A water suspension granulation process state recognition method based on a convolutional neural network

Abstract

The present application belongs to the field of water suspension granulation process control, and discloses a water suspension granulation process state recognition method based on a convolutional neural network; relevant image data is sorted and collected, data is preprocessed, and the model is trained; a convolutional neural network model is constructed, a trainer is configured, and a network model is trained; the acc curve and the loss curve of the training set and the verification set of the model are visualized, and the model training effect is judged. The water suspension granulation process state recognition method based on the convolutional neural network can realize the recognition of the water suspension granulation process state, and the classification is accurate; blind trial and error are avoided, the safety of the research process is improved, and a basis is provided for the unmanned and standardized preparation of the pressed explosive molding powder. The present application can get rid of excessive dependence on experienced production line workers, reduce the training investment for new people, and improve the efficiency in the molding powder preparation process.

Description

Technical Field

[0001] This invention relates to a method for identifying the state of a water suspension granulation process based on a convolutional neural network, and particularly to the identification of the state of a water suspension granulation process for press-loaded mixed explosives, belonging to the field of water suspension granulation process control. Background Technology

[0002] Currently, water suspension granulation technology is a common method for preparing molded powder for press-fitted mixed explosives. Water suspension granulation is essentially a process in which explosive powder gradually becomes wetted, nucleates, agglomerates, compacts, and breaks up under the action of a polymer binder. Many factors influence particle properties, such as the structure of the reaction vessel, stirring speed, feeding rate, temperature, and heating rate. To date, our understanding of the influence of these factors on the molded powder is limited, and the granulation process relies heavily on manual experience, requiring timely adjustments to process conditions based on various states. However, relying on manual identification of the various states during molded powder preparation involves long training periods, consumes significant financial, human, and material resources, is uneconomical, and poses safety risks; these problems seriously hinder the preparation of molded powder for press-fitted mixed explosives.

[0003] To avoid the above problems, a state recognition method for water suspension granulation process based on convolutional neural network is proposed. This method has the characteristics of high accuracy, no dependence on manual labor, high safety and good economy, and has been applied to the state recognition of the preparation process of NTO-based press-fit mixed explosive powder.

[0004] Convolutional Neural Networks (CNNs) are a class of feedforward neural networks that incorporate convolutional computations and have a deep structure. They are one of the representative algorithms in deep learning. CNNs have long been a core algorithm in the field of image recognition, often used to extract discriminative features from images for other classifiers to learn from.

[0005] This patent uses a large number of images of powder and granulation in two different states during the water suspension granulation process as input and output. It uses a convolutional neural network for supervised learning training to obtain a granulation process state recognition and prediction model with high accuracy, laying the foundation for unmanned research on water suspension granulation technology. Summary of the Invention

[0006] The purpose of this invention is to solve the problem that the observation of granulation state during the preparation of press-fit mixed explosive granulation powder is extremely dependent on manual labor, which is time-consuming and labor-intensive. This invention provides a state recognition method for the water suspension granulation process based on convolutional neural networks. This method creatively introduces convolutional neural networks into the research on the preparation and control of press-fit mixed explosive granulation powder, and proposes the idea and solution of replacing manual recognition with image recognition, which provides the possibility for realizing unmanned and standardized preparation of water suspension granulation powder.

[0007] The objective of this invention is achieved through the following technical solution.

[0008] A method for state recognition in water suspension granulation process based on convolutional neural networks, the specific steps of which are as follows:

[0009] Step 1: Organize and collect relevant image data, dividing the data into a training set (train) and a validation set (validation). 80% of the data will be used for training, and 20% for testing.

[0010] Step 2: Using the TensorFlow framework. First, preprocess the data. The training and validation sets from Step 1 are preprocessed separately. After the data is loaded into the generator, it is automatically converted to tensor (float32) format, dynamically generating the required batch data. Then, the image data is normalized to the (0-1) interval. The compressed image size is 64*64 for easier model training.

[0011] Step 3: Build a convolutional neural network model, configure the trainer, and train the network model.

[0012] Step 4: Demonstrate Results. Visualize the accuracy and loss curves of the model's training and validation sets to assess the training effectiveness. If the results are unsatisfactory, adjust the model parameters and retrain. If the results meet the requirements, save the model for future experiments.

[0013] Beneficial effects

[0014] 1. The present invention provides a method for identifying the state of a water suspension granulation process based on a convolutional neural network, which can identify the state of the water suspension granulation process and classify it accurately.

[0015] 2. The present invention provides a state recognition method for water suspension granulation process based on convolutional neural networks. This method extracts feature information from images using convolutional neural networks, making it less prone to overfitting and highly accurate. It can save significant experimental time, avoid blind trial and error, improve the safety of the research process, and accelerate the preparation of pressed mixed explosive molding powder, providing a foundation for the unmanned and standardized preparation of pressed mixed explosive molding powder.

[0016] 3. The water suspension granulation process state recognition method based on convolutional neural network of the present invention can get rid of the over-reliance on experienced production line workers, reduce the training investment for new people, and improve the efficiency and safety of the molding powder preparation process. Attached Figure Description

[0017] Figure 1 This is a flowchart of the state recognition process in this invention;

[0018] Figure 2This is a diagram of the convolutional neural network structure in this invention;

[0019] Figure 3 This is a graph showing the loss curves of the training set and validation set in this invention;

[0020] Figure 4 This is the acc curve diagram of the training set and validation set in this invention. Detailed Implementation

[0021] This invention provides a method for state recognition in a water suspension granulation process based on a convolutional neural network, such as... Figure 1 As shown, the specific steps are as follows:

[0022] Step 1: Based on the existing formula and process, NTO explosive molding powder was prepared using water suspension granulation technology. During the preparation process, images of the process status were collected, including 304 images of the powder state and 268 images of the granules state. The 264 powder state images and 208 granule state images (approximately 80%), and the 60 powder state images and 60 granule state images (approximately 20%), were respectively used as the training set and the validation set.

[0023] Step two involves using the Python os library to load image data, preprocess the image data, convert the image format, normalize the image so that the pixel values ​​are distributed between 0 and 1, and compress the image to improve the speed of model training.

[0024] Step 3: Construct a convolutional neural network model and configure the trainer. The structure of the convolutional neural network is as follows: Figure 2 As shown, the input image size is 64*64*3, the kernel size of the first convolutional layer is 3*3, and to obtain 32 feature maps, the activation layer uses the ReLU function. After one convolutional layer, 32 feature maps of size 62*62*3 are obtained, and their calculation formula is as follows:

[0025]

[0026] Where h and w are the height and width of the image, respectively (both 64 in this example), f is the kernel size (3 in this example), p is the padding size (no padding in this example), and s is the stride (1 in this example). Next, pooling is performed, resulting in 32 31*31*3 feature maps. The third convolutional layer has a 3*3 kernel, and the connected activation layer uses the ReLU function to obtain 64 29*29*3 feature maps. Then, pooling is performed again to obtain 64 14*14*3 feature maps. The fifth convolutional layer has a 3*3 kernel, and the connected activation layer uses the ReLU function to obtain 128 12*12*3 feature maps. Then, pooling is performed again to obtain 128 6*6*3 feature maps. Finally, a straightening operation is performed, and the image is fed into a fully connected layer for binary classification.

[0027] Step four: Analyze the loss curves and accuracy curves of the model's training and validation sets, such as... Figure 3 and Figure 4 As can be seen, the model training effect is very good.

[0028] This invention uses a state recognition method for water suspension granulation process based on convolutional neural networks. It is simple and convenient to operate, and the prediction results are accurate, reliable and repeatable.

[0029] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for state recognition in a water suspension granulation process based on a convolutional neural network, characterized in that: The specific steps are as follows: Step 1: Organize and collect relevant image data, and divide the data into a training set (train) and a validation set (validation), with 80% being the training set and 20% being the test set. Step 2: Using the TensorFlow framework, the data is first preprocessed. The training and validation sets from Step 1 are preprocessed separately. After the data is loaded into the generator, it is automatically converted to tensor float32 format, dynamically generating the required batch data. Then, the image data is normalized to the (0-1) interval, and the image size is compressed to 64. 64, which facilitates model training; Step 3: Build a convolutional neural network model, configure the trainer, and train the network model; Step 4: Demonstrate the results. Visualize the accuracy and loss curves of the model's training and validation sets to assess the training effect. If the effect is not ideal, continue to adjust the model parameters and retrain. If the effect meets the requirements, save the model for subsequent experiments. The kernel size of the first convolutional layer of the constructed convolutional neural network is 3.

3. To obtain 32 feature maps, the activation layers use the ReLU function; after one convolutional layer, we get 32 ​​62-bit feature maps. 62 The feature map of size 3 is then pooled to obtain 32 31-bit feature maps. 31 Feature map of size 3; kernel size of the third convolutional layer is 3.

3. The activation layer uses the ReLU function to obtain 64 29-bit cells. 29 3 Feature maps; then perform pooling of 64 x 14 pools. 14 3 feature maps; the kernel size of the fifth convolutional layer is 3.

3. The activation layer uses the ReLU function to obtain 128 x 12^n ... 12 3. Feature map; then perform pooling 128 x 6 = 128. 6 3. Feature map; finally, straighten the image and feed it into a fully connected layer for binary classification.

2. The method for state recognition in a water suspension granulation process based on a convolutional neural network according to claim 1, characterized in that: It can accurately identify the powder and granulation state during the water suspension granulation process of press-filled mixed explosives.

Images