Self-adaptive regulation and control method of intelligent aleurone layer grinding system

Abstract

The invention relates to the field of grinding, and particularly provides a self-adaptive regulation and control method of an aleurone layer intelligent grinding system, which comprises the following steps: acquiring original image data of wheat bran entering the grinding system in real time through optical imaging equipment, and preprocessing the acquired original image; performing feature extraction on the preprocessed image by using the trained deep learning model, and outputting an aleurone layer area proportion A, an average thickness T and a surface texture roughness R; dynamically calculating the spindle rotor rotating speed, the feeding rate and the airflow pressure of the grinding system according to the output data; and a control instruction is generated according to the calculated parameters, the control instruction is sent to the PLC control system, and the running state of the grinding system is adjusted. By means of the scheme, the technical problem that existing grinding equipment lacks intelligent regulation and control and real-time optimization is solved.

Description

Technical Field

[0001] The present invention relates to the field of grinding, and specifically provides an adaptive control method for an aleurone layer intelligent grinding system. Background Art

[0002] The wheat aleurone layer is the most nutritious part of wheat bran. It is rich in dietary fiber, protein, minerals, and multiple vitamins, and has important nutritional and health benefits. However, due to the close integration of the aleurone layer with other bran structures, traditional physical separation methods have the following problems: The material is unevenly distributed during the grinding process, resulting in a low endosperm extraction rate; traditional mechanical equipment lacks intelligent control means and cannot adjust process parameters in real time, making it difficult to meet the needs of different raw material characteristics; mechanical wear is severe, affecting equipment life and product safety; the production process has high energy consumption and is difficult to achieve fine grading.

[0003] Existing patents for multi-stage cyclone grinders and planetary mills have improved grinding efficiency to a certain extent, but they do not address the issues of intelligent control and real-time optimization. Furthermore, existing technologies fail to fully utilize modern digital technologies, resulting in a lack of flexibility and adaptability in the production process. Summary of the Invention

[0004] In order to overcome the above-mentioned defects, the present invention is proposed to provide a technical solution to the lack of intelligent control and real-time optimization of existing grinding equipment.

[0005] The present invention provides an adaptive control method for an aleurone layer intelligent grinding system, comprising the following steps: The original image data of wheat bran entering the grinding system is obtained in real time by optical imaging equipment, and the collected original images are preprocessed; The trained deep learning model is used to extract features from the preprocessed original image, and the aleurone area ratio A, average thickness T and surface texture roughness R are output; According to the aleurone layer area ratio A, average thickness T and surface texture roughness R, the spindle rotor speed, feed rate and air flow pressure of the grinding system are dynamically calculated; A control command is generated based on the calculated spindle rotor speed, feed rate, and air flow pressure, and the control command is sent to the PLC control system to adjust the operating status of the grinding system.

[0006] Furthermore, the pre-processing of the collected original image includes denoising, brightness normalization, contrast enhancement, color space conversion, edge detection and cropping and scaling.

[0007] further, The denoising process includes removing salt and pepper noise in the original image using median filtering; The brightness normalization includes unifying the pixel values of the denoised original image to the interval [0, 1], that is, wherein, represents the pixel value of a single pixel point after denoising of the original image, , represents the minimum and maximum values of all pixel values after denoising of the original image, represents the pixel value of the original image after normalization, ranging from 0 to 1; The contrast enhancement includes enhancing the local contrast by using the CLAHE algorithm, that is, wherein, represents the pixel value of the original image after contrast enhancement, and CLAHE represents the limited contrast adaptive histogram equalization algorithm; The color space conversion includes converting the original image after contrast enhancement from the RGB color space to the HSV color space, that is, wherein, represents the RGB color channel, represents the HSV color channel; The edge detection includes extracting the material edge by using the Canny operator for the image after color space conversion, that is, wherein, Canny represents an edge detector; The cropping and scaling process includes cropping and scaling the image to a specified size according to the model input requirements.

[0008] Further, the deep learning model uses a pre-trained ResNet50 as a feature extraction backbone network, removes the original classification head, retains the convolutional layer and residual block structure, and connects a CBAM module after the last residual block of ResNet50 to sequentially perform channel attention and spatial attention calculation on the output feature map.

[0009] Further, the output of the deep learning model is set to a classification branch and a segmentation branch, The classification branch includes averaging the CBAM output global feature map into a 512-dimensional vector, and outputting a two-class result of the flour layer or non-flour layer through a fully connected layer; The segmentation branch uses a U-Net structure decoder, fuses low-level features through twice upsampling and skip connection, and outputs a single-channel pixel-level segmentation mask.

[0010] Further, the loss function for training the deep learning model includes: Define the segmentation loss wherein, Spred represents the segmentation scalar value predicted by the model, Strue represents the real segmentation mask, represents a weight coefficient, BCELoss() represents a binary cross-entropy loss, and Dice Loss() represents a Dice coefficient loss; define the classification loss wherein, represents a classification scalar value predicted by the model, represents a real classification mask; total loss wherein, represents an indicator function, when, is 1, otherwise 0, represents a weight coefficient of the segmentation loss.

[0011] Further, the dynamic calculation of the spindle rotor speed, the feed rate and the air flow pressure of the grinding system according to the area proportion A of the paste powder layer, the average thickness T and the surface texture roughness R comprises: when the area proportion of the paste powder layer is less than 40%, the spindle rotor speed is set to 1200 rpm; when the area proportion of the paste powder layer is between 40% and 70%, the spindle rotor speed is set to 1400 rpm; when the area proportion of the paste powder layer is greater than 70%, the spindle rotor speed is set to 1600 rpm; calculate the feed rate wherein, and are weight coefficients; calculate the air flow pressure wherein, and are weight coefficients.

[0012] Further, the three-dimensional scanner is further used to collect the particle size distribution, the density and the humidity of the wheat bran raw material, the sensor is further used to monitor the running state and the environmental parameters of the equipment in the grinding system in real time, and the three-dimensional virtual model of the grinding system is constructed by using the machine learning algorithm based on the collected data.

[0013] Further, the running data in the production process is further uploaded to the cloud in real time, and the cloud data analysis platform is further used to store and analyze the production data.

[0014] Further, the grinding system comprises a grinding device, a grading and screening device and an energy management device.

[0015] Working principle and beneficial effects of the present application: In the technical solutions of the present application, according to the real-time collected original image of the wheat bran entering the grinding system, a trained deep learning model (such as ResNet50+CBAM attention mechanism) is used to extract features from the preprocessed original image, and the area ratio of the aleuron layer, the average thickness and the surface texture roughness are output. According to the output parameters, the rotor speed, the feed rate and the air flow pressure of the grinding system are dynamically adjusted. It ensures that the equipment always operates in the optimal working condition, avoiding over-grinding or under-grinding phenomenon. DETAILED DESCRIPTION

[0016] Some embodiments of the present application are described below. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application, and are not intended to limit the protection scope of the present application.

[0017] Embodiment 1

[0018] The self-adaptive control method of the aleuron layer intelligent grinding system in this embodiment mainly includes the following steps.

[0019] 1. Construct a digital twin model of the grinding system.

[0020] Use a three-dimensional scanner to collect the physical properties of the wheat bran raw material (such as particle size distribution, density, humidity, etc.), and use sensors to monitor the equipment operating state (speed, temperature, humidity, etc.) and environmental parameters (air pressure, workshop temperature and humidity) in real time. Based on the collected data, a virtual model of the grinding system is constructed using machine learning algorithms. This model can simulate various working conditions in the actual production process and generate optimal process parameters.

[0021] 2. Real-time acquisition of original image data of wheat bran by optical imaging equipment, preprocessing of collected original image, and extraction of aleuron layer region features.

[0022] Specifically, preprocessing includes denoising, brightness normalization, contrast enhancement, color space conversion, edge detection and cropping and scaling.

[0023] ① Denoising: use median filtering to remove salt and pepper noise in the original image.

[0024] Specifically, determine the convolution kernel parameters, and determine the sliding window size according to the convolution kernel parameters. Assuming that the convolution kernel is defined as 5x5, the sliding window is a square neighborhood window centered at pixel (x, y), and 5x5 represents that the width and height of the window are both 5 pixels. This window moves on the image and processes each pixel one by one. For each pixel (x, y) at the center of the window, take the brightness values of all pixel points in the window, sort them from small to large, and take the median as the new pixel value of the center point.

[0025] In this embodiment, median filtering can effectively eliminate salt and pepper noise (random black and white noise points on the image), which may occur during industrial camera acquisition. It is a nonlinear filtering that can better protect the edge information of the image while removing noise, which is crucial for subsequent edge detection.

[0026] ② Luminance normalization: unify the pixel values of the denoised original image to the interval [0, 1], eliminate the influence of light difference.

[0027]

[0028] wherein, denotes the pixel value of a single pixel point after denoising the original image, , denotes the minimum and maximum values of all pixel values after denoising the original image, denotes the pixel value of the original image after normalization, ranging from [0, 1].

[0029] In this embodiment, the pixel value range of an image is linearly mapped to a fixed interval [0, 1], ensuring that each image is independently stretched to its own full dynamic range. Eliminate the brightness difference between images collected at different times and under different lighting conditions. For example, the overall brightness of images taken in the morning and afternoon may be different, and after normalization, their contrast becomes consistent. Provide standardized input for deep learning models, which helps the stability and convergence speed of model training.

[0030] ③ Contrast enhancement: use CLAHE algorithm to enhance local contrast and improve image clarity.

[0031]

[0032] wherein, denotes the pixel value of the original image after contrast enhancement, and CLAHE denotes the limited contrast adaptive histogram equalization algorithm.

[0033] In this embodiment, CLAHE is to divide the image into multiple small blocks (tiles), and perform histogram equalization on each small block separately, thereby enhancing the local contrast of the region. The block size needs to be small enough to enhance local details and large enough to ensure statistical significance when defined. According to the actual effect, fine-tune. After calculating the small block histogram, a threshold will be set, and the part of the histogram that exceeds the threshold will be "clipped" and evenly distributed to the entire histogram interval, effectively suppressing the amplification of background noise. Through contrast enhancement, the local contrast between the powder layer and the background in the image can be significantly enhanced, making the features more obvious and facilitating model learning.

[0034] ④ Color space conversion: Convert the original image after contrast enhancement from RGB color space to HSV color space, keep H and S channels, and reduce the influence of illumination changes on color recognition.

[0035]

[0036] wherein, RGB represents the RGB color channel, HSV represents the HSV color channel.

[0037] In this embodiment, the RGB color space is composed of three primary colors of red, green and blue, and is very sensitive to illumination changes. In the HSV color space, H (Hue) represents the type of color (such as red, yellow, green), which is least affected by light and is a stable feature for identifying object color. S (Saturation) represents the purity of color, and V (Value) represents the brightness of color. Conversion to HSV can separate color information from brightness information, making the algorithm more robust to illumination changes. Regardless of light and dark, the "hue" of the powder layer is relatively stable.

[0038] The H and S channels are kept because the goal is to identify the powder layer (whose color is a stable feature), while the V channel represents brightness, which has been processed in the previous step and is still susceptible to light, so it can be discarded or only used as an auxiliary.

[0039] ⑤ Edge detection: For the image converted by color space conversion, use Canny operator to extract the material edge, which will assist in subsequent feature extraction and segmentation.

[0040]

[0041] wherein, Canny represents the edge detector. By setting a high threshold and a low threshold, pixel points with gradient values higher than the high threshold are determined as strong edges and are definitely retained. Pixel points with gradient values lower than the low threshold are discarded. Pixel points with gradient values between the two thresholds are considered as weak edges. Only when the weak edges are connected to the strong edges, they will be retained as true edges. Edge detection can highlight the outline and texture information of the powder layer area. These edge features can be combined with color features (HSV) to provide more rich and discriminative input information for the deep learning model, improving the segmentation accuracy.

[0042] ⑥ Cropping and scaling: According to the model input requirements, the image is cropped and scaled to a specified size (such as 512x512).

[0043]

[0044] wherein, Represents the pixel value after the original image cropping and scaling processing. By defining a rectangular region in advance, it is ensured that only the materials on the conveyor belt are included in the image, excluding unnecessary backgrounds (such as device frames, shadows), reducing interference. Scale to 512x512 to adapt to the deep learning model, which usually requires fixed-size input, ensuring that each input model image has the same size and proportion.

[0045] 3. Use the trained deep learning model (such as ResNet50+CBAM attention mechanism) to extract features from the preprocessed image, outputting the area ratio A of the flour paste layer, the average thickness T, and the surface texture roughness R.

[0046] In one embodiment, the deep learning model uses a pre-trained ResNet50 as the feature extraction backbone network, removes the original classification head, retains the convolutional layer and residual block structure, and connects a CBAM (Convolutional Block Attention Module) module after the last residual block of ResNet50. The module sequentially calculates the channel attention and spatial attention of the output feature map to enhance the saliency of key features. Channel attention is to generate channel weights (0~1) through global average pooling and fully connected layers to strengthen key channel features. Spatial attention is to generate a spatial weight mask through a convolutional layer to focus on the edge and texture features of the flour paste layer region.

[0047] In one embodiment, the output of the deep learning model is set to have a classification branch and a segmentation branch. The classification branch determines whether it is a flour paste layer, and the segmentation branch generates a pixel-level segmentation mask. Specifically: Classification branch: average pool the CBAM output global feature map into a 512-dimensional vector, and output a binary classification result (flour paste layer / non-flour paste layer) through a fully connected layer (nn.Linear(512 7 7, 1)).

[0048] Segmentation branch: use a U-Net structure decoder, fuse low-level features through two times of upsampling (UpBlock) and skip connection, and finally output a single-channel pixel-level segmentation mask.

[0049] In one embodiment, the deep learning model training includes: ① Obtain training images and preprocess and data augment the training images; Preprocessing: normalize the image brightness to the [0, 1] interval, and convert it to the HSV color space to reduce light interference.

[0050] Data augmentation: apply random flipping, rotation (±15°), and light transformation (random adjustment of brightness / contrast) to improve generalization and avoid overfitting.

[0051] ②Segmentation loss adopts Dice Loss + BCE Loss weighted combination: wherein, Spred represents the segmentation scalar value predicted by the model, Strue represents the real segmentation mask (0 or 1), represents the weight coefficient, used to balance the loss of two classes, usually set to 0.5.

[0052] Classification loss uses binary cross-entropy loss: wherein, represents the classification scalar value predicted by the model, represents the real classification mask (0 or 1), BCE Loss (Binary Cross Entropy Loss) refers to the binary cross-entropy loss, which is suitable for binary classification problems and measures the distance between the probability output by the sigmoid function and the real label. Dice Loss refers to the Dice coefficient loss, which is widely used in image segmentation tasks and measures the overlap between the predicted segmentation region and the real segmentation region.

[0053] Total loss wherein, represents the indicator function, when, is 1, otherwise 0, represents the weight coefficient of the segmentation loss, used to balance the two loss terms, usually set between 0.2 and 0.5.

[0054] 4. According to the model output and the preset target value, the running parameters of the grinding system are dynamically calculated, including the spindle rotor speed, the feed rate and the air flow pressure, specifically: The spindle rotor speed is dynamically adjusted according to the area ratio of the paste powder layer. When the area ratio of the paste powder layer is less than 40%, the spindle rotor speed is set to 1200 rpm; when the area ratio of the paste powder layer is between 40% and 70%, the spindle rotor speed is set to 1400 rpm; when the area ratio of the paste powder layer is greater than 70%, the spindle rotor speed is set to 1600 rpm.

[0055] The feed rate is calculated according to the area ratio A of the paste powder layer and the average thickness T wherein, and are weight coefficients, the value range of is 0.8-1.2, the value range of is 0.5-0.8.

[0056] The air flow pressure is calculated according to the average thickness T and the surface texture roughness R wherein, and the weight coefficient a is in the range of 0.6-1.0, the weight coefficient a is in the range of 0.6-1.0, the weight coefficient a is in the range of 0.6-1.0.

[0057] 5. Generate control instructions according to the operating parameters of the grinding system, send the control instructions to the PLC control system through the Modbus TCP / OPC UA protocol, adjust the operating state of the grinding system; at the same time, feedback real-time data to the digital twin model for state monitoring and visualization.

[0058] 6. The operating data (including energy consumption, yield, quality indicators, etc.) in the entire production process are uploaded to the cloud in real time to form a complete production record for historical data analysis, trend prediction, and process optimization. The cloud data analysis platform is used to store and analyze production data, and historical data analysis and trend prediction are used to provide a scientific basis for process improvement and new product development.

[0059] Based on the above steps, through the combination of digital twin technology and the grinding system, real-time monitoring and dynamic optimization of the grinding process are realized. According to the real-time collected original image of the wheat bran entering the grinding system, the trained deep learning model (such as ResNet50+CBAM attention mechanism) is used to extract features from the preprocessed original image, and the output is the area ratio of the aleuron layer, the average thickness, and the surface texture roughness. According to the output parameters, the rotor speed, feed rate, and air flow pressure of the grinding system are dynamically adjusted. This ensures that the equipment always operates in the optimal working condition, avoiding over-grinding or under-grinding. The application of digital twin technology seamlessly connects the virtual model with the actual production, with a prediction accuracy of over 95%, greatly reducing human operation errors. The energy consumption, yield, quality, and other data in the entire production process are uploaded to the cloud in real time to form a complete production record, which is convenient for later tracing and optimization. The digital twin platform supports historical data analysis and trend prediction, providing a scientific basis for process improvement and new product development.

[0060] It should be noted that although the above embodiments describe the steps in a specific order, those skilled in the art can understand that, in order to achieve the effect of the present application, the steps do not necessarily have to be executed in this order, they can be executed simultaneously (in parallel) or in other orders, and these changes are within the scope of protection of the present application.

[0061] Example 2

[0062] Based on the self-adaptive control method of the aleuron layer intelligent grinding system proposed in Example 1, the grinding system involved includes: 1. Grinding device The multi-stage stepped grinding chamber design, combined with the reverse rotating vortex shear assembly, improves grinding efficiency and product uniformity. The rotor blades are made of high-strength wear-resistant ceramic material, effectively extending the service life of the equipment and reducing the risk of product contamination due to wear and tear. The built-in cooling system reduces the temperature of the grinding chamber through circulating water, avoiding the decline in product quality due to overheating.

[0063] 2. Classification and screening device After grinding, a classification and screening device is added to separate materials of different particle sizes through centrifugal force and air flow, meeting the needs of different application scenarios. The classification device adopts a double-layer structure design, with the upper screen being 30-50 mesh and the lower screen being 100-110 mesh. The ground material is preliminarily separated by centrifugal force and air flow, with coarse particles returning to the grinding device for secondary processing, and fine particles being packaged after screening, ensuring uniform particle size and meeting the standards. The screening device is equipped with a built-in vibration motor, with amplitude and frequency automatically adjustable according to material characteristics, further improving separation accuracy.

[0064] 3. Energy management device The heat and kinetic energy generated during the grinding process are converted into usable energy through a heat exchanger to heat the feed bin or other auxiliary equipment. A dust collection and recycling system is provided to reduce resource waste and environmental pollution.

[0065] Thus, the technical solutions of the present application have been described, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without deviating from the principles of the present application, and the technical solutions after these changes or replacements will fall within the protection scope of the present application.

Claims

1. An adaptive control method for an aleurone layer intelligent grinding system, characterized in that: The following steps are involved: The original image data of wheat bran entering the grinding system is obtained in real time by optical imaging equipment, and the collected original images are preprocessed; The trained deep learning model is used to extract features from the preprocessed original image, and the aleurone area ratio A, average thickness T and surface texture roughness R are output; According to the aleurone layer area ratio A, average thickness T and surface texture roughness R, the spindle rotor speed, feed rate and air flow pressure of the grinding system are dynamically calculated; A control command is generated based on the calculated spindle rotor speed, feed rate, and air flow pressure, and the control command is sent to the PLC control system to adjust the operating status of the grinding system.

2. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The preprocessing of the collected original image includes denoising, brightness normalization, contrast enhancement, color space conversion, edge detection and cropping and scaling.

3. The adaptive control method of an aleurone layer intelligent grinding system according to claim 2, characterized in that: The denoising process includes removing salt and pepper noise in the original image using median filtering; The brightness normalization includes unifying the pixel values ​​of the original image after denoising to the interval [0, 1], that is, ,in, Represents the pixel value of a single pixel in the original image after denoising. 、 Represents the minimum and maximum values ​​of all pixel values ​​after denoising the original image. Represents the normalized pixel value of the original image, ranging from [0, 1]; The contrast enhancement includes using the CLAHE algorithm to enhance the local contrast, that is, ,in, represents the pixel value of the original image after contrast enhancement, and CLAHE represents the contrast-limited adaptive histogram equalization algorithm; The color space conversion includes converting the original image after contrast enhancement from RGB color space to HSV color space, that is, ,in, represents the RGB color channels, Represents HSV color channel; The edge detection includes extracting the edge of the material using the Canny operator for the image after color space conversion, that is, , where Canny represents the edge detector; The cropping and scaling process includes cropping and scaling the image to a specified size according to the model input requirements.

4. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The deep learning model uses the pre-trained ResNet50 as the feature extraction backbone network, removes the original classification head, retains the convolutional layer and residual block structure, connects the CBAM module after the last residual block of ResNet50, and performs channel attention and spatial attention calculations on the output feature map in sequence.

5. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The output of the deep learning model sets the classification branch and the segmentation branch, The classification branch involves averaging the global feature map output by CBAM into a 512-dimensional vector and outputting the binary classification results of aleurone or non-aleurone through a fully connected layer; The segmentation branch uses a U-Net structure decoder, which fuses low-level features through two upsampling and skip connections and outputs a single-channel pixel-level segmentation mask.

6. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The loss functions for deep learning model training include: Define segmentation loss ,in, Spred represents the scalar value of the split predicted by the model, Strue represents the true segmentation mask, Represents the weight coefficient, BCELoss() refers to the binary cross entropy loss, and Dice Loss() refers to the Dic coefficient loss; Defining classification loss ,in, represents the categorical scalar value predicted by the model, represents the true classification mask; Total loss ,in, represents the indicator function, hour, is 1, otherwise it is 0. Represents the weight coefficient of segmentation loss.

7. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The dynamic calculation of the spindle rotor speed, feed rate and air flow pressure of the grinding system according to the aleurone layer area ratio A, average thickness T and surface texture roughness R includes: When the aleurone layer area accounts for less than 40%, the spindle rotor speed is set to 1200 rpm; when the aleurone layer area accounts for between 40% and 70%, the spindle rotor speed is set to 1400 rpm; when the aleurone layer area accounts for more than 70%, the spindle rotor speed is set to 1600 rpm; Calculate feed rate ,in, and is the weight coefficient; Calculating airflow pressure ,in, and is the weight coefficient.

8. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: It also includes using a 3D scanner to collect the particle size distribution, density, and humidity of wheat bran raw materials, using sensors to monitor the equipment operating status and environmental parameters in the grinding system in real time, and using machine learning algorithms to build a 3D virtual model of the grinding system based on the collected data.

9. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: It also includes uploading the operating data of the production process to the cloud in real time, and using the cloud data analysis platform to store and analyze the production data.

10. The adaptive control method of an aleurone layer intelligent grinding system according to claim 1, characterized in that: The grinding system includes a grinding device, a grading and screening device, and an energy management device.