Method and device for monitoring health state of air supply controller

Abstract

The invention discloses a method and a device for monitoring the health state of a pneumatic conveying controller. The method comprises the following steps: acquiring operation data of a plurality of pneumatic conveying controllers in a pneumatic cut tobacco feeding system; the operation data comprises bus interface data and monitoring data; the monitoring data comprises servo motor working condition data, servo controller temperature rise data, regulating valve working condition data, tobacco shred air conveying speed data and negative pressure pipe working condition data; inputting the operation data of the plurality of air supply controllers into the trained monitoring network model for processing to obtain health states of the plurality of air supply controllers, and judging whether the air supply controllers fail or not; the monitoring network model comprises a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module which are connected in sequence; according to the invention, the operation data of a plurality of pneumatic conveying controllers in the pneumatic cut tobacco conveying system can be monitored and analyzed in real time, potential faults can be found in time, an accurate health state diagnosis result is provided, and the requirement of manual intervention is reduced.

Description

Technical Field

[0001] The invention relates to a method and a device for monitoring the health status of an air conveying controller, and belongs to the technical field of artificial intelligence. Background Art

[0002] The wind-driven wire feeding system is an important production equipment in cigarette factories. Its function is to transfer tobacco from the wire production line to the cigarette equipment through the pipeline. The wire feeding pipeline has the characteristics of long transmission distance and complex pipeline structure. In order to ensure the quality of tobacco delivery and reduce the tobacco fragmentation rate, the wind conveying controller is used to accurately control the wind conveying speed of tobacco (15-18m / s). Due to the wide installation area, hidden installation location, variable operating conditions, and difficulty in timely detection of mechanical equipment failures, the fluctuation of cigarette product quality caused by equipment operation failures has attracted widespread attention. Therefore, in order to ensure the stability of wind-driven wire feeding equipment operation, it is crucial to use intelligent fault diagnosis. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and device for monitoring the health status of an air conveying controller, which can monitor and analyze the operating data of multiple air conveying controllers in a wind wire feeding system in real time through a monitoring network model, promptly capture abnormal conditions in the operation of the air conveying controllers, promptly discover potential faults, provide accurate health status diagnosis results, and reduce the need for manual intervention.

[0004] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0005] In one aspect, the present invention provides a method for monitoring the health status of an air conveying controller, comprising:

[0006] Acquire the operation data of multiple air conveying controllers in the air-powered wire feeding system; the operation data includes: bus interface data and monitoring data; the monitoring data includes: servo motor working condition data, servo controller temperature rise data, regulating valve working condition data, tobacco air conveying speed data and negative pressure pipe working condition data;

[0007] The operating data of the multiple air conveying controllers are input into a trained monitoring network model for processing to obtain the health status of the multiple air conveying controllers and determine whether the air conveying controllers fail; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence.

[0008] Optionally, the operation data of the plurality of air conveyance controllers are input into a trained monitoring network model for processing to obtain the health status of the plurality of air conveyance controllers, including:

[0009] Inputting the operation data of multiple air conveying controllers into a multi-scale feature extraction module for feature extraction to obtain extracted multi-scale features;

[0010] The weight-adaptive feature fusion module is used to fuse the extracted multi-scale features to obtain the fused features;

[0011] The fused features are input into the fault detection module for identification to obtain the health status of multiple air conveying controllers; the fault detection module adopts LSTM network.

[0012] Optionally, the multi-scale feature extraction module includes: a first feature extraction channel submodule and a second feature extraction channel submodule; the first feature extraction channel submodule includes a first feature extraction network and an S-CPA attention unit; the second feature extraction channel submodule includes a second feature extraction network and an S-CPA attention unit.

[0013] Optionally, the first feature extraction network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer and a first fully connected layer; the second feature extraction network includes a convolutional layer one, a pooling layer one, a convolutional layer two, a pooling layer two, a convolutional layer three, a pooling layer three and a second fully connected layer; the S-CPA attention unit includes parallel spatial attention branches and channel attention branches.

[0014] Optionally, the convolution kernel sizes of the first convolution layer, the second convolution layer and the third convolution layer in the first feature extraction network are 32×1, 16×1 and 8×1 respectively, which are used to extract large-scale and shallow-level features; the convolution kernel sizes of the convolution layer one, the convolution layer two and the convolution layer three in the second feature extraction network are 7×1, 5×1 and 3×1 respectively, which are used for small-scale and deep-level features.

[0015] Optionally, the operation data of multiple air conveying controllers are input into a multi-scale feature extraction module for feature extraction to obtain extracted multi-scale features, including:

[0016] Input the operation data of multiple air conveying controllers into the first feature extraction channel submodule and the second feature extraction first feature extraction network channel submodule, perform multiple convolution pooling processes respectively, and output the results of the first feature extraction network and the second feature extraction network through the fully connected layer;

[0017] The results of the first feature extraction network and the second feature extraction network are respectively input into two S-CPA attention units, and processed by the parallel spatial attention branch and channel attention branch to obtain the output results of the first feature extraction channel submodule and the second feature extraction channel submodule, that is, the extracted multi-scale features.

[0018] Optionally, the weight adaptive feature fusion module is used to perform feature fusion on the extracted multi-scale features to obtain fused features, including:

[0019] Performing pooling processing on the output results of the first feature extraction channel submodule and the second feature extraction channel submodule respectively, and connecting the two pooled results to generate global representation information;

[0020] Input the global representation information into the third fully connected layer to obtain the global representation features;

[0021] The global representation features are normalized using the Softmax activation function to obtain the weights of the first feature extraction channel and the second feature extraction channel;

[0022] Based on the weights of the first feature extraction channel and the second feature extraction channel, the output results of the first feature extraction channel submodule and the second feature extraction channel submodule are weighted fused to obtain fused features.

[0023] Optional methods for obtaining a trained monitoring network model include:

[0024] Obtain operating data sets of multiple air conveying controllers in normal and fault states;

[0025] The running data set is input into the monitoring network model, and the cross entropy function is used as the loss function for training to obtain the optimal parameters of the model;

[0026] Based on the optimal parameters, a trained monitoring network model is obtained.

[0027] In another aspect, the present invention provides a device for monitoring the health status of an air conveying controller, comprising:

[0028] The data acquisition module is used to obtain the operating data of multiple air conveying controllers in the air-powered wire feeding system; the operating data includes: bus interface data and monitoring data; the monitoring data includes: servo motor operating data, servo controller temperature rise data, regulating valve operating data, tobacco air conveying speed data and negative pressure pipe operating data;

[0029] A data processing module, used to input the operation data of the plurality of air conveying controllers into a trained monitoring network model for processing, obtain the health status of the plurality of air conveying controllers, and determine whether the air conveying controllers have failed; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence;

[0030] The display module is used to receive the health status of multiple air conveying controllers through the host computer and display the location of the air conveying controller failure.

[0031] Optionally, the data acquisition module includes: a bus module and a monitoring module; the bus module is used to obtain bus interface data; the monitoring module is used to monitor servo motor operating data, servo controller temperature rise data, regulating valve operating data, tobacco wind delivery speed data and negative pressure pipe operating data.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention provides a method for monitoring the health status of an air conveyor controller. By monitoring the multi-scale feature extraction module, the weight adaptive feature fusion module and the fault detection module in the network model, the operating data of multiple air conveyor controllers in the wind wire feeding system are monitored and analyzed in real time, abnormal conditions in the operation of the air conveyor controllers are captured in time, potential faults are discovered in time, accurate health status diagnosis results are provided, and the need for manual intervention is reduced.

[0034] The present invention provides a device for monitoring the health status of an air conveying controller. Through the collection of data from an operating data processing module in a data acquisition module, the fusion diagnosis of the operating data by the data processing module, and the continuous monitoring and data analysis of the operating status of the air conveying controller by a display module, the device can predict the occurrence of potential faults, take maintenance measures in advance, and avoid downtime losses caused by faults. It can also provide data support for the optimization of production processes and further improve overall production efficiency.

[0035] The present invention provides a multi-scale feature extraction module of a monitoring network model in a method for monitoring the health status of a wind conveyor controller. The module extracts features through two channels to better adapt to monitoring requirements under different working conditions. The module also introduces the S-CPA attention mechanism (Split-and-Consolidate Pyramid Attention, combining the spatial and channel attention mechanisms) to simultaneously focus on the spatial position and channel information of the feature graph and suppress irrelevant or interfering features. The combination of multi-scale feature extraction and the S-CPA attention mechanism can more sensitively detect potential faults or abnormal conditions in the wind conveyor controller, provide more accurate early warning information, significantly improve monitoring performance and accuracy, and improve the level of intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor, among which:

[0037] Figure 1A flow chart of a method for monitoring the health status of an air conveying controller according to an embodiment of the present invention is shown;

[0038] Figure 2 The schematic diagram of the monitoring network model in the embodiment of the present invention is shown;

[0039] Figure 3 The schematic diagram of the multi-scale feature extraction module in the embodiment of the present invention is shown;

[0040] Figure 4 The schematic diagram of the working process of the S-CPA attention unit in the embodiment of the present invention is shown;

[0041] Figure 5 The figure shows a schematic diagram of the working process of the weight adaptive feature fusion module in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0043] The application principle of the present invention is described in detail below in conjunction with the accompanying drawings.

[0044] Example 1

[0045] like Figure 1-Figure 2 As shown, an embodiment of the present invention provides a method for monitoring the health status of an air conveying controller, comprising the following steps:

[0046] S01: Acquire the operation data of multiple air conveying controllers in the air-powered wire feeding system; the operation data includes: bus interface data and monitoring data; the monitoring data includes: servo motor working condition data, servo controller temperature rise data, regulating valve working condition data, tobacco air conveying speed data and negative pressure pipe working condition data;

[0047] S02: Input the operating data of the multiple air conveying controllers into the trained monitoring network model for processing, obtain the health status of the multiple air conveying controllers, and determine whether the air conveying controllers fail; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence.

[0048] Based on a method for monitoring the health status of an air conveyor controller in an embodiment of the present invention, the operating data of multiple air conveyor controllers in a wind wire feeding system are monitored and analyzed in real time through a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module in a monitoring network model, so as to timely capture abnormal conditions in the operation of the air conveyor controller, timely discover potential faults, provide accurate health status diagnosis results, and reduce the need for manual intervention.

[0049] In this embodiment, in step S02, the operation data of the plurality of air conveyance controllers are input into the trained monitoring network model for processing to obtain the health status of the plurality of air conveyance controllers, including:

[0050] S021: inputting the operation data of the plurality of air conveying controllers into a multi-scale feature extraction module for feature extraction to obtain extracted multi-scale features;

[0051] S022: using a weight adaptive feature fusion module, performing feature fusion on the extracted multi-scale features to obtain fused features;

[0052] S023: Input the fused features into the fault detection module for identification to obtain the health status of multiple air conveying controllers; wherein the fault detection module adopts an LSTM network.

[0053] In this embodiment, in step S021, the operation data of multiple air conveying controllers are input into the multi-scale feature extraction module for feature extraction to obtain the extracted multi-scale features, including:

[0054] Input the operation data of multiple air conveying controllers into the first feature extraction channel submodule and the second feature extraction first feature extraction network channel submodule, perform multiple convolution pooling processes respectively, and output the results of the first feature extraction network and the second feature extraction network through the fully connected layer;

[0055] The results of the first feature extraction network and the second feature extraction network are respectively input into two S-CPA attention units, and processed by the parallel spatial attention branch and channel attention branch to obtain the output results of the first feature extraction channel submodule and the second feature extraction channel submodule, that is, the extracted multi-scale features.

[0056] Among them, Figure 3 As shown, the multi-scale feature extraction module includes: a first feature extraction channel submodule and a second feature extraction channel submodule; the first feature extraction channel submodule includes a first feature extraction network and an S-CPA attention unit; the second feature extraction channel submodule includes a second feature extraction network and an S-CPA attention unit.

[0057] Specifically, the first feature extraction network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer and a first fully connected layer; the second feature extraction network includes a convolutional layer 1, a pooling layer 1, a convolutional layer 2, a pooling layer 2, a convolutional layer 3, a pooling layer 3 and a second fully connected layer; the S-CPA attention unit includes parallel spatial attention branches and channel attention branches.

[0058] Specifically, the convolution kernel sizes of the first convolution layer, the second convolution layer, and the third convolution layer in the first feature extraction network are 32×1, 16×1, and 8×1, respectively, which are used to extract large-scale and shallow-level features; the convolution kernel sizes of the first convolution layer, the second convolution layer, and the third convolution layer in the second feature extraction network are 7×1, 5×1, and 3×1, respectively, which are used for small-scale and deep-level features.

[0059] Among them, Figure 4 As shown, a schematic diagram of the S-CPA attention unit workflow; the result of the first feature extraction network or the second feature extraction network is input into the S-CPA attention unit and processed by the parallel spatial attention branch and channel attention branch;

[0060] Specifically, in the spatial attention branch, the similarity weight between each position in the spatial feature map and other positions is calculated, and the features of different positions are weighted and aggregated according to the similarity weight to output the spatial dimension features;

[0061] The spatial attention mechanism is expressed as:

[0062]

[0063] in: Represents input; It is a convolution operation, that is, the Conv (4×1) convolution operation with a convolution kernel size of 4×1 in the figure; and It is a fully connected layer, namely Dense1 and Dense2 in the figure. Indicates the calculation of the square root of each element of the tensor. (·) represents the SoftMax function.

[0064] The output of the spatial attention branch is:

[0065]

[0066] in: is a spatial multiplication operation, N is the dimension of the attention weight matrix, W is the attention weight, C is the number of channels, is a tensor space.

[0067] In the channel attention branch, the channel attention mechanism expression is:

[0068]

[0069] in: Represents input; is the convolution operation (shown in the figure); It is the average pooling operation, that is, Average Pooling in the figure; (·) is the SoftMax function.

[0070] The output of the channel attention branch is:

[0071]

[0072] in: is a spatial multiplication operation, N is the dimension of the attention weight matrix, and C is the number of channels.

[0073] The dimensions of the outputs of the two branches are not exactly the same. They are added after the dimensions are aligned, and tensor broadcasting is used to achieve dimension alignment and addition operations.

[0074] The final output of the S-CPA attention unit is the sum of the two branches, expressed as follows:

[0075]

[0076] like Figure 5 As shown, in this embodiment, in step S022, a weight adaptive feature fusion module is used to perform feature fusion on the extracted multi-scale features to obtain fused features, and the steps include the following:

[0077] Performing pooling processing on the output results of the first feature extraction channel submodule and the second feature extraction channel submodule respectively, and connecting the two pooled results to generate global representation information;

[0078] Input the global representation information into the third fully connected layer to obtain the global representation features;

[0079] The global representation features are normalized using the Softmax activation function to obtain the weights of the first feature extraction channel and the second feature extraction channel;

[0080] Based on the weights of the first feature extraction channel and the second feature extraction channel, the output results of the first feature extraction channel submodule and the second feature extraction channel submodule are weighted fused to obtain fused features.

[0081] Specifically, the features extracted by the first feature extraction channel and the second feature extraction channel submodules are pooled and compressed (global average pooling is used to compress the original spatial dimension, while the number of channels remains unchanged), and the features of the two channels after compression are obtained. and , which is expressed as follows:

[0082]

[0083] in: is the dimension of the feature space, and Represents the features extracted by the two channels, obtained by the aforementioned multi-scale feature extraction module.

[0084] Characteristics of the two channels after compression and Generate global representation information after connection , and then pass through the third fully connected layer to obtain the global representation feature , the formula is as follows;

[0085]

[0086]

[0087] in, and are the weights and biases set in the third fully connected layer.

[0088] The global representation information is input into the Softmax layer to obtain the weight of each channel feature. The formula is as follows:

[0089]

[0090] in, and are the weights of the first feature extraction channel and the second feature extraction channel, respectively. Represents a set of weights.

[0091] Using the weight of each channel feature, the spatial features of the two channels are weighted fused in the fusion layer to obtain the fused features. , the formula is as follows:

[0092]

[0093] In this embodiment, the method for obtaining the trained monitoring network model includes:

[0094] Obtain operating data sets of multiple air conveying controllers in normal and fault states;

[0095] The running data set is input into the monitoring network model, and the cross entropy function is used as the loss function for training to obtain the optimal parameters of the model;

[0096] Based on the optimal parameters, a trained monitoring network model is obtained.

[0097] Specifically, the cross entropy function formula is as follows:

[0098]

[0099] in: is the loss value, is the output sequence, is the probability distribution of the running dataset samples, is a symbolic function.

[0100] Example 2

[0101] The present invention provides a device for monitoring the health status of an air conveying controller, comprising:

[0102] The data acquisition module is used to obtain the operating data of multiple air conveying controllers in the air-powered wire feeding system; the operating data includes: bus interface data and monitoring data; the monitoring data includes: servo motor operating data, servo controller temperature rise data, regulating valve operating data, tobacco air conveying speed data and negative pressure pipe operating data;

[0103] A data processing module, used to input the operation data of the plurality of air conveying controllers into a trained monitoring network model for processing, obtain the health status of the plurality of air conveying controllers, and determine whether the air conveying controllers have failed; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence;

[0104] The display module is used to receive the health status of multiple air conveying controllers through the host computer and display the location of the air conveying controller failure.

[0105] Specifically, the data acquisition module includes: a bus module and a monitoring module; the bus module is used to obtain bus interface data; the monitoring module is used to monitor servo motor operating data, servo controller temperature rise data, regulating valve operating data, tobacco wind delivery speed data and negative pressure pipe operating data.

[0106] Specifically, the wind-driven wire feeding system consists of a main control system (host computer) and 20-30 independent wind-driven controller control units, which communicate through the PROFINET network. Each wind-driven controller has a separate CPU control unit to achieve wind-driven wire feeding control of a single unit; based on the bus interface data, the main control system power supply, hardware failure, and PROFINET network communication failure can be diagnosed;

[0107] For a single air delivery controller control unit, the faults that can be diagnosed include: valve blockage, abnormal wind speed (too high, too low), abnormal valve opening, valve stuck, air leakage when the valve is closed, wind pressure sensor failure, wind speed sensor failure, servo motor abnormality (overcurrent, temperature rise, etc.), duct surge, etc.

[0108] Example 3

[0109] The characteristics of a single controller failure in an air conveying controller are universal and robust.

[0110] Taking the valve zeroing error as an example, the steps of the air conveying controller health status monitoring method process in Example 1 are explained, especially how to use multi-scale feature extraction technology and adaptive fusion strategy to analyze the status of the air conveying control system and evaluate the health status of the system.

[0111] 1. Fault description

[0112] When the pneumatic wire feeding system is started or stopped, the valve should be completely closed. If the valve is not completely closed or is not closed tightly, the wind pressure or wind speed will exceed the set threshold, thus affecting the normal operation of the system. Fault diagnosis is performed by monitoring and analyzing system-related data.

[0113] 2. Main steps of health status monitoring:

[0114] (1) Multi-scale feature extraction:

[0115] Data collection: Collect data such as wind speed, wind pressure, valve opening, etc. through a number of corresponding sensors.

[0116] Feature extraction: Perform multi-scale analysis on data to extract feature information at different scales.

[0117] (2) Feature fusion:

[0118] Feature fusion: Features extracted from different scales are fused to form comprehensive health status features. This can be achieved through weighted averaging, principal component analysis (PCA), and other methods.

[0119] Adaptive adjustment: According to the actual operating status of the system, the weight of feature fusion is adaptively adjusted to improve the accuracy of health status monitoring.

[0120] (3) Health status assessment / fault detection:

[0121] Model construction: Build a fault detection module based on the fused feature information.

[0122] Fault detection and early warning: monitor the system status in real time and determine whether there are any abnormalities through the fault detection module.

[0123] (4) Real-time anomaly detection

[0124] Dynamic threshold detection: Learn dynamic thresholds through historical data and monitor real-time feature values. If the wind speed or wind pressure exceeds the set threshold, an abnormality is detected.

[0125] Anomaly Detection Algorithms: Use algorithms such as Isolation Forest, Autoencoders, etc. to detect abnormal patterns in the system and process complex feature data.

[0126] Specific:

[0127] 1. Monitor valve status:

[0128] Normal state: When the wind-powered wire feeding is powered on or the wind-powered tobacco feeding is stopped, the valve should be in the closed state.

[0129] Abnormal status: If the valve does not close under the above conditions, there may be a malfunction.

[0130] 2. Detect wind pressure and wind speed in the air duct:

[0131] Air leakage in the air duct: Even if the valve is closed, if there is a slight air leak in the air duct, the air pressure and air speed of the system may be affected.

[0132] Abnormal judgment: Set the thresholds for wind pressure and wind speed and monitor them in real time. If the wind pressure or wind speed exceeds the threshold, it means there is an abnormality, which may be caused by air leakage in the air duct due to the valve not closing properly.

[0133] If a valve zeroing error is detected, the fault diagnosis results and warning information of a single air delivery controller are sent to the main control system via the PROFINET network, and the main control system generates a fault report and notifies the operator.

[0134] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for monitoring the health status of an air conveying controller, characterized in that: include: Obtain the operating data of multiple air conveying controllers in the air wire feeding system; The operation data includes: bus interface data and monitoring data; the monitoring data includes: servo motor working condition data, servo controller temperature rise data, regulating valve working condition data, tobacco wind delivery speed data and negative pressure pipe working condition data; The operating data of the multiple air conveying controllers are input into a trained monitoring network model for processing to obtain the health status of the multiple air conveying controllers and determine whether the air conveying controllers fail; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence.

2. The method for monitoring the health status of an air conveying controller according to claim 1, characterized in that: The operation data of the multiple air conveyance controllers are input into the trained monitoring network model for processing to obtain the health status of the multiple air conveyance controllers, including: Inputting the operation data of multiple air conveying controllers into a multi-scale feature extraction module for feature extraction to obtain extracted multi-scale features; The weight-adaptive feature fusion module is used to fuse the extracted multi-scale features to obtain the fused features; The fused features are input into the fault detection module for identification to obtain the health status of multiple air conveying controllers; the fault detection module adopts LSTM network.

3. The method for monitoring the health status of an air conveying controller according to claim 2, characterized in that: The multi-scale feature extraction module includes: a first feature extraction channel submodule and a second feature extraction channel submodule; the first feature extraction channel submodule includes a first feature extraction network and an S-CPA attention unit; the second feature extraction channel submodule includes a second feature extraction network and an S-CPA attention unit.

4. The method for monitoring the health status of an air conveying controller according to claim 3, characterized in that: The first feature extraction network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer and a first fully connected layer; the second feature extraction network includes a convolutional layer 1, a pooling layer 1, a convolutional layer 2, a pooling layer 2, a convolutional layer 3, a pooling layer 3 and a second fully connected layer; the S-CPA attention unit includes parallel spatial attention branches and channel attention branches.

5. The method for monitoring the health status of an air conveying controller according to claim 4, characterized in that: The convolution kernel sizes of the first convolution layer, the second convolution layer and the third convolution layer in the first feature extraction network are 32×1, 16×1 and 8×1 respectively, which are used to extract large-scale and shallow-level features; the convolution kernel sizes of the first convolution layer, the second convolution layer and the third convolution layer in the second feature extraction network are 7×1, 5×1 and 3×1 respectively, which are used for small-scale and deep-level features.

6. The method for monitoring the health status of an air conveying controller according to claim 5, characterized in that: The operation data of multiple air conveying controllers are input into the multi-scale feature extraction module for feature extraction to obtain the extracted multi-scale features, including: Input the operation data of multiple air conveying controllers into the first feature extraction channel submodule and the second feature extraction first feature extraction network channel submodule, perform multiple convolution pooling processes respectively, and output the results of the first feature extraction network and the second feature extraction network through the fully connected layer; The results of the first feature extraction network and the second feature extraction network are respectively input into two S-CPA attention units, and processed by the parallel spatial attention branch and channel attention branch to obtain the output results of the first feature extraction channel submodule and the second feature extraction channel submodule, that is, the extracted multi-scale features.

7. The method for monitoring the health status of an air conveying controller according to claim 6, characterized in that: The weight-adaptive feature fusion module is used to fuse the extracted multi-scale features to obtain fused features, including: Performing pooling processing on the output results of the first feature extraction channel submodule and the second feature extraction channel submodule respectively, and connecting the two pooled results to generate global representation information; Input the global representation information into the third fully connected layer to obtain the global representation features; The global representation features are normalized using the Softmax activation function to obtain the weights of the first feature extraction channel and the second feature extraction channel; Based on the weights of the first feature extraction channel and the second feature extraction channel, the output results of the first feature extraction channel submodule and the second feature extraction channel submodule are weighted fused to obtain fused features.

8. The method for monitoring the health status of an air conveying controller according to claim 7, characterized in that: Methods for obtaining a trained monitoring network model include: Obtain operating data sets of multiple air conveying controllers in normal and fault states; The running data set is input into the monitoring network model, and the cross entropy function is used as the loss function for training to obtain the optimal parameters of the model; Based on the optimal parameters, a trained monitoring network model is obtained.

9. A device for monitoring the health status of an air conveying controller, characterized in that: include: A data acquisition module is used to obtain the operating data of multiple air conveying controllers in the air-powered wire feeding system; The operation data includes: bus interface data and monitoring data; the monitoring data includes: servo motor working condition data, servo controller temperature rise data, regulating valve working condition data, tobacco wind delivery speed data and negative pressure pipe working condition data; A data processing module, used to input the operation data of the plurality of air conveying controllers into a trained monitoring network model for processing, obtain the health status of the plurality of air conveying controllers, and determine whether the air conveying controllers have failed; the monitoring network model includes a multi-scale feature extraction module, a weight adaptive feature fusion module and a fault detection module connected in sequence; The display module is used to receive the health status of multiple air conveying controllers through the host computer and display the location of the air conveying controller failure.

10. The device for monitoring the health status of an air conveying controller according to claim 9, characterized in that: The data acquisition module includes: a bus module and a monitoring module; the bus module is used to obtain bus interface data; the monitoring module is used to monitor servo motor operating data, servo controller temperature rise data, regulating valve operating data, tobacco wind delivery speed data and negative pressure pipe operating data.

Images