A secondary extrusion device based on a deep learning model

By using a secondary extrusion device based on a deep learning model, utilizing an extrusion screw with equidistant cross-section changes and multi-sensor data feedback, combined with PID controller optimization, the problems of high energy consumption and rapid wear of the secondary extrusion machine are solved, achieving efficient extrusion and extending equipment life.

CN118418506BActive Publication Date: 2026-11-17BEIJING HANYUANSHENG NEW AGRI TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410470133.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-18
Publication Date
2026-11-17
Estimated Expiration
2044-04-18

AI Technical Summary

Technical Problem

Existing secondary extruders consume a lot of energy, generate a lot of noise, and wear out the extrusion rods quickly in sewage treatment, making it difficult to achieve efficient extrusion and extend the equipment's lifespan.

Method used

A secondary extrusion device based on a deep learning model is adopted. Through the extrusion screw with equidistant cross-section variation and multi-sensor data feedback, combined with the optimization of the PID controller, intelligent control is achieved, which optimizes the extrusion process to reduce energy consumption and extend equipment life.

Benefits of technology

It improves the material moisture extrusion rate, reduces energy consumption, extends the service life of the extrusion screw, and improves the accuracy of humidity control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118418506B_ABST
    Figure CN118418506B_ABST
Patent Text Reader

Abstract

A secondary extrusion device based on a deep learning model includes a receiving device (1), a power unit (2), a secondary extrusion unit (3), a PID controller (5), and an STM32 microcontroller (6). The receiving device (1) is located at the top, and the power unit (2) and the secondary extrusion unit (3) are located at the lower end of the receiving device (1). A support (4) is located at the lower end of the power unit (2) and the secondary extrusion unit (3). An extrusion cylinder (3-6) is located inside the secondary extrusion unit (3). An extrusion screw (3-7) for extruding materials is installed inside the extrusion cylinder (3-6). A humidity sensor (1-2) and a humidity sensor (3-6-3) and an image acquisition device (1-4) are located on the secondary extrusion unit (3). Its beneficial effects are: it can achieve stable extrusion of material moisture, reduce energy consumption, keep the PID controller (5) in the best working state, and maximize the extrusion of material moisture.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of deep learning and extrusion technology, specifically to a secondary extrusion device based on a deep learning model. Background Technology

[0002] The rapid growth of the livestock industry has also brought environmental pressures, with farm manure pollution becoming one of the main sources of agricultural non-point source pollution. To address this challenge, livestock farms commonly use manure treatment equipment, two of the most common being solid-liquid separators and secondary extruders.

[0003] Currently, considerable research has been conducted domestically on secondary cow manure utilization. For example, publication number CN216946685U discloses an intelligent production system for cattle farm bedding material based on cow manure recycling. This system includes a solid-liquid separator, a secondary extrusion dewatering machine, and an integrated intelligent aerobic fermentation device arranged sequentially. The integrated intelligent aerobic fermentation device comprises a negative-pressure aerobic fermentation chamber, a feeding area, an aerobic fermentation area, and a discharging area within the negative-pressure aerobic fermentation chamber. An aeration device is installed in the aerobic fermentation area of ​​the negative-pressure aerobic fermentation chamber. An operating robot for automatic material handling, mixing, and turning is also installed within the negative-pressure aerobic fermentation chamber. Furthermore, the negative-pressure aerobic fermentation chamber is equipped with online temperature monitoring devices, H2S online monitoring devices, and NH3 online monitoring devices, all connected to the aerobic fermentation chamber monitoring and control system. This invention realizes intelligent production of cattle farm bedding material based on cow manure recycling.

[0004] In addition, Chinese patent publication number "CN215403765U" discloses a secondary hydraulic screw extruder for sewage treatment. This equipment includes an extruder body, a hydraulic feeder, a hydraulic pressure feeder, and an extrusion screw. The extruder body has a feed inlet at the top and a discharge outlet on one side.

[0005] In manure treatment, secondary extrusion devices are frequently used to squeeze cow manure dry and regenerate it into bedding material for cattle farms. However, the secondary extrusion process is characterized by high energy consumption, high noise levels, and rapid wear of the extrusion rods. Therefore, a secondary extruder is needed that can achieve higher extrusion efficiency, reduce energy consumption, allow for adjustable extrusion speed, prevent excessive wear, and extend the service life of the equipment. Summary of the Invention

[0006] This invention provides a secondary extrusion device based on a deep learning model, which is used to achieve stable moisture extrusion of the extrusion device, improve the moisture extrusion rate of the material, reduce energy consumption, learn and simulate the behavior of the PID controller through the deep learning model, achieve efficient control of the system, optimize the parameters of the PID controller, and keep the device in the best working state.

[0007] To achieve the above objectives, the following technical solution is adopted: a secondary extrusion device based on a deep learning model, comprising a receiving device (1), a power unit (2), a secondary extrusion unit (3), a PID controller (5), and an STM32 microcontroller (6). The receiving device (1) is located at the top, and the lower end of the receiving device (1) is provided with the power unit (2) and the secondary extrusion unit (3). The lower ends of the power unit (2) and the secondary extrusion unit (3) are provided with a support (4). The secondary extrusion unit (3) is provided with an extrusion cylinder (3-6), and an extrusion screw (3-7) for extruding materials is installed inside the extrusion cylinder (3-6). The secondary extrusion unit (3) is provided with a humidity sensor (1-2) and a humidity sensor (3-6-3), as well as an image acquisition device (1-4). The extrusion screw (3-7) is an equidistant screw with a variable cross-section, and the two ends of the extrusion screw (3-7) are... The screw consists of a fine end (3-7-1) and a coarse end (3-7-2). The fine end (3-7-1) is connected to the power unit (2), and the coarse end (3-7-2) is connected to the pressing component of the extrusion device. The diameter of the coarse end (3-7-2) is 1.5-2 times the diameter of the fine end (3-7-1). The thickness D2 of the screw body (3-7-3) near the coarse end (3-7-2) is greater than that near the fine end (3-7-1). The thickness D1 of the screw body (3-7-3) is 1.5-2 times. The STM32 microcontroller (6) is equipped with a pre-trained deep learning model. The deep learning model outputs the Kp, Ki and Kd parameters of the PID controller (5) based on the data collected by the humidity sensor 1 (1-2) and humidity sensor 2 (3-6-3) and the image acquisition device (1-4), so that the PID controller (5) is in the best working state during the working process.

[0008] Furthermore, the training process of the pre-trained deep learning model within the STM32 microcontroller (6) is as follows:

[0009] S101 Data Collection: Collect humidity data from humidity sensor 1 (1-2), humidity sensor 3 (3-3-1) and humidity sensor 2 (3-6-3), as well as the Kp, Ki and Kd parameter data of the corresponding PID controller (5), and classify the image results data after the data collected by the image acquisition device (1-4) is classified by the pre-trained classification model.

[0010] S102 Data Preprocessing: The data from step S101 is classified and processed. The humidity data, the Kp, Ki and Kd parameter data of the PID controller (5) are normalized by the maximum and minimum values. The image result data is processed by one-hot encoding.

[0011] S103 constructs a neural network model: The neural network model used is a five-layer model, including an input layer, three hidden layers, and an output layer; the three hidden layers have 128, 64, and 32 neurons respectively, and the output layer has 3 neurons, which correspond to the Kp, Ki, and Kd parameters of the PID controller (5) respectively;

[0012] S104 Neural Network Training: The data is divided into training, validation, and test sets. The training objective of the neural network is to minimize the loss function by adjusting the weight values. The loss function is the mean squared error (MSE), the activation function is the ReLU function, and stochastic gradient descent (SGD) is used to optimize the model. Training is stopped after 10,000 iterations or when the loss function converges.

[0013] S105 Model Saving: Save the classification model generated after training in step S104.

[0014] Furthermore, the training process of the pre-trained classification model in step S101 is as follows:

[0015] S201 Image Acquisition: Collects images acquired by the image acquisition device (1-4);

[0016] S202 Image preprocessing: Adjust the images collected in step S201 to the same size, and normalize the pixel values ​​in the images to the range of [0,1].

[0017] The S203 builds the AlexNet model, which consists of 5 convolutional layers and 3 fully connected layers. The convolutional layers use the ReLU activation function, and the fully connected layers use Dropout.

[0018] The convolutional layer uses a 3x3 convolutional kernel with a stride of 1; a corresponding pooling layer is provided after the convolutional layer.

[0019] S204 training model: The cross-entropy loss function is used to measure the difference between the model prediction and the actual label, and the optimizer uses stochastic gradient descent to update the model parameters;

[0020] Repeat the steps of forward propagation, loss function calculation, back propagation and parameter update until 10,000 iterations are reached or the loss function converges and training stops.

[0021] S205 Save the classification model: Save the model generated in step S204.

[0022] Furthermore, the receiving device (1) includes a hopper (1-1) and a screw drive (1-3). Multiple humidity sensors (1-2) are evenly arranged inside the hopper (1-1). The screw drive (1-3) is located inside the hopper (1-1) and is used to transport the material extruded once to the unloading section (1-6). An image acquisition device (1-4) for monitoring the material inside the hopper (1-1) is provided at the edge of the hopper (1-1). A motor (1-5) for providing power to the screw drive (1-3) is provided on the outside of the hopper (1-1).

[0023] Furthermore, the power unit (2) includes a transmission connection box (2-1), a transmission box (2-2), and a motor.

[0024] (2-5) The transmission box (2-2) is connected to the support (4) through the transmission box bracket (2-4). The transmission box (2-2) is connected to the transmission connecting box (2-1) on the outside. The lower end of the transmission connecting box (2-1) is connected to the second motor (2-5). The second motor (2-5) is located at the lower end of the support surface of the support (4). The inner side of the transmission box (2-2) is connected to the shaft connector (2-3) for providing power to the secondary extrusion part (3).

[0025] Furthermore, the secondary extrusion section (3) includes an extrusion box (3-1) and a discharge box (3-2). The right side of the extrusion box (3-1) is connected to a feed cylinder (3-9), and the end of the feed cylinder (3-9) is connected to a receiving hopper (3-10). The receiving hopper (3-10) is connected to the discharge section (1-6) and is used to receive the material conveyed from the receiving device (1) to the secondary extrusion section (3). The lower end of the extrusion box (3-1) is provided with a water outlet (3-4) for collecting the water squeezed in the extrusion box (3-1).

[0026] Furthermore, the discharge box (3-2) is connected to the extrusion box (3-1). The lower end of the discharge box (3-2) is provided with a discharge baffle (3-3). The discharge box (3-2) is provided with a discharge port (3-5) for discharging the extruded material. The discharge port (3-5) is provided with a discharge baffle (3-3) on its outer side for directional discharge of the extruded material. The discharge baffle (3-3) is provided with multiple humidity sensors (3-3-1) for monitoring the moisture content of the material after extrusion. The outer side of the discharge box (3-2) is provided with a touch screen (3-8) for manual operation.

[0027] Furthermore, the extrusion box (3-1) and the discharge box (3-2) are equipped with a fixedly connected extrusion cylinder (3-6). The extrusion cylinder (3-6) is equipped with an extrusion screw (3-7). The extrusion cylinder (3-6) mainly includes an extrusion cylinder body (3-6-1). Multiple water outlet holes (3-6-2) are evenly distributed on the extrusion cylinder body (3-6-1) to discharge the water squeezed from the material. Multiple humidity sensors (3-6-3) are also evenly distributed between the water outlet holes (3-6-2) to monitor the moisture content of the material in each section of the extrusion cylinder (3-6).

[0028] Furthermore, the extrusion screw (3-7) is located inside the extrusion cylinder (3-6). The extrusion screw (3-7) includes a spiral blade (3-7-4). The spiral blade (3-7-4) is provided with a screw body (3-7-3). One end of the spiral blade (3-7-4) extends to the shaft connector (2-3) to receive power from the power unit (2). The other end is connected to the top pressing component on the discharge box (3-2). The top pressing component is an electric or hydraulic top pressing component, which is used to cooperate with the discharge box (3-2) to extrude and discharge the material at the end of the discharge box (3-2).

[0029] Furthermore, the humidity sensor 1 (1-2), image acquisition device (1-4), humidity sensor 2 (3-6-3), and humidity sensor 3 (3-3-1) feed back the real-time collected data to the STM32 microcontroller (6). The STM32 microcontroller (6) processes and outputs the Kp, Ki, and Kd parameters of the PID controller (5) through a pre-trained deep learning model, so that the PID controller (5) is in the best working state during the working process. Then, the PID controller (5) controls motor 1 (1-5) and motor 2 (2-5), thereby intelligently controlling the material in a secondary extrusion device based on a deep learning model to achieve optimal energy output and conversion into optimal moisture extrusion.

[0030] The beneficial effects of this invention are as follows:

[0031] 1. The extrusion screw (3-7) with varying cross-section is used to apply varying extrusion pressure to the material during operation. As the material is extruded forward, the moisture is gradually squeezed out, and the pressure on the material gradually increases, so that the moisture in the material can be squeezed out to a certain extent. At the same time, it can also reduce energy consumption: because the varying cross-section can transport the medium more effectively, energy consumption is reduced.

[0032] 2. Through deep learning model, the parameters of PID controller (5) are optimized to keep PID controller (5) in the best working state, realize the control of motor one (1-5) and motor two (2-5) to achieve higher extrusion efficiency, reduce energy consumption, and maximize the extraction of moisture from materials.

[0033] 3. Extend the service life of the extrusion screw (3-7). Automatic adjustment of the extrusion screw speed can avoid excessive wear and extend the service life of the equipment.

[0034] 4. Improve humidity control accuracy: By integrating data from multiple sensors and cameras, the secondary extruder can control humidity more accurately. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the extrusion device structure;

[0036] Figure 2 yes Figure 1 Left side structural diagram;

[0037] Figure 3 This is a schematic diagram of the front structure of the extrusion device;

[0038] Figure 4 yes Figure 3 bottom view

[0039] Figure 5 This is another three-dimensional structural diagram of the extrusion device;

[0040] Figure 6 This is a schematic diagram of the extrusion cylinder (3-6);

[0041] Figure 7 yes Figure 6 Enlarged view of point A in the middle;

[0042] Figure 8 This is a structural schematic diagram of the extrusion screw (3-7);

[0043] Figure 9 yes Figure 8 A three-dimensional image;

[0044] Figure 10 yes Figure 8 Another 3D view;

[0045] Figure 11 This is the circuit diagram of the extrusion device;

[0046] Figure 12 This is a flowchart of the training process for a deep learning model;

[0047] Figure 13 This is a flowchart of the training process for the classification model. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0049] A secondary extrusion device based on a deep learning model, such as Figure 1 As shown, it mainly includes a receiving device (1), a power unit (2), and a secondary extrusion unit (3). The receiving device (1) is located at the top and is used to receive materials from the primary extruder (i.e., a solid-liquid separator). The primary extruder mainly achieves solid-liquid separation, and the content of the material is generally around 78%. The power unit (2) and the secondary extrusion unit (3) are located at the lower end of the receiving device (1). A support (4) is located at the lower end of the power unit (2) and the secondary extrusion unit (3). The support (4) is used to support a secondary extrusion device based on a deep learning model. The power unit (2) is used to provide extrusion power to the secondary extrusion unit (3), which is directly connected to the receiving device (1).

[0050] like Figure 2 and 3 As shown, the receiving device (1) includes a hopper (1-1) and a screw drive (1-3). Multiple humidity sensors (1-2) are evenly arranged inside the hopper (1-1). The screw drive (1-3) is located inside the hopper (1-1) and is used to transport the material extruded once to the unloading section (1-6). An image acquisition device (1-4) for monitoring the material inside the hopper (1-1) is provided at the edge of the hopper (1-1). A motor (1-5) for providing power to the screw drive (1-3) is provided on the outside of the hopper (1-1).

[0051] like Figure 2 As shown, the power unit (2) includes a transmission connecting box (2-1), a transmission box (2-2), and a second motor (2-5). The transmission box (2-2) is connected to the support (4) via a transmission box bracket (2-4). The transmission connecting box (2-1) is connected to the outside of the transmission box (2-2). The lower end of the transmission connecting box (2-1) is connected to the second motor (2-5), which is located at the lower end of the support surface of the support (4). The inner side of the transmission box (2-2) is connected to a shaft connector (2-3) for providing power to the secondary extrusion section (3).

[0052] The secondary extrusion section (3) includes an extrusion box (3-1) and a discharge box (3-2). The right side of the extrusion box (3-1) is connected to a feed cylinder (3-9), and the end of the feed cylinder (3-9) is connected to a receiving hopper (3-10). The receiving hopper (3-10) is connected to the discharge section (1-6) and is used to receive the material transferred from the receiving device (1) to the secondary extrusion section (3). A water outlet (3-4) is provided at the lower end of the extrusion box (3-1) to collect the water squeezed in the extrusion box (3-1).

[0053] The discharge box (3-2) is connected to the extrusion box (3-1), and a discharge baffle (3-3) is provided at the lower end of the discharge box (3-2). Figure 4 As shown, the discharge box (3-2) is provided with a discharge port (3-5) for discharging the extruded material. The discharge port (3-5) is provided with a discharge baffle (3-3) on the outside for discharging the extruded material in a directional manner. Multiple humidity sensors (3-3-1) are provided on the discharge baffle (3-3) for monitoring the moisture content of the material after extrusion. A touch screen (3-8) for manual operation is provided on the outer side of the discharge box (3-2).

[0054] An extrusion cylinder (3-6) is fixedly connected inside the extrusion box (3-1) and the discharge box (3-2). An extrusion screw (3-7) is installed inside the extrusion cylinder (3-6). Figure 6-10 As shown, the extrusion cylinder (3-6) mainly includes an extrusion cylinder body (3-6-1). Multiple water outlet holes (3-6-2) are evenly distributed on the extrusion cylinder body (3-6-1) to discharge the water squeezed out of the material. Multiple humidity sensors (3-6-3) are also evenly distributed between the water outlet holes (3-6-2) to monitor the moisture content of the material in each section of the extrusion cylinder (3-6).

[0055] The extrusion screw (3-7) is located inside the extrusion cylinder (3-6) and is tightly fitted with the extrusion cylinder (3-6). The extrusion screw (3-7) includes a spiral blade (3-7-4) and a screw body (3-7-3) is provided on the spiral blade (3-7-4). One end of the spiral blade (3-7-4) extends to the shaft connector (2-3) to receive power from the power unit (2), and the other end is connected to the top pressing component on the discharge box (3-2). The top pressing component is mainly of two types: electric and hydraulic. It is used to cooperate with the discharge box (3-2) to extrude and discharge the material at the end of the discharge box (3-2).

[0056] The key innovation of this invention lies in making the extrusion screw (3-7) into a screw with equidistant spacing and varying cross-section, such as... Figure 8As shown, the extrusion screw (3-7) is divided into a thin end (3-7-1) and a thick end (3-7-2). The thin end (3-7-1) is connected to the power unit (2), and the thick end (3-7-2) is connected to the top pressure component. The diameter of the thick end (3-7-2) is 1.5-2 times the diameter of the thin end (3-7-1). The thickness D2 of the screw body (3-7-3) near the thick end (3-7-2) is 1.5-2 times the thickness D1 of the screw body (3-7-3) near the thin end (3-7-1). Using such a screw allows for varying extrusion pressure on the material during operation. As the material is extruded forward, moisture is gradually squeezed out, and the pressure on the material gradually increases, thus stabilizing the moisture content to a certain extent. Simultaneously, the changing cross-section allows for more efficient transport of the medium, reducing energy consumption.

[0057] like Figure 11 The diagram shows a structure in which various electrical components of a secondary extrusion device based on a deep learning model are interconnected. Humidity sensor 1 (1-2) and humidity sensor 2 (3-6-3) collect the moisture content of the material in real time and feed the data back to the STM32 microcontroller (6). An image acquisition device (1-4) installed on the hopper (1-1) transmits the collected material image to the STM32 microcontroller (6) in real time. The STM32 microcontroller (6) is equipped with a pre-trained deep learning model, which, based on the deep learning model... The data collected by humidity sensor 1 (1-2), humidity sensor 2 (3-6-3), and image acquisition device (1-4) outputs the Kp, Ki, and Kd parameters of PID controller (5), enabling PID controller (5) to operate in its optimal state during the working process. Then, PID controller (5) controls motor 1 (1-5) and motor 2 (2-5), thereby intelligently controlling the material in the moisture extraction of a deep learning model-based secondary extrusion device, achieving optimal energy output and optimal moisture extraction. Through the above working process, previously, it was necessary to manually adjust the three parameters of the PID controller (proportional term Kp, integral term Ki, and derivative term Kd) to optimize the working state of a deep learning model-based secondary extrusion device. Now, through the above adaptive PID controller, it can automatically adjust the PID parameters according to the dynamic characteristics of the system, and then learn and simulate the behavior of the PID controller through the deep learning model, achieving efficient control of a deep learning model-based secondary extrusion device.

[0058] like Figure 12 As shown, the training process of the pre-trained deep learning model in the STM32 microcontroller (6) is as follows:

[0059] S101 Data Collection: Collect humidity data from humidity sensor 1 (1-2), humidity sensor 3 (3-3-1) and humidity sensor 2 (3-6-3), as well as the Kp, Ki and Kd parameter data of the corresponding PID controller (5), and classify the image results data after the data collected by the image acquisition device (1-4) is classified by the pre-trained classification model.

[0060] S102 Data Preprocessing: The data from step S101 is classified and processed. The humidity data, the Kp, Ki and Kd parameter data of the PID controller (5) are normalized by the maximum and minimum values. The image result data is processed by one-hot encoding.

[0061] S103 Constructing the Neural Network Model: The neural network model used has five layers, including one input layer, three hidden layers, and one output layer; the three hidden layers have 128, 64, and 32 neurons respectively, and the output layer has 3 neurons, corresponding to the Kp, Ki, and Kd parameters of the PID controller (5); S104 Neural Network Training: The data is divided into training set, validation set, and test set. The training objective of the neural network is to minimize the loss function by adjusting the weight values. The loss function is mean squared error (MSE), the activation function is the ReLU function, and stochastic gradient descent (SGD) is used to optimize the model;

[0062] Iterative training: Repeat the steps of forward propagation, loss function calculation, backpropagation, and parameter update.

[0063] This continues until 10,000 iterations are completed or the loss function converges.

[0064] S105 Model Saving: Save the classification model generated after training in step S104 for later use.

[0065] like Figure 13 As shown, the training process of the pre-trained classification model is as follows:

[0066] S201 Image Acquisition: Collects images acquired by the image acquisition device (1-4);

[0067] S202 Image preprocessing: Adjust the images collected in step S201 to the same size, for example, 64x64 pixels;

[0068] Normalize the pixel values ​​in the image to the range [0,1].

[0069] S203 constructs the AlexNet model, and its model structure is as follows:

[0070] AlexNet consists of 5 convolutional layers and 3 fully connected layers;

[0071] ReLU activation function is used between convolutional layers, and Dropout is used between fully connected layers;

[0072] Convolutional layer: Uses convolutional kernels to extract features from the image, using 3x3 convolutional kernels with a stride of 1;

[0073] Pooling layer: Use max pooling or average pooling to reduce image size. Add a max pooling layer with a 2x2 pooling window.

[0074] Fully connected layer: The output of the convolutional layer is flattened into a one-dimensional vector. By adding one or more fully connected layers, it is used for classification. This invention classifies four types of materials: cow dung, pig dung, garbage, and weeds.

[0075] S204 training model:

[0076] The cross-entropy loss function is used to measure the difference between the model's predictions and the actual labels. The optimizer uses stochastic gradient descent (SGD) or other optimizers to update the model parameters.

[0077] Repeat the steps of forward propagation, loss function calculation, back propagation, and parameter update until 10,000 iterations are reached or the loss function converges;

[0078] S205 Save the classification model: Save the model generated in step S204.

[0079] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments.

Claims

1. A secondary extrusion device based on a deep learning model, comprising a receiving device (1), a power unit (2), a secondary extrusion unit (3), a PID controller (5), and an STM32 microcontroller (6), wherein the receiving device (1) is located at the top, the lower end of the receiving device (1) is provided with the power unit (2) and the secondary extrusion unit (3), the lower ends of the power unit (2) and the secondary extrusion unit (3) are provided with a support (4), the secondary extrusion unit (3) is provided with an extrusion cylinder (3-6), the extrusion cylinder (3-6) is installed with an extrusion screw (3-7) for extruding materials, the secondary extrusion unit (3) is provided with a humidity sensor (1-2), a humidity sensor (3-3-1), and a humidity sensor (3-6-3), and an image acquisition device (1-4), characterized in that, The extrusion screw (3-7) is an equidistant screw with a varying cross-section. The extrusion screw (3-7) is divided into a thin end (3-7-1) and a thick end (3-7-2). The thin end (3-7-1) is connected to the power unit (2), and the thick end (3-7-2) is connected to the pressing component of the extrusion device. The diameter of the thick end (3-7-2) is 1.5-2 times the diameter of the thin end (3-7-1). The thickness (D2) of the screw body (3-7-3) near the thick end (3-7-2) is close to... The thickness (D1) of the screw body (3-7-3) at the screw tip (3-7-1) is 1.5-2 times that of the screw body. The STM32 microcontroller (6) is equipped with a pre-trained deep learning model. The deep learning model outputs the Kp, Ki and Kd parameters of the PID controller (5) based on the data collected by the humidity sensor 1 (1-2), humidity sensor 3 (3-3-1) and humidity sensor 2 (3-6-3) and the image acquisition device (1-4), so as to realize that the PID controller (5) is in the best working state during the working process. The training process of the pre-trained deep learning model in the STM32 microcontroller (6) is as follows: S101 Data Collection: Collect humidity data from humidity sensor 1 (1-2), humidity sensor 3 (3-3-1) and humidity sensor 2 (3-6-3), as well as the Kp, Ki and Kd parameter data of the corresponding PID controller (5), and classify the image data collected by the image acquisition device (1-4) using a pre-trained classification model. S102 Data Preprocessing: The data from step S101 is classified and processed. The humidity data and the Kp, Ki and Kd parameter data of the PID controller (5) are normalized by the maximum and minimum values. The image result data is processed by one-hot encoding. The Kp, Ki and Kd parameter data of the PID controller (5) are the target feature data. S103 constructs a neural network model: The neural network model used is a five-layer model, including an input layer, three hidden layers, and an output layer; the three hidden layers have 128, 64, and 32 neurons respectively, and the output layer has 3 neurons, which correspond to the Kp, Ki, and Kd parameters of the PID controller (5) respectively; S104 Neural Network Training: The data is divided into training, validation, and test sets. The training objective of the neural network is to minimize the loss function by adjusting the weight values. The loss function is the mean squared error (MSE), the activation function is the ReLU function, and stochastic gradient descent (SGD) is used to optimize the model. Training is stopped after 10,000 iterations or when the loss function converges. S105 Model Saving: Save the classification model generated after training in step S104.

2. The secondary extrusion device based on a deep learning model according to claim 1, characterized in that: The training process of the pre-trained classification model in step S101 is as follows: S201 Image Acquisition: Collects images acquired by the image acquisition device (1-4); S202 Image preprocessing: Adjust the images collected in step S201 to the same size, and normalize the pixel values ​​in the images to the range of [0, 1]. The S203 builds the AlexNet model, which consists of 5 convolutional layers and 3 fully connected layers. The convolutional layers use the ReLU activation function, and the fully connected layers use Dropout. The convolutional layer uses a 3x3 convolutional kernel with a stride of 1; a corresponding pooling layer is provided after the convolutional layer. S204 training model: The cross-entropy loss function is used to measure the difference between the model prediction and the actual label, and the optimizer uses stochastic gradient descent to update the model parameters; Repeat the steps of forward propagation, loss function calculation, back propagation and parameter update until 10,000 iterations are reached or the loss function converges and training stops. S205 Save the classification model: Save the model generated in step S204.

3. A secondary extrusion device based on a deep learning model according to any one of claims 1-2, characterized in that: The receiving device (1) includes a hopper (1-1) and a screw drive (1-3). Multiple humidity sensors (1-2) are evenly arranged inside the hopper (1-1). The screw drive (1-3) is located inside the hopper (1-1) and is used to transport the material extruded once to the unloading section (1-6). An image acquisition device (1-4) for monitoring the material inside the hopper (1-1) is provided at the edge of the hopper (1-1). A motor (1-5) for providing power to the screw drive (1-3) is provided on the outside of the hopper (1-1).

4. The secondary extrusion device based on a deep learning model according to claim 3, characterized in that: The power unit (2) includes a transmission connecting box (2-1), a transmission box (2-2), and a second motor (2-5). The transmission box (2-2) is connected to the support (4) through a transmission box bracket (2-4). The transmission connecting box (2-1) is connected to the outside of the transmission box (2-2). The lower end of the transmission connecting box (2-1) is connected to the second motor (2-5). The second motor (2-5) is located at the lower end of the support surface of the support (4). The inner side of the transmission box (2-2) is connected to a shaft connector (2-3) for providing power to the secondary extrusion part (3).

5. The secondary extrusion device based on a deep learning model according to claim 4, characterized in that: The secondary extrusion section (3) includes an extrusion box (3-1) and a discharge box (3-2). The right side of the extrusion box (3-1) is connected to a feed cylinder (3-9), and the end of the feed cylinder (3-9) is connected to a receiving hopper (3-10). The receiving hopper (3-10) is connected to the discharge section (1-6) and is used to receive the material conveyed from the receiving device (1) to the secondary extrusion section (3). The lower end of the extrusion box (3-1) is provided with a water outlet (3-4) for collecting the water squeezed in the extrusion box (3-1).

6. The secondary extrusion device based on a deep learning model according to claim 5, characterized in that: The discharge box (3-2) is connected to the extrusion box (3-1). The lower end of the discharge box (3-2) is provided with a discharge baffle (3-3). The discharge box (3-2) is provided with a discharge port (3-5) for discharging the extruded material. The discharge port (3-5) is provided with a discharge baffle (3-3) on its outer side for directional discharge of the extruded material. The discharge baffle (3-3) is provided with multiple humidity sensors (3-3-1) for monitoring the moisture content of the material after extrusion. The outer side of the discharge box (3-2) is provided with a touch screen (3-8) for manual operation.

7. The secondary extrusion device based on a deep learning model according to claim 6, characterized in that: The extrusion box (3-1) and the discharge box (3-2) are equipped with a fixedly connected extrusion cylinder (3-6). The extrusion cylinder (3-6) is equipped with an extrusion screw (3-7). The extrusion cylinder (3-6) mainly includes an extrusion cylinder body (3-6-1). Multiple water outlet holes (3-6-2) are evenly distributed on the extrusion cylinder body (3-6-1) to discharge the water squeezed out of the material. Multiple humidity sensors (3-6-3) are also evenly distributed between the water outlet holes (3-6-2) to monitor the moisture content of the material in each section of the extrusion cylinder (3-6).

8. The secondary extrusion device based on a deep learning model according to claim 7, characterized in that: The extrusion screw (3-7) is located inside the extrusion cylinder (3-6). The extrusion screw (3-7) includes a spiral blade (3-7-4). The spiral blade (3-7-4) is provided with a screw body (3-7-3). One end of the spiral blade (3-7-4) extends to the shaft connector (2-3) to receive power from the power unit (2). The other end is connected to the top pressing component on the discharge box (3-2). The top pressing component is an electric or hydraulic top pressing component, which is used to cooperate with the discharge box (3-2) to extrude and discharge the material at the end of the discharge box (3-2).

9. The secondary extrusion device based on a deep learning model according to claim 8, characterized in that: The humidity sensor 1 (1-2), image acquisition device (1-4), humidity sensor 2 (3-6-3), and humidity sensor 3 (3-3-1) feed back the real-time collected data to the STM32 microcontroller (6). The STM32 microcontroller (6) processes and outputs the Kp, Ki, and Kd parameters of the PID controller (5) through a pre-trained deep learning model, so that the PID controller (5) is in the best working state during the working process. Then, the PID controller (5) controls the motor 1 (1-5) and the motor 2 (2-5), so as to intelligently control the moisture extrusion of the material in the extrusion device, so as to achieve the best energy output and convert it into the best moisture extrusion.

Citation Information

Patent Citations

  • Neural network self-adaptation control algorithm applied to tapered double-screw extruder

    CN103105777A

  • Secondary hydraulic screw extruder for excrement treatment

    CN215403765U

  • Screw press and operation method of screw press

    JP2021159878A

  • Device for removing water from livestock waste and livestock waste processing system using same

    WO2020139026A1