A method for fault diagnosis and monitoring of extruder based on CNN-MHA

Abstract

The present invention provides a method for fault diagnosis and monitoring of an extruder based on CNN-MHA, which comprehensively monitors the health status of the extruder by installing sensors, and transmits data collected by the sensors to a Raspberry Pi, which performs data conversion and transmits the data to the cloud, collects variables related to the operating conditions of the extruder during the model training period as sample data, and processes the sample data through data normalization, feature correlation analysis, feature importance analysis, and ADASYN oversampling method, and then obtains a high-precision optimal diagnosis model after model training, and finally displays the real-time operation data and fault analysis results of the extruder to the user in the form of charts through a Qt integrated fault diagnosis monitoring interface. The present invention realizes intelligent monitoring and high-precision fault diagnosis of an extruder through multi-sensor fusion technology and a CNN-MHA model, and provides intelligent maintenance suggestions, making equipment management more intelligent and automated.

Description

Technical Field

[0001] The invention belongs to the technical field of fault diagnosis, and in particular relates to a method for fault diagnosis and monitoring of an extruder based on CNN-MHA. Background Art

[0002] The feed industry is the material basis for the development of modern animal husbandry and breeding industry. Extrusion puffing technology, with its unparalleled advantages in feed processing, has become one of the fastest-growing new feed processing technologies at home and abroad. At present, the intelligence level of puffing machinery on the market is low. Once a failure occurs, especially a serious machine blockage failure, it is necessary to manually dismantle and clean the puffing chamber to check the cause of the failure. Since the puffing chamber is in a high temperature state, it may cause harm to the operator, posing a great safety risk. In view of the above risks, it is urgent to develop a highly intelligent, stable and reliable puffing machine fault diagnosis and monitoring system to reduce the risk of manual troubleshooting. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a method for fault diagnosis and monitoring of an extruder based on CNN-MHA. Various sensors are deployed on the extruder and in the feed to comprehensively collect the operating parameters of the extruder production. Through in-depth analysis of multi-source information such as the temperature, noise signal, main motor current and vibration signal of key components of the extruder, possible faults of the extruder can be discovered.

[0004] The present invention achieves the above technical objectives through the following technical means.

[0005] A method for fault diagnosis and monitoring of an extruder based on CNN-MHA includes the following steps:

[0006] Step 1: Install various sensors to comprehensively monitor the health status of the extruder;

[0007] Step 2: The signals collected by each sensor are transmitted to the Raspberry Pi, which performs data conversion and transmits the data to the cloud, and collects variables related to the operating conditions of the extruder during the model training period as sample data;

[0008] Step 3: Sample data processing: Label the collected sample data, divide it into a test set and a training set according to the preset fault type and a ratio of 7:3, and then process the sample data through data normalization, feature correlation analysis, feature importance analysis, and ADASYN oversampling method in turn;

[0009] Step 4: Model training and testing;

[0010] Step 5: Real-time data analysis;

[0011] Step 6: Qt interface integration.

[0012] Furthermore, in the step 1, the installation of the sensor includes: placing a temperature and humidity sensor in the feed raw material to collect temperature and humidity data of the feed raw material; placing an environmental temperature and humidity sensor in the middle of the extruder base to collect temperature and humidity data of the environment; placing a noise sensor near the cutter to collect noise data when the extruder is running; installing a single-phase current sensor at the connection between the main motor and the electric control cabinet to collect single-phase current data of the main motor; installing a weighing sensor at the product discharge port to collect weight data of the product; installing a temperature sensor at the wall of the heating chamber of the extruder to collect temperature data in the heating chamber of the extruder; and installing a vibration speed sensor on the side bracket of the feed port to collect vibration speed data of the extruder.

[0013] Furthermore, in step 3, the Pearson correlation coefficient is used to perform feature correlation analysis and the feature importance is calculated by the SHAP value, thereby screening the optimal features for extruder fault diagnosis, as follows:

[0014]

[0015]

[0016] Where r is the Pearson correlation coefficient; X and Y are two features, and r is used to measure the correlation between X and Y; X i and Y i is the i-th sample value of features X and Y; and is the sample mean of features X and Y; φ j is the SHAP value of feature j, indicating the importance of the feature; N is the set of features, |N|! is the factorial of the feature set N; S is the subset that does not contain feature j, f(S) is the model prediction output given the feature subset S, |S| is the number of elements in the set S, and |S|! represents the factorial of the set S.

[0017] Furthermore, in step 3, the ADASYN oversampling method is used to perform data enhancement processing on the sample data to solve the problem of data imbalance in the sample data, as follows:

[0018]

[0019] G m =r m ×G

[0020]

[0021] Among them, for the minority class sample x m , calculate the number of samples k, r belonging to the minority class among its k nearest neighborsm Represents the minority class sample x m The density distribution of G is the number of minority class samples that need to be generated overall, G m It means that in a specific minority class sample x m The number of synthetic samples that need to be generated in the local area; for each minority class sample x m , randomly select one of its k nearest neighbors x zm , and generate new samples σ is a random number between [0,1].

[0022] Furthermore, the specific process of step 4 is as follows:

[0023] A CNN-MHA model consisting of five convolutional layers and one multi-head attention layer was constructed. The learning rate was adjusted through experiments, and hyperparameters such as the number of iterations, batch size, and optimizer were selected. Multiple training and testing were performed. During the model training process, the cross-entropy loss function was used, and the training data was input into the model in batches in each iteration. The loss value was calculated and the model parameters were updated. The training loss, validation loss, and model convergence were recorded. After the training was completed, the model performance was evaluated on the test set, the cross-entropy loss value on the test set was calculated, and the F1 score was used to comprehensively evaluate the classification ability of the model. According to the test results, the model structure and hyperparameters were further adjusted to optimize the model performance, and finally a high-precision optimal fault diagnosis model was obtained.

[0024] Furthermore, the specific process of step 5 is as follows:

[0025] The real-time data collected in the cloud is first preprocessed and then input into the optimal fault diagnosis model pre-trained in step 4. The model analyzes the input data in real time, identifies the operating status of the equipment, and detects whether there are abnormalities or potential faults. Once a fault is detected, the model further analyzes the fault type and classifies it into a specific fault category based on the model's training data and preset fault types. For the detected fault type, the model automatically generates corresponding maintenance suggestions.

[0026] Furthermore, the specific process of step 6 is as follows:

[0027] The user interface is developed using the Qt framework, and the real-time data of the extruder operation is presented in the form of charts through the user interface. The detected faults are marked with corresponding colors and icons, and a fault analysis report is generated, including the fault type, occurrence time, possible causes and specific maintenance suggestions. At the same time, when a fault or abnormal situation is detected, the user interface automatically pops up a prompt box and sounds an alarm.

[0028] The present invention has the following beneficial effects:

[0029] The present invention realizes intelligent monitoring and high-precision fault diagnosis of the extruder through multi-sensor fusion technology and CNN-MHA model. The system can comprehensively collect and analyze the key data of the equipment, identify and classify the fault type in real time, and provide intelligent maintenance suggestions. Through data enhancement and model optimization, the system improves the accuracy of fault detection and the efficiency of equipment management. The user-friendly interface further enhances the operating experience, makes equipment management more intelligent and automated, and significantly improves the safety, stability and service life of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a framework diagram of the extruder fault diagnosis and monitoring method based on CNN-MHA according to the present invention;

[0031] Figure 2 This is the main view of the extruder data collection;

[0032] Figure 3 This is the structure diagram of the CNN-MHA neural network;

[0033] In the figure: 1. Feed temperature and humidity sensor; 2. Ambient temperature and humidity sensor; 3. Noise sensor; 4. Main motor single-phase current sensor; 5. Weighing sensor; 6. Cavity temperature sensor; 7. Vibration velocity sensor. DETAILED DESCRIPTION

[0034] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0035] Reference Figure 1 The CNN-MHA-based extruder fault diagnosis and monitoring method of the present invention comprises the following process:

[0036] Step 1: Sensor installation;

[0037] like Figure 2 As shown, a temperature and humidity sensor (1) is placed in the feed raw material to collect temperature and humidity data of the feed raw material; an environmental temperature and humidity sensor (2) is placed in the middle of the extruder base to collect temperature and humidity data of the environment; a noise sensor (3) is placed near the cutter to collect noise data when the extruder is running; a single-phase current sensor (4) is installed at the connection between the main motor and the electric control cabinet to collect single-phase current data of the main motor; a weighing sensor (5) is installed at the product discharge port to collect weight data of the product; a temperature sensor (6) is installed at the wall of the heating chamber of the extruder to collect temperature data in the heating chamber of the extruder; and a vibration speed sensor (7) is installed on the side bracket of the feed port to collect vibration speed data of the extruder.

[0038] Step 2: Data collection;

[0039] The signals collected by the sensors installed in step 1 are transmitted to the Raspberry Pi, which performs data conversion and transmits the data to the cloud through the wifi module. The variables related to the operating conditions of the extruder during the model training period are collected as sample data, including the temperature in the extruder chamber, feed temperature and humidity, ambient temperature and humidity, vibration signal, noise signal, main motor current, product weight, and feeding speed. The sampling interval of the sensor is 1 second.

[0040] Step 3: Data processing;

[0041] The collected sample data are labeled and divided into a test set and a training set according to the preset fault type and a ratio of 7:3. The specific data set composition is shown in Table 1:

[0042] Table 1 Composition of experimental dataset

[0043]

[0044]

[0045] In order to maintain the consistency of all data, the collected sample data is normalized, that is:

[0046]

[0047] Among them, for the original data x in the data set; x norm is the normalized data of the original data x, and its value range is [0, 1]; x max 、x min The maximum and minimum values ​​in the data set are

[0048] Then, the Pearson correlation coefficient was used to analyze the feature correlation and the SHAP value was used to calculate the feature importance. This method was used to screen the optimal features for extruder fault diagnosis, specifically:

[0049]

[0050]

[0051] Where r is the Pearson correlation coefficient; X and Y are two features, and r is used to measure the correlation between them; X i and Y i is the i-th sample value of features X and Y; and is the sample mean of features X and Y; φ jis the SHAP value of feature j, indicating the importance of the feature; N is the set of features, |N|! is the factorial of the feature set N; S is the subset that does not contain feature j, f(S) is the model prediction output given the feature subset S, |S| is the number of elements in the set S, and |S|! represents the factorial of the set S. The above formula indicates that for feature j, its SHAP value is the weighted average of the Shapley values ​​corresponding to the changes in the model prediction output under different combinations of feature subsets S.

[0052] Since the collected fault sample data has the characteristics of data imbalance, the ADASYN oversampling method is used to enhance its data, specifically:

[0053]

[0054] G m =r m ×G

[0055]

[0056] Among them, for the minority class sample x m , calculate the number of samples k, r belonging to the minority class among its k nearest neighbors m Represents the minority class sample x m The density distribution of G is the number of minority class samples that need to be generated overall. G in the formula m It means that in a specific minority class sample x m The number of synthetic samples that need to be generated in the local area; for each minority class sample x m , randomly select one of its k nearest neighbors x zm , and generate new samples σ is a random number between [0, 1].

[0057] Step 4: Model training and testing;

[0058] like Figure 3As shown in the figure, a CNN-MHA model consisting of five convolutional layers and one multi-head attention layer is constructed. The learning rate is adjusted through experiments, the appropriate number of iterations is selected to ensure that the model is fully trained while avoiding overfitting and underfitting, and the appropriate batch size is selected to balance computational efficiency and model performance. Different optimizers are tested to find the optimization algorithm that best suits the current model. During the model training process, the cross entropy loss function is used, and the training data is input into the model in batches in each iteration. The loss value is calculated and the model parameters are updated. The training loss and validation loss are recorded, and the model convergence is confirmed. After the training is completed, the model performance is evaluated on the test set, the cross entropy loss value on the test set is calculated, and the classification ability of the model is comprehensively evaluated using the F1 score. According to the test results, the model structure and hyperparameters are further adjusted to optimize the model performance, and finally a high-precision optimal fault diagnosis model is obtained.

[0059] Figure 3 In the figure, Convolution 3x3 means 3x3 convolution; Max Pooling 2x2 means 2x2 maximum pooling; Dropout 0.3 means 0.3 dropout unit; Flatten layer means flattening layer.

[0060] Step 5: Real-time data analysis;

[0061] The real-time data collected in the cloud is first preprocessed and then input into the pre-trained optimal fault diagnosis model. The model analyzes the input data in real time, identifies the operating status of the equipment, and detects whether there are abnormalities or potential faults. Once a fault is detected, the model will further analyze the fault type and classify it into a specific fault category based on the model's training data and preset fault types to ensure the accuracy and reliability of the classification. For the detected fault type, the system automatically generates corresponding maintenance suggestions.

[0062] Step 6: Qt interface integration;

[0063] The user interface is developed using the Qt framework. The user interface can present real-time data in the form of a line chart, clearly showing the changing trends of key parameters such as temperature, current, and vibration, so that users can monitor the operating status of the equipment at a glance and discover potential problems in a timely manner. For detected faults, the system marks them with eye-catching colors and icons, and generates a detailed fault analysis report, including the fault type, occurrence time, possible causes, and specific maintenance suggestions. The user interface design focuses on user experience, adopts a modern interface style and a simple layout, allowing users to easily navigate and operate. Users can adjust the chart view as needed to view data details for different time periods. In addition, the system integrates a real-time notification function. Once a fault or abnormality is detected, the user interface will pop up a prompt box and sound an alarm to ensure that users will not miss any important information. The user interface developed through the Qt framework not only intuitively displays the operating data and fault analysis results of the extruder, but also provides a friendly interactive method and efficient operation experience, allowing users to fully grasp the equipment status, perform maintenance and troubleshooting in a timely manner, and improve the intelligence and automation level of equipment management.

[0064] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A method for fault diagnosis and monitoring of an extruder based on CNN-MHA, characterized in that: The process includes the following: Step 1: Install various sensors to comprehensively monitor the health status of the extruder; Step 2: The signals collected by each sensor are transmitted to the Raspberry Pi, which performs data conversion and transmits the data to the cloud, and collects variables related to the operating conditions of the extruder during the model training period as sample data; Step 3: Sample data processing: Label the collected sample data, divide it into a test set and a training set according to the preset fault type and a ratio of 7:3, and then process the sample data through data normalization, feature correlation analysis, feature importance analysis, and ADASYN oversampling method in turn; Step 4: Model training and testing; Step 5: Real-time data analysis; Step 6: Qt interface integration; In step 3, the feature correlation analysis is performed using the Pearson correlation coefficient and the feature importance is calculated using the SHAP value, thereby screening the optimal features for extruder fault diagnosis, as follows: Where r is the Pearson correlation coefficient; X and Y are two features, and r is used to measure the correlation between X and Y; X i and Y i is the i-th sample value of features X and Y; and is the sample mean of features X and Y; φ j is the SHAP value of feature j, indicating the importance of the feature; N is the set of features, |N|! is the factorial of the feature set N; S is the subset that does not contain feature j, f(S) is the model prediction output given the feature subset S, |S| is the number of elements in the set S, and |S|! represents the factorial of the set S; In step 3, the ADASYN oversampling method is used to perform data enhancement processing on the sample data to solve the data imbalance characteristic of the fault sample data, as follows: G m =r m ×G Among them, for the minority class sample x m , calculate the number of samples k belonging to the minority class among its k nearest neighbors m , r m Represents the minority class sample x m The density distribution of G is the number of minority class samples that need to be generated overall, G m It means that in a specific minority class sample x m The number of synthetic samples that need to be generated in the local area; for each minority class sample x m , randomly select one of its k nearest neighbors x zm , and generate new samples σ is a random number between [0,1]; The specific process of step 4 is as follows: A CNN-MHA model consisting of five convolutional layers and one multi-head attention layer was constructed. The learning rate was adjusted experimentally, and hyperparameters such as the number of iterations, batch size, and optimizer were selected. Multiple training and testing were performed. During the model training process, the cross-entropy loss function was used, and the training data was input into the model in batches in each iteration. The loss value was calculated and the model parameters were updated. The training loss, validation loss, and model convergence were recorded. After the training was completed, the model performance was evaluated on the test set, the cross-entropy loss value on the test set was calculated, and the F1 score was used to comprehensively evaluate the classification ability of the model. Based on the test results, the model structure and hyperparameters were further adjusted to optimize the model performance, and finally a high-precision optimal fault diagnosis model was obtained. The specific process of step 5 is as follows: The real-time data collected in the cloud is first preprocessed and then input into the optimal fault diagnosis model pre-trained in step 4. The model analyzes the input data in real time, identifies the operating status of the equipment, and detects whether there are abnormalities or potential faults. Once a fault is detected, the model further analyzes the fault type and classifies it into a specific fault category based on the model's training data and preset fault types. For the detected fault type, the model automatically generates corresponding maintenance suggestions.

2. The method for fault diagnosis and monitoring of an extruder based on CNN-MHA according to claim 1, characterized in that: In the step 1, the installation of the sensor includes: placing a temperature and humidity sensor (1) in the feed raw material to collect temperature and humidity data of the feed raw material; placing an environmental temperature and humidity sensor (2) in the middle of the base of the extruder to collect temperature and humidity data of the environment; placing a noise sensor (3) near the cutter to collect noise data when the extruder is running; installing a single-phase current sensor (4) at the connection between the main motor and the electric control cabinet to collect single-phase current data of the main motor; installing a weighing sensor (5) at the product discharge port to collect weight data of the product; installing a temperature sensor (6) at the wall of the heating chamber of the extruder to collect temperature data in the heating chamber of the extruder; and installing a vibration speed sensor (7) on the side bracket of the feed port to collect vibration speed data of the extruder.

3. The method for fault diagnosis and monitoring of an extruder based on CNN-MHA according to claim 1, characterized in that: The specific process of step 6 is as follows: The user interface is developed using the Qt framework, and the real-time data of the extruder operation is presented in the form of charts through the user interface. The detected faults are marked with corresponding colors and icons, and a fault analysis report is generated, including the fault type, occurrence time, possible causes and specific maintenance suggestions. At the same time, when a fault or abnormal situation is detected, the user interface automatically pops up a prompt box and sounds an alarm.

Images