Elevator detection method and system

Abstract

The invention relates to the technical field of elevator detection, in particular to an elevator detection method and system.The method comprises the steps that multi-source data of an elevator is collected in real time, preprocessing and feature extraction are conducted on the multi-source data, and an initial feature vector is generated; the weight of the initial feature vector is dynamically adjusted according to the use scene and the operation mode of the elevator, fusion processing is conducted on the adjusted initial feature vector through a fusion algorithm, and a comprehensive feature vector is generated; and inputting the comprehensive feature vector into a risk prediction model, generating a comprehensive risk score, when the comprehensive risk score exceeds a dynamic threshold value, outputting a fault type and a maintenance suggestion through a prediction maintenance model, and visually displaying the fault type and the maintenance suggestion on a user interaction interface.

Description

Technical Field

[0001] This invention relates to the field of elevator testing technology, and more particularly to an elevator testing method and system. Background Technology

[0002] Currently, common elevator inspection methods mainly rely on various sensors to collect elevator operation data, such as speed, acceleration, vibration, and door system status, and analyze this data through preset algorithms to determine whether there are potential malfunctions in the elevator.

[0003] However, existing elevator detection technologies have significant shortcomings, failing to dynamically adjust data weights based on elevator operating modes and usage scenarios. Elevator operating modes and usage scenarios vary greatly across different locations. For example, in office buildings, high passenger flow during morning and evening rush hours leads to frequent elevator starts and stops, resulting in a primary operating mode of high-speed operation and frequent acceleration and deceleration. In hospitals, in addition to normal passenger transport, the entry and exit of stretchers and medical equipment necessitates specific requirements for elevator stability and space constraints, emphasizing a smooth, slow operating mode. In hotels, due to relatively concentrated guest check-in and check-out times and the presence of luggage, elevator operating modes differ further.

[0004] However, existing detection technologies cannot achieve such dynamic adjustments. In the different scenarios mentioned above, the same standard is used to measure data such as speed, acceleration, and vibration. This makes it difficult for elevator inspection to accurately and promptly detect potential faults under different operating modes and usage scenarios, reducing the accuracy and effectiveness of elevator inspection and posing risks to the safe operation of elevators. Summary of the Invention

[0005] To address the above problems, this invention provides an elevator inspection method and system.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An elevator inspection method, comprising: Real-time acquisition of multi-source data from the elevator, preprocessing and feature extraction of the multi-source data, and generation of initial feature vectors; The weights of the initial feature vectors are dynamically adjusted according to the elevator's usage scenarios and operating modes, and the adjusted initial feature vectors are fused using a fusion algorithm to generate a comprehensive feature vector. The comprehensive feature vector is input into the risk prediction model to generate a comprehensive risk score. When the comprehensive risk score exceeds the dynamic threshold, the fault type and maintenance suggestions are output through the predictive maintenance model, and the fault type and maintenance suggestions are visualized on the user interface.

[0007] Preferably, the step of inputting the comprehensive feature vector into the risk prediction model to generate a comprehensive risk score specifically includes: The comprehensive feature vector is input into a pre-trained risk assessment neural network; The risk assessment neural network calculates risk quantification values ​​in parallel across multiple dimensions, including at least the probability of failure, the severity level of failure, and the imminence of failure. The risk quantification values ​​from the multiple dimensions are combined using a weighted fusion function to generate a comprehensive risk score.

[0008] Preferably, when the comprehensive risk score exceeds the dynamic threshold, the fault type and maintenance suggestions are output through the predictive maintenance model, specifically including: When the comprehensive risk score exceeds the dynamic threshold, the gradient vector of the comprehensive feature vector is calculated through the prediction and maintenance model. Based on the gradient vector, the feature dimensions corresponding to the comprehensive risk score are identified by sorting them according to their contribution values. Based on a predefined feature mapping table, the identified feature dimensions are mapped to their corresponding physical sensor and signal features; based on a predefined defect knowledge base, the physical sensor and signal features are mapped to one or more possible defect types. Using the defect type as an index, query the preset maintenance knowledge base and obtain the corresponding maintenance operation information based on the maintenance knowledge base; Based on the fault type, confidence level, and maintenance operation information, one or more specific maintenance suggestions are automatically generated. Experts or maintenance personnel determine the rationality of the maintenance suggestions. When the maintenance suggestions are feasible, decision information containing the fault type and maintenance suggestions is output.

[0009] Preferably, the multi-source data includes mechanical vibration data, acoustic data, electrical operation data, time data, environmental data, and video data.

[0010] Preferably, the real-time acquisition of multi-source elevator data, the preprocessing and feature extraction of the multi-source data to generate an initial feature vector, specifically includes: Multi-source data is collected synchronously or asynchronously by sensor arrays or other data acquisition instruments deployed at different locations in the elevator; Perform data alignment, outlier detection, filtering and denoising, and data normalization on multi-source data to generate standardized data; Time-frequency domain features are extracted from standardized data, and a feature set is output. The feature set is assembled in a predetermined order, and the initial feature vector is output.

[0011] Preferably, the step of dynamically adjusting the weights of the initial feature vector based on the elevator's usage scenario and operating mode, and then fusing the adjusted initial feature vector using a fusion algorithm to generate a comprehensive feature vector, specifically includes: Based on real-time multi-source data of the elevator, identify the current usage scenario and operating mode of the elevator; Based on the identified use case and operating mode, the corresponding set of weight coefficients is called from the predefined strategy library; Using the set of weight coefficients, a weighted operation is performed on each feature component in the initial feature vector to generate a weighted feature vector; The weighted feature vectors are subjected to dimensionality reduction and aggregation processing using a fusion algorithm to generate a comprehensive feature vector.

[0012] Preferably, the usage scenarios include passenger elevators, freight elevators, medical elevators, dumbwaiters, and sightseeing elevators. The usage scenarios are classified into predefined scenario types based on one or more dimensions of the elevator's real-time load rate, operating frequency, and time, using a rule engine or classification model.

[0013] Preferably, the operating modes include peak hours, off-peak hours, and nighttime. The usage scenarios are based on one or more dimensions of the elevator's real-time load rate, operating frequency, and time, and the current mode is classified into a predefined operating mode type through a rule engine or classification model.

[0014] An elevator detection system, comprising: The data acquisition module is used to collect multi-source data from the elevator in real time, preprocess the multi-source data and extract features to generate an initial feature vector; The data processing module is used to dynamically adjust the weights of the initial feature vector according to the elevator's usage scenario and operating mode, and to fuse the adjusted initial feature vector through a fusion algorithm to generate a comprehensive feature vector. The suggestion generation module is used to input the comprehensive feature vector into the risk prediction model to generate a comprehensive risk score. When the comprehensive risk score exceeds the dynamic threshold, the prediction and maintenance model outputs the fault type and maintenance suggestion, and displays the fault type and maintenance suggestion on the user interface.

[0015] Preferably, the data processing module includes: An edge computing gateway is deployed at the elevator site and communicates with the data acquisition module to implement the functions of the data processing module. The cloud platform server is connected to the edge computing gateway via network communication and is used to carry the risk prediction model and the prediction maintenance model.

[0016] The beneficial effects of this invention are as follows: 1. This invention is based on one or more dimensions of elevator data, such as real-time load rate, operating frequency, and time. It uses a rule engine or classification model to classify the current scenario or operating mode into a predefined scenario type or operating mode type. The weight of the initial feature vector is dynamically adjusted according to the elevator's usage scenario and operating mode to accurately and timely detect potential faults, thereby improving the accuracy and effectiveness of elevator detection.

[0017] 2. This invention can identify potential fault hazards and issue early warnings a long period before a fault occurs by conducting in-depth analysis and trend prediction of multi-source data and using a predictive maintenance model. This transforms the operation and maintenance mode from traditional post-fault maintenance to predictive maintenance, reducing unexpected downtime and avoiding the serious consequences that may be caused when a fault occurs. Attached Figure Description

[0018] Figure 1 This is a flowchart of the elevator detection method in a specific embodiment of the present invention; Figure 2 This is a block diagram of the elevator detection system in a specific embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0020] Please see Figure 1 As shown, the present invention relates to an elevator inspection method, comprising: Multi-source data is collected synchronously or asynchronously by sensor arrays or other acquisition instruments deployed at different locations in the elevator. This multi-source data includes mechanical vibration data, acoustic data, electrical operation data, time data, environmental data, and video data.

[0021] Data alignment, outlier detection, filtering and denoising, and data normalization are performed on multi-source data to generate standardized data. Time-frequency domain features are extracted from the standardized data, and a feature set is output. The feature set is then concatenated in a predetermined order to output an initial feature vector. The weights of the initial feature vector are dynamically adjusted based on the elevator's usage scenario and operating mode. A fusion algorithm is then used to fuse the adjusted initial feature vectors to generate a comprehensive feature vector, which includes: Based on real-time multi-source data of the elevator, identify the current usage scenario and operating mode of the elevator; Based on the identified use case and operating mode, the corresponding set of weight coefficients is called from the predefined strategy library; Using the set of weight coefficients, a weighted operation is performed on each feature component in the initial feature vector to generate a weighted feature vector; The weighted feature vectors are subjected to dimensionality reduction and aggregation processing using a fusion algorithm to generate a comprehensive feature vector.

[0022] Specifically, the use cases and operating modes of this application are based on one or more dimensions of elevator data, such as real-time load rate, operating frequency, and time. The current scenario is classified into a predefined scenario type or operating mode type through a rule engine or classification model. Specifically, the usage scenarios include passenger elevators, freight elevators, medical elevators, dumbwaiters, and sightseeing elevators, and the operating modes include peak hours, off-peak hours, and nighttime. When the scenario is identified as a "high-frequency start-stop scenario", that is, the operating mode is during peak period, the weight of sensor data or data features related to the door machine system and acceleration is increased; When the scenario is identified as a "heavy load operation scenario", that is, when the usage scenario is a freight elevator, the weight of sensor data or data features related to traction machine current and main unit temperature is increased. When the scenario is identified as a "stable operation scenario", even if the operation mode is during off-peak hours, the weight of sensor data or data features related to vibration and noise is increased.

[0023] The comprehensive feature vector is input into a pre-trained risk assessment neural network; The risk assessment neural network calculates risk quantification values ​​in parallel across multiple dimensions, including at least the probability of failure, the severity level of failure, and the imminence of failure. The risk quantification values ​​from the multiple dimensions are combined using a weighted fusion function to generate a comprehensive risk score; Based on the comprehensive risk score, the gradient vector of the comprehensive feature vector is calculated through the predictive maintenance model. Based on the gradient vector, the feature dimensions corresponding to the comprehensive risk score are identified by sorting them according to their contribution values. Based on a predefined feature mapping table, the identified feature dimensions are mapped to their corresponding physical sensor and signal features; based on a predefined defect knowledge base, the physical sensor and signal features are mapped to one or more possible defect types. Using the defect type as an index, query the preset maintenance knowledge base and obtain the corresponding maintenance operation information based on the maintenance knowledge base; Based on the fault type, confidence level, and maintenance operation information, one or more specific maintenance suggestions are automatically generated. Experts or maintenance personnel determine the rationality of the maintenance suggestions. When the maintenance suggestions are feasible, decision information containing the fault type and maintenance suggestions is output.

[0024] When the comprehensive risk score exceeds the dynamic threshold, the predictive maintenance model outputs the fault type and maintenance suggestions, and then visualizes the fault type and maintenance suggestions on the user interface.

[0025] See Figure 2 As shown, further, in a second aspect of this application, an elevator detection system is proposed, comprising: The data acquisition module is used to collect multi-source data from the elevator in real time, preprocess the multi-source data and extract features to generate an initial feature vector; The data processing module is used to dynamically adjust the weights of the initial feature vector according to the elevator's usage scenario and operating mode, and to fuse the adjusted initial feature vector through a fusion algorithm to generate a comprehensive feature vector. The suggestion generation module is used to input the comprehensive feature vector into the risk prediction model to generate a comprehensive risk score. When the comprehensive risk score exceeds the dynamic threshold, the prediction and maintenance model outputs the fault type and maintenance suggestion, and displays the fault type and maintenance suggestion on the user interface.

[0026] The data processing module includes: An edge computing gateway, deployed at the elevator site, communicates with the data acquisition module to implement the functions of the data processing module; The cloud platform server is connected to the edge computing gateway via network communication and is used to carry the risk prediction model and the prediction maintenance model. Example 2

[0027] Based on the above embodiment 1, a multimodal fault fingerprint database is constructed in this embodiment to store the characteristic combination of vibration spectrum, current waveform and temperature curve under typical faults; The time-frequency domain features of sensor data are extracted in real time through edge computing nodes and then matched with the fingerprint database for similarity. When the matching degree exceeds the first threshold, a primary warning is triggered and compressed data is uploaded to the cloud; The cloud-based system uses a GNN model to analyze the interrelationships between components and outputs root cause diagnosis results and maintenance priority ranking. Repair each component according to its priority.

[0028] Specifically, in this application, the data acquisition modules are respectively located in various subsystems of the elevator, including: Traction system: Installed on the base of the traction machine main unit, used to collect vibration, noise and temperature data of the main unit; Guiding system: Installed on the car frame, used to collect vibration data relative to the guide rail; Car system: Located on the top of the car, used to collect comprehensive vibration and noise data of the car; Electrical control system: Installed in the form of a current clamp on the traction motor power line inside the control cabinet, used to collect operating current data; Door system: Installed on the car door operator to collect the current data of the operator's motor.

[0029] This invention can identify potential faults and issue early warnings a considerable period (such as days or weeks) before a failure occurs through in-depth analysis and trend prediction of multi-source data. This transforms the operation and maintenance model from traditional post-failure repair to predictive maintenance, greatly reducing unexpected downtime and avoiding potentially serious consequences when a failure occurs.

[0030] By deploying various sensors, such as vibration sensors, noise sensors, current and voltage sensors, and infrared thermal imagers, on key parts of the elevator (traction machine, car, guide rails, and door operator), multi-dimensional physical data of elevator operation are collected simultaneously.

[0031] The collected multidimensional physical data is filtered, denoised, and standardized, and various feature vectors in the time domain, frequency domain, and time-frequency domain are extracted to form a high-dimensional feature set.

[0032] Artificial intelligence-based fusion diagnostics: inputting multi-source feature vectors into a pre-trained deep learning model (such as a convolutional neural network CNN, a long short-term memory network LSTM, or a hybrid model thereof).

[0033] This model can automatically learn the complex nonlinear mapping relationship between different features and fault types, and output a comprehensive health status score and specific fault type identification results.

[0034] Based on historical health status data, time series prediction algorithms (such as Prophet and LSTM) are used to predict the performance degradation trend of key components. When the health score falls below a threshold or the predicted trend indicates an impending failure, the system automatically sends tiered early warning information (such as attention, warning, and danger) to the management platform.

[0035] Specifically, "Note" indicates an early potential problem with slight performance degradation; "Warning" indicates a defect is developing with moderate performance degradation, requiring planned maintenance; and "Danger" indicates a serious defect with a potential for functional failure at any time, requiring immediate shutdown. Furthermore, in this embodiment, taking the traction system, guide system, and door system as examples, the following are included: 1. Traction system: Warning level: Caution; Possible system diagnostic conclusion: Slight bearing wear is possible; Recommendation: During the next planned maintenance, increase the inspection of the traction machine bearings, listen for any abnormal sounds during operation, and record the vibration trend.

[0036] Warning level: Caution; Possible system diagnostic conclusion: bearing outer ring failure (75% confidence level), characteristic frequency amplitude increased; Recommendation: Schedule maintenance (ideally within one week). Check the lubrication of the traction machine bearings. Conduct vibration spectrum analysis to confirm the fault characteristics and prepare spare parts; Warning level: Danger Possible system diagnostic conclusion: Severe bearing damage (90% confidence level), accompanied by high temperature rise; Recommendation: Stop the machine immediately for inspection! There is a risk of jamming. Check the traction machine bearings for cracks and the cage for damage. Contact a professional maintenance person immediately to replace the bearings.

[0037] 2. Guidance System: Warning level: Caution; Possible system diagnostic conclusion: slight wear on the guide shoe liner; Recommendation: Observe the smoothness of the elevator's operation and pay attention to any abnormal friction noises. During the next maintenance, focus on measuring the remaining thickness of the guide shoe gaskets.

[0038] Warning Level: Warning Possible system diagnostic conclusion: Increased wear of guide shoes (80% confidence level) leads to excessive vertical vibration of the car; Recommendation: Schedule maintenance (ideally within two weeks), check the wear of all guide shoes, and measure the lubrication condition of the guide rails. Replace any guide shoe gaskets with excessive wear.

[0039] Warning level: Danger Possible system diagnostic conclusion: Guide shoe liner depleted, metal in direct contact with guide rail; Recommendation: Stop the elevator immediately and replace the guide shoes. Metal friction has damaged the guide rails. Check the guide rail contact surfaces for scratches. All guide shoe gaskets that have exceeded the wear limit must be replaced immediately.

[0040] 3. Door system Warning Level: Caution Possible system diagnostic conclusion: Slight decrease in gantry crane belt tension; Recommendation: Observe whether there is slippage or abnormal noise during the opening and closing of the door. Check and adjust the belt tension during the next maintenance.

[0041] Warning level: Warning; Possible system diagnostic conclusion: Door operator belt slippage (70% confidence level), resulting in abnormal door opening and closing position; Recommendation: Schedule maintenance (ideally within one week) to check belt wear, aging, and tension. Clean the belt and pulleys. Replace the belt if it is severely worn.

[0042] Warning level: Dangerous; Possible system diagnostic conclusion: The gantry crane belt is severely aged and may break at any time; Recommendation: Stop the elevator immediately and replace the belt. The impending belt breakage will cause the door operator to malfunction and trap passengers. Replace the door operator belt immediately and check other components of the door operator system.

[0043] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An elevator detection method, characterized by, The application comprises the following steps: Real-time acquisition of multi-source data of the elevator, preprocessing and feature extraction of the multi-source data, and generation of an initial feature vector; Dynamic adjustment of the weight of the initial feature vector according to the use scenario and operation mode of the elevator, fusion processing of the adjusted initial feature vector through a fusion algorithm, and generation of a comprehensive feature vector; Inputting the comprehensive feature vector into a risk prediction model to generate a comprehensive risk score, outputting a fault type and a maintenance suggestion through a predictive maintenance model when the comprehensive risk score exceeds a dynamic threshold, and visualizing the fault type and the maintenance suggestion on a user interface.

2. The elevator detection method according to claim 1, characterized by The inputting of the comprehensive feature vector into the risk prediction model to generate the comprehensive risk score specifically comprises: Inputting the comprehensive feature vector into a pre-trained risk assessment neural network; Parallelly calculating risk quantization values in multiple dimensions through the risk assessment neural network, wherein the dimensions at least include a fault occurrence probability, a fault severity level, and a fault proximity; Comprehensively generating a comprehensive risk score through a weighted fusion function based on the risk quantization values in multiple dimensions.

3. The elevator detection method according to claim 1, characterized by, The outputting of the fault type and the maintenance suggestion through the predictive maintenance model when the comprehensive risk score exceeds the dynamic threshold specifically comprises: When the comprehensive risk score exceeds the dynamic threshold, calculating a gradient vector of the comprehensive feature vector through the predictive maintenance model, sorting the feature dimensions corresponding to the comprehensive risk score according to the contribution values based on the gradient vector, and identifying the feature dimensions; Mapping the identified feature dimensions to corresponding physical sensors and signal features based on a pre-defined feature mapping table, and mapping the physical sensors and signal features to one or more possible defect types based on a pre-defined defect knowledge base; Taking the defect types as indexes to query a pre-set maintenance knowledge base, and obtaining corresponding maintenance operation information based on the maintenance knowledge base; Based on the fault type, the confidence, and the maintenance operation information, automatically generating one or more specific maintenance suggestions, determining the rationality of the maintenance suggestions by an expert or a maintenance personnel, and outputting decision information containing the fault type and the maintenance suggestion when the maintenance suggestion is feasible.

4. The elevator detection method according to claim 1, characterized by The multi-source data includes mechanical vibration data, acoustic data, electrical operation data, time data, environmental data, and video data.

5. The elevator detection method according to claim 1, characterized by The real-time acquisition of the multi-source data of the elevator, the preprocessing and feature extraction of the multi-source data, and the generation of the initial feature vector specifically comprise the following steps: Synchronously or asynchronously acquiring the multi-source data through a sensor array or other acquisition instruments deployed at different positions of the elevator; Performing data alignment, outlier processing, filtering and denoising, and data normalization on the multi-source data to generate standardized data; Performing time-frequency domain feature extraction on the standardized data to output a feature set; Splicing the feature set in a predetermined order to output an initial feature vector.

6. The elevator detection method according to claim 1, characterized by The dynamic adjustment of the weight of the initial feature vector according to the use scenario and operation mode of the elevator, and the fusion processing of the adjusted initial feature vector through the fusion algorithm to generate the comprehensive feature vector specifically comprise the following steps: Identifying the use scenario and operation mode to which the elevator currently belongs based on real-time multi-source data of the elevator; According to the identified use scenario and operation mode, a corresponding set of weight coefficients is called from a predefined policy library; Using the set of weight coefficients, each feature component in the initial feature vector is subjected to a weighting operation to generate a weighted feature vector; Through a fusion algorithm, the weighted feature vector is subjected to dimension reduction and aggregation processing to generate a comprehensive feature vector.

7. The elevator detection method according to claim 6, characterized by The use scenarios include passenger elevators, cargo elevators, medical elevators, miscellaneous elevators, and sightseeing elevators, and the use scenarios are classified into predefined scenario types by a rule engine or a classification model based on one or more dimensional data of real-time load rate, operation frequency, and time of the elevators.

8. The elevator detection method according to claim 6, characterized by The operation modes include peak hours, flat peak hours, and night hours, and the use scenarios are classified into predefined operation mode types by a rule engine or a classification model based on one or more dimensional data of real-time load rate, operation frequency, and time of the elevators.

9. An elevator detection system characterized by It comprises: A data acquisition module for real-time acquisition of multi-source data of the elevator, pre-processing and feature extraction of the multi-source data, and generation of an initial feature vector; A data processing module for dynamically adjusting the weight of the initial feature vector according to the use scenario and operation mode of the elevator, and performing fusion processing on the adjusted initial feature vector through a fusion algorithm to generate a comprehensive feature vector; A suggestion generation module for inputting the comprehensive feature vector into a risk prediction model to generate a comprehensive risk score, and outputting a fault type and maintenance suggestion through a predictive maintenance model when the comprehensive risk score exceeds a dynamic threshold, and visualizing the fault type and maintenance suggestion on a user interaction interface.

10. The elevator detection system of claim 9, wherein, The data processing module comprises: An edge computing gateway deployed at the elevator site and in communication connection with the data acquisition module for realizing the functions of the data processing module; A cloud platform server in network communication connection with the edge computing gateway for carrying the risk prediction model and the predictive maintenance model.

Images