A method for pre-diagnosing elevator failure
By classifying elevator malfunctions and modeling them in multiple dimensions, and utilizing the Logistic model, machine learning model, and Cox proportional hazards model, the accuracy and applicability of elevator malfunction prediction were addressed, achieving efficient elevator malfunction prediction and early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SJEC CORP
- Filing Date
- 2023-05-17
- Publication Date
- 2026-07-31
AI Technical Summary
Existing elevator fault prediction technologies suffer from low accuracy and limited applicability, making it difficult to effectively predict potential elevator faults.
By classifying elevator fault codes, extracting structured data, performing variance analysis and variable screening, establishing Logistic and machine learning models, and combining them with the Cox proportional hazards model, we can predict possible future Class C faults of elevators. We can also use elevator operation log data for data cleaning and feature engineering to achieve multi-dimensional fault prediction.
It improves the accuracy, recall rate, and correctness of elevator malfunction prediction, enabling early prediction of elevator malfunctions and reducing the probability of safety accidents.
Smart Images
Figure CN116812698B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of elevator fault prediction technology, and in particular to a method for pre-diagnosing elevator faults. Background Technology
[0002] The elevator industry in China has developed for over forty years. With continuous advancements in science and technology, the total number of elevators has increased exponentially. China is now the world's largest elevator producer, consumer, and exporter. As a special type of equipment closely related to people's lives, the safety of elevators directly impacts the lives and property of users. With the acceleration of urbanization, the number of elevators in newly built residential communities and commercial facilities in China has increased dramatically. Consequently, the number of aging elevators has also increased, making elevator safety issues increasingly prominent. Therefore, elevator fault prediction is particularly important. Existing elevator fault prediction technologies all rely on a single dimension, resulting in low accuracy and limited applicability.
[0003] The above background information is provided only to assist in understanding the inventive concept and technical solution of this invention. It does not necessarily belong to the prior art of this patent application, nor does it necessarily provide technical teaching. In the absence of clear evidence that the above information was disclosed before the filing date of this patent application, the above background information should not be used to evaluate the novelty and inventiveness of this application. Summary of the Invention
[0004] To overcome the shortcomings of existing technologies, this invention provides a method for pre-diagnosing elevator faults, the specific technical solution of which is as follows:
[0005] An elevator fault pre-diagnosis method is provided, including the following steps: classifying elevator faults according to the severity of the fault based on the elevator fault code. The elevator fault types include Class A faults, Class B faults, and Class C faults, wherein Class A faults are automatic repair faults, Class B faults are potential elevator stop faults, and Class C faults are direct elevator stop faults.
[0006] Log data from multiple elevators is extracted and preprocessed to form structured data. The fault types of each elevator are statistically analyzed. Feature variables related to the elevator's operating status are extracted from the structured data and used as corresponding variables. The occurrence of Class C faults is used as a factor, with occurrence and non-occurrence as two levels for analysis of variance. The analysis of variance results determine whether the corresponding feature variables affect Class C faults, thus identifying the variables influencing the occurrence of Class C faults. Combined with correlation analysis results, key variables affecting the occurrence of Class C faults are selected. A fault prediction model is then established and trained. If the accuracy, recall, precision, and F1 score of the fault prediction model all meet the corresponding preset requirements, the fault prediction model is output to predict whether a Class C fault will occur in the elevator in the future.
[0007] Furthermore, the fault prediction model includes a Logistic model, where the prediction result for the C-type fault is 0 or 1, where 0 indicates that the C-type fault did not occur and 1 indicates that the C-type fault occurred.
[0008] The structured data dataset is divided into a training set and a test set. The Logistic model is optimized using the training set and then tested using the test set to obtain result datasets with predicted values of 0, 1, 1, and 0, and thus its predictive ability is judged.
[0009] Furthermore, the fault prediction model includes a machine learning model, where the prediction result for the C-type fault is 0 or 1, where 0 indicates that the C-type fault did not occur and 1 indicates that the C-type fault occurred.
[0010] The structured data dataset is divided into a training set and a test set. The machine learning model is optimized using the training set and then tested using the test set to obtain result datasets with predicted values of 0, 1, 1, and 0, and thus its predictive ability is judged.
[0011] Furthermore, the fault prediction model also includes a machine learning model, using the result datasets with predicted values of 0 and / or 1 in the Logistic model as the training set of the machine learning model, and using the result datasets with 0 predicted as 1 and 1 predicted as 0 as the test set of the machine learning model.
[0012] Furthermore, the machine learning model is modeled using random forest, decision tree, support vector machine, XGBoost, or LightGBM.
[0013] Furthermore, the fault types and lifespan of each elevator were statistically analyzed. The lifespan of an elevator refers to the time interval between two consecutive stops. The time characteristics of the lifespan of different Class C faults were analyzed, and the frequency of occurrence of different lifespan times was statistically analyzed.
[0014] The fault prediction model includes a Cox proportional hazards model. By calling SAS to filter variables, key variables affecting lifespan are obtained, and then a corresponding Cox proportional hazards model is established. The Cox proportional hazards model can calculate the corresponding survival probability, which serves as a probability reference for elevator prediction results.
[0015] Furthermore, if the elevator prediction result indicates a Class C fault, and the corresponding probability value is lower than the first preset standard, an alarm will be triggered; if the elevator prediction result indicates a Class C fault, and the corresponding probability value is higher than the first preset standard, model optimization is required; if the elevator prediction result indicates no Class C fault, and the corresponding probability value is higher than the second preset standard, no alarm will be triggered; if the elevator prediction result indicates no Class C fault, and the corresponding probability value is lower than the second preset standard, model optimization is required.
[0016] Furthermore, the data extraction from the log data includes the following steps:
[0017] Extract all data information from predefined fields in the elevator operation log;
[0018] Convert byte data according to different protocol versions;
[0019] Extract data from each elevator and merge data from different protocol versions of the same elevator;
[0020] Convert text-formatted data into tabular format;
[0021] To examine the time interval between two consecutive failures of a single elevator, all data within a single day from the elevator's data can be merged into a single day, and the frequency of various fault codes can be counted.
[0022] Furthermore, the data preprocessing of the log data includes: handling outliers and filling in missing values in the data extracted from the log data, thereby achieving data cleaning.
[0023] Furthermore, in elevator malfunction statistics, it is necessary to remove artificially simulated elevator malfunctions. The distinction between real and simulated malfunctions includes the following steps:
[0024] Faults that have occurred under simulated conditions are recorded as test faults;
[0025] If the test fault occurs during elevator maintenance, the corresponding test fault will be recorded as a simulated fault.
[0026] If the test fault is not accompanied by any other elevator fault, the corresponding test fault shall be recorded as a real fault.
[0027] Compared with the prior art, the present invention has the following advantages: it quantifies the key factors that cause elevator malfunctions from different dimensions, thereby realizing the pre-diagnosis of elevator malfunctions. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the pre-diagnosis system structure in the elevator fault pre-diagnosis method provided in this embodiment of the invention;
[0029] Figure 2 This is a schematic diagram illustrating the application of the pre-diagnosis system in the elevator fault pre-diagnosis method provided in this embodiment of the invention;
[0030] Figure 3 This is a schematic diagram of the elevator fault pre-diagnosis method provided in an embodiment of the present invention;
[0031] Figure 4 This is a schematic representation of elevator fault code classification in the elevator fault pre-diagnosis method provided in this embodiment of the invention;
[0032] Figure 5 This is a schematic diagram illustrating the overall evaluation index performance of the model under different training sets in the elevator fault pre-diagnosis method provided in this embodiment of the invention. Detailed Implementation
[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.
[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.
[0035] In one embodiment of the present invention, an elevator fault pre-diagnosis system is provided, see [link to relevant documentation]. Figure 1 and Figure 2The pre-diagnostic system includes an elevator logic controller, elevator fault acquisition and analysis equipment, and a server with elevator fault prediction modeling capabilities. The elevator fault acquisition and analysis equipment is installed in the computer room and communicates with the elevator logic controller in real time. When an elevator malfunctions, the elevator fault acquisition and analysis equipment will detect the fault immediately and upload the elevator fault information and the elevator's status at the time of the fault to the server with elevator fault prediction modeling capabilities.
[0036] The elevator logic controller reads elevator fault and status information in real time and transmits it to the elevator fault acquisition and analysis equipment. The elevator fault acquisition and analysis equipment has a built-in wireless communication module that wirelessly uploads elevator fault information and the elevator's status at the time of the fault to the server in real time. (See also...) Figure 3 The server records historical faults and extracts historical elevator status and event data. Through methods such as locating field bytes, data processing and transformation, segmented extraction by elevator, merging different protocols, and text data conversion, outlier handling and missing value filling are performed on the elevator historical logs, achieving data cleaning. The cleaned data is then input into mathematical models such as the Logistic model, machine learning model, and Cox proportional hazards model. Through calculation, the probability of elevator faults occurring in the next day and in the next 2-7 days is determined. Based on the feedback of whether elevator faults occur, the accuracy of the algorithm is optimized and improved. The server can generate elevator fault prediction accuracy reports for different time periods. Through various statistical analysis methods and big data analysis techniques such as data cleaning and feature engineering, the key factors causing elevator faults are quantified from different dimensions, thereby achieving pre-diagnosis of elevator faults.
[0037] In one embodiment of the present invention, an elevator fault pre-diagnosis method is provided, which is implemented through a pre-diagnosis system and includes the following steps:
[0038] Based on elevator fault codes, elevator faults are classified according to their severity. (See [link / reference]) Figure 4 The elevator malfunction types include Class A, Class B, and Class C malfunctions. Class A malfunctions are self-healing malfunctions, which are relatively common and have little impact on normal elevator operation (i.e., they will recover automatically and do not affect the normal use of the elevator). Class B malfunctions are potential elevator stop malfunctions, which can cause low-speed operation, delayed operation, etc., but will recover on their own (i.e., more serious malfunctions that require serious attention and handling, but do not necessarily require immediate on-site handling). Class C malfunctions are direct elevator stop malfunctions, which directly cause the elevator to stop (i.e., elevator stop malfunctions that require immediate on-site handling).
[0039] Log data from multiple elevators is extracted and preprocessed to form structured data. The fault types of each elevator are statistically analyzed. Feature variables related to the elevator's operating status are extracted from the structured data and used as corresponding variables. The occurrence of a Class C fault is considered as the outcome, with occurrence and non-occurrence as two levels for analysis of variance. The analysis of variance results determine whether the corresponding feature variables affect Class C faults, thus identifying the variables influencing their occurrence. Combined with correlation analysis results, key variables affecting Class C faults are selected, and a fault prediction model is established and trained. Key variables include voltage, current, operating speed, and load. If the accuracy, recall, precision, and F1 score of the fault prediction model all meet the corresponding preset requirements, the model is output to predict whether a Class C fault will occur in the elevator in the future.
[0040] In a preferred embodiment, the fault prediction model includes a Logistic model, where the prediction result for the C-type fault is 0 or 1, where 0 indicates that the C-type fault did not occur and 1 indicates that the C-type fault occurred.
[0041] The structured data dataset is divided into a training set and a test set. The Logistic model is optimized using the training set and then tested using the test set, resulting in datasets with predicted values of 0, 1, 1, and 0, thus assessing its predictive ability. The Logistic model obtains influencing variables through variable compression and predictive ability filtering; it uses forward and backward regression methods for modeling, further removing insignificant variables. The results are similar to those of stepwise regression, and the maximum likelihood estimates of the parameters and corresponding statistical test results are presented.
[0042] Building upon the previous embodiment, the fault prediction model further includes a machine learning model. The result datasets with predicted values of 0 and / or 1 from the Logistic model are used as the training set for the machine learning model, while the result datasets with 0 predicted as 1 and 1 predicted as 0 are used as the test set. Specifically, this embodiment employs Logistic regression modeling and the random forest algorithm from the machine learning algorithm, combined with the Self-spaced Ensemble imbalanced learning framework, to perform two-stage modeling of the data, thereby achieving elevator fault prediction.
[0043] After the logistic regression model, the random forest model was trained separately using three methods: datasets with predicted values of 1, datasets with predicted values of 0, and a combination of both. The test dataset consisted of datasets where 1 was predicted as 0 and 0 as 1. After combining the two stages of modeling, from... Figure 5 It can be seen that the two-stage modeling of the three methods all perform very well, with an accuracy of at least 95%, a recall of at least 93%, a precision of over 99%, and an F1 score of at least 96%, thus demonstrating that two-stage modeling can greatly improve the prediction effect.
[0044] In a preferred embodiment, the fault prediction model includes a machine learning model, wherein the prediction result of the C-type fault is 0 or 1, where 0 indicates that the C-type fault has not occurred and 1 indicates that the C-type fault has occurred; the machine learning model is modeled using random forest, decision tree, support vector machine, XGBoost or LightGBM.
[0045] The structured data dataset is divided into a training set and a test set. The machine learning model is optimized using the training set and then tested using the test set to obtain result datasets with predicted values of 0, 1, 1, and 0, and thus its predictive ability is judged.
[0046] In a preferred embodiment, the fault type and lifespan of each elevator are statistically analyzed. The elevator's lifespan refers to the time interval between two consecutive stops. The temporal characteristics of the lifespan of different Class C faults are analyzed, and the frequency of occurrence of different lifespan durations is statistically analyzed. The fault prediction model also includes a Cox proportional hazards model. Variables are filtered using SAS to obtain key variables affecting lifespan, and a corresponding Cox proportional hazards model is established. The Cox proportional hazards model can calculate the corresponding survival probability as a probabilistic reference for the elevator prediction result. This probability value can add a probabilistic basis to the subsequent judgment of elevator fault occurrence. If the elevator prediction result is a Class C fault and its corresponding probability value is lower than a first preset standard, an alarm is triggered. If the elevator prediction result is a Class C fault and its corresponding probability value is higher than the first preset standard, model optimization is required. If the elevator prediction result is no Class C fault and its corresponding probability value is higher than a second preset standard, no alarm is triggered. If the elevator prediction result is no Class C fault and its corresponding probability value is lower than the second preset standard, model optimization is required.
[0047] Specifically, the basic calculation formula for the Cox proportional hazards model is as follows:
[0048]
[0049] In the formula, x1, x2, ..., x p For factors related to survival time, β1, β2, ..., β3 are the estimated parameter values corresponding to these factors in the survival analysis; h(t) is the hazard function at time t, and h0(t) is the hazard function at time t when all influencing factors are zero, called the baseline hazard function. Therefore, the baseline survival function is as follows:
[0050] S0(t)=exp(-∫0 t h0(u)du)
[0051] The elevator survival probability can be obtained through this function; by calling the PHReg procedure in SAS to filter the variables step by step, the key variables are obtained, and survival analysis is performed to improve the function parameters.
[0052] In one embodiment of the present invention, in the statistics of elevator malfunctions and lifespan, it is necessary to remove artificially simulated elevator malfunctions. The distinction between real malfunctions and simulated malfunctions includes the following steps:
[0053] Faults that have been simulated are recorded as test faults. Class C faults within these test faults are categorized as follows: if the test fault occurs during elevator maintenance, it is recorded as a simulated fault; if it does not occur during any other elevator faults, it is recorded as a real fault. If the test fault occurs during other Class C elevator faults, it is considered an elevator fault, but the test fault is not recorded in the corresponding count.
[0054] In one embodiment of the present invention, the data extraction of the log data includes the following steps:
[0055] The process involves extracting all data from predefined fields in the elevator operation logs; converting byte data according to different protocol versions; extracting data for each elevator and merging data from different protocol versions for the same elevator; converting text-formatted data into tabular format; examining the time interval between two consecutive faults for a single elevator; merging all data from a single elevator within a single day into a single day's data; and counting the frequency of various fault codes. The data preprocessing of the log data includes outlier removal and missing value imputation on the extracted data, thereby achieving data cleaning.
[0056] The elevator fault pre-diagnosis method provided by this invention uses historical elevator operation data to predict elevator faults and remaining lifespan, so that relevant personnel can understand the elevator's status in advance, maintain the elevator in a timely manner, and reduce the probability of elevator safety accidents.
[0057] The above description is merely a preferred embodiment of the present invention and does not limit its patent scope. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, whether directly or indirectly applied to other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. A method for pre-diagnosing an elevator failure, characterized by, Includes the following steps: According to the elevator fault codes, elevator faults are classified according to their severity. The elevator fault types include Class A faults, Class B faults, and Class C faults. Among them, Class A faults are automatic repair faults, Class B faults are potential elevator stop faults, and Class C faults are direct elevator stop faults. Log data from multiple elevators is extracted and preprocessed to form structured data. The fault types of each elevator are statistically analyzed. Feature variables related to the elevator's operating status are extracted from the structured data. These feature variables are used as corresponding variables. The occurrence of Class C faults is taken as the result, with occurrence and non-occurrence as two levels for analysis of variance. The analysis of variance results determine whether the corresponding feature variables affect Class C faults, thus identifying the variables influencing the occurrence of Class C faults. Combined with the results of variable correlation analysis, key variables affecting the occurrence of Class C faults are selected. A fault prediction model is then established and trained. These key variables include voltage, current, operating speed, and load. The fault prediction model includes a Logistic model and a machine learning model. The prediction result for Class C faults is set to 0 or 1, where 0 indicates no Class C fault occurred and 1 indicates a Class C fault occurred. The dataset formed by the structured data is divided into a training set and a test set. The datasets with predicted values of 0 and 1 from the Logistic model are used as the training set for the machine learning model, and the datasets with 0 predicted as 1 and 1 predicted as 0 are used as the test set. A two-stage modeling process is performed using Logistic regression modeling and the random forest algorithm from the machine learning algorithm, combined with the Self-spaced Ensemble imbalanced learning framework. If the accuracy, recall, precision, and F1 score of the fault prediction model all meet the corresponding preset requirements, the fault prediction model is output to predict whether the elevator will experience a Class C fault in the future. If the elevator prediction indicates a Class C fault and the corresponding probability value is lower than the first preset standard, an alarm will be triggered. If the elevator prediction indicates a Class C fault and the corresponding probability value is higher than the first preset standard, model optimization is required. If the elevator prediction indicates no Class C fault has occurred and the corresponding probability value is higher than the second preset standard, no alarm will be triggered. If the elevator prediction indicates no Class C fault has occurred and the corresponding probability value is lower than the second preset standard, model optimization is required. In elevator fault statistics, it is necessary to remove artificially simulated elevator faults. The distinction between real faults and simulated faults includes the following steps: faults that have occurred under artificial simulation are recorded as test faults; if the test fault occurs in conjunction with elevator maintenance, the corresponding test fault is recorded as a simulated fault; if the test fault does not occur in conjunction with any other elevator faults, the corresponding test fault is recorded as a real fault.
2. The elevator failure pre-diagnosis method according to claim 1, characterized by, The Logistic model is optimized using a training set and then tested using a test set to obtain result datasets with predicted values of 0, 1, 1 (predicting 0 as 1), and 0 (predicting 1 as 0), thereby determining its predictive ability.
3. The elevator failure pre-diagnosis method according to claim 1, characterized by, The machine learning model is optimized using a training set and then tested using a test set to obtain result datasets with predicted values of 0, 1, 1 (predicting 0 as 1), and 0 (predicting 1 as 0), thereby determining its predictive ability.
4. The elevator failure pre-diagnosis method according to claim 1, characterized by, The machine learning model is modeled using random forest, decision tree, support vector machine, XGBoost, or LightGBM.
5. The elevator failure pre-diagnosis method according to claim 1, characterized by, The data extraction from the log data includes the following steps: Extract all data information from predefined fields in the elevator operation log; Convert byte data according to different protocol versions; Extract data from each elevator and merge data from different protocol versions of the same elevator; Convert text-formatted data into tabular format; To examine the time interval between two consecutive failures of a single elevator, all data from a single elevator within a single day are merged into one day, and the frequency of various fault codes is counted.
6. The elevator failure pre-diagnosis method according to claim 5, characterized by, The data preprocessing of the log data includes: removing outliers and filling in missing values in the data extracted from the log data, thereby achieving data cleaning.