Conveyor fault early warning method, control device and conveyor

Abstract

The embodiment of the invention provides a conveyor fault early warning method, a control device and a conveyor, and the method comprises the steps that S1, conveyor state data are collected, and the state data comprise a conveyor belt edge displacement image, a driving device vibration signal and material distribution data; s2, a conveyor fault early warning model is constructed, the model comprises a visual feature extraction module, and the visual feature extraction module adopts hole convolution to capture the edge displacement features of the conveyor belt; the vibration feature extraction module is used for extracting vibration features by adopting a bidirectional LSTM (Long Short Term Memory) structure; the density sensing module is used for acquiring a material distribution condition by adopting a graph convolutional network; and S3, the state data are input into the fault early warning model, fault type prediction is obtained, and fault types comprise conveyer belt oscillation, driving mechanism unbalance and load distribution abnormity. According to the method, the fault prediction precision can be improved, the false report and missing report rate is reduced, and accurate early warning is realized.

Description

Technical Field

[0001] The present application relates to the field of fault warning, and particularly to a conveyor fault warning method, a control device and a conveyor. Background Art

[0002] As a key transmission device in the modern industrial production system, conveyors are widely used in various fields, and their operating stability is directly related to production efficiency and safety. If equipment failures may lead to production interruptions, material losses or even safety accidents, so fault warning technology has important value for ensuring continuous production and reducing maintenance costs. However, there are significant deficiencies in existing conveyor fault warning methods. Existing technologies mostly rely on single-type sensors for single-parameter threshold judgment, making it difficult to capture the multi-dimensional coupling characteristics of the equipment operating state, resulting in a high false negative rate for early compound faults. Existing feature extraction methods have poor adaptability to non-stationary signals and cannot effectively associate the spatio-temporal relationships between multi-modal data such as vision and vibration. Existing models lack the dynamic response ability to complex working conditions, and fixed threshold setting is prone to false alarms. In addition, multi-sensor data is often processed in isolation, and the effective fusion of cross-modal features cannot be achieved, restricting the accuracy of fault location and type identification. These problems seriously restrict the reliability and practicality of the warning system.

[0003] Therefore, there is an urgent need for a conveyor fault warning method to overcome the above problems. Summary of the Invention

[0004] The present application provides a conveyor fault warning method, a control device and a conveyor. This method improves the fault prediction accuracy, reduces the false alarm and false negative rates, and realizes accurate warning through multi-modal data collaborative analysis and a fault warning model.

[0005] In a first aspect, a conveyor fault warning method is provided, and the method includes:

[0006] S1: Collect conveyor status data, where the status data includes: the edge displacement image of the conveyor belt, the vibration signal of the driving device, and the material distribution data;

[0007] S2: Construct a conveyor fault warning model, and the model includes:

[0008] A visual feature extraction module that uses dilated convolution to capture the edge displacement features of the conveyor belt;

[0009] A vibration feature extraction module that uses a bidirectional LSTM structure to extract vibration features;

[0010] A density perception module that uses a graph convolutional network to obtain the material distribution;

[0011] S3: Input the status data into the fault warning model to obtain the predicted fault types, which include: conveyor belt oscillation, driving mechanism imbalance, and abnormal load distribution.

[0012] It should be understood that the visual feature extraction module uses dilated convolution to capture millimeter-level displacement changes of the conveyor belt. The vibration feature extraction module extracts the temporal features of the vibration signal through bidirectional LSTM. The density perception module analyzes the spatial correlation of material distribution based on graph convolution. The three modules work together to effectively identify the coupled features of faults and improve the prediction accuracy.

[0013] Combined with the first aspect, in some implementation manners of the first aspect, the visual feature extraction module includes multiple dilated convolution layers with different dilation rates, and the dilated convolution layers with different dilation rates are used to capture detailed features of different scales.

[0014] It should be understood that dilated convolution expands the receptive field by interval sampling and preserves the resolution. In this application, multiple dilated convolution layers with different dilation rates are adopted. The layer with a small dilation rate captures the fine displacement details at the edge of the conveyor belt, and the layer with a large dilation rate extracts the overall deformation trend. The multi-scale feature fusion enhances the ability to identify the early conveyor belt oscillation features and improves the detection accuracy of small displacements compared with the traditional convolutional network.

[0015] Combined with the first aspect, in some implementation manners of the first aspect, the vibration feature extraction module includes:

[0016] A forward LSTM structure, which is used to extract the forward features of the vibration signal of the driving device;

[0017] A backward LSTM structure, which is used to extract the backward features of the vibration signal of the driving device.

[0018] It should be understood that bidirectional LSTM processes the forward and backward temporal features of the vibration signal simultaneously. The forward LSTM captures the propagation characteristics of impact events, and the backward LSTM analyzes the vibration attenuation law. The two work together to extract the complete dynamic features of the driving mechanism imbalance, enhance the ability to distinguish instantaneous impact and continuous vibration, and improve the recognition rate of driving mechanism imbalance faults.

[0019] Combined with the first aspect, in some implementation manners of the first aspect, the construction method of the density perception module includes:

[0020] Construct a topological map of the conveyor belt partition. The load status of each partition corresponding to the longitudinally equal segments of the conveyor belt is used as the node feature. Edge connections are established between adjacent partitions and weight coefficients are assigned;

[0021] Perform two-level graph convolution operations. The first-level graph convolution aggregates the features of the node itself and its directly adjacent nodes, and the second-level graph convolution extends the feature propagation to indirectly adjacent nodes through the power operation of the adjacency matrix.

[0022] It should be understood that the graph convolution models the conveyor belt partition as the nodes of a topological graph. The first-level aggregation directly adjacent to the load characteristics of the sections to detect local accumulation; the second-level extends to indirectly associated sections through the power operation of the adjacency matrix to identify the abnormal propagation path of the load. The two-level operation takes into account both local details and global distribution, improves the accuracy of abnormal load distribution detection, and reduces the false alarm rate by 18%.

[0023] Combined with the first aspect, in some implementation manners of the first aspect, the fault warning model includes a cross-modal attention fusion module, and the cross-modal attention fusion module is used to dynamically weight the output feature maps of each module.

[0024] Combined with the first aspect, in some implementation manners of the first aspect, the cross-modal attention fusion module uses a cross-attention mechanism to fuse the outputs of each module.

[0025] It should be understood that this module dynamically adjusts the fusion weights of visual, vibration, and density features through a spatial-channel dual visual mechanism. The cross-attention mechanism establishes the spatio-temporal correlation between the vibration signal and the conveyor belt displacement, enhances the representation ability of composite fault features, and improves the prediction accuracy.

[0026] Combined with the first aspect, in some implementation manners of the first aspect, step S3 includes:

[0027] S301: Set the first threshold for each fault type according to the historical fault data of the conveyor.

[0028] S302: Input the status data into the fault warning model to obtain the fault type prediction and the corresponding fault probability.

[0029] S303: Compare the fault probability with the first threshold. If the fault probability is higher than the first threshold, a fault warning is issued.

[0030] Combined with the first aspect, in some implementation manners of the first aspect, the method for setting the first threshold is: based on the fault probability distribution curve of historical fault samples, select the basic threshold.

[0031] It should be understood that the cumulative probability quantile is selected as the basic threshold based on the historical fault probability distribution to ensure that the warning standard conforms to the actual operation law. This setting method keeps the warning accuracy stable and avoids the subjective deviation of manually setting the threshold.

[0032] In a second aspect, a control device is provided. The control device includes a processor and a memory. The processor is coupled to the memory. The memory is used to store computer programs or instructions. The processor is used to execute the computer programs or instructions in the memory so that any implementation manner described in the first aspect is executed.

[0033] In a third aspect, a conveyor is provided, which includes the control device as described in the second aspect. Description of the Drawings

[0034] Figure 1 FIG. is a flowchart for implementing a conveyor fault warning method provided by an embodiment of the present application.

[0035] Figure 2 FIG. is a schematic structural diagram of a fault warning model provided by an embodiment of the present application.

[0036] Figure 3 FIG. is a flowchart for implementing a fault prediction method provided by an embodiment of the present application. Detailed Embodiments

[0037] The terms used in the following embodiments are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and appended claims of the present application, the singular forms "a", "an", "", "the foregoing", "said", and "this" are also intended to include, for example, the expression "one or more", unless the context clearly indicates otherwise. It should also be understood that in the following embodiments of the present application, "at least one" and "one or more" mean one, two, or more than two. The term "and / or" is used to describe the association relationship of associated objects and indicates that three relationships can exist; for example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship.

[0038] Reference to "an embodiment" or "some embodiments" etc. described in this specification means that a specific feature, structure, or characteristic described in combination with the embodiment is included in one or more embodiments of the present application. Thus, statements such as "in an embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments" etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0039] As a key transmission device in the modern industrial production system, conveyors are widely used in various fields, and their operating stability is directly related to production efficiency and safety. However, the existing conveyor fault warning methods cannot effectively cope with the increasingly complex working scenarios of conveyors.

[0040] The embodiments of the present application provide a conveyor fault warning method, a control device, and a conveyor. This method uses multi-modal data collaborative analysis and a fault warning model to improve the accuracy of fault prediction, reduce the false alarm and missed alarm rates, and achieve accurate warning.

[0041] The technical solutions of the embodiments of the present application will be described below with reference to the accompanying drawings.

[0042] Figure 1 It is a flowchart of the implementation of a conveyor fault warning method provided by the embodiments of the present application. Figure 2 It is a schematic diagram of the structure of a fault warning model provided by the embodiments of the present application.

[0043] In some examples, the method includes:

[0044] S1: Collect conveyor status data, where the status data includes: conveyor belt edge displacement images, drive device vibration signals, and material distribution data;

[0045] S2: Construct a conveyor fault warning model, and the model includes:

[0046] A visual feature extraction module that uses dilated convolution to capture conveyor belt edge displacement features;

[0047] A vibration feature extraction module that uses a bidirectional LSTM structure to extract vibration features;

[0048] A density perception module that uses a graph convolutional network to obtain the material distribution;

[0049] S3: Input the status data into the fault warning model to obtain a fault type prediction, where the fault types include: conveyor belt oscillation, drive mechanism imbalance, and abnormal load distribution.

[0050] In some examples, the visual feature extraction module includes multiple dilated convolutional layers with different dilation rates, and the dilated convolutional layers with different dilation rates are used to capture detail features of different scales.

[0051] In a possible implementation, the visual feature extraction module contains a cascaded set of dilated convolutional layer groups. The first layer uses a 3×3 convolutional kernel with a dilation rate of 2 to capture the local deformation details of the conveyor belt edge, and the second layer is configured with a convolutional kernel of the same size with a dilation rate of 4 to extract the overall displacement trend with a larger span. After the outputs of each layer are batch-normalized and ReLU-activated, the shallow high-resolution features and the deep semantic features are fused through residual connections to form a multi-scale displacement perception ability, where the dilation sampling interval of the dilated convolution is kept constant by zero-padding to maintain the size of the feature map.

[0052] In some examples, the vibration feature extraction module includes:

[0053] Forward LSTM structure, which is used to extract the forward features of the vibration signal of the driving device;

[0054] Backward LSTM structure, which is used to extract the backward features of the vibration signal of the driving device.

[0055] In a possible implementation, the bidirectional LSTM in the vibration feature extraction module consists of two independent LSTM units, a forward one and a backward one. Each unit is set with 128 hidden layer neurons. The input sequence is the framed data of the vibration signal sampled at 256Hz (each frame is 50ms). The output gate uses the sigmoid activation function, and the initial bias of the forget gate is set to 1.0 to alleviate the vanishing gradient. The forward LSTM processes the original time series signal, and the backward LSTM processes the reverse sequence. After their hidden states are concatenated in the time dimension, they are connected to the time attention mechanism layer to calculate the weights of each time step, and finally a weighted fusion vibration feature vector is output.

[0056] In some examples, the construction method of the density perception module includes:

[0057] Construct a conveyor belt partition topology map, where the load state of each partition corresponding to the longitudinally equal segments of the conveyor belt is used as the node feature, and edge connections are established between adjacent partitions and weight coefficients are assigned;

[0058] Perform two-level graph convolution operations. The first-level graph convolution aggregates the features of the node itself and its directly adjacent nodes, and the second-level graph convolution propagates the features to indirectly adjacent nodes through the power operation of the adjacency matrix.

[0059] In some examples, the fault warning model includes a cross-modal attention fusion module, which is used to dynamically weight the output feature maps of each module.

[0060] In some examples, the cross-modal attention fusion module uses the cross-attention mechanism to fuse the outputs of each module.

[0061] Figure 3 This is a flowchart for implementing a fault prediction method provided by an embodiment of this application.

[0062] In some examples, step S3 includes:

[0063] S301: Set the first threshold for each fault type according to the historical fault data of the conveyor;

[0064] S302: Input the status data into the fault warning model to obtain the fault type prediction and the corresponding fault probability;

[0065] S303: Compare the fault probability with the first threshold. If the fault probability is higher than the first threshold, a fault warning is issued.

[0066] In some examples, the method for setting the first threshold is as follows: based on the failure probability distribution curve of historical failure samples, a basic threshold is selected.

[0067] In a possible implementation, the cumulative probability 95% quantile is selected as the basic threshold based on the historical failure probability distribution to ensure that the warning standard conforms to the actual operation law.

[0068] The embodiment of the present application provides a control device, which includes a processor and a memory. The processor is coupled to the memory. The memory is used to store computer programs or instructions. The processor is used to execute the computer programs or instructions in the memory, so that any of the methods in the foregoing examples is executed.

[0069] The embodiment of the present application further provides a conveyor, which includes the control device as described above.

[0070] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. Any equivalent modification or change made by those of ordinary skill in the art according to the disclosure of the present invention shall be included in the protection scope recorded in the claims.

Claims

1. A conveyor failure early warning method, characterized in that: The method includes: S1: Collecting conveyor status data, the status data including: conveyor belt edge displacement image, drive device vibration signal and material distribution data; S2: Constructing a conveyor failure warning model, the model includes: A visual feature extraction module, wherein the visual feature extraction module uses a dilated convolution to capture the edge displacement features of the conveyor belt; A vibration feature extraction module, wherein the vibration feature extraction module uses a bidirectional LSTM structure to extract vibration features; A density perception module, which uses a graph convolutional network to obtain material distribution; S3: Inputting the state data into the fault warning model to obtain a fault type prediction, wherein the fault types include: conveyor belt oscillation, drive mechanism imbalance and abnormal load distribution.

2. The method according to claim 1, characterized in that The visual feature extraction module includes a plurality of dilated convolutional layers with different dilation rates, and the dilated convolutional layers with different dilation rates are used to capture detail features of different scales.

3. The method according to claim 1, characterized in that The vibration feature extraction module comprises: A forward LSTM structure, wherein the forward LSTM structure is used to extract positive features of the vibration signal of the driving device; A backward LSTM structure is used to extract reverse features of the vibration signal of the driving device.

4. The method according to claim 1, characterized in that: The method for constructing the density perception module includes: Construct a conveyor belt partition topology map. The load state of each partition corresponding to the longitudinal equal segment of the conveyor belt is used as the node feature. Edge connections are established between adjacent partitions and weight coefficients are assigned. A two-level graph convolution operation is performed. The first level graph convolution aggregates the features of the node itself and its direct neighboring nodes, and the second level graph convolution expands the feature propagation to indirect neighboring nodes through adjacency matrix power operation.

5. The method according to claim 1, characterized in that: The fault warning model includes a cross-modal attention fusion module, which is used to dynamically weight the output feature maps of each module.

6. The method according to claim 5, characterized in that The cross-modal attention fusion module adopts the cross-attention mechanism to fuse the outputs of each module.

7. The method according to claim 1, characterized in that The step S3 comprises: S301: setting a first threshold value for each fault type according to the historical fault data of the conveyor; S302: Inputting the state data into the fault warning model to obtain a fault type prediction and a corresponding fault probability; S303: Compare the failure probability with the first threshold, and if the failure probability is higher than the first threshold, issue a failure warning.

8. The method according to claim 7, characterized in that The method for setting the first threshold is: selecting a basic threshold based on a failure probability distribution curve of historical failure samples.

9. A control device, characterized in that: The method comprises a processor and a memory, wherein the processor is coupled to the memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions in the memory, so that the method according to any one of claims 1 to 8 is executed.

10. A conveyor, characterized in that: The conveyor comprises a control device as claimed in claim 9.

Images