Detection method of multi-disc elevator brake based on two-level contrast fusion network

Abstract

The present invention provides a fault detection method, system, medium and device for a multi-caliper disc elevator brake based on a two-level contrast fusion network. The fault detection method comprises: obtaining a first vibration health feature and a first torque health feature according to a first encoder and a third encoder of a first-level contrast fusion module and a vibration signal and a torque signal; obtaining a second vibration health feature and a second torque health feature according to a second encoder and a fourth encoder of a second-level contrast fusion module and the first vibration health feature and the first torque health feature; calculating the health status feature similarity between brake calipers according to the second vibration health feature and the second torque health feature; calculating the wear amount of the corresponding brake caliper according to the captured image; calculating the total health status value of the brake caliper according to the health status feature similarity and the wear amount; and determining the faulty brake caliper according to the total health status value. The present invention can obtain more comprehensive fault feature information and improve the accuracy and reliability of fault diagnosis.

Description

Technical Field

[0001] The invention relates to the field of fault diagnosis of multi-clamp disc elevator brakes, and in particular to a detection method of multi-clamp disc elevator brakes based on a two-level comparison fusion network. Background Art

[0002] High-speed elevators are mainly used in high-rise or super-high-rise buildings, such as commercial buildings, high-rise residential buildings, hotels, medical institutions and urban subway stations. These places need to carry a large number of people quickly and efficiently, and their operating speed can reach 10m / s. The traction machine of an ordinary elevator consists of an electric motor driving a traction wheel and a brake through a reducer, while the traction machine of a high-speed elevator consists of an electric motor directly connected to a traction wheel and a brake. The multi-caliper disc brake is an important braking component in a high-speed elevator, and its working condition plays a vital role in the safety of the high-speed elevator. The existing fault detection methods for the multi-caliper disc brake of a high-speed elevator mainly include the following, and there are corresponding problems:

[0003] 1) Key performance data monitoring based on physical sensors: Monitor the performance data of the brake, such as braking torque, relative position, brake clearance and temperature, to evaluate the performance and status of the brake. However, this method is easily affected by environmental factors and the accuracy of detection equipment and tools, resulting in inaccurate measurement data and unable to accurately characterize the state changes of high-speed elevator multi-caliper disc brakes.

[0004] 2) Remaining life prediction technology: By establishing a multivariate matrix and a remaining life graph, obtaining a feature matrix and an adjacency matrix, and then obtaining a feature association matrix, a specific prediction model is used to obtain a performance degradation curve to predict the remaining life of the brake. This method can improve the safety of elevator use, but requires complex data processing and model training, and requires professional knowledge and skills to process and analyze data. In addition, the prediction model may not cover all factors that may affect brake performance, such as sudden mechanical failures or extreme usage conditions.

[0005] Moreover, most of the current comprehensive brake models for elevators only consider the overall performance of the brakes, while ignoring the characteristics of its internal components, making it difficult to accurately diagnose and locate faults. Specifically, on the one hand, there is a lack of research on multi-caliper disc brakes for elevators in China, and more research is focused on disc brakes for automobiles. The performance testing of multi-caliper disc brakes for high-speed elevators mainly focuses on their braking force and braking distance. Failure samples of multi-caliper disc brakes for elevators are extremely scarce, which poses a major challenge to deep learning methods. At the same time, it is costly to conduct simulation experiments to obtain failure samples. On the other hand, the measured data of multi-caliper disc elevator brakes usually includes multiple sensor data, such as brake pad friction, brake disc torque, brake disc clearance and temperature. It is challenging to fuse information from these data to extract fault features. Summary of the invention

[0006] In order to achieve accuracy and positioning efficiency in fault diagnosis of multi-disc elevator brakes, the embodiments of the present invention provide a fault detection method, system, medium, device and computer program product for multi-disc elevator brakes based on a two-level comparison fusion network. By collecting key performance parameters of high-speed elevator multi-disc brakes and performing multi-data fusion to obtain more comprehensive fault feature information, the accuracy and reliability of fault diagnosis are improved, thereby ensuring the safe operation of elevators.

[0007] According to a first aspect of the present invention, an embodiment of the present invention provides a fault detection method for a multi-caliper disc elevator brake based on a two-level contrast fusion network, the method comprising: obtaining an original data set of the brake, the original data set comprising: a vibration signal of each brake caliper, a torque signal of each brake caliper, and a captured image of each brake caliper; obtaining a first vibration health feature and a first torque health feature according to a first encoder and a third encoder of a first-level contrast fusion module and the vibration signal and the torque signal; obtaining a second vibration health feature and a second torque health feature according to a second encoder and a fourth encoder of a second-level contrast fusion module and the first vibration health feature and the first torque health feature; calculating the health status feature similarity between the brake calipers according to the second vibration health feature and the second torque health feature; calculating the wear amount of the corresponding brake caliper according to the captured image; calculating the total health status value of the brake caliper according to the health status feature similarity and the wear amount; and determining the faulty brake caliper according to the total health status value.

[0008] The above-mentioned embodiment of the present invention collects the vibration signal, torque signal and captured image of each brake caliper in the high-speed elevator multi-caliper disc brake as the original data set, and performs multi-data fusion through a two-level comparison fusion network to obtain the similarity of health status characteristics between the brake calipers, and then determines the total health status value of the brake caliper according to the health status characteristic similarity and the wear amount calculated based on the captured image for fault diagnosis, which can improve the accuracy and reliability of fault diagnosis.

[0009] In some embodiments of the present invention, the vibration signal of each brake caliper is collected by a vibration sensor; the torque signal of each brake caliper is collected by a torque sensor; and the captured image of each brake caliper is collected by a CMOS image sensor.

[0010] In some embodiments of the present invention, the fault detection method for a multi-caliper disc elevator brake based on a two-level contrast fusion network also includes: determining a first vibration health feature matrix, a first torque health feature matrix, a second vibration health feature matrix, and a second torque health feature matrix according to the first vibration health feature, the first torque health feature, the second vibration health feature, and the second torque health feature; calculating the similarity loss of vibration signal features and the similarity loss of torque signal features between brake calipers according to the first vibration health feature matrix and the first torque health feature matrix; calculating the contrast loss between the vibration signal features and the torque signal features of the brake caliper according to the second vibration health feature matrix and the second torque health feature matrix; and training the two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module according to the similarity loss and contrast loss.

[0011] In some embodiments of the present invention, calculating the wear amount of the corresponding brake caliper according to the captured image includes: acquiring a segmented image according to the captured image; calculating the gap circumference and the gap area of ​​the brake caliper according to the segmented image and its corresponding mask image; calculating the difference judgment value of the brake caliper according to the gap circumference and the gap area; and calculating the wear amount of the brake caliper according to the difference judgment value of the brake caliper and the difference judgment value of an unworn brake caliper.

[0012] In some embodiments of the present invention, calculating the total health status value of the brake caliper based on the health status feature similarity and the wear amount includes: obtaining the total health status value of the brake caliper by weighted summing the inverse of the wear amount and the health status feature similarity; determining the faulty brake caliper based on the total health status value includes: when the total health status value of a brake caliper is lower than the health status threshold, determining that the brake caliper is faulty.

[0013] According to a second aspect of the present invention, an embodiment of the present invention provides a fault detection system for a multi-caliper disc elevator brake based on a two-level contrast fusion network, wherein the brake includes a plurality of brake calipers, and the fault detection system includes: an original data set acquisition module, used to acquire the original data set of the brake, wherein the original data set includes: a vibration signal of each brake caliper, a torque signal of each brake caliper, and a captured image of each brake caliper; a first feature extraction module, used to acquire a first vibration health feature and a first torque health feature according to a first encoder and a third encoder of a first-level contrast fusion module and the vibration signal and the torque signal; a second feature extraction module, used to acquire a second vibration health feature and a second torque health feature according to a second encoder and a fourth encoder of a second-level contrast fusion module and the first vibration health feature and the first torque health feature; a health status feature similarity calculation module, used to calculate the health status feature similarity between the brake calipers according to the second vibration health feature and the second torque health feature; a wear amount calculation module, used to calculate the wear amount of the corresponding brake caliper according to the captured image; a health status value calculation module, used to calculate the total health status value of the brake caliper according to the health status feature similarity and the wear amount; a fault determination module, used to determine the faulty brake caliper according to the total health status value.

[0014] The above-mentioned embodiment of the present invention collects the vibration signal, torque signal and captured image of each brake caliper in the high-speed elevator multi-caliper disc brake as the original data set, and performs multi-data fusion through a two-level comparison fusion network to obtain the similarity of health status characteristics between the brake calipers, and then determines the total health status value of the brake caliper according to the health status characteristic similarity and the wear amount calculated based on the captured image for fault diagnosis, which can improve the accuracy and reliability of fault diagnosis.

[0015] In some embodiments of the present invention, the fault detection system for a multi-caliper disc elevator brake based on a two-level contrast fusion network also includes a network training module for performing the following operations: determining a first vibration health feature matrix, a first torque health feature matrix, a second vibration health feature matrix, and a second torque health feature matrix according to the first vibration health feature, the first torque health feature, the second vibration health feature, and the second torque health feature; calculating the similarity loss of vibration signal features and the similarity loss of torque signal features between brake calipers according to the first vibration health feature matrix and the first torque health feature matrix; calculating the contrast loss between the vibration signal features and the torque signal features of the brake caliper according to the second vibration health feature matrix and the second torque health feature matrix; and training the two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module according to the similarity loss and contrast loss.

[0016] According to the third aspect of the present invention, an embodiment of the present invention provides a computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the computer performs the following operations: the operations include the steps included in the fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network as described in any of the above embodiments.

[0017] According to a fourth aspect of the present invention, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory is used to store one or more computer-readable instructions, wherein the one or more computer-readable instructions, when executed by the processor, can implement the fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network as described in any of the above embodiments.

[0018] According to a fifth aspect of the present invention, an embodiment of the present invention provides a computer program product comprising a computer program, which, when executed by a processor, implements the fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network as described in any of the above embodiments.

[0019] From the above, it can be seen that the fault detection method, system, medium, equipment and computer program product of the multi-caliper disc elevator brake based on the two-level contrast fusion network provided in the embodiment of the present invention collects the vibration signal, torque signal and captured image of each brake caliper in the high-speed elevator multi-caliper disc brake as the original data set, and performs multi-data fusion through the two-level contrast fusion network to obtain the similarity of health status characteristics between the brake calipers, and then determines the total health status value of the brake caliper according to the similarity of health status characteristics and the wear amount calculated based on the captured images for fault diagnosis, which can improve the accuracy and reliability of fault diagnosis, realize the precise positioning of brake caliper faults, and thus ensure the safe operation of the elevator. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network according to an embodiment of the present invention;

[0021] Figure 2 is a front view schematic diagram of the structure of a multi-caliper brake and a torque sensor according to an embodiment of the present invention;

[0022] Figure 3 is a left schematic diagram of the structure of a clamping disc support, a vibration sensor and a CMOS image sensor according to an embodiment of the present invention;

[0023] Figure 4 is a partial front view schematic diagram of the structure of a clamping disc support, a vibration sensor and a CMOS image sensor according to an embodiment of the present invention;

[0024] Figure 5 is a schematic diagram of a two-stage comparison fusion network structure for fault detection of a multi-caliper disc elevator brake according to an embodiment of the present invention;

[0025] Figure 6 is a schematic diagram of a two-stage learning network architecture according to an embodiment of the present invention;

[0026] Figure 7 is a structural schematic diagram of an improved U-type network (ReUNet) for edge feature extraction of an image captured by a CMOS image sensor according to an embodiment of the present invention;

[0027] Figure 8 yes Figure 7 Schematic diagram of the processing flow of the adaptive channel enhancement module introduced in the improved U-type network (ReUNet);

[0028] Fig. 9 1 is an architecture diagram of a fault detection system for a multi-caliper disc elevator brake based on a two-level comparison fusion network according to an embodiment of the present invention.

[0029] The reference numerals are explained as follows: 1-brake caliper, 2-vibration sensor, 3-CMOS image sensor, 4-torque sensor, 5-brake bracket, 6-caliper disc bracket, 7-transmission shaft. DETAILED DESCRIPTION

[0030] The various aspects of the present invention are described in detail below in conjunction with the accompanying drawings and specific embodiments. Among them, well-known modules, units and their connections, links, communications or operations are not shown or described in detail. In addition, the described features, architectures or functions can be combined in any way in one or more embodiments. It should be understood by those skilled in the art that the various embodiments described below are only used for illustration and are not intended to limit the scope of protection of the present invention. It can also be easily understood that the modules or units or processing methods in each embodiment described herein and shown in the accompanying drawings can be combined and designed in various different configurations.

[0031] The following is a brief description of the terms used in this article.

[0032] CMOS image sensor: Complementary Metal Oxide Semiconductor Image Sensor, a typical solid-state imaging sensor.

[0033] TCF-Net: Two level contrast fusion network, two level contrast fusion network.

[0034] U-Net: A convolutional neural network (CNN) architecture.

[0035] Mask image: Also known as mask image, a binary image or grayscale image with the same size (height and width) as the original image.

[0036] Torque health characteristics: The stable or normal characteristics of torque when mechanical equipment is running.

[0037] Vibration health characteristics: The stable or normal characteristics of vibration when mechanical equipment is running.

[0038] Active layer: A hidden layer in a neural network that performs data processing and feature extraction tasks between the input layer and the output layer.

[0039] Figure 1 It is a flow chart of a fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network according to an embodiment of the present invention.

[0040] like Figure 1 As shown, in one embodiment of the present invention, the brake includes a plurality of brake calipers, and the fault detection method may include: step S11, step S12, step S13, step S14, step S15, step S16 and step S17, and the above steps are described in detail below.

[0041] In step S11, the original data set of the brake is obtained, and the original data set includes: the vibration signal of each brake caliper, the torque signal of each brake caliper, and the captured image of each brake caliper. In some embodiments, during the operation of the high-speed elevator, the vibration signal of each brake caliper is collected by a vibration sensor; the torque signal of each brake caliper is collected by a torque sensor; and the captured image of each brake caliper is collected by a CMOS image sensor. High-precision sensors in the art are used to obtain performance data such as the wear amount of brake pads, brake caliper torque, and brake caliper vibration state of each brake caliper of a high-speed elevator multi-caliper disc as the original data set.

[0042] The present invention provides a specific example of a collection device for collecting raw data sets of a multi-caliper disc elevator brake, wherein the multi-caliper disc elevator brake has five brake calipers, and the collection device includes: a vibration sensor, a torque sensor, and a CMOS image sensor. It should be understood that the present invention is not limited to this, and other collection devices known in the art can be used to collect raw data sets. Figures 2 to 4 The examples are described in detail.

[0043] Figure 2is a front view schematic diagram of the structure of a multi-caliper brake and a torque sensor according to an embodiment of the present invention; Figure 3 is a left schematic diagram of the structure of a clamping disc support, a vibration sensor and a CMOS image sensor according to an embodiment of the present invention; Figure 4 It is a partial front view schematic diagram of the structure of a clamping disc support, a vibration sensor and a CMOS image sensor according to an embodiment of the present invention.

[0044] like Figures 2 to 4 As shown, the multi-caliper disc brake has five brake calipers 1 , and each brake caliper 1 is provided with a vibration sensor 2 , a CMOS image sensor 3 and a torque sensor 4 .

[0045] like Figure 2 and Figure 3 As shown, the brake bracket 5 is used to support and fix the elevator brake to ensure the stable operation of the brake caliper; 5 brake calipers 1 are arranged on the caliper disc bracket 6, and the vibration sensor 2 arranged on each brake caliper is used to collect the vibration signal of each brake caliper; the CMOS image sensor 3 installed on the brake caliper is used for wear detection; the torque sensor 4 installed on the brake caliper collects the torque signal of each brake caliper. Specifically, the wear degree of the brake pad is judged after processing the image output by the CMOS image sensor, and then the wear amount of the brake pad is known; the torque sensor 4 adopts a photoelectric torque sensor, and each brake caliper is respectively installed with the torque sensor 4, and a grating bar of the torque sensor is connected to the transmission shaft 7. When the transmission shaft 7 is slightly deformed by the torsion, the grating bar connected thereto will have a slight displacement relative to the fixed grating, and the displacement will change the optical signal transmission between the gratings, and then generate an optical signal proportional to the torque. The torque sensor converts the optical signal into an electrical signal, thereby measuring the braking torque on each brake caliper when the elevator is braked.

[0046] In some embodiments, a wireless transmission module is wirelessly connected to the aforementioned acquisition instrument including a vibration sensor, a torque sensor, and a CMOS image sensor. The wireless transmission module is used to wirelessly receive and summarize data from the acquisition card, and then transmit the data to the next processing level for subsequent processing. An edge processing unit is integrated in the wireless transmission module to store the collected data in a database and convert it into a corresponding supporting data format, thereby ensuring the effective storage and transmission of multi-signal data. The corresponding port is opened at the data center end so that the client can smoothly read and obtain the data stored in the database.

[0047] In step S12, the first vibration health feature and the first torque health feature are obtained according to the first encoder and the third encoder of the first level comparison and fusion module and the vibration signal and the torque signal. In this embodiment, the signal collected by the sensor is preprocessed, for example, the collected signal is amplified by an amplifier, and then processed by a filter to reduce noise and interference, and the processed signal is input into the network for subsequent processing.

[0048] In step S13, the second vibration health feature and the second torque health feature are obtained according to the second encoder and the fourth encoder of the second level comparison and fusion module and the first vibration health feature and the first torque health feature. Through cross-sensor comparison learning, features from different sensors can be fused.

[0049] In this embodiment, the vibration and torque health features are obtained through a two-level contrast fusion network structure (TCF-Net). In a further implementation, the first vibration health feature matrix, the first torque health feature matrix, the second vibration health feature matrix, and the second torque health feature matrix are determined according to the first vibration health feature, the first torque health feature, the second vibration health feature, and the second torque health feature; the similarity loss of the vibration signal features between the brake calipers and the similarity loss of the torque signal features are calculated according to the first vibration health feature matrix and the first torque health feature matrix; the contrast loss between the vibration signal features and the torque signal features of the brake caliper is calculated according to the second vibration health feature matrix and the second torque health feature matrix; and the two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module is trained according to the similarity loss and contrast loss.

[0050] The two-level contrast fusion network structure includes a first-level contrast fusion stage and a second-level contrast fusion stage. In the first-level contrast fusion stage, the encoder 1 of the first-level contrast fusion module is used to encode the signal collected by the vibration sensor to generate a first vibration health feature matrix, and the similarity loss of the vibration signal features between different brake calipers is calculated based on the first vibration health feature matrix. Similarly, the torque signal is encoded by the encoder 3 to generate a first torque health feature matrix, and the similarity loss of the torque signal features between different brake calipers is calculated based on the first torque health feature matrix. In the second-level contrast fusion stage, the vibration signal and the torque signal from the vibration sensor and the torque sensor are processed in depth by the encoder 2 and the encoder 4 of the second-level contrast fusion module respectively to extract deep vibration and torque features (i.e., the second vibration health feature and the second torque health feature), and the contrast loss between the vibration signal and the torque signal features is calculated based on the second vibration health feature and the second torque health feature. Further, the joint two-level contrast fusion network training is performed according to the similarity loss and contrast loss.

[0051] In step S14, the health status feature similarity between the brake calipers is calculated according to the second vibration health feature and the second torque health feature.

[0052] The present invention provides a specific example of a two-level comparison fusion network structure for fault detection of a multi-caliper disc elevator brake. Figure 5 and Figure 6 It should be understood that the present invention is not limited to this specific example, and other neural network architectures for data fusion known in the art may be used.

[0053] Figure 5 is a schematic diagram of a two-stage comparison fusion network structure for fault detection of a multi-caliper disc elevator brake according to an embodiment of the present invention; Figure 6 Schematic diagram of a two-stage learning network architecture according to an embodiment of the present invention.

[0054] like Figure 5 As shown in the figure, the two-level contrast fusion network structure (TCF-Net) consists of four main parts: sensor data input, first-level contrast fusion, second-level contrast fusion, and fault diagnosis. Sensor data input, first-level contrast fusion, and second-level contrast fusion are used for neural network training; the fault diagnosis part is used for multi-clamp fault diagnosis reasoning.

[0055] Among them, the first-level contrast fusion and the second-level contrast fusion parts constitute the two-level learning (TLL) part, which is used to optimize the network training process and improve the interpretability of the network. Figure 6 The two-level learning network framework of the two-level learning part is shown as Figure 6 As shown, TLL consists of two-stage network connections, where the vibration signal and torque signal is the input of the first-level network, The output of the first-level network is also the input of the second-level network. is the output of the second-level network. Each layer of the network is associated with its own network learning loss function, where represents the loss function of the first layer network, Represents the loss function of the second layer network.

[0056] In the process of neural network parameter learning, the error term of the network layer It is expressed by the following formula (1):

[0057] (1)

[0058] In formula (1), from the network Action layer to The error propagation matrix of the active layer is recorded as ∈ ( > ),in and They are The effect layer and The number of neurons in the active layer; ∈ Represents the extended co-location matrix, which consists of the co-location matrix and zeros. It can effectively prevent the gradient from disappearing during network training, thereby improving the convergence speed; and Represent the number of layers of the first layer network and the second layer network respectively.

[0059] like Figure 5 As shown in the figure, the first-level contrast fusion module is the first component module of the TCF network. The first-level contrast fusion module consists of two parts: the first-level vibration signal fusion and the first-level torque signal fusion. It extracts features that are sensitive to health conditions by comparing the vibration signal features and torque signal features of different brake calipers in the brake disc. In this example, the first feature extraction is performed using the following formulas (2) and (3):

[0060] (2)

[0061] (3)

[0062] Among them, the health feature of the vibration signal (the first vibration health feature) = [ , , ... , ]; Health characteristics of torque signal (first torque health characteristics) = [ , , ... , ]; , ∈ , represents the number of brake calipers, 𝑁 represents the number of brake calipers, 𝐿 is the length of the eigenvector, represents the set of real numbers; , ∈ They represent the input vibration signal and torque signal respectively, and Encode represents the encoder code.

[0063] In this example, the similarity loss of the vibration signal features and the similarity loss of the torque signal features between the brake calipers are calculated by the following formulas (4) and (5):

[0064] (4)

[0065] (5)

[0066] in: , They represent the similarity loss of vibration signal features and the similarity loss of torque signal features respectively.

[0067] The second-level contrast fusion module includes building two encoders (encoders 2 and 4) to extract deep vibration and torque features (i.e., second vibration health features and second torque health features) through the sensor. In this example, the second feature extraction is performed using the following formulas (6) and (7):

[0068] (6)

[0069] (7)

[0070] in, and They represent the first vibration health feature and the first torque health feature output by the first-level fusion module respectively; Encode represents the encoder code; =[ , , ... , ]and =[ , , ..., ] represent the second-level vibration and torque characteristics (second vibration health characteristics and second torque health characteristics), , ∈ .

[0071] After the first-level contrast fusion module extracts the first health-sensitive feature, the second-level contrast fusion module is used to extract the single caliper / brake caliper feature. Its purpose is to embed more empirical knowledge into the model by learning the different features of the samples, thereby enhancing the generalization ability of the model. This solves the problem in the prior art that the network trained to extract single-level features is often difficult to achieve satisfactory convergence due to the scarcity of fault samples.

[0072] In this example, the contrast loss between the vibration signal characteristics and the torque signal characteristics of the brake caliper is calculated by the following formula (8):

[0073] (8)

[0074] in, N b Right now N batch ; ∈ is a label matrix whose diagonal elements represent the labels of each brake caliper and the remaining elements are set to 0; Tr(…) represents the trace of the matrix; and, the formula (8) is calculated by the following formulas (9) and (10): and :

[0075] (9)

[0076] (10)

[0077] in, and They are vibration characteristic loss and moment characteristic loss respectively; f T , f V They are vibration feature loss matrix and torque feature loss matrix respectively; the softmax function is a normalized exponential function.

[0078] By calculating the contrast loss, the coupling relationship between vibration features and torque features can be learned through second-level contrastive learning, thereby achieving second-level contrastive fusion.

[0079] In the fault diagnosis part, the health-related features are extracted through the two-level contrast fusion network structure of the above example, and the health status similarity of multiple brake calipers (clusters) is calculated by the following formulas (11) and (12):

[0080] (11)

[0081] (12)

[0082] in, H V and H T are the vibration health state similarity and torque health state similarity of the i-th brake caliper respectively.

[0083] In a further embodiment, the range of the brake caliper that may be faulty is determined based on the similarity of the health status characteristics of the vibration and torque signals, and the total health status value of the brake caliper that may be faulty is subsequently calculated in combination with the wear amount of the brake caliper obtained by the CMOS image sensor to determine which specific one or several brake calipers have a problem, thereby achieving accurate fault location of the multi-caliper disc brake of a high-speed elevator. Among them, since the brake calipers in the same cluster usually operate in the same environment for a long time, the health status similarity of the faulty brake caliper will be significantly lower than that of other brake calipers. For example, when the difference value between the health status similarity of one brake caliper and the health status similarity of other brake calipers is greater than 10%, it is determined that the brake caliper is faulty.

[0084] In step S15 , the wear amount of the corresponding brake caliper is calculated based on the captured image.

[0085] In some embodiments, calculating the wear amount of the corresponding brake caliper based on the captured image includes: acquiring a segmented image based on the captured image; calculating the gap circumference and the gap area of ​​the brake caliper based on the segmented image and its corresponding mask image; calculating the difference judgment value of the brake caliper based on the gap circumference and the gap area; and calculating the wear amount of the brake caliper based on the difference judgment value of the brake caliper and the difference judgment value of an unworn brake caliper.

[0086] The present invention provides a specific example of extracting edge features from captured images collected by a CMOS image sensor to obtain the wear amount / wear state of a brake caliper, which calculates the wear amount of the brake caliper based on an improved U-type network (ReUNet). Figure 7 and Figure 8 To explain in detail:

[0087] Figure 7 is a structural schematic diagram of an improved U-type network (ReUNet) for edge feature extraction of an image captured by a CMOS image sensor according to an embodiment of the present invention; Figure 8 yes Figure 7 Schematic diagram of the processing flow of the Adaptive Channel Enhancement Module (ACEM) introduced in the improved U-type network (ReUNet).

[0088] like Figure 7 As shown in the figure, the ReUNet network architecture starts with a dual down sampling module (DDSM). The DDSM module generates a feature map rich in multi-scale information by combining the parallel processing paths of the dilated convolution and the conventional convolution, thereby fully capturing the global and local features of the input image (i.e., the captured image of each brake caliper). The Adaptive Channel Enhancement Module (ACEM) is introduced in the jump connection stage of the ReUNet network. The ACEM module performs a channel attention mechanism on the aforementioned feature map to enhance the correlation between channels and reduce the loss of key information. Finally, the ReUNet network outputs a segmented image to perform a post-processing step, and the wear condition of the brake caliper is accurately determined by evaluating the similarity between the segmented image result and the reference mask image.

[0089] Among them, Figure 7As shown in the figure, the feature extraction stage of the ReUNet network first performs a preliminary convolution operation on the input image (i.e., the captured image of each brake caliper). Specifically, the image first passes through a conventional 3×3 convolution layer to increase the dimension to 64 channels before being downsampled by the dual downsampling module DDSM. Then, the convolutional feature map is further fed into the dual downsampling path of the dual downsampling module DDSM for in-depth feature extraction.

[0090] In the dual downsampling path, the initial step is to perform a max pooling operation on the input feature map (the convolutional feature map), thereby reducing the size of the feature map by half. The reduced-size feature map is then input to the Dual Down Sampling Module (DDSM). The dual downsampling path consists of four DDSM layers stacked together, each of which processes the dilated convolution and normal convolution of the input feature map, thereby effectively extracting global and local features from the input image.

[0091] At the same time, in the improved U-net ReUNet example, in order to enhance the interaction between channels and reduce the loss of key channel information, the Adaptive Channel Enhancement Module (ACEM) is introduced on the basis of the U-Net architecture. The ACEM module compresses the feature map in the channel dimension to generate a spatial attention feature map. By integrating the generated spatial attention feature map into different levels of jump connections, the model can focus more on shallow edge details and deep spatial context information.

[0092] Specifically, if Figure 8 As shown, the ACEM module processes the input feature map In the process of inputting the image, the size of the feature map is defined as height H and width W, thereby reducing the dimension of the entire feature map to one The characteristic vector of The Xth element of the eigenvector of is , which is obtained by applying the maximum pooling operation expressed by the following formula (13) on the Xth channel:

[0093] (13)

[0094] in, is the compression function; The feature map The middle position ( ) specific elements.

[0095] Secondly, an adaptive channel enhancement operation is performed, which learns the importance of features by introducing a set of weight parameters of the fully connected layer and adjusts the weights of each channel accordingly. Specifically, the adaptive channel enhancement operation includes the following steps a and b: Step a: In the feature compression stage, a fully connected layer (through Figure 8 The first feature map in the full connection layer is concatenated to obtain the fully connected layer) to reduce the dimension of the feature vector from X to X / s, where s is the predetermined reduction ratio. Then, the compressed feature vector is nonlinearly transformed by applying the ReLU function represented by the following formula (14) to obtain the intermediate vector ,

[0096] (14)

[0097] in, is the intermediate function; is the ReLU function; is the compression weight, .

[0098] Step b: Process through a fully connected layer (corresponding to Figure 8 The second feature map concatenation process in is performed to restore the dimension of the intermediate vector from the X / s dimension to the original X dimension. Then, the Sigmoid activation function expressed by the following formula (15) is applied to the output of the fully connected layer to perform an element-by-element activation operation, thereby obtaining a vector reflecting the weight of each channel. , :

[0099] (15)

[0100] in, is the activation function; is the Sigmoid function; is the incentive weight, .

[0101] Finally, the weight vector is converted into With input features Y Input result function graph Multiply to get the spatial attention feature map (Right now Figure 8 The output image processed by the ACEM module shown in the figure) completes the channel attention operation:

[0102] (16)

[0103] in, is the weight vector With the input feature map Y The input result function.

[0104] In a further embodiment, by evaluating the segmented image (i.e., Figure 7 The weighted calculated difference (i.e., the difference discrimination value) between the final output image obtained by the improved U-shaped network processing shown in FIG. 1 and the corresponding mask image in the circumference of the brake caliper gap and the gap area ), to accurately determine the wear state of the caliper disc. Specifically, the calculation method involves extracting features from the perimeter and area of ​​the brake caliper gap, and comprehensively considering the features through set weight parameters (perimeter weight and area weight of the gap), thereby calculating the similarity difference between the two (i.e., the difference discrimination value ). This similarity comparison algorithm can provide a quantitative assessment of the wear degree of the brake caliper, thereby providing a scientific basis for fault diagnosis and subsequent maintenance decisions.

[0105] In this embodiment, the similarity difference is calculated by the following steps: First, the circumference L and area S of each brake caliper gap in the segmented image are calculated by using the function for calculating the contour circumference and area in the OpenCV library (OpenSource Computer Vision Library is an open source library for computer vision tasks), and the gap circumference L and the gap area S are weightedly summed by the following formula (17) to obtain the difference discrimination value: :

[0106] (17)

[0107] Wherein, α is the gap perimeter weight; β is the gap area weight. In an exemplary embodiment, α=0.25, β=0.75.

[0108] Secondly, the current brake caliper sample wear amount is calculated by the following formula (18): Size:

[0109] (18)

[0110] in. This is the difference judgment value for an unworn brake caliper.

[0111] In an alternative embodiment, directly based on the wear amount The size of the brake caliper can be used to determine whether the brake caliper is normal or abnormally worn. Specifically, if the wear value of a brake caliper is significantly higher than the wear value of other brake calipers by 0.025 or more, the brake caliper is confirmed to be abnormal.

[0112] In step S16, the total health state value of the brake caliper is calculated according to the health state feature similarity and the wear amount.

[0113] In some embodiments, calculating the total health state value of the brake caliper according to the health state feature similarity and the wear amount includes: obtaining the total health state value of the brake caliper by weighted summing the inverse of the wear amount and the health state feature similarity. In this embodiment, the total health state value of the brake caliper is calculated by the following formulas (19) and (20):

[0114] (19)

[0115] (20)

[0116] Where Q is the total health status value of the brake caliper; is the wear of the brake caliper; H is the health status feature similarity; a is the wear weight; b is the health status feature similarity weight. In an exemplary embodiment, a=0.3, b=0.7. Optionally, under the premise of ensuring the reliability and stability of the data, for a high-speed elevator multi-caliper disc brake, the possible failure is not only caused by the wear of the caliper, but also may be caused by other reasons. Therefore, the value of the health status feature similarity weight b is greater than the value of the wear weight a.

[0117] In step S17 , a faulty brake caliper is determined based on the total health state value.

[0118] In some embodiments, determining a faulty brake caliper according to the total health status value includes: determining that the brake caliper is faulty when the total health status value of a brake caliper is lower than a health status threshold, wherein the health status threshold is the total health status value of a healthy caliper.

[0119] The above fault detection method of the embodiment of the present invention is adopted, by collecting the vibration signal, torque signal and captured image of each brake caliper in the high-speed elevator multi-caliper disc brake as the original data set, and performing multi-data fusion through a two-level comparison fusion network to obtain the health status feature similarity between the brake calipers, and then determining the total health status value of the brake caliper according to the health status feature similarity and the wear amount calculated based on the captured image for fault diagnosis, which can improve the accuracy and reliability of fault diagnosis, realize the precise positioning of the brake caliper fault, and thus ensure the safety of elevator operation. In addition, the fault detection method does not rely on existing fault samples, but only relies on the real-time measured performance parameters to realize the fault diagnosis and positioning of the faulty caliper disc / brake caliper, and integrates multi-sensor data, solving the problems of lack of failure samples and difficulty in feature extraction in the prior art.

[0120] Fig. 9is an architecture diagram of a fault detection system for a multi-caliper disc elevator brake based on a two-level comparison fusion network according to an embodiment of the present invention, wherein the brake includes a plurality of brake calipers. Fig. 9 As shown, the fault detection system comprises:

[0121] The original data set acquisition module 910 is used to acquire the original data set of the brake, and the original data set includes: a vibration signal of each brake caliper, a torque signal of each brake caliper, and a captured image of each brake caliper.

[0122] The first feature extraction module 920 is used to obtain a first vibration health feature and a first torque health feature according to the first encoder and the third encoder of the first-level comparison and fusion module and the vibration signal and the torque signal.

[0123] The second feature extraction module 930 is used to obtain a second vibration health feature and a second torque health feature according to the second encoder and the fourth encoder of the second-level comparison and fusion module and the first vibration health feature and the first torque health feature.

[0124] The health status feature similarity calculation module 940 is used to calculate the health status feature similarity between the brake calipers according to the second vibration health feature and the second torque health feature.

[0125] The wear amount calculation module 950 is used to calculate the wear amount of the corresponding brake caliper according to the captured image.

[0126] The health state value calculation module 960 is used to calculate the total health state value of the brake caliper according to the health state feature similarity and the wear amount.

[0127] The fault determination module 970 is configured to determine a faulty brake caliper according to the total health status value.

[0128] The network training module 980 is used to perform the following operations: determine the first vibration health feature matrix, the first torque health feature matrix, the second vibration health feature matrix, and the second torque health feature matrix according to the first vibration health feature, the first torque health feature, the second vibration health feature, and the second torque health feature; calculate the similarity loss of the vibration signal features and the similarity loss of the torque signal features between the brake calipers according to the first vibration health feature matrix and the first torque health feature matrix; calculate the contrast loss between the vibration signal features and the torque signal features of the brake caliper according to the second vibration health feature matrix and the second torque health feature matrix; train the two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module according to the similarity loss and contrast loss.

[0129] The above fault detection system of the embodiment of the present invention collects the vibration signal, torque signal and captured image of each brake caliper in the high-speed elevator multi-caliper disc brake as the original data set, and performs multi-data fusion through a two-level comparison fusion network to obtain the similarity of the health status characteristics between the brake calipers, and then determines the total health status value of the brake caliper according to the health status characteristic similarity and the wear amount calculated based on the captured image for fault diagnosis, which can improve the accuracy and reliability of fault diagnosis, realize the precise positioning of the brake caliper fault, and thus ensure the safety of elevator operation. In addition, the fault detection method does not rely on existing fault samples, but only relies on the performance parameters measured in real time to realize the fault diagnosis and positioning of the faulty caliper disc / brake caliper, and integrates multi-sensor data, solving the problems of lack of failure samples and difficulty in feature extraction in the prior art.

[0130] Through the description of the above implementation modes, those skilled in the art can clearly understand that the present invention can be implemented by combining software with a hardware platform. Based on such an understanding, all or part of the contribution of the technical solution of the present invention to the background technology can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation modes of the present invention or some parts of the implementation modes.

[0131] Correspondingly, an embodiment of the present invention further provides a computer-readable storage medium, on which computer-readable instructions or programs are stored, and when the computer-readable instructions or programs are executed by a processor, the computer performs the following operations: the operations include the steps included in the fault detection method described in any of the above embodiments, which are not repeated here. The storage medium may include, for example, a CD, a hard disk, a floppy disk, a flash memory, a magnetic tape, etc.

[0132] In addition, an embodiment of the present invention further provides a computer device including a memory and a processor, wherein the memory is used to store one or more computer-readable instructions or programs, wherein the one or more computer-readable instructions or programs can implement the fault detection method described in any of the above embodiments when executed by the processor. The computer device can be, for example, a server, a desktop computer, a notebook computer, a tablet computer, etc.

[0133] The present invention also provides a computer program product including a computer program, wherein the computer program includes program code for executing the fault detection method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the fault detection method provided by the embodiment of the present disclosure.

[0134] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention is described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention. Therefore, the protection scope of the present invention shall be subject to the claims.

Claims

1. A fault detection method for a multi-caliper disc elevator brake based on a two-level contrast fusion network, wherein the brake comprises a plurality of brake calipers, characterized in that: The fault detection method comprises: Acquire an original data set of the brake, wherein the original data set includes: a vibration signal of each brake caliper, a torque signal of each brake caliper, and a captured image of each brake caliper; Acquire a first vibration health feature and a first torque health feature according to the first encoder and the third encoder of the first-level comparison and fusion module and the vibration signal and the torque signal; Acquire a second vibration health feature and a second torque health feature according to the second encoder and the fourth encoder of the second-level comparison and fusion module and the first vibration health feature and the first torque health feature; Calculating the health status feature similarity between the brake calipers according to the second vibration health feature and the second torque health feature; calculating the wear amount of the corresponding brake caliper according to the captured image; Calculating the total health state value of the brake caliper according to the health state characteristic similarity and the wear amount; A faulty brake caliper is determined based on the total health state value.

2. The fault detection method according to claim 1, characterized in that: collecting a vibration signal of each brake caliper by a vibration sensor; Collecting the torque signal of each brake caliper through a torque sensor; The captured images of each brake caliper are collected by a CMOS image sensor.

3. The fault detection method according to claim 1, characterized in that: The fault detection method further comprises: Determine a first vibration health feature matrix, a first moment health feature matrix, a second vibration health feature matrix, and a second moment health feature matrix according to the first vibration health feature, the first moment health feature, the second vibration health feature, and the second moment health feature; Calculating the similarity loss of vibration signal features and the similarity loss of torque signal features between brake calipers according to the first vibration health feature matrix and the first torque health feature matrix; Calculating the contrast loss between the vibration signal feature and the torque signal feature of the brake caliper according to the second vibration health feature matrix and the second torque health feature matrix; A two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module is trained according to the similarity loss and the contrast loss.

4. The fault detection method according to claim 1, characterized in that: Calculating the wear amount of the corresponding brake caliper according to the captured image includes: Acquire a segmented image according to the captured image; Calculating the gap perimeter and gap area of ​​the brake caliper according to the segmented image and its corresponding mask image; Calculating a difference discrimination value of the brake caliper according to the gap perimeter and the gap area; The wear amount of the brake caliper is calculated based on the difference determination value of the brake caliper and the difference determination value of the unworn brake caliper.

5. The fault detection method according to claim 1, characterized in that: Calculating the total health state value of the brake caliper according to the health state feature similarity and the wear amount includes: obtaining the total health state value of the brake caliper by weighted summing the inverse of the wear amount and the health state feature similarity; Determining a faulty brake caliper based on the total health state value includes: determining that the brake caliper is faulty when the total health state value of the brake caliper is lower than a health state threshold.

6. A fault detection system for a multi-caliper disc elevator brake based on a two-level contrast fusion network, wherein the brake comprises a plurality of brake calipers, characterized in that: The fault detection system comprises: A raw data set acquisition module, used to acquire the raw data set of the brake, wherein the raw data set includes: a vibration signal of each brake caliper, a torque signal of each brake caliper, and a photographed image of each brake caliper; A first feature extraction module, used for acquiring a first vibration health feature and a first torque health feature according to the first encoder and the third encoder of the first-level comparison and fusion module and the vibration signal and the torque signal; A second feature extraction module, configured to obtain a second vibration health feature and a second torque health feature according to the second encoder and the fourth encoder of the second-level comparison and fusion module and the first vibration health feature and the first torque health feature; a health status feature similarity calculation module, configured to calculate the health status feature similarity between brake calipers according to the second vibration health feature and the second torque health feature; a wear amount calculation module, used for calculating the wear amount of the corresponding brake caliper according to the captured image; A health status value calculation module, used for calculating the total health status value of the brake caliper according to the health status feature similarity and the wear amount; A fault determination module is used to determine a faulty brake caliper according to the total health state value.

7. The fault detection system according to claim 6, characterized in that: The fault detection system further includes a network training module, which is used to perform the following operations: Determine a first vibration health feature matrix, a first moment health feature matrix, a second vibration health feature matrix, and a second moment health feature matrix according to the first vibration health feature, the first moment health feature, the second vibration health feature, and the second moment health feature; Calculating the similarity loss of vibration signal features and the similarity loss of torque signal features between brake calipers according to the first vibration health feature matrix and the first torque health feature matrix; Calculating the contrast loss between the vibration signal feature and the torque signal feature of the brake caliper according to the second vibration health feature matrix and the second torque health feature matrix; A two-level contrast fusion network including the first-level contrast fusion module and the second-level contrast fusion module is trained according to the similarity loss and the contrast loss.

8. A computer-readable storage medium storing computer-readable instructions, characterized in that: The computer-readable instructions are executed by a processor to implement a fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network as described in any one of claims 1-5.

9. A computer device comprising a memory and a processor, The memory stores computer-readable instructions, characterized in that: The processor executes the computer-readable instructions to implement the fault detection method for a multi-caliper disc elevator brake based on a two-level comparison fusion network according to any one of claims 1 to 5.

Images