Fault prediction and early warning method, device and equipment for elevator brake and medium

Abstract

The invention provides an elevator brake fault prediction and early warning method, device and equipment and a medium, and relates to the technical field of elevator maintaining.The method comprises the steps that original vibration characteristic data detected by an FBG vibration sensor installed on a to-be-predicted elevator brake is obtained, inputting the original vibration characteristic data into a preset DSMRA model which is constructed by taking a model loss function formed by a reconstruction error and a manifold regularization item as a target function and taking model parameters of each layer of encoder and each layer of decoder as optimization variables for characteristic extraction, and obtaining reconstructed vibration characteristic data; inputting the reconstructed vibration characteristic data into a preset DAGU model constructed based on a channel attention mechanism, a space attention mechanism and a gating mechanism, and outputting a predicted signal characteristic value; if the variance characteristic value of the calculated and predicted signal characteristic value is larger than the preset threshold value, the fault early warning prompt signal is sent, the effectiveness and efficiency of characteristic extraction are improved, the calculated amount is reduced, and the accuracy of fault prediction of the elevator brake is effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of elevator maintenance, and particularly relates to a method, device, equipment and medium for predicting and warning elevator brake failures. Background Art

[0002] The products in the elevator industry are mainly used to facilitate the vertical transportation of people and goods between different floors of a building. In the actual application of elevators, elevators are always in a high-frequency and heavy-load working condition, and certain wear will occur on the electrical protection devices and mechanical parts inside the elevator. Compared with other various safety protection devices of the elevator, the working state of the brake is the most frequent. The elevator needs the corresponding opening and braking of the brake when starting and stopping. The long-term frequent operation of the brake will inevitably lead to problems such as wear, aging, and loosening of the components. The failure of the elevator brake will inevitably cause serious elevator accidents. If the brake fails, there will be a huge safety hazard in the elevator. Since it cannot be braked normally, it may even crash, posing a great threat to the lives of both passengers and maintenance personnel. Therefore, developing an elevator brake fault warning system is of great significance. By analyzing the vibration signal of the brake, it is possible to determine when the brake will fail, help passengers take the elevator safely, and assist maintenance personnel in maintaining elevator safety, and monitor the elevator operation process in real time to eliminate faults in advance.

[0003] There are the following problems in the existing elevator fault diagnosis and fault warning methods: (1) The sensitivity and frequency range of traditional vibration sensors are limited. They can neither detect the subtle changes during micro-vibrations nor work only within a specific frequency range. These drawbacks cannot meet the current elevator's requirements for safety and maintenance and also limit the accuracy of diagnosis. (2) In the traditional electrical brake fault warning, the extraction of characteristic data usually relies on manual operation. This method is not only inefficient but also results in uneven data quality due to the knowledge differences among different personnel. Therefore, in modern elevator maintenance methods, relying on manual feature extraction cannot fully guarantee the safe operation of the elevator. (3) The fault warning model cannot effectively process high-dimensional data, often has difficulty distinguishing which features are highly correlated with equipment failures, easily leads to information redundancy or poor feature selection, and at the same time has difficulty effectively capturing the dynamically changing key features and cannot adapt to the changes in the equipment state at different time points.

[0004] In summary, there is an urgent need to propose an elevator brake fault prediction and warning method that can effectively improve the effectiveness and efficiency of feature extraction and the accuracy of elevator brake fault prediction. Summary of the Invention

[0005] The object of the present application is to address the above problems and provide a method, device, equipment, and medium for predicting and warning elevator brake failures, so as to improve the effectiveness and efficiency of feature extraction and the accuracy of elevator brake failure prediction.

[0006] In a first aspect, the present application provides a method for predicting and warning elevator brake failures, including:

[0007] Obtain the original vibration feature data detected by an FBG vibration sensor installed on the elevator brake to be predicted;

[0008] Input the original vibration feature data into a preset DSMRA model to perform feature extraction on the original vibration feature data, and obtain reconstructed vibration feature data; the preset DSMRA model is constructed with the model loss function composed of the reconstruction error and the manifold regularization term as the objective function and the model parameters of each layer of encoder and each layer of decoder as the optimization variables;

[0009] Input the reconstructed vibration feature data into a preset DAGU model, and output the predicted signal eigenvalue; the preset DAGU model is constructed based on the channel attention mechanism, the spatial attention mechanism, and the gating mechanism;

[0010] Calculate the variance eigenvalue of the predicted signal eigenvalue. If the variance eigenvalue is greater than a preset threshold, send a fault warning prompt signal.

[0011] According to the technical solution provided by the present application, the step of inputting the original vibration feature data into a preset DSMRA model to perform feature extraction on the original vibration feature data and obtain reconstructed vibration feature data includes:

[0012] Successively encode the original vibration feature data through each layer of encoder of the preset DSMRA model to obtain hidden layer feature data;

[0013] Calculate the manifold regularization term according to the hidden layer feature data and the feature similarity graph;

[0014] Successively decode the hidden layer feature data through each layer of decoder of the preset DSMRA model to obtain output feature data;

[0015] Calculate the reconstruction error according to the original vibration feature data and the output feature data;

[0016] Construct the model loss function according to the manifold regularization term and the reconstruction error, and use the model loss function as the objective function and the model parameters of each layer of encoder and each layer of decoder as the optimization variables to perform model optimization training. When the model loss function meets the convergence condition, take the obtained output feature data as the reconstructed vibration feature data.

[0017] According to the technical solution provided by the present application, before inputting the original vibration feature data into a preset DSMRA model to extract features from the original vibration feature data and obtain reconstructed vibration feature data, the method further includes:

[0018] Construct an encoding mapping function for each layer of encoder based on an encoding weight matrix, an encoding bias vector, and an encoding activation function;

[0019] Construct a decoding mapping function for each layer of decoder based on a decoding weight matrix, a decoding bias vector, and a decoding activation function.

[0020] According to the technical solution provided by the present application, the method further includes:

[0021] Construct a channel attention mechanism function based on a channel weight matrix, a channel bias vector, and a channel activation function;

[0022] Construct a spatial attention mechanism function based on a spatial projection matrix, a scaling factor, and a spatial activation function;

[0023] Construct a gating mechanism function based on a gating weight matrix and a gating bias vector according to the channel attention mechanism function and the spatial attention mechanism function;

[0024] Construct the preset DAGU model according to the gating mechanism function, the channel attention mechanism function, and the spatial attention mechanism function.

[0025] According to the technical solution provided by the present application, calculating the manifold regularization term according to the hidden layer feature data and the feature similarity graph includes:

[0026] Obtain a similarity weight matrix according to the feature similarity graph;

[0027] According to the hidden layer feature data and the similarity weight matrix, according to the formula:

[0028]

[0029] Calculate the manifold regularization term; where, Ω(f) is the manifold regularization term; W ij is the similarity weight matrix; x i is the i-th hidden layer feature data; x j is the j-th hidden layer feature data; Pf(x i ) - f(x j )P 2 is the distance between the i-th hidden layer feature data x i and the j-th hidden layer feature data x j in the feature space.

[0030] According to the technical solution provided by the present application, based on the channel attention mechanism function and the spatial attention mechanism function, a gating mechanism function is constructed based on a gating weight matrix and a gating bias vector, including:

[0031] According to the formula:

[0032]

[0033] Construct the gating mechanism function; where G is the output of the gating mechanism function; σ(g) is a non-linear activation function; W g is the gating weight matrix; is the hidden layer feature data; is the output of the channel attention mechanism function; b g is the gating bias vector; W c is the channel weight matrix; b c is the channel bias vector; is the output of the spatial attention mechanism function; sofemax(g) is a probability distribution conversion function; W q ,W k ,W v are all spatial projection matrices; d k is a scaling factor; is a concatenation matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

[0034] According to the technical solution provided by the present application, based on the gating mechanism function, the channel attention mechanism function and the spatial attention mechanism function, the preset DAGU model is constructed, including:

[0035] According to the formula:

[0036]

[0037] Construct the preset DAGU model; where O is the predicted signal eigenvalue output by the preset DAGU model; G is the output of the gating mechanism function; is a concatenation matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

[0038] In a second aspect, the present application provides an elevator brake fault prediction and warning device, including:

[0039] An acquisition module, configured to acquire original vibration feature data detected by an FBG vibration sensor installed on an elevator brake to be predicted;

[0040] A processing module, configured to input the original vibration feature data into a preset DSMRA model to perform feature extraction on the original vibration feature data, and obtain reconstructed vibration feature data; the preset DSMRA model is constructed by taking a model loss function composed of a reconstruction error and a manifold regularization term as an objective function, and taking the model parameters of each layer of encoder and each layer of decoder as optimization variables;

[0041] A prediction module, configured to input the reconstructed vibration feature data into a preset DAGU model and output a predicted signal eigenvalue; the preset DAGU model is constructed based on a channel attention mechanism, a spatial attention mechanism, and a gating mechanism;

[0042] An early warning module, configured to calculate a variance eigenvalue of the predicted signal eigenvalue, and if the variance eigenvalue is greater than a preset threshold, send a fault early warning prompt signal.

[0043] In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where when the processor executes the computer program, the method described above is implemented.

[0044] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described above is implemented.

[0045] Compared with the prior art, the beneficial effects of the present application: The elevator brake fault prediction and early warning method, device, equipment and medium provided by the present application obtain the original vibration characteristic data detected by the FBG vibration sensor installed on the elevator brake to be predicted, and input the original vibration characteristic data into a preset DSMRA model constructed with the model loss function composed of the reconstruction error and the manifold regularization term as the objective function and the model parameters of each layer of encoder and each layer of decoder as the optimization variables for feature extraction to obtain the reconstructed vibration characteristic data; input the reconstructed vibration characteristic data into a preset DAGU model constructed based on the channel attention mechanism, the spatial attention mechanism and the gating mechanism, and output the predicted signal eigenvalue; if the variance eigenvalue of the calculated predicted signal eigenvalue is greater than the preset threshold, a fault early warning prompt signal is sent. First, the characteristics of the FBG sensor are used to accurately detect the original vibration characteristic data. The improved DSMRA network model proposed in this application adds the manifold regularization, maintains the local geometric structure of the data in the feature space, enhances the effectiveness of feature extraction, and the multi-layer encoder extracts features layer by layer, enhancing the expression ability of the model, reducing the dependence on labeled data, retaining the data distribution structure inside the hidden layer of the encoder, and reducing the manual extraction steps that must be based on expert experience. This not only reduces the manual extraction steps, but also improves the effectiveness and efficiency of feature extraction. And an improved DAGU model is proposed, which uses a dual attention mechanism, namely spatial attention and channel attention. The channel attention mechanism automatically discovers the correlation of different types of vibration data, so as to be able to capture the changes in different time periods during the operation of the equipment. The spatial attention mechanism can automatically identify the vibration data more relevant to the fault early warning, realize trend prediction, ignore irrelevant information, reduce the amount of calculation, and effectively improve the accuracy of elevator brake fault prediction. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions in the embodiments, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0047] Figure 1 It is a flowchart of a method for predicting and early warning elevator brake faults provided by an embodiment of the present application;

[0048] Figure 2 It is a schematic installation diagram of the FBG vibration sensor provided by an embodiment of the present application;

[0049] Figure 3 It is a schematic structural diagram of the FBG vibration sensor provided by an embodiment of the present application;

[0050] Figure 4Schematic diagram of the preset DAGU model provided by the embodiment of the present application;

[0051] Figure 5 Schematic diagram of the preset DSMRA model provided by the embodiment of the present application;

[0052] Figure 6 Schematic diagram of an elevator brake fault prediction and early warning device provided by the embodiment of the present application;

[0053] Figure 7 Schematic diagram of the computer system of an electronic device provided by the embodiment of the present application. Detailed implementation manners

[0054] In order to enable those skilled in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. The description in this part is only exemplary and explanatory, and should not have any restrictive effect on the protection scope of the present application. Specifically, the described embodiments are only some of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0055] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0056] In order to make the technical solutions of the present application clearer and easier to understand, the elevator brake fault prediction and early warning method provided by the embodiment of the present application will be introduced below.

[0057] As Figure 1 shown, this figure is a flowchart of an elevator brake fault prediction and early warning method provided by this embodiment. The method includes the following steps:

[0058] S101. Obtain the original vibration characteristic data detected by the FBG vibration sensor installed on the elevator brake to be predicted;

[0059] Specifically, as Figure 2As shown in the figure, in order to ensure the capture of the original vibration data of the elevator brake to be predicted, the Fiber Bragg Gratting (FBG) vibration sensor 1 is installed on the fixed bracket of the elevator brake to be predicted in advance. Before installation, ensure that the surface is flat, clean and free of oil stains. The installation surface can be polished with sandpaper and cleaned with alcohol, and the FBG vibration sensor 1 is fixed to the selected metal installation surface using a high-strength and vibration-resistant adhesive to ensure a tight fit with the installation surface and avoid loosening, so as to ensure the pasting effect of the FBG vibration sensor 1. Among them, the special adhesive for the FBG vibration sensor can be used to ensure that the temperature, mechanical stress and vibration endured by the FBG vibration sensor 1 during operation will not affect its bonding strength, so as to capture the original vibration data of the entire elevator brake to be predicted. As Figure 3As shown, the FBG vibration sensor 1 includes two FBGs 2, two cross springs 3, two bases 4, two L-shaped bases 5, two rigid beams 6, and two mass blocks 7. By integrating two FBGs into a compact structure, the reasonable configuration of the FBG 2, cross spring 3, base 4, rigid beam 6, and mass block 7 makes the whole small and light, suitable for installation in the elevator space. The cross spring 3 and the rigid beam 6 accurately conduct vibration or mechanical strain to the FBG 2, thus realizing high-sensitivity vibration detection; the two mass blocks 7 help to reduce external interference noise and optimize the vibration response, improving the anti-interference ability through structural design to ensure accurate measurement in a complex environment. The two FBGs can achieve bidirectional vibration detection, eliminating errors that may occur in a single grating. The complementarity of the two gratings improves the measurement accuracy and reliability; the L-shaped base 5 and the base 4 are easy to fix and install, adapting to the elevator environment; the two symmetric cross springs 3 can reduce the influence of friction and lateral crosstalk. When affected by external vibration, the mass block 7 will deflect vertically with the inertial force, and the rigid beam 6 converts the vertical deflection of the mass block 7 into an axial deflection. Due to the micro-vibration of the mass block 7 around the center of the cross spring 3, strain is generated, which in turn affects the central wavelength of its reflected light. By measuring the change in the central wavelength of the reflected light, the magnitude of the acceleration can be determined, thus realizing the measurement of vibration. Among them, the change in the central wavelength of the reflected light is positively correlated with both the linear displacement and the acceleration of the mass block 7. The above-mentioned multiple precision components included in the structure of the FBG vibration sensor need to be tightly fixed to prevent damage during vibration or impact. Use high-density ethylene-vinyl acetate copolymer (EVA) material to fix each component at its specific position and isolate them from each other to avoid friction and collision. The outer package should use a durable and compression-resistant hard material such as a high-strength plastic box, which can effectively prevent internal damage caused by external impact and compression. The FBG vibration sensor has the advantages of anti-electromagnetic interference characteristics, high sensitivity, easy networking for multi-point monitoring, simple structure, reliable measurement data results, corrosion resistance, etc., and can effectively monitor the elevator brake in real time. Compared with other traditional sensors such as inductive displacement vibration sensors, this application uses the FBG vibration sensor to accurately sense the fault vibration signal, achieving the effect of accurately collecting data by using the FBG vibration sensor. The installation position of the FBG vibration sensor is selected on the key components of the brake, such as the brake block, brake disc, or connection part, to capture the most obvious vibration signal, thereby capturing abnormal signals that may appear and obtaining sufficient and complete original vibration data.

[0060] S102. Input the original vibration feature data into a preset DSMRA model to extract features from the original vibration feature data, and obtain reconstructed vibration feature data. The preset DSMRA model is constructed with the model loss function composed of reconstruction error and manifold regularization term as the objective function, and the model parameters of each layer of encoder and each layer of decoder as the optimization variables.

[0061] Specifically, encode the original vibration feature data successively through each layer of encoder of the preset Denoising stackmanifold regularization autoencoder (DSMRA) model to obtain hidden layer feature data. Then, calculate the manifold regularization term according to the hidden layer feature data and the feature similarity graph. Next, decode the hidden layer feature data successively through each layer of decoder of the preset DSMRA model to obtain output feature data. Further, calculate the reconstruction error according to the original vibration feature data and the output feature data. Finally, construct the model loss function according to the manifold regularization term and the reconstruction error, and use the model loss function as the objective function, and the model parameters of each layer of encoder and each layer of decoder as the optimization variables to perform model optimization training until the model loss function meets the convergence condition to obtain the reconstructed vibration feature data, realizing effective noise reduction in an unsupervised manner.

[0062] S103. Input the reconstructed vibration feature data into a preset DAGU model to output the predicted signal feature value. The preset DAGU model is constructed based on the channel attention mechanism, spatial attention mechanism and gating mechanism.

[0063] Specifically, construct the channel attention mechanism function based on the channel weight matrix, channel bias vector and channel activation function. And construct the spatial attention mechanism function based on the spatial projection matrix, scaling factor and spatial activation function. Then, construct the gating mechanism function based on the channel attention mechanism function and the spatial attention mechanism function, based on the gating weight matrix and gating bias vector. Next, construct the preset network combined with the Double attention gatedunit (DAGU) model according to the gating mechanism function and the channel attention mechanism function and the spatial attention mechanism function. After completing the construction of the preset DAGU model, it can be as Figure 4The reconstructed vibration feature data is input into a preset DAGU model that combines a channel attention mechanism, a spatial attention mechanism, and a gating mechanism, and a predicted signal eigenvalue (i.e., a fault prediction result) is output. Among them, DAGU is a model structure simplified based on the Long Short-Term Memory (LSTM) and the Gated Recurrent Unit (GRU). Among them, LSTM is a variant problem proposed to solve the problems of gradient disappearance and gradient explosion that easily occur in the training process of the Recurrent Neural Network (RNN). Since the GRU model cannot distinguish the importance of transmitted information during the training process, information redundancy will be generated, affecting the training effect. Therefore, this application provides an improved DAGU model. Based on the DAGU model, after adding a dual attention mechanism, it can distinguish data with different importance levels, remove useless information, transmit useful information, and achieve the technical effects of improving processing ability and reducing the amount of calculation.

[0064] S104. Calculate the variance eigenvalue of the predicted signal eigenvalue. If the variance eigenvalue is greater than a preset threshold, a fault warning prompt signal is sent.

[0065] Specifically, calculate the variance eigenvalue of the predicted signal eigenvalue. The variance eigenvalue is used to characterize the volatility and stability of the signal. Compare the calculated variance eigenvalue with the preset threshold. If the variance eigenvalue exceeds the preset threshold, the system will automatically trigger an alarm and send a fault warning prompt signal to remind relevant personnel to check the equipment to prevent potential faults or abnormal conditions. Among them, the preset threshold can be set and adjusted according to historical data or equipment standards, and no specific limitation is made here.

[0066] On the basis of the above embodiments, further, the inputting the original vibration feature data into a preset DSMRA model to extract features from the original vibration feature data to obtain reconstructed vibration feature data includes:

[0067] Successively encode the original vibration feature data through each layer encoder of the preset DSMRA model to obtain hidden layer feature data;

[0068] Specifically, DSMRA is a complex neural network structure that integrates popular regularization methods and Stacked Denoising Autoencoders (SDAE). Its goal is to achieve efficient feature extraction while considering the noise and potential manifold structure in the data. As a variant of the autoencoder, the Denoising Autoencoder (DAE) can be trained to reconstruct the original data from corrupted input data, thereby enhancing the model's noise resistance. Under the framework of stacked DAE, multiple autoencoder layers are trained sequentially, and each layer compresses the data into a low-dimensional representation. Based on this, the present invention proposes an improved version of the preset DSMRA network. The structural schematic diagram of the improved DSMRA network structure is as shown in Figure 5 As shown, by means of the manifold regularization method, the structural integrity of the data distribution in the hidden layer is effectively maintained, and the robustness and effectiveness of feature extraction are significantly improved. Therefore, in each level of encoder of the preset DSMRA model, the input data is compressed into low-dimensional features by the encoder, reducing the data dimension and extracting the key feature information therein. The noisy original vibration feature data x is encoded to obtain the hidden layer feature data h (L) .

[0069] Calculate the manifold regularization term according to the hidden layer feature data and the feature similarity graph;

[0070] Specifically, in order to maintain the local geometric structure of the data, similar data points are closer in the high-dimensional space. The manifold regularization method ensures that these similar data points can also remain similar in the low-dimensional feature space through the feature similarity graph structure, which helps to prevent the feature points from being scattered in the feature space, making the features more compact and structured. Therefore, the manifold regularization term can be calculated according to the hidden layer feature data and the feature similarity graph. The manifold regularization term runs through each layer of the encoder, and by penalizing the situation where the gap between adjacent data points in the feature space is too large, it ensures that the local geometric structure is retained during feature extraction.

[0071] Successively decode the hidden layer feature data through each layer of decoder of the preset DSMRA model to obtain the output feature data;

[0072] Specifically, through the decoder, the low-dimensional features are restored. The more accurate the feature data output by the decoder, the better the features extracted by the encoder. Therefore, the data can be decoded layer by layer to check whether the encoder has extracted the correct features. Further, the model learning can be guided by calculating the reconstruction error according to the original vibration feature data and the output feature data to ensure that the low-dimensional features extracted by the encoder can retain the information of the input data to the greatest extent.

[0073] Calculate the reconstruction error according to the original vibration characteristic data and the output characteristic data;

[0074] Specifically, according to the formula

[0075]

[0076] calculate the reconstruction error; where, L reconstruction is the reconstruction error, representing the difference between the original vibration characteristic data x and the output characteristic data ; x is the original vibration characteristic data; is the output characteristic data after being decoded by the decoder, P·P 2 is the Euclidean distance, used to measure the difference between the original vibration characteristic data x and the output characteristic data ;

[0077] Construct the model loss function according to the manifold regularization term and the reconstruction error, and use the model loss function as the objective function, and use the model parameters of each layer of encoder and each layer of decoder as the optimization variables to perform model optimization training until the model loss function meets the convergence condition to obtain the reconstructed vibration characteristic data.

[0078] Specifically, the model loss function of the preset DSMRA model consists of two parts: the reconstruction error and the manifold regularization term. The optimization goal of the loss function is to check that the error between the output characteristic data reconstructed by the encoding-decoding process and the original vibration characteristic data is as small as possible, while maintaining the local geometric structure of the data. The reconstruction error of each layer of encoder will affect the entire model loss function, and the loss function jointly guides the training of the model through manifold regularization and reconstruction error. During the decoding and reconstruction process, the model loss function not only quantifies the reconstruction quality, guides the model training, balances reconstruction and regularization, but also promotes the generalization ability of the model. By continuously optimizing the loss function, the autoencoder can effectively extract, encode and decode data features. Among them, based on the manifold regularization term and the reconstruction error, according to the formula:

[0079] L = L reconstruction + λΩ(f)

[0080] Construct the model loss function, where, L is the loss function, L reconstructionLet \(E\) be the reconstruction error, \(\Omega(f)\) be the manifold regularization term; \(\lambda\) be the weight coefficient of the regularization term. Then, taking the model loss function as the objective function and minimizing the objective function as the optimization goal, the loss function is gradually reduced, and the model parameters of each layer of the encoder and each layer of the decoder are updated, so that the features extracted by the encoder and the reconstruction accuracy of the decoder are maximized. Through layer-by-layer optimization, the encoding weight matrix, encoding bias vector of each layer of the encoder, and the decoding weight matrix and decoding bias vector of each layer of the decoder will be updated, and the feature extraction and reconstruction capabilities of each level will be improved until the output value of the loss function converges to an optimal value.

[0081] On the basis of the above embodiments, further, before inputting the original vibration feature data into the preset DSMRA model to extract features from the original vibration feature data and obtain the reconstructed vibration feature data, the method further includes:

[0082] Construct the encoding mapping functions of each layer of the encoder based on the encoding weight matrix, encoding bias vector, and encoding activation function;

[0083] Specifically, based on the encoding weight matrix, encoding bias vector, and encoding activation function, according to the formula:

[0084] h (L) =f(W (L) x + b (L) )

[0085] Construct the encoding mapping functions of each layer of the encoder; where \(x\) is the original vibration feature data; \(W\) (l) is the encoding weight matrix of the \(l\)-th layer; \(b\) (l) is the encoding bias vector; \(f(·)\) is the encoding activation function, generally the non-linear activation function Sigmoid.

[0086] Construct the decoding mapping functions of each layer of the decoder based on the decoding weight matrix, decoding bias vector, and decoding activation function.

[0087] Specifically, based on the decoding weight matrix, decoding bias vector, and decoding activation function, according to the formula:

[0088]

[0089] Construct the decoding mapping functions of each layer of the decoder; where is the output feature data decoded by the \(L\)-th layer of the decoder; \(W'\) (L) is the decoding weight matrix of the \(L\)-th layer of the decoder; \(h\) (L) is the hidden layer feature data output by the \(L\)-th layer of the encoder; \(b'\) (L)is the decoding bias vector for the L-th layer decoder; g(·) is the decoding activation function of the decoder, usually a linear function, Sigmoid.

[0090] On the basis of the above embodiments, further, the method further includes:

[0091] Construct a channel attention mechanism function based on the channel weight matrix, channel bias vector, and channel activation function;

[0092] Specifically, the channel attention mechanism (Coordinate Attention, CA) focuses on the importance between each channel of the feature. By assigning different weights to different feature channels, it adjusts the importance of each feature channel in the input. Specifically, based on the channel weight matrix, channel bias vector, and channel activation function, according to the formula:

[0093]

[0094] Construct the channel attention mechanism function; where, is the channel attention mechanism function; W c is the channel weight matrix; b c is the channel bias vector; σ(g) is the non-linear activation function Sigmoid.

[0095] Construct a spatial attention mechanism function based on the spatial projection matrix, scaling factor, and spatial activation function;

[0096] Specifically, the spatial attention mechanism (Spatial Attention, SA) focuses on the correlation of each time step or spatial position of the input, that is, the relationship between different positions. By calculating the similarity between each time step or spatial position of the input, it adjusts the information transmission between different positions, so that the model can better capture the temporal or spatial dependencies. Specifically, based on the spatial projection matrix, scaling factor, and spatial activation function, according to the formula:

[0097]

[0098] Construct the spatial attention mechanism function; where, is the output of the spatial attention mechanism function; sofemax(g) is the probability distribution conversion function; are respectively Query, key, and Value, W q ,W k ,W v are all spatial projection matrices; d k is the scaling factor; usually take d k= d, for stabilizing the gradient. It should be noted that the probability distribution conversion function sofemax(g) serves to convert each row of into a probability distribution for weight normalization.

[0099] Based on the channel attention mechanism function and the spatial attention mechanism function, construct a gating mechanism function based on the gating weight matrix and the gating bias vector;

[0100] Specifically, the gating mechanism is used to control the output of the attention to ensure that only the most valuable information is retained. The commonly used gating mechanism can draw on the structures in LSTM or GRU. Based on the channel attention mechanism function and the spatial attention mechanism function, based on the gating weight matrix and the gating bias vector, according to the formula:

[0101]

[0102] Construct the gating mechanism function; where G is the output of the gating mechanism function; σ(g) is the non-linear activation function; W g is the gating weight matrix; is the hidden layer feature data; is the output of the channel attention mechanism function; is the output of the spatial attention mechanism function; b g is the gating bias vector; is the concatenated matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

[0103] Construct the preset DAGU model according to the gating mechanism function, the channel attention mechanism function, and the spatial attention mechanism function.

[0104] Specifically, construct the preset DAGU model according to the gating mechanism function, the channel attention mechanism function, and the spatial attention mechanism function, so that the output of the preset DAGU model combines channel attention, spatial attention, and gating signals. The final output is the feature weighted by the dual attention mechanism, which contains the spatial and channel information of the input data. After being screened by the gating mechanism, a more refined feature representation is obtained.

[0105] Further on the basis of the above embodiments, the calculating the manifold regularization term according to the hidden layer feature data and the feature similarity graph includes:

[0106] Obtain the similarity weight matrix according to the feature similarity graph;

[0107] According to the hidden layer feature data and the similarity weight matrix, according to the formula:

[0108]

[0109] Calculate the manifold regularization term; where, Ω(f) is the manifold regularization term; W ij is the similarity weight matrix; x i is the feature data of the i-th hidden layer; x j is the feature data of the j-th hidden layer; Pf(x i ) - f(x j )P 2 is the distance in the feature space between the feature data x i of the i-th hidden layer and the feature data x j of the j-th hidden layer.

[0110] Specifically, the manifold regularization term Ω(f) represents the degree of preservation of the local structure of the data; the similarity weight matrix W ij represents the similarity between the feature data x i of the i-th hidden layer and the feature data x j of the j-th hidden layer; Pf(x i ) - f(x j )P 2 is the square of the Euclidean distance in the feature space between the feature data x i of the i-th hidden layer and the feature data x j of the j-th hidden layer, and is used to measure their similarity.

[0111] On the basis of the above embodiments, further, according to the channel attention mechanism function and the spatial attention mechanism function, a gating mechanism function is constructed based on a gating weight matrix and a gating bias vector, including:

[0112] According to the formula:

[0113]

[0114] Construct the gating mechanism function; where, G is the output of the gating mechanism function; σ(g) is a non-linear activation function; W g is the gating weight matrix; is the feature data of the hidden layer; is the output of the channel attention mechanism function; W c is the channel weight matrix; b c is the channel bias vector; is the output of the spatial attention mechanism function; sofemax(g) is a probability distribution conversion function; W q , W k , W v are all spatial projection matrices; d k is a scaling factor; It is a splicing matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function. It should be noted that the function of the probability distribution conversion function sofemax(g) is to convert each row of into a probability distribution for weight normalization.

[0115] On the basis of the above embodiments, further, according to the gating mechanism function, the channel attention mechanism function and the spatial attention mechanism function, the preset DAGU model is constructed, including:

[0116] According to the formula:

[0117]

[0118] The preset DAGU model is constructed; where, O is the predicted signal eigenvalue output by the preset DAGU model; G is the output of the gating mechanism function; It is a splicing matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

[0119] As described above in combination with Figures 1-5 The method for predicting and warning elevator brake failures provided in the embodiments of the present application has been introduced in detail. Next, the elevator brake failure prediction and warning device, electronic device and computer-readable storage medium provided in the embodiments of the present application will be introduced with reference to the accompanying drawings.

[0120] As Figure 6 shown, this figure is a schematic diagram of the elevator brake failure prediction and warning device provided by the present application. The device includes:

[0121] An acquisition module 201, configured to acquire the original vibration feature data detected by the FBG vibration sensor installed on the elevator brake to be predicted;

[0122] A processing module 202, configured to input the original vibration feature data into a preset DSMRA model to perform feature extraction on the original vibration feature data to obtain reconstructed vibration feature data; the preset DSMRA model is constructed with the model loss function composed of the reconstruction error and the manifold regularization term as the objective function and the model parameters of each layer of encoder and each layer of decoder as the optimization variables;

[0123] A prediction module 203, configured to input the reconstructed vibration feature data into a preset DAGU model and output a predicted signal eigenvalue; the preset DAGU model is constructed based on the channel attention mechanism, the spatial attention mechanism and the gating mechanism;

[0124] An early warning module 204, configured to calculate the variance eigenvalue of the predicted signal eigenvalue. If the variance eigenvalue is greater than a preset threshold, a fault early warning prompt signal is sent.

[0125] The elevator brake fault prediction and early warning device provided by the embodiments of the present application can correspond to the implementation of the elevator brake fault prediction and early warning method described in the embodiments of the present application, and the above functions of each module of the device correspond to the realization of Figure 1 the corresponding processes of the method shown, and for the sake of brevity, they will not be described in detail here.

[0126] The embodiments of the present application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the elevator brake fault prediction and early warning method as described in the above embodiments.

[0127] As Figure 7 shown, the computer system 300 of the electronic device includes a CPU 301, which can execute various appropriate actions and processes according to the program stored in the ROM 302 or the program loaded from the storage section 308 into the RAM 303. In the RAM 303, various programs and data required for system operation are also stored. The CPU 301, ROM 302, and RAM 303 are connected to each other via a bus 304. The I / O interface 305 is also connected to the bus 304. Among them, the CPU 301 represents the central processing unit, the ROM 302 represents the read-only memory, the RAM 303 represents the random access memory, and the I / O represents the input / output.

[0128] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, a mouse, etc.; an output section 307 including a cathode ray tube, a liquid crystal display, etc. and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. The drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed, so that the computer program read from it can be installed into the storage section 308 as needed.

[0129] In particular, the process of the elevator brake fault prediction and warning method described in the above embodiments can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains program codes for executing the elevator brake fault prediction and warning method described in the above embodiments. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 309, and / or installed from the removable medium 311. When the computer program is executed by the CPU 301, the above functions defined in the computer system 300 are executed.

[0130] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the elevator brake fault prediction and warning method as described in the above embodiments.

[0131] Specifically, the computer-readable storage medium may be included in the electronic device described in the above embodiments; or it may exist alone without being assembled into the electronic device. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the elevator brake fault prediction and warning method as described in the above embodiments.

[0132] It should be noted that the computer-readable storage medium shown in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. In the present application, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above.

[0133] In this text, specific examples are used to illustrate the principles and implementation modes of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. The above is only the preferred implementation mode of the present application. It should be noted that due to the limited nature of written expression and objectively infinite specific structures, for those of ordinary skill in the art, without departing from the principles of the present invention, several improvements, embellishments or changes can be made, or the above technical features can be combined in an appropriate manner; these improvements, embellishments, changes or combinations, or directly applying the concept and technical solution of the invention to other occasions without improvement, shall all be regarded as the protection scope of the present application.

Claims

1. An elevator brake fault prediction and early warning method, characterized in that: include: Acquire original vibration characteristic data detected by the FBG vibration sensor installed on the elevator brake to be predicted; Inputting the original vibration feature data into a preset DSMRA model to extract features of the original vibration feature data to obtain reconstructed vibration feature data; the preset DSMRA model is constructed with a model loss function consisting of a reconstruction error and a manifold regularization term as an objective function and model parameters of each layer of encoders and each layer of decoders as optimization variables; Inputting the reconstructed vibration feature data into a preset DAGU model, and outputting a predicted signal feature value; the preset DAGU model is constructed based on a channel attention mechanism, a spatial attention mechanism, and a gating mechanism; The variance characteristic value of the predicted signal characteristic value is calculated, and if the variance characteristic value is greater than a preset threshold, a fault warning prompt signal is sent.

2. The method according to claim 1, characterized in that The step of inputting the original vibration characteristic data into a preset DSMRA model to extract characteristics of the original vibration characteristic data to obtain reconstructed vibration characteristic data comprises: The original vibration feature data is encoded successively by each layer encoder of the preset DSMRA model to obtain hidden layer feature data; Calculating the manifold regularization term according to the hidden layer feature data and the feature similarity graph; The hidden layer feature data is decoded one by one by each layer decoder of the preset DSMRA model to obtain output feature data; Calculating the reconstruction error according to the original vibration characteristic data and the output characteristic data; The model loss function is constructed according to the manifold regularization term and the reconstruction error, and the model loss function is used as the objective function. The model parameters of the encoders and decoders at each layer are used as optimization variables to perform model optimization training until the model loss function meets the convergence condition, and the output feature data is obtained as the reconstructed vibration feature data.

3. The method according to claim 2, characterized in that Before inputting the original vibration characteristic data into a preset DSMRA model to extract characteristics of the original vibration characteristic data to obtain reconstructed vibration characteristic data, the method further includes: Based on the encoding weight matrix, the encoding bias vector and the encoding activation function, the encoding mapping function of each layer encoder is constructed; Based on the decoding weight matrix, decoding bias vector and decoding activation function, the decoding mapping function of each layer decoder is constructed.

4. The method according to claim 2, characterized in that: The method further comprises: Construct a channel attention mechanism function based on the channel weight matrix, channel bias vector and channel activation function; Construct a spatial attention mechanism function based on the spatial projection matrix, scaling factor, and spatial activation function; According to the channel attention mechanism function and the spatial attention mechanism function, a gating mechanism function is constructed based on a gating weight matrix and a gating bias vector; The preset DAGU model is constructed according to the gating mechanism function, the channel attention mechanism function and the spatial attention mechanism function.

5. The method according to claim 2, characterized in that: The step of calculating the manifold regularization term according to the hidden layer feature data and the feature similarity graph comprises: Obtaining a similarity weight matrix according to the feature similarity graph; According to the hidden layer feature data and the similarity weight matrix, according to the formula: Calculate the manifold regularization term; where Ω(f) is the manifold regularization term; W ij is the similarity weight matrix; x i is the feature data of the i-th hidden layer; x j is the jth hidden layer feature data; Pf(x i )-f(x j ) 2 is the i-th hidden layer feature data x i and the jth hidden layer feature data x j The distance in feature space.

6. The method according to claim 4, characterized in that According to the channel attention mechanism function and the spatial attention mechanism function, based on the gating weight matrix and the gating bias vector, a gating mechanism function is constructed, including: According to the formula: Construct a gating mechanism function; where G is the output of the gating mechanism function; σ(g) is a nonlinear activation function; W g is the gating weight matrix; is the hidden layer feature data; is the output of the channel attention mechanism function; b g is the gate bias vector; W c is the channel weight matrix; b c is the channel bias vector; is the output of the spatial attention mechanism function; sofemax(g) is the probability distribution conversion function; W q ,W k ,W v are spatial projection matrices; d k is the scaling factor; It is a concatenated matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

7. The method according to claim 6, characterized in that According to the gating mechanism function, the channel attention mechanism function and the spatial attention mechanism function, the preset DAGU model is constructed, including: According to the formula: Constructing the preset DAGU model; wherein O is the predicted signal characteristic value output by the preset DAGU model; G is the output of the gating mechanism function; It is a concatenated matrix composed of the output of the channel attention mechanism function and the output of the spatial attention mechanism function.

8. An elevator brake fault prediction and warning device, characterized in that: include: An acquisition module, used for acquiring original vibration characteristic data detected by a FBG vibration sensor installed on the elevator brake to be predicted; A processing module, used for inputting the original vibration feature data into a preset DSMRA model to perform feature extraction on the original vibration feature data to obtain reconstructed vibration feature data; the preset DSMRA model is constructed with a model loss function consisting of a reconstruction error and a manifold regularization term as an objective function and model parameters of each layer of encoders and each layer of decoders as optimization variables; A prediction module, used for inputting the reconstructed vibration feature data into a preset DAGU model and outputting a predicted signal feature value; the preset DAGU model is constructed based on a channel attention mechanism, a spatial attention mechanism and a gating mechanism; The early warning module is used to calculate the variance characteristic value of the predicted signal characteristic value, and send a fault early warning prompt signal if the variance characteristic value is greater than a preset threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images