Elevator lever drum brake fault detection method, device and equipment

Abstract

The invention discloses a fault detection method, device and equipment for an elevator lever drum brake, and relates to the field of fault detection.The method comprises the steps that original signal data of the elevator lever drum brake are obtained; carrying out sparse processing on the weight of a filter of each convolutional layer in the ResNet fault diagnosis network according to the global pruning rate to obtain a sparse ResNet fault diagnosis network; pruning a filter of each convolutional layer in the sparse ResNet fault diagnosis network by adopting a K-means clustering algorithm to obtain a pruned ResNet fault diagnosis network; and inputting the original signal data into the pruned ResNet fault diagnosis network to obtain a fault detection result of the elevator lever drum brake. According to the invention, the fault detection efficiency of the elevator lever drum brake is improved.

Description

Technical Field

[0001] This application relates to the field of fault detection, and in particular to a method, apparatus and equipment for fault detection of elevator lever drum brake. Background Technology

[0002] With the continuous development of elevator lever drum brake fault diagnosis technology, in elevator operation and maintenance scenarios, facing the huge amount of collected parameters and calculation requirements, the traditional centralized data processing method (transmitting all data to the cloud for processing) faces many problems such as insufficient bandwidth, high latency, and excessive cloud load. In order to handle and repair faults in a timely manner and avoid major faults, the requirements for data real-time performance are becoming increasingly higher. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, and equipment for detecting faults in elevator lever drum brakes. By performing sparse processing and filter pruning on the fault diagnosis network, the number of parameters and computational load of the fault diagnosis network are effectively reduced while ensuring the performance of the fault diagnosis network, making it possible to deploy it on resource-constrained edge devices and effectively improving the efficiency of fault diagnosis.

[0004] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for detecting faults in an elevator lever drum brake, including: Acquire the raw signal data of the elevator lever drum brake; The weights of the filters in each convolutional layer of the ResNet fault diagnosis network are sparsified according to the global pruning rate to obtain a sparse ResNet fault diagnosis network. The ResNet fault diagnosis network includes cascaded residual blocks. Each residual block includes a first convolutional module and a second convolutional module connected in sequence. The first convolutional module includes a convolutional layer, a normalization layer, and an activation layer connected in sequence. The second convolutional module includes a convolutional layer, a normalization layer, and a reduction operation layer connected in sequence. The K-means clustering algorithm is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network to obtain the pruned ResNet fault diagnosis network. The original signal data is input into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake.

[0005] Secondly, this application provides a fault detection device for an elevator lever drum brake, comprising: The raw signal data acquisition module is used to acquire the raw signal data of the elevator lever drum brake. A sparse module is used to sparsify the weights of the filters in each convolutional layer of the ResNet fault diagnosis network according to the global pruning rate, resulting in a sparsed ResNet fault diagnosis network. The ResNet fault diagnosis network includes cascaded residual blocks. Each residual block includes a first convolutional module and a second convolutional module connected in sequence. The first convolutional module includes a convolutional layer, a normalization layer, and an activation layer connected in sequence. The second convolutional module includes a convolutional layer, a normalization layer, and a reduction operation layer connected in sequence. The pruning module is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network using the K-means clustering algorithm, so as to obtain the pruned ResNet fault diagnosis network. The fault detection module is used to input the original signal data into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake.

[0006] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described elevator lever drum brake fault detection method.

[0007] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, apparatus, and device for detecting faults in elevator lever drum brakes. By sparsifying the weights of filters in each convolutional layer of a ResNet fault diagnosis network, unimportant weights are removed and identified, thereby reducing the inference latency of the ResNet fault diagnosis network. K-means clustering is then used to prune the filters in each convolutional layer of the sparsified ResNet fault diagnosis network, concentrating similar filters within each convolutional layer and removing redundant filters. This application further improves the accuracy of fault diagnosis while preserving the collective characteristics of the layer filters. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating a method for detecting faults in an elevator lever drum brake according to an embodiment of this application; Figure 2A schematic diagram of the ResNet fault diagnosis network structure for a fault detection method of an elevator lever drum brake provided in an embodiment of this application; Figure 3 A schematic diagram of a pruned ResNet fault diagnosis network for a fault detection method of an elevator lever drum brake provided in an embodiment of this application; Figure 4 A schematic diagram of the overall structure of an elevator lever drum brake traction machine test bench for a fault detection method of an elevator lever drum brake provided in an embodiment of this application. Figure 5 This is a schematic diagram showing the sparsification of the filter weights in any convolutional layer of a fault detection method for an elevator lever drum brake provided in an embodiment of this application. Figure 6 A schematic diagram of filter pruning in any convolutional layer of an elevator lever drum brake fault detection method provided in an embodiment of this application; Figure 7 A functional module schematic diagram of a fault device for an elevator lever drum brake provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] In one exemplary embodiment, such as Figure 1 As shown, a method for detecting faults in an elevator lever drum brake is provided. This method is executed by a computer device and includes steps 101 to 104. Wherein: Step 101: Obtain the raw signal data of the elevator lever drum brake.

[0013] Step 102: Sparsify the weights of the filters in each convolutional layer of the ResNet fault diagnosis network according to the global pruning rate to obtain the sparsed ResNet fault diagnosis network. The structure of the ResNet fault diagnosis network is as follows: Figure 2 As shown.

[0014] The ResNet fault diagnosis network includes cascaded residual blocks; each residual block includes a first convolutional module and a second convolutional module connected in sequence; the first convolutional module includes a convolutional layer, a normalization layer and an activation layer connected in sequence; the second convolutional module includes a convolutional layer (conv), a normalization layer (BN) and a shrinkage operation layer (Shrinkage) connected in sequence.

[0015] Step 103: The K-means clustering algorithm is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network, resulting in the pruned ResNet fault diagnosis network, as shown below. Figure 3 As shown.

[0016] Step 104: Input the original signal data into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake.

[0017] By implementing steps 101 to 104 above, this application can improve the accuracy and efficiency of fault diagnosis.

[0018] In another exemplary embodiment of this application, the data of this application is generated using a self-built system such as... Figure 4 The elevator brake condition monitoring test bench shown consists of two WYJ103-05.4 permanent magnet synchronous traction machines. The brake wheel diameter is 653mm, the traction machine speed at the drive end is 151.9r / min, the rated load is 1350kg, the rated elevator speed is 1.75m / s, and the rated torque is 950N·m. The left side of the test bench is the loading end, used to provide the load, and the right side is the drive end, used to simulate the actual operation of the elevator. The dataset in this application includes data collected by three different types of sensors: torque and speed sensors, temperature sensors, and acceleration sensors. It includes eight states: normal state, partial contact surface wear, full contact surface wear, surface oil contamination, surface foreign matter, insufficient braking force, excessive gap, and loose fit.

[0019] Before step 101, the method further includes: obtaining a training sample set and a test sample set, preprocessing the training sample set and the test sample set, and training the ResNet fault diagnosis network using the preprocessed training sample set and the preprocessed test sample set.

[0020] The specific process for obtaining the training and testing sample sets is as follows: Multiple raw signal data points containing fault categories C and their corresponding fault labels are collected using M sensors. Each raw signal data point is preprocessed, and the preprocessed W raw signal data points and their corresponding fault labels are combined to form the training sample set. The preprocessed V raw signal data and their corresponding fault labels constitute a test sample set. Specifically, M=5, C=8, W=150, V=40.

[0021] The specific preprocessing process for each raw signal data is as follows: Each raw signal data is divided into fixed-length data segments using a sliding window sampling method, and each data segment is standardized to obtain Q preprocessed raw signal data segments, where Q=4096. Each raw signal data segment includes the raw signal data and a label. This example uses one-hot encoding to encode the category labels, mapping each fault category to an independent binary vector, where only the position corresponding to that category is 1, and the remaining positions are 0.

[0022] The preprocessed W original signal data segments and their corresponding fault labels are used to form a training sample set. The preprocessed V segments of the original signal data were used as the test sample set. .in, This represents the w-th preprocessed original signal data segment in the training sample set. , express The corresponding fault label. This represents the v-th preprocessed original signal data segment. , express The corresponding fault label.

[0023] In another exemplary embodiment of this application, before obtaining the training sample set and the test sample set, the method further includes: building a ResNet fault diagnosis network, such as... Figure 2 As shown, the network is sparsed and pruned to obtain the pruned ResNet fault diagnosis network, as follows. Figure 3 As shown.

[0024] In the ResNet fault diagnosis network, such as Figure 2 As shown, to better extract fault feature information, the ResNet fault diagnosis network also includes an input module and an output module. The input module includes an input layer, a convolutional layer, a normalization layer, and a ReLU activation layer. The output module includes a ReLU activation layer, a global average pooling (GAP) layer, a fully connected (FC) layer, and a softmax activation layer.

[0025] The threshold is learned through a Global Average Pooling (GAP) layer and a Fully Connected (FC) layer, using the formula... Soft thresholding is applied to the feature map. Specifically, This refers to the original feature data before soft thresholding. The learned threshold, For the absolute value of the characteristic, This is feature data that has undergone soft thresholding. When the absolute value of the feature... hour, ,at this time An output of 0 corresponds to suppressing the feature, i.e., treating it as noise or invalid information; when When, output This process preserves effective features and scales their amplitude. Furthermore, residual connections within the residual block handle dimension matching, ensuring that the residual input and output dimensions are consistent. Reducing the operational layer enhances feature discriminativity and adapts to 64×64 single-channel sensor data.

[0026] After constructing the ResNet fault diagnosis network, this application initializes the parameters of the ResNet fault diagnosis network. The ResNet fault diagnosis network is then initialized and trained in batches of 50, defining cross-entropy loss, the Adam optimizer, and StepLR scheduling. Specifically, the Adam optimizer has an initial learning rate of LR=0.001 and a weight decay of 0.0001. The StepLR scheduler decays the weights by 0.9 every 25 epochs.

[0027] training sample set The test sample set is used as input for training the ResNet fault diagnosis network. The data is used as input to the ResNet fault diagnosis network for testing, and the accuracy of the ResNet fault diagnosis network is obtained. The weights of the ResNet fault diagnosis network with the highest accuracy are saved for the next step of ResNet fault diagnosis network sparsity and pruning.

[0028] In another exemplary embodiment of this application, the weights of the ResNet fault diagnosis network with the highest accuracy are obtained, such as... Figure 5 As shown, step 102 above is replaced by steps 201 to 205: Step 201: For any filter in any convolutional layer of the ResNet fault diagnosis network, obtain the importance of the filter weights.

[0029] Step 202: Sort the importance of the filter weights to obtain the importance of the arranged weights of the filter.

[0030] Step 203: Determine the weight sparsity boundary of the filter based on the global pruning rate and the importance of the weights arranged in the filter.

[0031] Step 204: Based on the sparse boundary of the filter weights, the weights of the filter are sparsified to obtain the sparsed weights of the filter.

[0032] Step 205: Based on the weights of the sparsed filters of each convolutional layer, the sparsed ResNet fault diagnosis network is obtained.

[0033] In another exemplary embodiment of this application, for the i-th convolutional layer in a ResNet fault diagnosis network, the i-th convolutional layer includes n filters; the importance of the q-th weight of the b-th filter in the i-th convolutional layer is determined using the following formula: .

[0034] in, Let q be the importance of the b-th weight in the i-th convolutional layer. Let q be the weight of the b-th filter in the i-th convolutional layer.

[0035] In another exemplary embodiment of this application, the q-th weight of the b-th filter after sparsening in the i-th convolutional layer is obtained using the following formula: .

[0036] in, Let q be the weight of the b-th filter after sparsening in the i-th convolutional layer. Let b be the sparse boundary of the weights of the b-th filter.

[0037] This application performs unstructured pruning in steps 201-205, removing elements below the weight sparse boundary from the filters of each convolutional layer in the sparsed ResNet fault diagnosis network. The weights are set to zero, and weights with importance below the sparse boundary are pruned. Through global weight pruning, each filter can be transformed into a sparse filter.

[0038] In another exemplary embodiment of this application, such as Figure 6 As shown, step 103 above is replaced by steps 301 to 302: Step 301: For the s-th iteration of any n filters in any convolutional layer of the sparsed ResNet fault diagnosis network, calculate the distance between any filter and the s-th cluster center of j clusters, and assign the filter to a cluster according to the distance; wherein, j filters out of n filters are taken as the first cluster center of the j clusters. ; .

[0039] Step 302: Determine whether s is greater than a set threshold; if yes, calculate the importance score of each filter in the j clusters, and filter the filters in the j clusters according to the importance score to obtain the filtered filters. Based on the filtered filters, obtain the pruned ResNet fault diagnosis network; if no, calculate the (s+1)th cluster center of the j clusters according to the filters in the j clusters, and perform the (s+1)th iteration.

[0040] In another exemplary embodiment of this application, the distance between the Lth filter and the center of the sth cluster of the Jth cluster is determined using the following formula: .

[0041] in, The center of the s-th cluster of the J-th cluster in the i-th convolutional layer. For the L-th filter of the i-th convolutional layer, Let be the distance between the L-th filter and the s-th cluster center of the J-th cluster, where i is the index of the convolutional layer, J is the index of the cluster in the i-th convolutional layer, L is the index of the filter in the i-th convolutional layer, and s is the number of times the cluster center has been updated. , .

[0042] In another exemplary embodiment of this application, the Lth filter is assigned to the cluster using the following formula: .

[0043] in, To assign the Lth filter to the cluster, The center of the s-th cluster of the J-th cluster in the i-th convolutional layer. For the L-th filter of the i-th convolutional layer, Let be the distance between the Lth filter and the sth cluster center of the Jth cluster, where i is the index of the convolutional layer, J is the index of the cluster in the i-th convolutional layer, L is the index of the filter in the i-th convolutional layer, s is the number of times the cluster center is updated, and j is the number of clusters.

[0044] In another exemplary embodiment of this application, the (s+1)th cluster center of the j clusters is calculated based on the filters in the j clusters using the following formula: .

[0045] in, The center of the (s+1)th cluster of the Jth cluster in the i-th convolutional layer. To assign the Lth filter to the cluster, Let L be the Lth filter in the i-th convolutional layer, where L is the index of the filter in the i-th convolutional layer.

[0046] Steps 301 and 302 are performed. When a specified number of iterations is reached or the cluster centers no longer change significantly, it indicates that a good clustering effect has been achieved, thus assigning each filter in the convolutional layer to the optimal cluster. The K-means clustering algorithm is used to cluster the filters in each convolutional layer. The filters in each cluster are analyzed and compared, and filters with lower importance are removed, while those with higher importance are retained. Subsequently, the remaining filters in each cluster are integrated, thus completing the filter pruning process.

[0047] This application also uses the intra-cluster sum of squares as an evaluation metric. By plotting the relationship between clustering error and J value under different J values, the inflection point of the curve is found, and the J value corresponding to this point is the optimal number of filter clusters. The intra-cluster sum of squares is determined using the following formula: .

[0048] After determining the optimal J value, the filters of each convolutional layer are pruned. In the ResNet fault diagnosis network that has converged or reached the preset number of iterations, all filters except the cluster center are pruned.

[0049] After obtaining the pruned ResNet fault diagnosis network, this application uses the training set to train the parameters of the pruned ResNet fault diagnosis network and then fine-tunes the pruned ResNet fault diagnosis network to restore some of its accuracy, resulting in a lightweight ResNet fault diagnosis network.

[0050] This application inputs the required global pruning rate p and inputs the original signal data into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake. With a global pruning rate p=0.85, even at a high pruning rate, the accuracy A of the pruned ResNet fault diagnosis network can still reach 98.6%.

[0051] Based on the same inventive concept, such as Figure 7 As shown, a fault detection device for an elevator lever drum brake is provided, comprising: The raw signal data acquisition module 701 is used to acquire the raw signal data of the elevator lever drum brake.

[0052] The sparse module 702 is used to sparsify the weights of the filters in each convolutional layer of the ResNet fault diagnosis network according to the global pruning rate, so as to obtain a sparse ResNet fault diagnosis network. The ResNet fault diagnosis network includes cascaded residual blocks. The residual blocks include a first convolutional module and a second convolutional module connected in sequence. The first convolutional module includes a convolutional layer, a normalization layer and an activation layer connected in sequence. The second convolutional module includes a convolutional layer, a normalization layer and a reduction operation layer connected in sequence.

[0053] The pruning module 703 is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network using the K-means clustering algorithm to obtain the pruned ResNet fault diagnosis network.

[0054] The fault detection module 704 is used to input the original signal data into the pruned ResNet fault diagnosis network to obtain the fault detection result of the elevator lever drum brake.

[0055] The hardware platform used in this application is: Intel(R) Core(TM) i5-14400F CPU, 16GB memory, NVIDIA GeForce RTX 4060 Ti 16GB GPU; Windows 11 operating system; and Python 3.8.11 and Torch 1.10.0 software platform.

[0056] This application, through experiments conducted on 10 experiments, achieved average accuracy rates of 100% before and 98.5% after pruning. This demonstrates that after lightweight pruning of the ResNet fault diagnosis network, this application can still retain the original ResNet fault diagnosis network's ability to extract fault features and maintain high accuracy.

[0057] This application identifies and removes unimportant weights based on a global pruning rate to reduce inference latency in the ResNet fault diagnosis network. It employs clustering to classify filters within each convolutional layer. Filters in each cluster are similar, and redundant filters within the cluster are removed at a predetermined pruning rate. Simultaneously, this application preserves the collective characteristics of filters within convolutional layers, further improving the accuracy of fault diagnosis.

[0058] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the raw signal data of the elevator lever drum brake. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a fault detection method for an elevator lever drum brake.

[0059] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0060] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0061] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0062] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0063] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for detecting faults in an elevator lever drum brake, characterized in that, The method includes: Acquire the raw signal data of the elevator lever drum brake; The weights of the filters in each convolutional layer of the ResNet fault diagnosis network are sparsified according to the global pruning rate to obtain a sparse ResNet fault diagnosis network. The ResNet fault diagnosis network includes cascaded residual blocks. Each residual block includes a first convolutional module and a second convolutional module connected in sequence. The first convolutional module includes a convolutional layer, a normalization layer, and an activation layer connected in sequence. The second convolutional module includes a convolutional layer, a normalization layer, and a reduction operation layer connected in sequence. The K-means clustering algorithm is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network to obtain the pruned ResNet fault diagnosis network. The original signal data is input into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake.

2. The elevator lever drum brake fault detection method according to claim 1, characterized in that, The weights of the filters in each convolutional layer of the ResNet fault diagnosis network are sparsified according to the global pruning rate to obtain the sparsed ResNet fault diagnosis network, specifically including: For any filter in any convolutional layer of the ResNet fault diagnosis network, obtain the importance of the filter weights; The importance of the filter weights is sorted to obtain the importance of the filter weights after sorting. Based on the global pruning rate and the importance of the weights arranged in the filter, the weight sparsity boundary of the filter is determined. The weights of the filter are sparsified based on the sparse boundary of the filter weights to obtain the sparsed weights of the filter. Based on the weights of the sparsed filters in each convolutional layer, a sparsed ResNet fault diagnosis network is obtained.

3. The elevator lever drum brake fault detection method according to claim 2, characterized in that, For the i-th convolutional layer in the ResNet fault diagnosis network, the i-th convolutional layer includes n filters; The importance of the q-th weight of the b-th filter in the i-th convolutional layer is determined using the following formula: ； in, Let q be the importance of the b-th weight in the i-th convolutional layer. Let q be the weight of the b-th filter in the i-th convolutional layer.

4. The elevator lever drum brake fault detection method according to claim 3, characterized in that, The q-th weight of the b-th filter after sparsening in the i-th convolutional layer is obtained using the following formula: ； in, Let q be the weight of the b-th filter after sparsening in the i-th convolutional layer. Let b be the weight sparse boundary of the b-th filter.

5. The elevator lever drum brake fault detection method according to claim 1, characterized in that, The K-means clustering algorithm is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network, resulting in a pruned ResNet fault diagnosis network, specifically including: For the s-th iteration of n filters in any convolutional layer of the sparsed ResNet fault diagnosis network, the distance between any filter and the s-th cluster center of j clusters is calculated, and the filter is assigned to a cluster according to the distance; wherein, the j filters out of the n filters are taken as the first cluster center of the j clusters. ; ; Determine if s is greater than a set threshold; if yes, calculate the importance score of each filter in j clusters, and filter the filters in j clusters according to the importance score to obtain the filtered filters. Based on the filtered filters, obtain the pruned ResNet fault diagnosis network; if no, calculate the (s+1)th cluster center of j clusters according to the filters in j clusters, and perform the (s+1)th iteration.

6. The elevator lever drum brake fault detection method according to claim 5, characterized in that, The distance between the Lth filter and the center of the sth cluster of the Jth cluster is determined using the following formula: ； in, The center of the s-th cluster of the J-th cluster in the i-th convolutional layer. For the L-th filter of the i-th convolutional layer, Let be the distance between the Lth filter and the sth cluster center of the Jth cluster, where i is the index of the convolutional layer, J is the index of the cluster in the i-th convolutional layer, L is the index of the filter in the i-th convolutional layer, and s is the number of times the cluster center has been updated.

7. The elevator lever drum brake fault detection method according to claim 5, characterized in that, The Lth filter is assigned to the cluster using the following formula: ； in, To assign the Lth filter to the cluster, The center of the s-th cluster of the J-th cluster in the i-th convolutional layer. For the L-th filter of the i-th convolutional layer, Let be the distance between the Lth filter and the sth cluster center of the Jth cluster, where i is the index of the convolutional layer, J is the index of the cluster in the i-th convolutional layer, L is the index of the filter in the i-th convolutional layer, s is the number of times the cluster center is updated, and j is the number of clusters.

8. The elevator lever drum brake fault detection method according to claim 5, characterized in that, The (s+1)th cluster center of the j clusters is calculated using the following formula based on the filters in the j clusters: ； in, The center of the (s+1)th cluster of the Jth cluster in the i-th convolutional layer. To assign the Lth filter to the cluster, Let L be the Lth filter in the i-th convolutional layer, where L is the index of the filter in the i-th convolutional layer.

9. A fault detection device for an elevator lever drum brake, characterized in that, The elevator lever drum brake fault detection method according to any one of claims 1-8, the apparatus comprising: The raw signal data acquisition module is used to acquire the raw signal data of the elevator lever drum brake. A sparse module is used to sparsify the weights of the filters in each convolutional layer of the ResNet fault diagnosis network according to the global pruning rate, resulting in a sparsed ResNet fault diagnosis network. The ResNet fault diagnosis network includes cascaded residual blocks. Each residual block includes a first convolutional module and a second convolutional module connected in sequence. The first convolutional module includes a convolutional layer, a normalization layer, and an activation layer connected in sequence. The second convolutional module includes a convolutional layer, a normalization layer, and a reduction operation layer connected in sequence. The pruning module is used to prune the filters of each convolutional layer in the sparsed ResNet fault diagnosis network using the K-means clustering algorithm, so as to obtain the pruned ResNet fault diagnosis network. The fault detection module is used to input the original signal data into the pruned ResNet fault diagnosis network to obtain the fault detection results of the elevator lever drum brake.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the elevator lever drum brake fault detection method according to any one of claims 1-8.

Images