A motor fault diagnosis method for cigarette machine dust removal equipment based on deep transfer learning

Abstract

This application relates to the field of mechanical condition monitoring, and specifically discloses a method for diagnosing motor faults in dust removal equipment of cigarette making machines based on deep transfer learning. The method collects vibration and temperature signals of motor equipment in dust removal workshops in real time, performs data preprocessing, and uses convolutional neural networks to construct a transfer learning model for feature extraction and training to predict vibration prediction results. prd At the same time, the temperature warning model M is constructed by performing secondary dynamic threshold judgment on the pre-processed motor temperature data. temp , and perform diagnosis to obtain temperature warning results t prd The present invention uses vibration prediction and temperature early warning models to diagnose motor faults and push status levels and fault information to the client, thereby improving equipment management efficiency and operational stability.

Description

Technical Field

[0001] The present application relates to the field of mechanical condition monitoring, and more specifically, to a method for diagnosing motor faults in dust removal equipment of cigarette making machines based on deep transfer learning. Background Art

[0002] In the tobacco manufacturing industry, the widespread use and frequent operation of rotating machinery such as motors and fans make it particularly important to monitor the mechanical health of these devices.

[0003] To ensure the reliability of these critical components, vibration monitoring is often employed. Advanced signal processing and spectrum analysis techniques can accurately assess the operating status of these devices. This vibration monitoring technology is widely used in the tobacco industry, including but not limited to dust removal systems, cigarette production lines, and packaging equipment. These monitoring methods are crucial for maintaining continuous production line operation and ensuring operational safety.

[0004] However, traditional signal processing and spectrum analysis typically require maintenance personnel to possess a high level of expertise and experience in order to correctly interpret vibration data and spectrum analysis results. This limits the widespread adoption and application of monitoring technology, as not all maintenance personnel possess this expertise.

[0005] With the continuous advancement of machine learning and data mining technologies, a new, data-driven fault diagnosis method is becoming mainstream. This method reduces the need for highly skilled maintenance personnel, making the fault diagnosis process more efficient and easier to operate. It has been widely used in industrial production, especially in online equipment monitoring.

[0006] However, data collection in industrial environments often faces challenges, such as a lack of sufficient negative samples and the data is often unlabeled, which is a problem for machine learning and deep learning models that rely on large amounts of labeled data.

[0007] Therefore, a motor fault diagnosis method for dust removal equipment of a cigarette making machine based on deep transfer learning is provided. Summary of the Invention

[0008] This application is proposed to address the above-mentioned technical problems. It provides a method for diagnosing motor faults in cigarette machine dust removal equipment based on deep transfer learning. This method utilizes the vibration and temperature signals of the motor of the cigarette machine dust removal equipment for data analysis and modeling. This method adapts a data model trained based on known domain vibration data to the fault diagnosis of the motor of the cigarette machine dust removal equipment. It also incorporates real-time early warning data of the motor temperature to improve the accuracy of fault prediction.

[0009] Specifically, according to one aspect of the present application, a method for diagnosing motor faults in a dust removal device of a cigarette making machine based on deep transfer learning is provided, which includes:

[0010] S1. Collecting vibration signals and temperature signals of the motor equipment in the dust removal workshop in real time during operation, and storing the vibration signals and temperature signals in a local database respectively;

[0011] S2. performing data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and performing data preprocessing operations for null values ​​and abnormal values ​​on the temperature signal;

[0012] S3. Use convolutional neural network as the core architecture of the transfer learning model. In addition to the output layer, define a 6-layer neural network as the feature extraction layer of the transfer learning network;

[0013] S4, training the transfer learning model to obtain a prediction model M T ;

[0014] S5, using the prediction model M T Predict the vibration data and get the vibration prediction result y prd ;

[0015] S6, perform secondary dynamic threshold judgment on the pre-processed motor temperature data and build a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd ;

[0016] S7, based on the vibration prediction result y prd and the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results;

[0017] Wherein, said S4 includes:

[0018] S41. Divide the source domain dataset D s , target domain dataset D T , for the source domain dataset D s , the target domain dataset D T Perform data preprocessing operations respectively;

[0019] S42. Set the initial network parameters for model training, wherein the learning rate ∈=1e-3 is set, the Step method is selected as the learning rate adjustment strategy, and the SGD method is selected as the optimization strategy;

[0020] S43, the source domain dataset D sInput the transfer learning model and use batch normalization technology to accelerate model convergence. After repeated iterations, the optimal pre-training model M is obtained. S , and save it locally;

[0021] S44, the target domain dataset D T Input pre-trained model M S , and use batch normalization technology to fine-tune the model, and after repeated iterations, obtain the prediction model M that adapts to the target domain data T .

[0022] Optionally, the S2 includes: S21, resampling the vibration signal, eliminating missing values, and setting the batch size to 128; S22, adjusting the sliding window length L and overlap rate r of the resampled data, wherein the sliding window length L = 128 and the overlap rate r = 0.5; S23, applying Z-Score normalization to the resampled vibration signal to eliminate dimensionality effects and accelerate model convergence, wherein the Z-Score calculation formula is as follows:

[0023]

[0024] Among them, X scaler is the standardized vibration data, X is the resampled vibration signal, μ is the mean of the data under this feature dimension, and σ is the standard deviation of the data under this feature; S24, remove abnormal values ​​and null values ​​from the temperature signal.

[0025] In particular, the S3 includes: introducing a customized linear layer into the output layer as a classification layer, and using the Dropout technology to reduce the overfitting phenomenon of the model.

[0026] Optionally, the S41 includes: S411, using a vibration test bench equipped with a vibration temperature sensor to collect vibration data of a motor using a faulty bearing and a normal bearing, and using the obtained labeled data as a source domain dataset D S S412, install wireless vibration sensors on the motor equipment in the cigarette dust removal workshop, collect the vibration data of the motor in the new field within a certain period of time, and store the data in the database as the target domain data set D T ; S413, the source domain dataset D S , the target domain dataset D T Execute step S2 to process the data into a format suitable for the neural network.

[0027] Optionally, the S7 includes: the motor status level includes a third-level state, a second-level state and a first-level state.

[0028] Optionally, S7 includes: if the vibration warning and temperature warning appear at the same time, pushing the third-level status to the client; if there is only a vibration or temperature warning, pushing the second-level status to the client; if there is no warning, pushing the first-level status to the client, and displaying the equipment fault diagnosis information according to the prediction results.

[0029] Specifically, according to another aspect of the present application, a motor fault diagnosis system for dust removal equipment of a cigarette making machine based on deep transfer learning is provided, which includes:

[0030] The motor data acquisition module is used to collect the vibration signal and temperature signal of the motor equipment in the dust removal workshop in real time during operation, and store the vibration signal and temperature signal in the local database respectively;

[0031] A motor data preprocessing module, configured to perform data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and perform data preprocessing operations for null values ​​and abnormal values ​​on the temperature signal;

[0032] The transfer learning model building module is used to use the convolutional neural network as the core architecture of the transfer learning model. In addition to the output layer, a 6-layer neural network is defined as the feature extraction layer of the transfer learning network;

[0033] The transfer learning model training module is used to train the transfer learning model to obtain the prediction model M T ;

[0034] Motor vibration prediction module, used to use the prediction model M T Predict the vibration data and get the vibration prediction result y prd ;

[0035] The motor temperature warning module is used to perform secondary dynamic threshold judgment on the pre-processed motor temperature data and build a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd ;

[0036] The motor fault prediction module is used to predict the motor fault based on the vibration prediction result y prd and the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results;

[0037] The transfer learning model training module includes:

[0038] Divide the source domain dataset D s , target domain dataset D T , for the source domain dataset Ds , the target domain dataset D T Perform data preprocessing operations respectively;

[0039] Set the initial network parameters for model training, where the learning rate is set to ∈ = 1e-3, the Step method is selected for the learning rate adjustment strategy, and the SGD method is selected for the optimization strategy;

[0040] The source domain dataset D s Input the transfer learning model and use batch normalization technology to accelerate model convergence. After repeated iterations, the optimal pre-training model M is obtained. S , and save it locally;

[0041] The target domain dataset D T Input pre-trained model M S , and use batch normalization technology to fine-tune the model, and after repeated iterations, obtain the prediction model M that adapts to the target domain data T .

[0042] Optionally, the motor fault prediction module includes: the motor state level includes a third state, a second state and a first state.

[0043] Compared with the prior art, the present application provides a method for diagnosing motor faults in dust removal equipment for cigarette making machines based on deep transfer learning. This method uses a pre-trained model in the source domain to perform feature extraction and preliminary classification when the target domain data set lacks sufficient labels, effectively addressing the problem of training prediction models caused by the scarcity of field data labels. Moreover, by using transfer learning, the model can learn common features from the source domain data and apply these features to the target domain, so that it can maintain a high recognition accuracy under variable working conditions, thereby improving the model's cross-working condition recognition capability. In addition, the method further enhances the model's prediction accuracy and reliability under diverse operating conditions by combining the motor temperature warning results. This method does not require high technical skills from the operator, supports online automatic monitoring, can quickly identify potential equipment problems, reduces the need for manual on-site inspections, saves human resources, and promotes the construction of digital factories, improving the management efficiency and quality of dust removal equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0045] Figure 1The figure illustrates a flow chart of a prediction model for a motor fault diagnosis method for a dust removal device of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0046] Figure 2 The figure illustrates a flow chart of a method for diagnosing motor faults in a dust removal device of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0047] Figure 3 The figure illustrates a flowchart of step S2 in the motor fault diagnosis method of the dust removal equipment of the cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0048] Figure 4 The figure illustrates a neural network architecture diagram of a method for diagnosing motor faults in a dust removal device of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0049] Figure 5 The figure illustrates a flowchart of step S4 in the method for diagnosing motor faults of dust removal equipment of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0050] Figure 6 The figure illustrates a flowchart of step S41 in step S4 of the method for diagnosing motor faults of dust removal equipment of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0051] Figure 7 The figure illustrates a training loss curve of a motor fault diagnosis method for a dust removal device of a cigarette making machine based on deep transfer learning according to an embodiment of the present application.

[0052] Figure 8 The figure illustrates the accuracy change trend of the cigarette machine dust removal equipment motor fault diagnosis method based on deep transfer learning according to an embodiment of the present application.

[0053] Figure 9 The figure shows a schematic diagram of a motor fault diagnosis system for dust removal equipment of a cigarette making machine based on deep transfer learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] With the continuous advancement of transfer learning technology, it is now possible to initially train a model using existing labeled data, then apply it to new datasets and fine-tune the model to improve its fault diagnosis capabilities for this new data. This approach has demonstrated its effectiveness in multiple engineering projects, particularly in improving the model's predictive accuracy for new equipment data and its ability to adapt to different operating conditions.

[0055] Transfer learning technology offers an effective solution to the data collection challenges faced by tobacco machinery, such as the difficulty in obtaining data, a lack of negative examples, and the ever-changing operating conditions of the equipment. By building a data-driven motor fault diagnosis model, the status of key tobacco machinery equipment can be effectively monitored, thereby improving management efficiency and stability. This technology has broad application prospects in the tobacco industry.

[0056] Therefore, a motor fault diagnosis method for dust removal equipment of a cigarette making machine based on deep transfer learning is proposed to overcome the problems mentioned in the above background technology.

[0057] Specifically, Figure 1 The figure shows a flow chart of the prediction model of the motor fault diagnosis method of the dust removal equipment of the cigarette machine based on deep transfer learning according to the embodiment of the present application. Figure 1 As shown, this method collects vibration and temperature signal data from motor equipment in dust removal workshops. Through data preprocessing, network construction, model training, and model fine-tuning, it builds a motor vibration prediction model. Simultaneously, it builds a motor temperature warning model by preprocessing motor temperature data and identifying operating conditions. In this way, transfer learning and multidimensional information fusion techniques are used to predict motor equipment faults and provide warnings for abnormal motor temperatures under varying operating conditions, enabling real-time status monitoring of motor equipment.

[0058] Below, the embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0059] Example 1:

[0060] The data set used in this embodiment is the vibration test bench experimental data and the vibration signal data collected from the key motor equipment in the dust removal workshop. The vibration test bench data is used as the source domain data set D S There are 4 operating conditions, 4 major categories and 10 minor fault types. The data collected from the motor equipment in the dust removal workshop is used as the target domain dataset D T , a total of 4 working conditions.

[0061] This embodiment adopts a method for diagnosing motor faults in a cigarette machine dust removal device based on deep transfer learning. The specific process is as follows:

[0062] Figure 2 The figure shows a flow chart of a method for diagnosing motor faults of a dust removal device of a cigarette making machine based on deep transfer learning according to an embodiment of the present application. Figure 2As shown, according to the embodiment of the present application, the motor fault diagnosis method of the dust removal equipment of the cigarette making machine based on deep transfer learning includes: S1, real-time collection of vibration signals and temperature signals of the motor equipment in the dust removal workshop during operation, and storing the vibration signals and the temperature signals in a local database respectively; S2, performing data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and performing data preprocessing operations on null values ​​and abnormal values ​​on the temperature signal; S3, using a convolutional neural network as the core architecture of the transfer learning model, and defining a 6-layer neural network as the feature extraction layer of the transfer learning network in addition to the output layer; S4, training the transfer learning model to obtain a prediction model M T ; S5, using the prediction model M T Predict the vibration data and get the vibration prediction result y prd ; S6, perform secondary dynamic threshold judgment on the pre-processed motor temperature data and build a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd ; S7, based on the vibration prediction result y prd and the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results.

[0063] In the embodiment of the present application, step S1 collects the vibration signal and temperature signal of the motor equipment in the dust removal workshop in real time during operation, and stores the vibration signal and temperature signal in a local database. It should be understood that vibration and temperature are important indicators of the equipment's operating status. Combining vibration and temperature signals can monitor the equipment's status from multiple dimensions, improving the comprehensiveness of fault detection. Therefore, a vibration and temperature sensor is installed to sample the vibration and temperature signals of the motor equipment during operation.

[0064] In an embodiment of the present application, the step S2 is to perform data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and to perform data preprocessing operations on null values ​​and abnormal values ​​on the temperature signal. It should be understood that the vibration signal as raw data is not in the format required by the network model. Therefore, before using the vibration data of the motor for model training, the vibration signal is first preprocessed by data resampling, window segmentation, and data standardization to convert it into the format required by the model. Furthermore, considering that sensor failure or data transmission errors may cause abnormal values, missing values, etc. to appear in the temperature signal, based on this, the temperature signal is preprocessed by data of null values ​​and abnormal values ​​to ensure the accuracy and integrity of the data.

[0065] Specifically, Figure 3 The figure shows a flow chart of step S2 in the method for diagnosing motor faults of dust removal equipment of cigarette making machine based on deep transfer learning according to an embodiment of the present application. Figure 3 As shown, the step S2 includes: S21, resampling the vibration signal, eliminating missing values, and setting the batch size to 128; S22, adjusting the sliding window length L and overlap rate r of the resampled data, wherein the sliding window length L = 128, and the overlap rate r = 0.5; S23, applying Z-Score normalization to the resampled vibration signal to eliminate dimensionality effects and accelerate model convergence, wherein the Z-Score calculation formula is as follows:

[0066]

[0067] Among them, X scaler is the standardized vibration data, X is the resampled vibration signal, μ is the mean of the data under this feature dimension, and σ is the standard deviation of the data under this feature; S24, removing abnormal values ​​and null values ​​from the temperature signal.

[0068] In an embodiment of the present application, step S3 adopts a convolutional neural network as the core architecture of the transfer learning model. In addition to the output layer, a 6-layer neural network is defined as the feature extraction layer of the transfer learning network. It should be understood that transfer learning allows the model to use knowledge trained in one field (source domain) to improve learning efficiency and performance in another related field (target domain). However, it may be difficult to obtain a large amount of labeled data in the target domain (such as specific equipment in the tobacco manufacturing field). Convolutional neural networks (CNNs) are good at automatically extracting features from raw data, reducing dependence on professional knowledge and manual feature engineering, which is especially valuable when the data volume is large and complex. Based on this, a convolutional neural network (CNN) is used as the core architecture of the transfer learning model to construct a feature extraction layer. In this way, the learning efficiency and performance of the model can be improved when the data is limited and complex.

[0069] Specifically, the step S3 includes: introducing a custom linear layer into the output layer as a classification layer, and reducing the overfitting phenomenon of the model by using the Dropout technology.

[0070] In particular, Figure 4 The figure shows the neural network architecture of the method for diagnosing motor faults of dust removal equipment of cigarette making machine based on deep transfer learning according to the embodiment of the present application. Figure 4As shown in Figure 1, the architecture of the neural network includes convolutional layers (the basic components for feature extraction), batch normalization layers (used to stabilize and accelerate the training process of the neural network), maximum pooling layers (used to reduce the spatial size of features and increase the level of abstraction of features while retaining important information), average pooling layers (another pooling technique that calculates the average value within a region and can retain more feature information than the maximum pooling layer), and fully connected layers (the last few layers of the CNN, used to map features to the final classification results). This architecture is the core of the transfer learning model, specifically for feature extraction and classification.

[0071] In the embodiment of the present application, the step S4 is to train the transfer learning model to obtain the prediction model M T It should be understood that the training process enables the model to adapt to the specific working environment and conditions in the tobacco manufacturing field. In other words, model training is a key step in implementing the method for diagnosing motor faults in cigarette machine dust removal equipment based on deep transfer learning in the embodiment of the present application. It enables the model to fully utilize existing knowledge, adapt to specific environments, and improve the accuracy and efficiency of fault diagnosis, thereby bringing significant economic and operational benefits to the tobacco manufacturing field. Therefore, the transfer learning model is further trained.

[0072] Specifically, Figure 5 The figure shows a flow chart of step S4 in the method for diagnosing motor faults of dust removal equipment of cigarette making machine based on deep transfer learning according to an embodiment of the present application. Figure 5 As shown, the step S4 includes: S41, dividing the source domain data set D s , target domain dataset D T , for the source domain dataset D s , the target domain dataset D T Perform data preprocessing operations respectively; S42, set the initial network parameters for model training, wherein the learning rate ∈=1e-3 is set, the learning rate adjustment strategy selects the Step method, and the optimization strategy selects the SGD method; S43, the source domain dataset D s Input the transfer learning model and use batch normalization technology to accelerate model convergence. After repeated iterations, the optimal pre-training model M is obtained. S , and save it locally; S44, the target domain dataset D T Input pre-trained model M S , and use batch normalization technology to fine-tune the model, and after repeated iterations, obtain the prediction model M that adapts to the target domain data T .

[0073] That is, this method is used in the target domain dataset D T In the absence of sufficient labels, the pre-trained model M of the source domain is used SThis method performs feature extraction and preliminary classification, effectively addressing the difficulty of predictive model training caused by the scarcity of field data labels. Furthermore, by using transfer learning, the model can learn common features from the source domain data and apply these features to the target domain, enabling it to maintain high recognition accuracy even under varying operating conditions, thereby improving the model's cross-condition recognition capabilities.

[0074] In particular, Figure 6 The figure shows a flow chart of step S41 in step S4 of the method for diagnosing motor faults of dust removal equipment of cigarette making machine based on deep transfer learning according to an embodiment of the present application. Figure 6 As shown, the step S41 includes: S411, using a vibration test bench equipped with a vibration temperature sensor to collect vibration data of the motor using the faulty bearing and the normal bearing, and using the obtained labeled data as the source domain data set D S ; S412, install wireless vibration sensors on the motor equipment in the cigarette dust removal workshop, collect the motor vibration data of the new field within a certain period of time, and store the data in the database as the target domain dataset D T ; S413, the source domain dataset D S , the target domain dataset D T Execute step S2 to process the data into a format suitable for the neural network.

[0075] In particular, in the embodiment of the present application, 100 iterations of training are performed on the source domain and target domain training sets, and each iteration corresponds to an update of the model parameters. Specifically, Figure 7 The figure shows a training loss curve of the motor fault diagnosis method of the cigarette machine dust removal equipment based on deep transfer learning according to an embodiment of the present application. Figure 7 As shown, the training loss generally shows a downward trend as the number of iterations increases. This indicates that the model gradually learns the characteristics of the data during training and its predictive performance is improving. In other words, the motor fault diagnosis method for dust removal equipment in cigarette making machines based on deep transfer learning in this embodiment of the application demonstrated good performance improvement and stability during training. The model training strategy is effective, and the final model is expected to achieve high fault diagnosis accuracy in practical applications.

[0076] In particular, in the embodiment of the present application, an accuracy test is performed on the target domain verification set. Specifically, Figure 8 The figure shows the accuracy change trend of the method for diagnosing motor faults of dust removal equipment of cigarette making machine based on deep transfer learning according to the embodiment of the present application. Figure 8As shown in , it specifically shows the changes in two accuracy indicators, namely the target domain training accuracy (targettrainacc) and the target domain test accuracy (targettestacc), which reflect the performance of the model on training data and unseen data respectively. Figure 8 It is a visual representation of the effect of the fine-tuning process in the evaluation step S44. By observing the changes in the target domain training accuracy (target train acc) and test accuracy (target test acc), we can understand the learning progress and final performance of the model on the target domain. Figure 8 As can be seen from the figure, as training progresses, both the target domain training accuracy and the target test accuracy show an increasing trend. This indicates that the model gradually learns the characteristics of the target domain data during training and has good generalization ability on unseen data.

[0077] In the embodiment of the present application, the step S5 uses the prediction model M T Predict the vibration data and get the vibration prediction result y prd It should be understood that after the model training, the next step is to predict the vibration data. Through the predictive analysis of the vibration data, predictive maintenance can be performed before the failure occurs to prevent potential equipment failure. Therefore, further use of the prediction model M T Predict the vibration data and get the vibration prediction result y prd .

[0078] In the embodiment of the present application, the step S6 is to perform a secondary dynamic threshold judgment on the pre-processed motor temperature data to construct a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd It should be understood that the motor will generate heat during operation, and continuous temperature monitoring is essential to prevent overheating and failure. Through dynamic monitoring and early warning of temperature data, measures can be taken before a failure occurs to avoid potential damage. Furthermore, considering that the secondary dynamic threshold judgment can adjust the early warning threshold according to real-time data, it can be adjusted according to real-time data to reduce false alarms and missed alarms, and improve the accuracy of the early warning system. Based on this, the secondary dynamic threshold judgment is performed on the pre-processed motor temperature data and a temperature early warning model M is constructed. temp This helps to achieve automated temperature monitoring and early warning, thereby reducing the need for manual on-site inspections and saving human resources.

[0079] In the embodiment of the present application, the step S7 is based on the vibration prediction result y prd and the temperature warning result t prd , pushing the motor status level to the client and displaying device fault diagnosis information based on the predicted results. It should be understood that delivering motor status to the client in real time based on the predicted results is crucial to ensuring the effectiveness and practicality of motor fault diagnosis methods. By defining different status levels, the severity of faults can be quantified. Providing clear motor status information and fault diagnosis results to the client helps them implement a graded response strategy, avoid faults, and reduce unplanned downtime.

[0080] Specifically, the motor status levels include three levels: three, two, and one. If both a vibration warning and a temperature warning occur, a three-level status (alarm) is pushed to the client. If only a vibration or temperature warning occurs, a two-level status (warning) is pushed to the client. If no warning occurs, a one-level status (normal) is pushed to the client. Device fault diagnosis information is displayed based on the prediction results.

[0081] In summary, the method for diagnosing motor faults in dust removal equipment of cigarette making machines based on deep transfer learning according to the embodiment of the present application is explained. The method collects vibration and temperature signals of motor equipment in dust removal workshop in real time, performs data preprocessing, and uses convolutional neural network to construct a transfer learning model for feature extraction and training to predict vibration prediction results y prd At the same time, the temperature warning model M is constructed by performing secondary dynamic threshold judgment on the pre-processed motor temperature data. temp , and perform diagnosis to obtain temperature warning results t prd The present invention uses vibration prediction and temperature early warning models to diagnose motor faults and push status levels and fault information to the client, thereby improving equipment management efficiency and operational stability.

[0082] It should also be noted that while this deep transfer learning-based motor fault diagnosis method for cigarette machine dust removal equipment is designed and optimized for cigarette machine dust removal equipment, it does not necessarily apply exclusively to cigarette machine dust removal equipment. The core advantage of transfer learning models lies in their ability to apply knowledge learned from one domain (source domain) to another (target domain). This means that as long as the data features of the new equipment have a certain degree of similarity with the data features used to train the model, the model can be adapted to the new equipment through appropriate adjustments. Furthermore, the various components of this method (such as data acquisition, preprocessing, feature extraction, model training, and fault diagnosis) can be considered relatively independent modules. This modular design allows the method to be adjusted and restructured to meet the needs of different equipment. Therefore, the method's deep transfer learning-based technical framework and modular design make it potentially adaptable and applicable to fault diagnosis of motors in other industrial equipment. Of course, this may require retraining and adjustments to the model to adapt to the specific characteristics and needs of the new equipment.

[0083] Example 2:

[0084] Figure 9 The figure shows a schematic diagram of a motor fault diagnosis system for a cigarette machine dust removal device based on deep transfer learning according to an embodiment of the present application. Figure 9 As shown, the motor fault diagnosis system 10 for dust removal equipment of a cigarette making machine based on deep transfer learning according to an embodiment of the present application includes: a motor data acquisition module 100, which is used to collect the vibration signal and temperature signal of the motor equipment in the dust removal workshop in real time during operation, and store the vibration signal and the temperature signal in a local database respectively; a motor data preprocessing module 200, which is used to perform data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal and perform data preprocessing operations on null values ​​and abnormal values ​​on the temperature signal; a transfer learning model building module 300, which is used to adopt a convolutional neural network as the core architecture of the transfer learning model, and define a 6-layer neural network as the feature extraction layer of the transfer learning network in addition to the output layer; a transfer learning model training module 400, which is used to train the transfer learning model to obtain a prediction model M T Motor vibration prediction module 500, for using the prediction model M T Predict the vibration data and get the vibration prediction result y prd ; Motor temperature warning module 600, for the pre-processed motor temperature data for secondary dynamic threshold judgment, build a temperature warning model M temp , and using the temperature warning model M temg Diagnose the temperature data and get the temperature warning result t prd Motor fault prediction module 700, for predicting the vibration result based on y prdand the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results.

[0085] Here, those skilled in the art will appreciate that the specific functions and operations of each step in the above-mentioned cigarette machine dust removal equipment motor fault diagnosis system 10 based on deep transfer learning have been referenced above. Figure 2 The description of the motor fault diagnosis method of the cigarette machine dust removal equipment based on deep transfer learning has been introduced in detail, and therefore, its repeated description will be omitted.

[0086] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit of the technical solutions of the present invention.

Claims

1. A method for diagnosing motor faults in dust removal equipment of cigarette making machines based on deep transfer learning, characterized in that: include: S1. Collecting vibration signals and temperature signals of the motor equipment in the dust removal workshop in real time during operation, and storing the vibration signals and temperature signals in a local database respectively; S2. performing data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and performing data preprocessing operations for null values ​​and abnormal values ​​on the temperature signal; S3. Use convolutional neural network as the core architecture of the transfer learning model. In addition to the output layer, define a 6-layer neural network as the feature extraction layer of the transfer learning network; S4, training the transfer learning model to obtain a prediction model M T ; S5, using the prediction model M T Predict the vibration data and get the vibration prediction result y prd ; S6, perform secondary dynamic threshold judgment on the pre-processed motor temperature data and build a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd ; S7, based on the vibration prediction result y prd and the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results; Wherein, said S4 includes: S41. Divide the source domain dataset D s , target domain dataset D T , for the source domain dataset D s , the target domain dataset D T Perform data preprocessing operations respectively; S42. Set the initial network parameters for model training, wherein the learning rate ∈=1e-3 is set, the Step method is selected as the learning rate adjustment strategy, and the SGD method is selected as the optimization strategy; S43, the source domain dataset D s Input the transfer learning model and use batch normalization technology to accelerate model convergence. After repeated iterations, the optimal pre-training model M is obtained. S , and save it locally; S44, the target domain dataset D T Input pre-trained model M S , and use batch normalization technology to fine-tune the model, and after repeated iterations, obtain the prediction model M that adapts to the target domain data T .

2. The method for diagnosing motor faults of dust removal equipment of cigarette making machines based on deep transfer learning according to claim 1 is characterized in that: Said S2 comprises: S21, resample the vibration signal, eliminate missing values, and set the batch size to 128; S22, adjusting the sliding window length L and overlap ratio r of the resampled data, wherein the sliding window length L=128 and the overlap ratio r=0.5; S23. Apply Z-Score normalization to the resampled vibration signal to eliminate the dimension effect and accelerate the convergence of the model. The Z-Score calculation formula is as follows: Among them, X scaler is the standardized vibration data, X is the resampled vibration signal, μ is the mean of the data under this feature dimension, and σ is the standard deviation of the data under this feature; S24: Eliminate abnormal values ​​and null values ​​of the temperature signal.

3. The method for diagnosing motor faults of dust removal equipment of cigarette making machines based on deep transfer learning according to claim 2 is characterized in that: The S3 includes: introducing a custom linear layer into the output layer as a classification layer, and using the Dropout technology to reduce the overfitting phenomenon of the model.

4. The method for diagnosing motor faults of dust removal equipment of cigarette making machines based on deep transfer learning according to claim 3 is characterized in that: The S41 includes: S411, use a vibration test bench equipped with a vibration temperature sensor to collect vibration data of the motor using the faulty bearing and the normal bearing, and use the obtained labeled data as the source domain dataset D S ; S412, install wireless vibration sensors on the motor equipment in the cigarette dust removal workshop, collect the vibration data of the motor in the new field within a certain period of time, and store the data in the database as the target domain dataset D T ; S413: the source domain dataset D S , the target domain dataset D T Execute step S2 to process the data into a format suitable for the neural network.

5. The method for diagnosing motor faults of dust removal equipment of cigarette making machines based on deep transfer learning according to claim 4 is characterized in that: The S7 includes: the motor status levels include a third-level state, a second-level state, and a first-level state.

6. The method for diagnosing motor faults of dust removal equipment of cigarette making machines based on deep transfer learning according to claim 5 is characterized in that: The S7 includes: if the vibration warning and the temperature warning appear at the same time, the third-level status is pushed to the client; if only the vibration or temperature warning appears, the second-level status is pushed to the client; if there is no warning, the first-level status is pushed to the client, and the equipment fault diagnosis information is displayed according to the prediction results.

7. A motor fault diagnosis system for dust removal equipment of cigarette making machines based on deep transfer learning, characterized in that: include: The motor data acquisition module is used to collect the vibration signal and temperature signal of the motor equipment in the dust removal workshop in real time during operation, and store the vibration signal and temperature signal in the local database respectively; A motor data preprocessing module, configured to perform data resampling, window segmentation, and data standardization preprocessing operations on the vibration signal, and perform data preprocessing operations for null values ​​and abnormal values ​​on the temperature signal; The transfer learning model building module is used to use the convolutional neural network as the core architecture of the transfer learning model. In addition to the output layer, a 6-layer neural network is defined as the feature extraction layer of the transfer learning network; The transfer learning model training module is used to train the transfer learning model to obtain the prediction model M T ; Motor vibration prediction module, used to use the prediction model M T Predict the vibration data and get the vibration prediction result y prd ; The motor temperature warning module is used to perform secondary dynamic threshold judgment on the pre-processed motor temperature data and build a temperature warning model M temp , and using the temperature warning model M temp Diagnose the temperature data and get the temperature warning result t prd ; The motor fault prediction module is used to predict the motor fault based on the vibration prediction result y prd and the temperature warning result t prd , push the motor status level to the client and display the equipment fault diagnosis information based on the prediction results; The transfer learning model training module includes: Divide the source domain dataset D s , target domain dataset D T , for the source domain dataset D s , the target domain dataset D T Perform data preprocessing operations respectively; Set the initial network parameters for model training, where the learning rate is set to ∈ = 1e-3, the Step method is selected for the learning rate adjustment strategy, and the SGD method is selected for the optimization strategy; The source domain dataset D s Input the transfer learning model and use batch normalization technology to accelerate model convergence. After repeated iterations, the optimal pre-training model M is obtained. S , and save it locally; The target domain dataset D T Input pre-trained model M S , and use batch normalization technology to fine-tune the model, and after repeated iterations, obtain the prediction model M that adapts to the target domain data T .

8. The cigarette making machine dust removal equipment motor fault diagnosis system based on deep transfer learning according to claim 7 is characterized in that: The motor fault prediction module includes: the motor state levels include a third state, a second state, and a first state.

Images