CN-121765489-B - Elevator rotating part fault detection method, electronic equipment and program product

CN121765489BCN 121765489 BCN121765489 BCN 121765489BCN-121765489-B

Abstract

The application provides a fault detection method for an elevator rotating component, electronic equipment and a program product. The method comprises the steps of obtaining vibration data to be detected of a rotating component of an elevator to be detected, extracting frequency domain features of the preprocessed vibration data to be detected to obtain features to be detected, mapping the features to be detected from a target domain to a mapping feature of a source domain based on a pre-established cross-domain mapping relation through a self-encoder model, inputting the mapping feature into a trained classification model to obtain a detection result of the rotating component of the elevator to be detected, wherein the classification model is obtained through training of a first vibration data set of the source structure, and the first vibration data set comprises a fault data set and a fault-free data set of the source structure. Therefore, the fault detection of the elevator rotating component can be realized without using the fault data of the elevator rotating component to carry out model training, and the problem of false alarm and missing report caused by the fact that the model cannot learn the fault characteristics due to the lack of the fault data of the elevator rotating component is avoided.

Inventors

LIU YU
Ao Shiming
QUAN YONGLIN
LIU ZHONGYAN
YUAN JINGJIE
LV XIAO
LUO HENG
CHEN ZHUO
HE YANG
WAN SHUAI
LI JUN
ZOU TONGFENG

Assignees

重庆市特种设备检测研究院(重庆市特种设备事故应急调查处理中心)

Dates

Publication Date: 20260508
Application Date: 20260303

Claims (9)

1. A method for detecting a failure of a rotating member of an elevator, the method comprising: Obtaining vibration data to be detected of the elevator rotating component to be detected; Carrying out frequency domain feature extraction on the preprocessed vibration data to be detected to obtain features to be detected; mapping the feature to be detected from a target domain to a feature to be detected of a source domain based on a pre-established cross-domain mapping relation through a self-encoder model, wherein the target domain represents an elevator rotating component, the source domain represents a source structure, the source structure is an experimental test platform with the same type of feature as the elevator rotating component, the cross-domain mapping relation is obtained based on a fault-free data set and a second vibration data set in a first vibration data set of the source structure, and the second vibration data set is vibration data of the elevator rotating component in a fault-free state; Inputting the mapping characteristics into a trained classification model to obtain a detection result of the elevator rotating component to be detected, wherein the detection result comprises a result indicating that no fault exists or the fault exists, the classification model is obtained through training of a first vibration data set of the source structure, and the first vibration data set comprises a fault data set and a fault-free data set of the source structure; wherein, before the obtaining the vibration data to be measured of the elevator rotating component to be measured, the method further comprises: Acquiring a first vibration data set of the source structure, a second vibration data set of the elevator rotating component in a fault-free state, wherein the first vibration data set comprises a fault-free data set representing that the source structure vibrates in the fault-free state and a fault data set representing that the source structure vibrates in the fault state; Frequency domain feature extraction is carried out on the preprocessed first vibration data set and the preprocessed second vibration data set to respectively obtain a first feature set and a second feature set, wherein the first feature set comprises a fault feature set corresponding to the fault data set and a fault-free feature set corresponding to the fault-free data set; Taking the second feature set as input data and the non-fault feature set as tag data, and establishing a cross-domain mapping relation through a self-encoder model to convert the second feature set into feature sets consistent with the distribution of the non-fault feature set; and training the classification model through the fault feature set and the non-fault feature set in the first feature set until the classification model converges to obtain a trained classification model.
2. The method of claim 1, wherein between the step of obtaining a first vibration dataset of the source structure, a second vibration dataset of the elevator rotating component in a fault-free state, and the step of frequency domain feature extraction of the preprocessed first vibration dataset and second vibration dataset, the method further comprises: Preprocessing the first vibration data set and the second vibration data set to obtain a preprocessed first vibration data set and a preprocessed second vibration data set, wherein the preprocessing comprises: Filtering and denoising the first vibration data set and the second vibration data set to obtain a denoised first vibration data set and a denoised second vibration data set; Dividing the denoised first vibration data set and the denoised second vibration data set according to a plurality of load levels of the elevator to obtain a sample subset corresponding to each load level; and carrying out sample expansion on the sample subset by adopting a data enhancement strategy, wherein the data enhancement strategy comprises at least one operation of time stretching, gao Sijia noise and time domain translation on part of data in the sample subset.
3. The method of claim 1, wherein performing frequency domain feature extraction on the preprocessed first vibration data set and the preprocessed second vibration data set to obtain the first feature set and the second feature set, respectively, comprises: Performing fast fourier transform on vibration data of any vibration signal in the first vibration data set and the second vibration data set after preprocessing to obtain frequency domain signal amplitude, wherein the frequency domain signal amplitude is expressed as: ; in the formula, The frequency is the frequency domain signal amplitude, f represents the frequency, N is the time sequence index, and N is more than or equal to 0 and less than or equal to N, and N is the number of signal sampling points; Vibration data of any vibration signal at an nth sampling point; j is an imaginary unit; Is a natural constant; If the vibration signal is a signal in a fault state, extracting all frequency domain signal amplitudes in a preset frequency band from all frequency domain signal amplitudes of the vibration signal according to a preset frequency band representing fault sensitivity to form a fault feature; If the vibration signal is a signal in a fault-free state, taking the frequency domain signal amplitude corresponding to the vibration signal as a fault-free characteristic; All fault features and all fault-free features obtained based on the first vibration data set are used as a first feature set, and all fault-free features obtained based on the second vibration data set are used as a second feature set.
4. The method of claim 1, wherein the self-encoder model comprises an encoder and a decoder; the establishing a cross-domain mapping relationship by using the second feature set as input data and the fault-free feature set as tag data through a self-encoder model comprises the following steps: Inputting the corresponding characteristics of each vibration signal in the second characteristic set into the encoder to compress the characteristics of the target domain into the latent space to obtain the characteristics of the latent space, and reconstructing the characteristics consistent with the characteristic distribution of the source domain by the decoder according to the characteristics of the latent space to serve as reconstruction characteristics; Optimizing parameters of the self-encoder model based on a first preset loss function according to the reconstruction features and the corresponding tag data, and iteratively training the self-encoder model according to the second feature set and the fault-free feature set until the self-encoder model converges, so that the self-encoder model has a function of converting target domain features into features aligned with source domain feature distribution as an established cross-domain mapping relationship, wherein the first preset loss function is as follows: ; in the formula, The method comprises the steps of reconstructing errors, wherein X refers to the reconstruction features, Y refers to label data corresponding to the reconstruction features and is a non-fault feature of a source domain, and M, Q is the number of samples in the non-fault feature set and the number of samples in the second feature set respectively; Referring to the reconstruction feature corresponding to the ith feature in the second feature set; Refers to the jth feature in the failure-free feature set.
5. The method of claim 1, wherein training the classification model with the set of fault features and the set of non-fault features in the first set of features until the classification model converges, results in a trained classification model, comprising: A41, dividing the first feature set into a training set, a verification set and a test set according to a preset proportion by adopting a hierarchical sampling method; A42, iteratively training the classification model through the training set to obtain a primarily trained classification model; A43, traversing adjustable parameters in the classification model by adopting a grid search method to obtain a plurality of parameter sets, wherein the adjustable parameters comprise kernel parameters and regularization parameters, each parameter set and the primarily trained classification model form a candidate classification model, parameter tuning is performed on the candidate classification model through the verification set, and the candidate classification model with the highest fault recognition accuracy is used as an optimized classification model; a44, testing the optimized classification model through the test set to obtain a test result which indicates whether the performance of the classification model is qualified; A45, when the test result is unqualified, repeating the steps A43 and A44 until the test result is qualified, and obtaining a trained classification model; a46, obtaining a trained classification model when the test result is qualified.
6. The method of claim 5, wherein the iterative training process comprises: selecting a group of batch samples from the training set, wherein the batch samples comprise fault-free features and fault features with balanced distribution; taking the batch of samples as an input feature vector of the current round, and inputting an initial classification model; And calling a radial basis function through a classification model, and calculating the similarity between each feature in the batch of samples and the reference sample feature one by one, wherein the radial basis function has the formula: ; in the formula, Is that And (3) with Is a similarity value of (1); Currently input features in the batch of samples; the characteristic is the pre-stored ith reference sample, and gamma is a nuclear parameter; Is an exponential function; Mapping the characteristics in the batch of samples to a high-dimensional space based on the corresponding similarity of the batch of samples through a classification model to obtain high-dimensional characteristics; Determining an optimal decision boundary in a high-dimensional space by minimizing an objective function, the objective function being: ; The constraint conditions of the objective function are as follows: ; in the formula, For a high-dimensional spatial weight, Mapping the features in the batch of samples to high-dimensional features; Is a decision boundary offset; In order to relax the variables of the variables, For the total number of samples in the training set, Is a regularization parameter; based on the optimal decision boundary, a decision function is obtained, wherein the decision function is as follows: ; wherein f (x) represents the value of the decision function, When the fault state is determined to be a fault-free state, When the fault state is determined; u is the total number of support vectors constructed by the participation decision boundary; Training the obtained Lagrangian coefficient for the optimal decision boundary; And (3) reversely propagating the loss value of the classification model by adopting a gradient descent method, updating the related parameters of the classification model until a preset stopping condition is met, so as to obtain the classification model for preliminary training, wherein the preset stopping condition comprises that the single-round loss value is lower than a first preset threshold value or the descending amplitude of the continuous R-round loss value is lower than a second preset threshold value, R is a preset integer which is greater than or equal to 5, and the loss value is obtained based on the value output by the decision function.
7. The method of claim 1, wherein the elevator rotating component comprises at least one of a traction machine, a gantry system, a guide sheave, and a counterweight sheave in an elevator, the method further comprising: And when the detection result shows that a fault exists, sending a fault prompt to the management terminal.
8. An electronic device comprising a processor and a memory coupled to each other, the memory storing a computer program that, when executed by the processor, causes the electronic device to perform the method of any one of claims 1-7.
9. A program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1 to 7.

Description

Elevator rotating part fault detection method, electronic equipment and program product Technical Field The invention relates to the technical field of elevator detection, in particular to an elevator rotating part fault detection method, electronic equipment and a program product. Background The elevator is used as urban vertical transportation facilities, and the health states of rotating parts such as a traction machine, a door machine system, a guide wheel and the like in the elevator are directly related to running safety. Under the condition of long-term dynamic load, the rotating parts of the elevator are easy to have faults such as bearing abrasion, gear damage, unbalanced rotor and the like, and if the rotating parts are not detected in time, shutdown or safety accidents can be caused. The fault of the elevator rotating component can be detected by collecting the vibration signal of the rotating component and utilizing a corresponding deep learning model. Because the fault data quantity of the elevator rotating part is small, enough fault data cannot be collected generally, so that the deep learning model detects faults, and the problems of false alarm, missing alarm and the like are easy to occur. Disclosure of Invention In view of the foregoing, an object of an embodiment of the present application is to provide a method, an electronic device, and a program product for detecting a failure of an elevator rotating member, which can improve the problem that the detection of the elevator rotating member is prone to false alarm and missing alarm. In order to achieve the technical purpose, the application adopts the following technical scheme: in a first aspect, an embodiment of the present application provides a method for detecting a fault in a rotating member of an elevator, the method including: Obtaining vibration data to be detected of the elevator rotating component to be detected; Carrying out frequency domain feature extraction on the preprocessed vibration data to be detected to obtain features to be detected; mapping the feature to be detected from a target domain to a feature to be detected of a source domain based on a pre-established cross-domain mapping relation through a self-encoder model, wherein the target domain represents an elevator rotating component, the source domain represents a source structure, the source structure is an experimental test platform with the same type of feature as the elevator rotating component, the cross-domain mapping relation is obtained based on a fault-free data set and a second vibration data set in a first vibration data set of the source structure, and the second vibration data set is vibration data of the elevator rotating component in a fault-free state; and inputting the mapping characteristics into a trained classification model to obtain a detection result of the elevator rotating component to be detected, wherein the detection result comprises a result indicating that no fault exists or the fault exists, the classification model is obtained through training of a first vibration data set of the source structure, and the first vibration data set comprises a fault data set and a fault-free data set of the source structure. With reference to the first aspect, in some optional embodiments, before the acquiring the vibration data to be measured of the rotating component of the elevator to be measured, the method further includes: Acquiring a first vibration data set of the source structure, a second vibration data set of the elevator rotating component in a fault-free state, wherein the first vibration data set comprises a fault-free data set representing that the source structure vibrates in the fault-free state and a fault data set representing that the source structure vibrates in the fault state; Frequency domain feature extraction is carried out on the preprocessed first vibration data set and the preprocessed second vibration data set to respectively obtain a first feature set and a second feature set, wherein the first feature set comprises a fault feature set corresponding to the fault data set and a fault-free feature set corresponding to the fault-free data set; Taking the second feature set as input data and the non-fault feature set as tag data, and establishing a cross-domain mapping relation through a self-encoder model to convert the second feature set into feature sets consistent with the distribution of the non-fault feature set; and training the classification model through the fault feature set and the non-fault feature set in the first feature set until the classification model converges to obtain a trained classification model. With reference to the first aspect, in some optional embodiments, between the acquiring of the first vibration data set of the source structure and the second vibration data set of the elevator rotating component in the fault-free state in step and the performing of frequency domain feature extraction on t