An elevator intelligent maintenance profile generation system
Patent Information
- Application Number
- CN202610871144.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-16
- Publication Date
- 2026-09-25
AI Technical Summary
[0002]在电梯维保中，需要根据电梯的运行状态制定相应的维保计划，目前一般根据电梯的使用年限和使用环境选用不同级别的维保计划，而不同电梯的性能和实际状态均不同，这会导致维保计划无法很好地适配目标电梯，影响对目标电梯的维保质量，而要根据电梯的实际健康状态制定精准的维保计划，则需要大量的数据采集工作，且需要针对数据进行全面的分析，同时还需要丰富的经验，才可以制定出精准的维保计划，工作量较大且存在很大的不确定性
[0012]本发明的有益效果在于：通过自动采集电梯运行数据，并对电梯运行数据进行时域特征和频域特征的提取，结合进行批量训练和增量训练的神经网络模型，以此生成电梯的健康状态画像，实现对电梯健康状态的准确评估，以此生成更匹配目标电梯的维保计划，提高对电梯的维保效率。
Smart Images
Figure CN122820170A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of elevator maintenance technology, specifically to an intelligent elevator maintenance profile generation system. Background Technology
[0002] In elevator maintenance, a corresponding maintenance plan needs to be developed based on the elevator's operating status. Currently, different levels of maintenance plans are generally selected based on the elevator's service life and operating environment. However, different elevators have different performance and actual conditions, which can lead to maintenance plans not being well adapted to the target elevator, affecting the quality of maintenance. To develop an accurate maintenance plan based on the elevator's actual health status, a large amount of data collection and comprehensive data analysis are required, along with extensive experience. This involves a significant workload and considerable uncertainty. Summary of the Invention
[0003] The purpose of this invention is to provide an intelligent elevator maintenance profile generation system, specifically a system that can automatically acquire elevator operation data and generate an elevator health status profile to generate an accurate elevator maintenance plan.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent elevator maintenance profile generation system, comprising a data acquisition module, a data preprocessing module, a feature engineering processing module, a model training processing module, an elevator health status characterization module, a performance comparison and analysis module for similar equipment groups, a dynamic optimal operating state reference model module, and a maintenance decision information output module; the data acquisition module is used to collect elevator operating data; the data preprocessing module is used to clean and normalize the raw data collected by the data acquisition module to generate structured data; the feature engineering processing module is used to extract time-domain and frequency-domain features from the structured data to obtain multi-dimensional feature structured data; the model training processing module is used to train the elevator health status profile model based on the batch training dataset of the multi-dimensional feature structured data. The system trains and receives multi-dimensional structured data from the feature engineering processing module to generate elevator health status feature vectors. The elevator health status characterization module integrates these feature vectors to generate a health status profile for each elevator. The performance comparison and analysis module for similar equipment groups receives elevator health status feature vectors from different elevators, clusters them, performs performance difference analysis on each cluster, and generates performance comparison analysis results. The dynamic optimal operating state reference model module selects the performance comparison analysis results of the corresponding cluster based on the target elevator's health status profile, thereby generating ideal parameter ranges, trend predictions, and warning thresholds for the target elevator. The maintenance decision information output module displays maintenance recommendations based on the ideal parameter ranges, trend predictions, and warning thresholds to the operators.
[0005] Specifically, the data acquisition module includes a data acquisition intelligent agent framework module and a data acquisition skill plug-in module. The data acquisition intelligent agent framework module is equipped with a visualization skill module configuration interface and an expansion interface that can connect to multiple data acquisition skill plug-in modules. The data acquisition skill plug-in module is used to connect to the elevator's operation sensors to collect the elevator's operation data.
[0006] Specifically, the data preprocessing module cleans the raw data by filling in missing values and removing outliers, and normalizes the raw data by converting the data from each sensor to a uniform numerical range.
[0007] Specifically, the feature engineering processing module extracts time-domain and frequency-domain features from structured data to obtain multidimensional feature structured data. This is achieved by extracting time-domain and frequency-domain features from the structured data respectively, and then combining the time-domain and frequency-domain feature sets to obtain multidimensional feature structured data with multidimensional time-domain and frequency-domain features.
[0008] Specifically, after receiving a set number of multidimensional feature structured data, the model training processing module uses all the received multidimensional feature structured data to update the full model parameters; after receiving a single multidimensional feature structured data and outputting the elevator health status feature vector, the model training processing module calculates the deviation between the elevator health status feature vector and the historical elevator health status feature vector, and performs local model parameter fine-tuning when the deviation exceeds a set threshold.
[0009] Specifically, when the elevator health status representation module integrates the elevator health status feature vector, it converts the elevator health status feature vector into specific elevator health indicators.
[0010] Specifically, the performance comparison and analysis module for similar equipment groups performs group performance difference analysis for each cluster group, and the generated performance comparison and analysis results include the confidence intervals of each parameter of all equipment in a single cluster group.
[0011] Specifically, the dynamic optimal operating state reference model module is a model trained by iteratively learning normal data. By inputting the health status profile of the target elevator and the performance comparison analysis results, it outputs the ideal parameter range, trend prediction, and early warning threshold of the target elevator.
[0012] The beneficial effects of this invention are as follows: by automatically collecting elevator operation data and extracting time-domain and frequency-domain features from the elevator operation data, and combining it with a neural network model that performs batch training and incremental training, a health status profile of the elevator is generated, thereby achieving an accurate assessment of the elevator's health status and generating a maintenance plan that is more suitable for the target elevator, thus improving the efficiency of elevator maintenance. Attached Figure Description
[0013] Appendix Figure 1 This is a schematic diagram of the overall connection principle of an intelligent elevator maintenance profile generation system in an embodiment. Detailed Implementation
[0014] Example 1, referring to Figure 1 An intelligent elevator maintenance profile generation system includes a data acquisition module, a data preprocessing module, a feature engineering module, a model training module, an elevator health status characterization module, a performance comparison and analysis module for similar equipment groups, a dynamic optimal operating status reference model module, and a maintenance decision information output module. The data acquisition module collects elevator operating data. The data preprocessing module cleans and normalizes the raw data collected by the data acquisition module to generate structured data. The feature engineering module extracts time-domain and frequency-domain features from the structured data to obtain multi-dimensional feature structured data. The model training module trains the elevator health status profile model based on a batch training dataset of the multi-dimensional feature structured data and outputs the model data from the feature engineering module. The module receives multi-dimensional structured data to generate elevator health status feature vectors; the elevator health status characterization module integrates these feature vectors to generate a health status profile for a single elevator; the performance comparison and analysis module for similar equipment groups receives elevator health status feature vectors from different elevators and performs clustering, analyzing performance differences for each cluster and generating performance comparison analysis results; the dynamic optimal operating state reference module selects the performance comparison analysis results of the corresponding cluster based on the target elevator's health status profile to generate an optimal operating state reference model for the target elevator; and the maintenance decision information output module generates maintenance suggestions based on the optimal operating state reference model and the target elevator's structured data, displaying these suggestions to operators.
[0015] Specifically, the data acquisition module includes a data acquisition intelligent agent framework module and a data acquisition skill plugin module. The data acquisition intelligent agent framework module has a visual skill module configuration interface and an extension interface that can connect to multiple data acquisition skill plugin modules. The data acquisition skill plugin modules are used to connect to the elevator's operation sensors to collect elevator operation data. The data acquisition intelligent agent framework module is deployed on a local server cluster, and its intelligent agent framework can use a currently available open-source data acquisition intelligent agent framework; more specifically, this embodiment uses Qianwen Intelligent Agent version 3.5. The data acquisition skill plugin module connects to the elevator's main control room and periodically acquires elevator operation data collected by the elevator's built-in sensors to obtain elevator operation data and transmits it to the data preprocessing module.
[0016] Specifically, the data preprocessing module cleans the raw data, including filling in missing values and removing outliers. The normalization process converts the sensor data to a uniform numerical range. When filling in missing values, linear interpolation can be used; for severely missing values, the corresponding portion of the elevator operation data is directly removed. When removing outliers, the Z-Score (3σ criterion) and Interquartile Range (IQR) algorithm, combined with physical thresholds for elevator operation (reasonable upper and lower limits for speed and current), can be used to quickly identify and delete outliers in the raw data.
[0017] Specifically, the feature engineering processing module extracts time-domain and frequency-domain features from structured data to obtain multidimensional feature structured data. This is achieved by extracting time-domain and frequency-domain features from the structured data respectively, and then combining the time-domain and frequency-domain feature sets to obtain multidimensional feature structured data with multidimensional time-domain and frequency-domain features. Specifically, when the feature engineering processing module performs time-domain feature extraction, it extracts statistical quantities from the time-series signals of sensors such as vibration and current in the structured data, including but not limited to: mean, standard deviation, peak value, root mean square (RMS), peak-to-peak value, waveform factor, impulse factor, margin factor, skewness, kurtosis, etc., to obtain an 8-12 dimensional feature set. When performing frequency-domain feature extraction, it first converts the time-series signals of sensors such as vibration and current in the structured data to the frequency domain, and then extracts the feature quantities, including but not limited to: centroid frequency, mean square frequency, frequency variance, spectral peak value, and frequency band energy (such as 0-50Hz power frequency energy), to form a 5-10 dimensional frequency-domain feature set.
[0018] In addition, after completing the initial training, the model training module can receive multi-dimensional structured data from the feature engineering module and generate elevator health status feature vectors. These feature vectors include feature vectors reflecting both mechanical and electrical health indicators. Specifically, mechanical health indicators include vibration amplitude, operating noise in decibels, leveling accuracy error, traction machine bearing temperature, and guide rail clearance; electrical health indicators include contactor / relay engagement time, insulation resistance value, control cabinet temperature, inverter output current harmonic distortion rate, and door operator current curve anomaly. Correspondingly, when the elevator health status characterization module integrates these feature vectors, it converts them into specific elevator health indicators (i.e., the aforementioned mechanical and electrical health indicators).
[0019] Specifically, after receiving a set amount of multidimensional feature structured data, the model training processing module performs a full update of the model parameters using all received multidimensional feature structured data. When a sufficient amount of multidimensional feature structured data is received, this full model parameter update allows the model to better adapt to the target elevator, improving the accuracy of the output elevator health status feature vector. Furthermore, after receiving a single set of multidimensional feature structured data and outputting an elevator health status feature vector, the model training processing module calculates the deviation between this feature vector and historical elevator health status feature vectors. Historical elevator health status feature vectors can be obtained from the elevator's historical elevator health status feature vectors, specifically those with a certain time interval from the current time. A large deviation indicates a significant change in the target elevator's health status, requiring model parameter fine-tuning to adapt to the current elevator state. Specifically, model parameter fine-tuning is performed using the deviation between the current elevator health status feature vector and historical elevator health status feature vectors.
[0020] Specifically, the performance comparison and analysis module for similar equipment groups performs group performance difference analysis for each cluster. The generated performance comparison and analysis results include confidence intervals for each parameter of all equipment within a single cluster. When generating confidence intervals, statistical analysis (such as mean, variance, percentile) and clustering algorithms (such as K-means) can be performed based on all historical operating data of all equipment within the cluster to calculate the confidence intervals for each parameter under normal operating conditions (which can be set to a 95% confidence level).
[0021] Specifically, the dynamic optimal operating state reference model module is a model trained by iteratively learning normal data. By inputting the health status profile of the target elevator and the performance comparison analysis results, it outputs the ideal parameter range, trend prediction, and early warning threshold of the target elevator.
[0022] Of course, the above are only preferred embodiments of the present invention and are not intended to limit the scope of application of the present invention. Therefore, any equivalent changes made to the principle of the present invention should be included within the protection scope of the present invention.
Claims
1. An intelligent elevator maintenance profile generation system, characterized in that: It includes a data acquisition module, a data preprocessing module, a feature engineering processing module, a model training processing module, an elevator health status characterization module, a performance comparison and analysis module for similar equipment groups, a dynamic optimal operating status reference model module, and a maintenance decision information output module; the data acquisition module is used to collect elevator operating data; The data preprocessing module is used to clean and normalize the raw data collected by the data acquisition module to generate structured data; The feature engineering processing module is used to extract time-domain and frequency-domain features from the structured data to obtain multi-dimensional feature structured data. The model training module trains the elevator health status profile model using a batch training dataset of multidimensional feature structured data, and receives multidimensional feature structured data from the feature engineering module to generate elevator health status feature vectors. The elevator health status characterization module integrates the elevator health status feature vectors to generate a health status profile for a single elevator. The performance comparison and analysis module for similar equipment groups receives elevator health status feature vectors from different elevators and performs clustering, analyzes the group performance differences for each cluster, and generates performance comparison and analysis results. The Dynamic Optimal Operating State Reference Model module is used to select the performance comparison analysis results of the corresponding clustering group based on the health status profile of the target elevator, thereby generating the ideal parameter range, trend prediction and early warning threshold for the target elevator. The maintenance decision information output module is used to display maintenance recommendations, including ideal parameter ranges, trend predictions, and early warning thresholds, to the operators.
2. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: The data acquisition module includes a data acquisition intelligent agent framework module and a data acquisition skill plug-in module. The data acquisition intelligent agent framework module is equipped with a visualization skill module configuration interface and an expansion interface that can be connected to multiple data acquisition skill plug-in modules. The data acquisition skill plug-in module is used to connect with the elevator's operation sensors to collect the elevator's operation data.
3. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: The data preprocessing module cleans the raw data by filling in missing values, removing outliers, and normalizing the data by converting the sensor data to a uniform numerical range.
4. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: The feature engineering processing module extracts time-domain and frequency-domain features from structured data to obtain multidimensional feature structured data. This is achieved by extracting time-domain and frequency-domain features from the structured data respectively, and then combining the time-domain and frequency-domain feature sets to obtain multidimensional feature structured data with multidimensional time-domain and frequency-domain features.
5. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: After receiving a set number of multidimensional feature structured data, the model training processing module updates all model parameters using all received multidimensional feature structured data. After receiving a single multidimensional feature structured data and outputting an elevator health status feature vector, the model training processing module calculates the deviation between the elevator health status feature vector and the historical elevator health status feature vector. When the deviation exceeds a set threshold, local model parameter fine-tuning is performed.
6. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: When the elevator health status characterization module integrates the elevator health status feature vector, it converts the elevator health status feature vector into specific elevator health indicators.
7. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: The performance comparison and analysis module for the same type of equipment performs a group performance difference analysis for each cluster group, and the generated performance comparison and analysis results include the confidence intervals of each parameter of all equipment in a single cluster group.
8. The elevator intelligent maintenance profile generation system according to claim 1, characterized in that: The dynamic optimal operating state reference model module is a model trained by iteratively learning normal data. By inputting the health status profile of the target elevator and the performance comparison analysis results, it outputs the ideal parameter range, trend prediction and warning threshold of the target elevator.