitative index of image clarity. If the gradient variance value is greater than the gradient variance threshold, it is judged as a clear image. If the gradient variance value is less than the gradient variance threshold, it is judged as a blurred image and the blurred image is removed. Based on the coating surface image with the blurred image removed, the sub-pixel level contour of the coating edge is extracted, and the pixel area of ​​the coating is obtained by calculating the contour area. Combined with the focal length parameters of the preset focusing system, the thickness value of the coating is calculated, and the thickness value is used as the feedback control quantity of the coater to achieve online control of the coating thickness.

2. The method according to claim 1, characterized in that The real-time acquisition of new energy battery coating images, obtaining a high-speed continuous image sequence with an image frame frequency greater than a target frequency, extracting unstable images from the continuous image sequence, and using an adaptive histogram equalization algorithm to perform contrast enhancement on the unstable images to obtain a stable image with a contrast variation range narrowed to a preset variation range, includes: Acquire grayscale matrix data according to the coordinates of the center point of the coating image acquisition area, process the grayscale matrix data using a Gaussian filter, and obtain a filtered image when the mean square error of the grayscale matrix data is less than a mean square error threshold; Performing a histogram statistical operation on the filtered image, decomposing the histogram data by wavelet transform, clustering the decomposed data by an adaptive threshold segmentation algorithm, and determining the image as the unstable image if the inter-class variance obtained by the clustering operation is greater than a preset inter-class variance threshold; An adaptive histogram equalization parameter matrix is ​​constructed for the unstable image, a block enhancement method is used to adjust the contrast of the unstable image, and a block boundary is smoothed by a bilinear interpolation algorithm to obtain an equalized image; Contrast range statistics are performed on the equalized image, pixel points exceeding a preset contrast range are adjusted using a region growing algorithm, and the adjusted image is smoothed by a median filter to obtain a stable image.

3. The method according to claim 1, characterized in that The chromaticity value range of the preset coating color is used to segment the stable image into a coating area and a non-coating area, the chromaticity value of the coating area is within the chromaticity value range of the preset coating color, and the segmented coating area is binarized to obtain a coating binary image, including: The color space transformation matrix is ​​used to transform the original image into a color space, and the image chromaticity value distribution histogram is obtained by calculating the color components; According to the pixel points in the chromaticity value distribution histogram that are within the preset coating color chromaticity value range, a coating area segmentation threshold is calculated using a maximum inter-class variance algorithm to obtain an initial coating area image; For the initial coating area image, a morphological opening operation is used to correct the boundary of the marked area, and the corrected area is expanded by a neighborhood growing algorithm to obtain the complete coating area image; Grayscale values ​​of the coating area in the complete coating area image are counted. If the grayscale value of the pixel is greater than the optimal binarization threshold, a value of 255 is assigned. If the grayscale value of the pixel is less than the optimal binarization threshold, a value of 0 is assigned to obtain the coating binarization image.

4. The method according to claim 1, characterized in that: The coating region is closed according to the segmented coating binary image to fill small holes inside the coating region, and noise outside the coating region is removed by opening operation to obtain a morphologically processed coating region image, including: For the connected domain features in the coating binary image, the connected domain features are marked by using an area screening method, the total number of pixels in each connected domain is calculated, and the non-coating pixels are identified by using a four-neighborhood connectivity analysis method. The coating image to be processed is obtained by calculating the ratio of the area to the perimeter of the connected domain formed by the non-coating pixels; A morphological closing operation is used to process the holes to be filled in the coating image to be processed, and a closed operation coating image is obtained by setting the size of the structural element to be larger than the diameter of the inscribed circle corresponding to the preset hole feature ratio; According to the closed operation coating image, the coating area edge is smoothed by using a morphological open operation, and the coating area image after morphological processing is obtained by setting the size of the structural element smaller than the diameter of the circumscribed circle corresponding to the preset minimum coating area.

5. The method according to claim 1, characterized in that The method includes identifying local over-exposed and under-exposed areas of the coating area image, suppressing the brightness of the local over-exposed area and enhancing the brightness of the local under-exposed area according to the brightness distribution of the coating area, obtaining a coating image with uniform brightness, transmitting the coating image with uniform brightness to a preset focusing system, extracting a color feature vector of the coating image, using the color feature vector of the coating image as a feedback control signal of the focusing system, and adjusting the focal length of the liquid lens in real time, including: According to the brightness histogram distribution of the coating area image, an adaptive segmentation algorithm is used to divide the brightness value into intervals, and the brightness abnormality image is obtained by calculating the pixel points in the coating area that exceed the preset overexposure threshold and are lower than the preset underexposure threshold; For the abnormal brightness image, a gamma transform function is used to attenuate overexposed pixels, a block enhancement algorithm is used to compensate for underexposed pixels, and a bilinear interpolation algorithm is used to smooth the block boundaries to obtain the coating image with uniform brightness; According to the red, green and blue three-channel data of the coating image with uniform brightness, a color space conversion matrix is ​​used to perform chromaticity calculation, and a chromaticity mean vector and a chromaticity gradient vector are obtained through a deep feature extraction network to obtain the color feature vector; For the color feature vector, a feedback controller is used to calculate the Euclidean distance with a preset standard feature vector, and a driving voltage is applied to the liquid lens through a proportional integral control algorithm to obtain a focal length adjustment signal.

6. The method according to claim 1, characterized in that The method of establishing a color change motion model based on historical focusing data, predicting the color change trend of the next frame of image, and using the predicted value as a feedforward control signal of the focusing system to suppress the oscillation of the focusing system includes: The color feature vector sequence collected by the focusing system is segmented by using a preset window length, and the change speed and change acceleration of the feature vector are calculated by the least square method to obtain the color motion feature vector; For the color motion feature vector, a motion model of the color change is constructed by a recursive neural network, and the feature vector at the next moment is predicted according to the motion model to obtain a motion prediction vector; According to the deviation value between the motion prediction vector and the current color feature vector, a proportional integral differential controller is used to calculate the correction parameter, and the corrected prediction vector is smoothed by an exponential weighted average method to obtain a feedforward compensation vector; For the feedforward compensation vector, high frequency suppression is performed through a low-pass filter, and the amplitude of the feedforward compensation vector is constrained by a nonlinear limiter to obtain a focusing anti-vibration signal.

7. The method according to claim 1, characterized in that The coating surface image obtained by the preset focusing system is evaluated for quality, the gradient variance value of the image is calculated, and the gradient variance value is used as a quantitative index of image clarity. If the gradient variance value is greater than the gradient variance threshold, it is judged as a clear image, and if the gradient variance value is less than the gradient variance threshold, it is judged as a blurred image, and the blurred image is removed, including: The Sobel operator is used to calculate the gradients in the horizontal and vertical directions of the coating surface image obtained by the focusing system, and the gradient amplitude matrix is ​​obtained through square and square root operations. The image is divided into several sub-regions by adopting a regional block method for the gradient amplitude matrix, and the regional gradient variance value is obtained by calculating the local variance of the gradient amplitude in each sub-region and performing normalization processing; Perform weighted average fusion according to the spatial distribution of the regional gradient variance value, wherein the weight coefficient is proportional to the sub-region area, and standardize the fused variance value to obtain the image gradient variance value; If the image gradient variance value is greater than the preset clarity threshold, it is determined to be a clear image and retained; if the image gradient variance value is less than the preset clarity threshold, it is determined to be a blurred image and discarded.

8. The method according to claim 1, characterized in that The method extracts the sub-pixel contour of the coating edge based on the coating surface image after removing the blurred image, obtains the pixel area of ​​the coating by calculating the contour area, calculates the coating thickness value in combination with the focal length parameter of the preset focusing system, and uses the thickness value as the feedback control amount of the coating machine to realize the online control of the coating thickness, including: The Gaussian template is used to perform convolution operation on the coating edge, and the grayscale centroid coordinates of the edge point are calculated by sub-pixel interpolation to obtain the sub-pixel edge point sequence; For the sub-pixel edge point sequence, the coating contour area is calculated using the rectangular integration method, the interior of the contour is filled using the region growing algorithm, and the coating area value is obtained according to the pixel count of the filled area; According to the coating area value and the focal length value of the focusing system, the spatial size of the coating area is calculated using an optical imaging formula, and the coating thickness value is obtained by correction through preset calibration parameters; The coating thickness is adjusted by adjusting the coating control parameters by using a proportional-integral controller to calculate the coating control voltage.

Images