CN-121974217-A - Elevator vibration fault detection and predictive maintenance system and method

CN121974217ACN 121974217 ACN121974217 ACN 121974217ACN-121974217-A

Abstract

The application relates to an elevator vibration fault detection and predictive maintenance system and method, which relate to the field of elevator safety monitoring and comprise a vibration and operation data acquisition module, a threshold-based safety detection module, a health mapping and self-learning module, a degradation trend analysis and predictive maintenance module, a maintenance and supervision feedback module and a maintenance and supervision feedback module, wherein the vibration and operation data acquisition module is used for acquiring vibration signals and operation data of a single complete operation period, the threshold-based safety detection module is used for executing a multi-layer threshold system comprising a safety threshold and an experience threshold, the health mapping and self-learning module is used for extracting feature vectors, constructing a health mapping model and calculating health index HI, and simultaneously restraining model updating through a sample screening mechanism, the degradation trend analysis and predictive maintenance module is used for analyzing HI time sequences and outputting predictive maintenance suggestions, and the maintenance and supervision feedback module is used for realizing model freezing, correction and alarm constraint. According to the scheme provided by the application, the accurate detection of the elevator vibration fault and the predictive maintenance of key components can be realized on the premise of not replacing the existing safety criterion.

Inventors

Du Huiran
CHEN GANG
TANG QIWEI

Assignees

日立楼宇技术（广州）有限公司

Dates

Publication Date: 20260505
Application Date: 20260320

Claims (10)

1. An elevator vibration fault detection and predictive maintenance system, comprising: The vibration and operation data acquisition module is used for acquiring triaxial acceleration vibration signals and operation state information in the operation process of the elevator and organizing the acquired data according to a single complete operation period, wherein the elevator is lifting equipment; The safety detection module based on the threshold value is used for executing a multi-layer threshold system, the multi-layer threshold system comprises a safety threshold value and an experience threshold value, the safety threshold value has the highest priority and is used for judging vibration amplitude and frequency band energy indexes, the experience threshold value is used as engineering constraint, constraint learning sample selection is carried out, and abnormal operation trend is judged in an auxiliary mode; The health mapping and self-learning module is used for extracting feature vectors from single operation period data, setting a learning sample screening mechanism, constructing an unsupervised learning health mapping model and calculating health index HI, wherein the health index HI is the reconstruction error or deviation degree of the feature vectors in a health feature subspace and represents the deviation degree of an operation state relative to a health structure; The degradation trend analysis and predictive maintenance module takes a time sequence of a health index HI as a trend analysis object, analyzes the HI sequence and outputs predictive maintenance advice; The maintenance and supervision feedback module comprises a maintenance event processing unit and a supervision feedback constraint unit and is used for realizing model management and data updating in the maintenance process and alarm constraint and result application in a supervision qualified state.
2. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein: The lifting device is used for transporting goods or personnel in a vertical direction or an inclined direction; The running state information at least comprises a running direction, a speed signal and running start-stop time; the single complete operating cycle corresponds to each complete start-stop process or fixed operating duration of the lifting device.
3. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein: The safety thresholds are used for judging vibration amplitude and frequency band energy indexes, and when any safety threshold is triggered, fault alarm information is directly output without depending on a health mapping and self-learning module, trend analysis and predictive maintenance advice; The empirical threshold is used for describing an empirical vibration range of normal operation of the elevator, and comprises an empirical upper limit of energy of a specific frequency band, an empirical interval of vibration stability in an operation period and characteristic variation amplitude related to a specific component, and the empirical threshold is used for assisting in judging abnormal operation trend and restraining learning sample selection of a health mapping model, so that safety alarm is not triggered directly.
4. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein: The feature vector extracted from the single operation cycle by the health mapping and self-learning module comprises at least one of multiband vibration energy features, harmonic structure related features, statistical consistency features in the operation cycle, and normalized features related to the operation direction and travel.
5. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein: the health mapping model is an unsupervised learning model and comprises any one of a Principal Component Analysis (PCA) model, a linear self-encoder model and an incremental principal component analysis model, and the health mapping model constructs a health feature subspace of the elevator in the current state through learning a history stable operation sample.
6. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein the learning sample screening mechanism comprises: the single run-time period data is used for health mapping model updates only when the following conditions are satisfied simultaneously: No safety threshold is triggered; no obvious violation of the empirical threshold; The health index HI changes steadily in continuous and repeated operation; Has consistency with the historical operating conditions; through the learning sample screening mechanism, distortion learning of the health mapping model due to short-term disturbance or slow degradation is avoided.
7. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein: The method for analyzing the HI sequence by the degradation trend analysis and predictive maintenance module comprises sliding window statistics, monotonicity judgment and health degree change rate estimation, wherein the sliding window statistics calculates statistics on HI values in each sliding window, filters HI value accidental fluctuation caused by short-time disturbance of an elevator, extracts real HI change trends caused by component degradation, the monotonicity judgment identifies whether degradation of key components of the elevator enters an irreversible continuous development stage or not through calculating difference values or growth rates of HI characteristic values of the continuous windows, eliminates false degradation of the HI values, locks real degradation trends needing to be concerned, and the health degree change rate estimation quantitatively calculates the ascending rate of the HI values on the basis of monotonicity judgment and confirmation of HI continuous ascending, analyzes whether the HI values represent acceleration trends or not, and provides numerical basis for threshold triggering probability in a subsequent predictive time window.
8. The elevator vibration fault detection and predictive maintenance system according to claim 1, wherein the analyzing the HI sequence outputs predictive maintenance advice comprising: outputting predictive maintenance suggestions when the HI presents a continuous rising trend and the HI change rate is within a predictive time window, wherein the probability of triggering an experience threshold or a safety threshold is remarkably improved; the predictive maintenance suggestion does not participate in fault judgment, and only provides accurate maintenance time and maintenance component direction for elevator maintenance, thereby realizing on-demand maintenance.
9. The elevator vibration fault detection and predictive maintenance system of claim 1, wherein the maintenance and supervision feedback module comprises: The maintenance event processing unit freezes the current health mapping model when detecting the occurrence of the maintenance event; taking the HI track before maintenance as a degradation reference record; Re-acquiring stable operation data after maintenance; Constructing or correcting a health feature subspace based on the re-acquired steady operation data; when the supervision feedback constraint unit confirms that the elevator is in a qualified running state, even if the health index HI is higher, a fault alarm is not output, and only the health mapping result is used for trend analysis and long-term statistics.
10. An elevator vibration fault detection and predictive maintenance method, applying the elevator vibration fault detection and predictive maintenance system according to any one of claims 1 to 9, characterized by comprising: The method comprises the steps of collecting triaxial acceleration vibration signals and running state information of elevator running, and organizing the triaxial acceleration vibration signals and the running state information into an independent analysis unit according to a single complete running period, wherein the elevator is lifting equipment, and the lifting equipment is equipment for transporting goods or personnel in a vertical direction or an inclined direction; based on a multi-layer threshold system, detecting a safety threshold and an empirical threshold, if the safety threshold is triggered, directly outputting a fault alarm, if the safety threshold is not triggered, transmitting data into a health mapping model, and if the safety threshold is not triggered, the empirical threshold is only used as engineering constraint and does not trigger the alarm; Extracting feature vectors from single operation period data, constructing a health feature subspace by using an unsupervised learning model, calculating a health index HI, and selecting an operation data updating model conforming to conditions through a sample screening mechanism, wherein the calculated health index HI comprises extracted multiband vibration energy features, harmonic structure related features, operation period internal statistics consistency features and feature vector combinations related to operation directions and strokes; carrying out sliding window statistics, monotonicity judgment and change rate estimation on the HI time sequence through degradation trend analysis, and outputting predictive maintenance suggestions if the maintenance criteria are met; and after the maintenance action is detected, model freezing, degradation recording and model correction are carried out, and constraint fault warning is carried out when the elevator running state is judged to be qualified, and only the health mapping result is used for trend analysis and long-term statistics.

Description

Elevator vibration fault detection and predictive maintenance system and method Technical Field The application relates to the technical field of elevator safety monitoring, in particular to an elevator vibration fault detection and predictive maintenance system and method. Background The existing elevator vibration monitoring system generally adopts a detection method based on a threshold value, mainly sets a safety threshold value according to national standard or industry standard, monitors indexes such as amplitude, frequency band energy and the like of vibration signals in real time, and ensures equipment operation safety. Meanwhile, by combining engineering practice experience, an experience threshold is set for key components such as a traction machine, a guide rail, a bearing and the like, and is used for assisting in judging abnormal running states. However, in practical application, since the fault occurrence probability of the elevator equipment is low, a large number of anomalies in long-term operation are represented by short-time disturbance or working condition change, a fixed threshold is difficult to distinguish real faults from normal fluctuation, false alarm is easy to generate, and maintenance efficiency and user experience are affected. And the deterioration of key components of the elevator usually shows a long-term and slow change trend, and is difficult to effectively identify before reaching a safety threshold, so that potential risks cannot be early warned in advance, and the optimal opportunity of predictive maintenance is missed. The empirical threshold depends on manual setting, is difficult to adapt to the changes of different elevator models, service environments and running states, lacks dynamic adjustment capability, and limits the applicability and accuracy of the elevator under complex working conditions. Although the existing elevator vibration monitoring system introduces an abnormality detection machine learning model to perform state evaluation, the method often tries to replace the existing safety criterion, cannot meet the requirements of the elevator industry on safety responsibility division and supervision compliance, and is difficult to actually land. Therefore, a technology for realizing elevator vibration fault detection and predictive maintenance by combining vibration signals, experience constraint and a self-learning health mapping model on the premise of not replacing the existing safety threshold system is needed, and the accuracy of fault identification and the scientificity of maintenance decisions are improved. Disclosure of Invention In order to solve or partially solve the problems existing in the related art, the application provides an elevator vibration fault detection and predictive maintenance system and method, and aims to solve the problems that false alarm is easy to occur and early recognition capability for degradation trend is lacking under the condition of fixed threshold detection of an elevator vibration monitoring system. A first aspect of the present application provides an elevator vibration fault detection and predictive maintenance system comprising: the vibration and operation data acquisition module is used for acquiring triaxial acceleration vibration signals and operation state information in the operation process of the elevator and organizing the acquired data according to a single complete operation period, wherein the elevator is lifting equipment; The safety detection module based on the threshold value is used for executing a multi-layer threshold system, the multi-layer threshold system comprises a safety threshold value and an experience threshold value, the safety threshold value has the highest priority and is used for judging vibration amplitude and frequency band energy indexes, the experience threshold value is used as engineering constraint, constraint learning sample selection is carried out, and abnormal operation trend is judged in an auxiliary mode; The health mapping and self-learning module is used for extracting feature vectors from single operation period data, setting a learning sample screening mechanism, constructing an unsupervised learning health mapping model and calculating health index HI, wherein the health index HI is the reconstruction error or deviation degree of the feature vectors in a health feature subspace and represents the deviation degree of an operation state relative to a health structure; The degradation trend analysis and predictive maintenance module takes a time sequence of a health index HI as a trend analysis object, analyzes the HI sequence and outputs predictive maintenance advice; The maintenance and supervision feedback module comprises a maintenance event processing unit and a supervision feedback constraint unit and is used for realizing model management and data updating in the maintenance process and alarm constraint and result application in a supervision qualified state