An elevator maintenance method and system

Abstract

The application relates to the elevator technical field, especially to an elevator maintenance method and system, the method comprises the following steps: acquiring real-time operation data of the elevator, comparing normal parameter range to generate an abnormal label; obtaining a fault core feature set through associated analysis of the abnormal label, combining deep learning to construct a fault diagnosis and maintenance recommendation model; inputting the core feature set to obtain a diagnosis result and a preliminary maintenance scheme, verifying whether the scheme meets the standard, generating a final maintenance plan if the scheme meets the standard, and predicting the next maintenance interval combined with historical records. The system comprises data acquisition, judgment, label generation and other modules. The application helps to solve the problem of lack of pertinence in traditional fixed cycle maintenance, improves the maintenance accuracy through real-time data and dynamic diagnosis, optimizes the maintenance interval combined with historical data, guarantees the safety of elevator operation, and reduces the cost of excessive maintenance and the risk of hidden danger omission.

Description

TECHNICAL FIELD

[0001] The present application relates to the elevator technical field, especially relates to an elevator maintenance method and system. BACKGROUND

[0002] The elevator is a vertical transportation equipment composed of a traction system, a door system, a safety protection system and the like, and is widely used in residential buildings, commercial buildings, hospitals and the like, and is an indispensable traffic carrier for modern city operation. Its operation safety is directly related to the life safety of users, and maintenance is the key to ensuring the safe operation of the elevator and prolonging the service life, and is of great significance for reducing operation and maintenance costs.

[0003] In the traditional method, the elevator maintenance is mostly in the fixed cycle mode, that is, the maintenance personnel carries out on-site inspection, component lubrication and debugging at fixed intervals such as 15 days and every month according to the preset, and the fault diagnosis is highly dependent on the personal experience of the maintenance personnel, and the equipment state is judged by visual observation and manual detection.

[0004] However, this fixed cycle mode has a core problem: it cannot adjust the maintenance frequency according to the actual running load and state of the elevator. For the elevator with low load and good state, fixed cycle maintenance is easy to cause over-maintenance and increase unnecessary cost; for the elevator with high load operation and potential faults, it may miss the opportunity to check hidden troubles due to too long maintenance interval, and then may cause equipment damage or safety accidents, which is difficult to meet the needs of modern elevator precise maintenance. Therefore, there is an urgent need for an elevator maintenance method to solve the problem of lack of pertinence of fixed cycle maintenance. SUMMARY

[0005] In a first aspect, the present application provides an elevator maintenance method, which adopts the following technical solution:

[0006] An elevator maintenance method comprises:

[0007] obtaining real-time running data of the elevator;

[0008] judging whether the real-time running data is in a normal interval based on a preset normal running parameter range of the elevator;

[0009] generating a corresponding abnormal label if the real-time running data is out of the normal interval;

[0010] performing relevance analysis on the abnormal label to obtain a core feature set of elevator faults;

[0011] constructing an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm;

[0012] inputting the core feature set into the fault diagnosis and maintenance recommendation model to output a fault diagnosis result and a preliminary maintenance scheme;

[0013] determine whether the preliminary maintenance scheme meets a preset maintenance effect standard;

[0014] If the maintenance effect standard is met, a final maintenance plan is generated based on the preliminary maintenance scheme;

[0015] Based on the final maintenance plan and the elevator historical maintenance record, the time interval for the next maintenance of the elevator is predicted.

[0016] By adopting the above technical solution, first, real-time running data such as elevator running speed, car vibration value, traction machine current and door machine switch time are obtained through sensors, and these data are compared with the preset normal running parameter range of the elevator. If it exceeds the normal range, an abnormal label is generated. Then, the abnormal label is analyzed for relevance, the parameter set representing the essence of the fault is extracted to form a core feature set, and a fault diagnosis and maintenance recommendation model is constructed based on a deep learning algorithm. After the core feature set is input into the model, the fault diagnosis result and the preliminary maintenance scheme are output. Then, it is determined whether the preliminary maintenance scheme meets the preset maintenance effect standard. If it meets the standard, a final maintenance plan containing execution details is generated. Finally, the final maintenance plan and the elevator historical maintenance record are combined, the normal running time of the elevator after the same maintenance in the past is referred to, and the current usage frequency is combined to predict the time interval for the next maintenance.

[0017] Through real-time data monitoring and dynamic fault diagnosis, the traditional fixed-cycle maintenance mode is replaced, which can be targeted to match the actual running state and the degree of wear of the elevator, effectively solving the problem that the traditional mode lacks pertinence and easily leads to over-maintenance or hidden danger. At the same time, the deep learning model is used to improve the fault recognition accuracy, further improve the accuracy of elevator maintenance, reduce the risk of faults caused by improper maintenance, and ultimately ensure the safety of elevator operation.

[0018] Optionally, the relevance analysis of the abnormal labels includes:

[0019] extracting the abnormal type, abnormal occurrence position and abnormal duration corresponding to different abnormal labels;

[0020] calculating the coincidence degree of abnormal types, the proximity degree of abnormal occurrence positions and the correlation degree of abnormal durations between different abnormal labels;

[0021] respectively assigning weight values corresponding to the coincidence degree, proximity degree and correlation degree;

[0022] calculating the comprehensive correlation coefficient between different abnormal labels based on the coincidence degree, proximity degree, correlation degree and corresponding weight values;

[0023] determining whether the comprehensive correlation coefficient exceeds a preset correlation coefficient threshold;

[0024] If the correlation coefficient threshold is exceeded, the corresponding abnormal label is classified into the same failure correlation group;

[0025] The most frequent abnormal type, the most concentrated abnormal occurrence position and the longest abnormal duration in the failure correlation group are extracted as the failure key information;

[0026] Based on the failure key information, parameters representing the nature of the failure are screened out to form a core feature set of the elevator failure.

[0027] Optionally, the calculation of the coincidence degree of abnormal types, the proximity of abnormal occurrence positions and the correlation degree of abnormal durations between different abnormal labels comprises:

[0028] The total number of abnormalities of the abnormal types corresponding to the first abnormal label and the second abnormal label and the number of single abnormalities of the same type in the abnormal types are counted;

[0029] Based on the total number of abnormalities and the number of single abnormalities, the coincidence degree of abnormal types is obtained;

[0030] The actual distance of abnormal occurrence positions of the first abnormal label and the second abnormal label and the maximum component spacing of the elevator are obtained;

[0031] Based on the actual distance and the maximum component spacing of the elevator, the proximity of abnormal occurrence positions is obtained;

[0032] The intersection length and the union length of the abnormal duration of the first abnormal label and the second abnormal label are calculated;

[0033] Based on the intersection length and the union length, the correlation degree of abnormal durations is obtained.

[0034] Optionally, the judgment of whether the preliminary maintenance scheme meets the preset maintenance effect standard comprises:

[0035] From the elevator historical maintenance database, the historical maintenance records of the same failure type as the preliminary maintenance scheme are retrieved;

[0036] The recurrence rate after maintenance, the running stability score after maintenance and the maintenance cost data in the historical maintenance records are extracted;

[0037] The average value of the recurrence rate corresponding to the recurrence rate, the average value of the running stability score corresponding to the running stability score and the average value of the maintenance cost corresponding to the maintenance cost are calculated;

[0038] The average value of the recurrence rate is compared with the preset recurrence rate threshold, the average value of the running stability score is compared with the preset stability threshold, and the average value of the maintenance cost is compared with the preset cost threshold;

[0039] If the average recurrence rate is lower than the preset recurrence rate threshold, the average operation stability score is higher than the preset stability threshold, and the average maintenance cost is lower than the preset cost threshold, it is determined that the preliminary maintenance scheme meets the maintenance effect standard;

[0040] If any one does not meet, mark the preliminary maintenance scheme as a to-be-optimized scheme, and record the index that does not meet the threshold;

[0041] For the to-be-optimized scheme, based on the index that does not meet the threshold, adjust the component replacement frequency or maintenance process parameters in the maintenance step;

[0042] Recalculate the recurrence rate, stability score and cost data corresponding to the adjusted scheme, and compare them with the preset threshold again until all standards are met.

[0043] Optionally, the adjustment of the component replacement frequency or maintenance process parameters in the maintenance step based on the index that does not meet the threshold for the to-be-optimized scheme comprises:

[0044] If only the post-maintenance failure recurrence rate is higher than the preset recurrence rate threshold, the vulnerable component corresponding to the fault type in the historical maintenance record is checked;

[0045] The replacement frequency of the vulnerable component is increased, or the same type of component with higher durability is replaced;

[0046] If only the post-maintenance elevator operation stability score is lower than the preset stability threshold, the component calibration link involved in the maintenance step is analyzed;

[0047] Optimize the calibration parameters, extend the duration of the calibration operation, or increase the number of calibrations;

[0048] If only the average maintenance cost is higher than the preset cost threshold, the link with the highest cost proportion in the maintenance step is screened;

[0049] Under the premise of ensuring the maintenance effect, replace the maintenance material with lower cost and the same function, or simplify the non-critical maintenance step;

[0050] If multiple indexes do not meet, the parameter that has the greatest impact on the failure recurrence rate is adjusted first, and then the stability and cost related parameters are adjusted in turn.

[0051] Optionally, the prediction of the time interval at which the elevator needs to be maintained next time based on the final maintenance plan and the historical maintenance record of the elevator comprises:

[0052] From the historical maintenance record of the elevator, extract multiple groups of maintenance records for the same fault type in the near future and the current final maintenance plan;

[0053] Calculate the duration of the normal operation of the elevator after each maintenance to obtain multiple groups of normal operation duration data;

[0054] Removing extreme values in the normal runtime data, calculating the average value of the remaining data as the reference runtime;

[0055] Obtaining the runtime of the current elevator, subtracting the runtime from the reference runtime to obtain a preliminary estimated interval;

[0056] Determining whether the use frequency of the current elevator is higher than the average use frequency of the recent multiple maintenance groups;

[0057] If the use frequency is higher than the average use frequency, multiplying the preliminary estimated interval by a preset down-regulation coefficient to obtain the final next maintenance time interval;

[0058] If the use frequency is not higher than the average use frequency, multiplying the preliminary estimated interval by a preset up-regulation coefficient to obtain the final next maintenance time interval;

[0059] If the number of historical records cannot be extracted, estimating the next maintenance time interval based on the rated service life of the elevator and the used service life in proportion.

[0060] Optionally, the determination of whether the use frequency of the current elevator is higher than the average use frequency of the recent multiple maintenance groups comprises:

[0061] Counting the daily operation times in the fixed period of the current elevator, calculating the average value of the daily operation times as the current use frequency;

[0062] From the historical maintenance records, extracting the daily operation times in the fixed period after the recent multiple maintenance groups, and calculating the fixed period average operation times of each group respectively;

[0063] Adding the fixed period average operation times of the multiple groups and taking the average value to obtain the average use frequency of the recent multiple maintenance groups;

[0064] Comparing the current use frequency with the average use frequency of the recent multiple maintenance groups;

[0065] If the current use frequency is greater than the average use frequency of the recent multiple maintenance groups, it is determined that the use frequency of the current elevator is higher than the average use frequency of the recent multiple maintenance groups;

[0066] If the current use frequency is less than or equal to the average use frequency of the recent multiple maintenance groups, it is determined that the use frequency of the current elevator is not higher than the average use frequency of the recent multiple maintenance groups.

[0067] Optionally, after the real-time operation data of the elevator is obtained, the method further comprises:

[0068] Preprocessing the real-time operation data, comparing the preprocessed real-time operation data with the standard operation data of the elevator when it is shipped, and calculating the data deviation value;

[0069] determining whether the data deviation value exceeds a preset deviation warning threshold value;

[0070] If the deviation warning threshold value is not exceeded, the real-time operation data of the elevator is continuously monitored;

[0071] If the deviation warning threshold value is exceeded, a data deviation warning information is generated and sent to a maintenance personnel terminal;

[0072] After receiving the warning information, the maintenance personnel performs a preliminary inspection on the elevator within a preset time and records the inspection results;

[0073] If the inspection results show no abnormalities, the deviation warning threshold value is updated;

[0074] If the inspection results show potential faults, a fault diagnosis process is started in advance.

[0075] Optionally, the elevator fault diagnosis and maintenance recommendation model is constructed based on a deep learning algorithm, and the method comprises the following steps:

[0076] Collecting operation data samples corresponding to different fault types of the elevator;

[0077] Dividing the sample data into a training set and a test set according to a preset ratio;

[0078] Selecting a preset deep learning model structure, inputting the training set data, setting a preset learning rate and iteration number for model training;

[0079] After each iteration, the model diagnosis accuracy is verified using the test set data, and if the accuracy does not improve continuously for a preset number of iterations, the training is stopped;

[0080] Saving the trained model and calculating the average diagnosis accuracy of the model on the test set;

[0081] If the average diagnosis accuracy is higher than a preset accuracy threshold value, the model is used as the elevator fault diagnosis and maintenance recommendation model;

[0082] If the average diagnosis accuracy is lower than the preset accuracy threshold value, the number of sample data is increased, the model learning rate is adjusted, and the training is restarted until the accuracy meets the standard.

[0083] In a second aspect, the application also discloses an elevator maintenance system, which adopts the following technical scheme:

[0084] An elevator maintenance system comprises:

[0085] A first acquisition module is configured to acquire real-time operation data of an elevator;

[0086] The first judging module is configured to judge whether the real-time operation data is in a normal range based on a preset normal operation parameter range of the elevator;

[0087] The label generating module is configured to generate a corresponding abnormal label if the real-time operation data is out of the normal range;

[0088] The second acquiring module is configured to perform correlation analysis on the abnormal label and acquire a core feature set of elevator faults;

[0089] The model constructing module is configured to construct an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm;

[0090] The scheme outputting module is configured to input the core feature set into the fault diagnosis and maintenance recommendation model and output a fault diagnosis result and a preliminary maintenance scheme;

[0091] The second judging module is configured to judge whether the preliminary maintenance scheme meets a preset maintenance effect standard;

[0092] The plan generating module is configured to generate a final maintenance plan based on the preliminary maintenance scheme if the maintenance effect standard is met;

[0093] The interval predicting module is configured to predict a time interval at which the elevator needs to be maintained next time based on the final maintenance plan and a historical maintenance record of the elevator.

[0094] In summary, the present application has the following beneficial technical effects:

[0095] Through real-time data monitoring and dynamic fault diagnosis, the traditional fixed-cycle maintenance mode is replaced, the actual operation state and the degree of wear of the elevator can be targetedly matched, the problems that the traditional mode lacks pertinence and is prone to cause excessive maintenance or hidden danger omission are effectively solved, the fault recognition precision is improved with the aid of the deep learning model, the elevator maintenance precision is further improved, the fault risk caused by improper maintenance is reduced, and finally the elevator operation safety is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0096] Figure 1 FIG. 1 is a main flowchart of an elevator maintenance method according to an embodiment of the present application;

[0097] Figure 2 FIG. 4 is a flowchart of acquiring a core feature set of elevator faults;

[0098] Figure 3 FIG. 6 is a flowchart of calculating abnormal type coincidence degrees, abnormal occurrence position proximity degrees and abnormal duration correlation degrees among different abnormal labels;

[0099] Figure 4is a flow chart for judging whether the preliminary maintenance scheme meets the preset maintenance effect standard;

[0100] Figure 5 is a flow chart for adjusting the component replacement frequency or maintenance process parameters in the maintenance step based on the index that does not meet the threshold value;

[0101] Figure 6 is a flow chart for predicting the next maintenance time interval of the elevator based on the final maintenance plan and the historical maintenance records of the elevator;

[0102] Figure 7 is a flow chart for judging whether the current elevator usage frequency is higher than the average usage frequency of multiple recent maintenance groups;

[0103] Figure 8 is a flow chart for pre-processing, deviation warning and subsequent processing of real-time running data of the elevator;

[0104] Figure 9 is a flow chart for constructing an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm;

[0105] Figure 10 is a module diagram of an elevator maintenance system according to an embodiment of the present application.

[0106] Explanation of reference signs:

[0107] 1, first acquisition module; 2, first judgment module; 3, label generation module; 4, second acquisition module; 5, model construction module; 6, scheme output module; 7, second judgment module; 8, plan generation module; 9, interval prediction module. DETAILED DESCRIPTION

[0108] In a first aspect, the present application discloses an elevator maintenance method.

[0109] Reference Figure 1 An elevator maintenance method includes steps S101 to S109:

[0110] Step S101: Obtain real-time running data of the elevator.

[0111] Specifically, in the present embodiment, the real-time running data is collected by sensors installed on key components of the elevator, including running speed, car vibration value, traction machine current, and door machine switch duration, etc.

[0112] Step S102: Based on the preset normal running parameter range of the elevator, judge whether the real-time running data is in the normal interval.

[0113] Specifically, in this embodiment, the preset normal operation parameter range of the elevator is determined based on the technical manual of the elevator, for example, for an elevator with a rated speed of 2 m / s, the normal range of the running speed is 1.9-2.1 m / s (±5% fluctuation); the car vibration value is less than or equal to 0.15 m / s (to avoid obvious discomfort of passengers); the current of the traction machine is less than or equal to 110% of the rated current; and the door machine is turned on and off for 2-4 seconds (too short is easy to pinch people, and too long affects efficiency). Whether the data is within the range is compared in real time to determine whether it is abnormal. 2

[0114] Step S103: If the real-time operation data is out of the normal range, a corresponding abnormal label is generated.

[0115] Specifically, in this embodiment, the abnormal label includes an abnormal type (such as speed overage, excessive vibration, current overload, and door machine jamming), an abnormal occurrence position (accurate to a specific component, such as a phase winding of the traction machine A or a 3-layer door machine motor), and an abnormal duration (from the occurrence of the abnormality to the recovery of the normality, such as 10:05:20-10:05:45, for a total of 25 seconds). The label is automatically associated with the original data of the corresponding time after being generated, which is convenient for tracing back.

[0116] Step S104: Correlation analysis is performed on the abnormal label to obtain a core feature set of the elevator fault.

[0117] Specifically, the correlation analysis mines the potential relationship between the abnormal labels through an algorithm, for example, the door machine jamming and the current overload may be caused by the aging of the door machine motor. In this embodiment, the core feature set is a key parameter extracted from the associated abnormality, such as a door machine resistance coefficient, a motor temperature, and a current fluctuation amplitude.

[0118] Step S105: Based on a deep learning algorithm, an elevator fault diagnosis and maintenance recommendation model is constructed.

[0119] Specifically, a combination model of a convolutional neural network (CNN) and a long short-term memory network (LSTM) in deep learning is used. The CNN is used to extract local features (such as the peak value of the vibration waveform) from the real-time operation data, and the LSTM is used to analyze the time sequence features (such as the change trend of the current over time). In this embodiment, the model input is the core feature set, and the output is the fault type and the maintenance suggestion.

[0120] Step S106: The core feature set is input into the fault diagnosis and maintenance recommendation model, and the fault diagnosis result and the preliminary maintenance scheme are output.

[0121] ​Specifically, the core feature set is input into the built model, the model analyzes through internal algorithms, and outputs the fault diagnosis result; at the same time, the model combines the solutions of the same type of fault in the historical maintenance data to generate a preliminary maintenance plan, and the plan content includes the parts to be replaced, the maintenance steps, the tools required and the estimated time length; in this embodiment, the fault diagnosis result refers to the determination result of the model on the elevator fault type, fault position and fault severity, which provides a clear direction for subsequent maintenance, and the fault level is usually divided into slight (does not affect normal operation, needs to be observed regularly), medium (affects operation efficiency, needs to be maintained as soon as possible), and serious (exists safety hazard, needs to be maintained immediately); the preliminary maintenance plan refers to the maintenance plan generated based on the fault diagnosis result, which has not been verified for effectiveness, and contains the core elements required for maintenance, but there may be optimization space.

[0122] Step S107: Determine whether the preliminary maintenance plan meets the preset maintenance effect standard.

[0123] Specifically, from the elevator historical maintenance database, all historical maintenance records (not less than 10) for the same fault type in the past 2 years are called; the fault recurrence rate after maintenance, the elevator operation stability score (full score 10 points, converted from the running parameter fluctuation collected by the sensor, the smaller the fluctuation, the higher the score) and the maintenance cost data (including component procurement cost, labor service cost) of each record are extracted; the average value of the three indexes is calculated; the average value is compared with the preset threshold value (preset recurrence rate threshold 10%, stability score threshold 8 points, cost threshold 1000 yuan): if all three meet, it is determined that the scheme meets the standard; if any one does not meet, it is marked as a to-be-optimized scheme, and the unmet index is recorded; for the to-be-optimized scheme, adjust the component brand (such as replacing it with a domestic bearing with higher cost performance) or simplify non-critical steps (such as reducing the number of subsequent reviews), recalculate the indexes and compare them until all thresholds are met. In this embodiment, the maintenance effect standard refers to the index system preset to measure the effectiveness and economy of the maintenance plan, which usually includes three core indexes: fault recurrence rate, operation stability, and maintenance cost, to ensure that the elevator can run stably for a long time after maintenance and the cost is controllable.

[0124] Step S108: If the maintenance effect standard is met, generate a final maintenance plan based on the preliminary maintenance plan.

[0125] Specifically, based on the preliminary maintenance plan, specific execution information is supplemented, including determining the maintenance execution date, assigning maintenance personnel, specifying safety protection measures, and adding emergency plans; at the same time, the component model, tool list, and step sequence in the plan are finally confirmed to form a final maintenance plan that can be directly executed, and is synchronized to the maintenance personnel terminal and the elevator management system; in the embodiment, the final maintenance plan refers to a maintenance plan that can be directly implemented after the effect verification and the supplement of complete execution information, which is more operable and safer than the preliminary plan, and is the basis for the maintenance personnel to execute.

[0126] Step S109: predicting the time interval for the next maintenance of the elevator based on the final maintenance plan and the historical maintenance records of the elevator.

[0127] Specifically, in the embodiment, the time interval refers to the time interval from the current time node to the next maintenance.

[0128] The elevator maintenance method provided in the embodiment first acquires real-time running data such as the running speed of the elevator, the car vibration value, the traction machine current, and the door machine switch time length through the sensor, compares these data with the preset normal running parameter range of the elevator, and generates an abnormal label if it exceeds the normal range; then, the correlation analysis is performed on the abnormal label, the parameter set representing the nature of the fault is extracted to form a core feature set, and a fault diagnosis and maintenance recommendation model is constructed based on a deep learning algorithm, the core feature set is input into the model, and the fault diagnosis result and the preliminary maintenance plan are output; then, it is judged whether the preliminary maintenance plan meets the preset maintenance effect standard, if it meets, a final maintenance plan containing execution details is generated, and finally, the next maintenance time interval is predicted by combining the final maintenance plan with the historical maintenance records of the elevator, referring to the normal running time of the elevator after the same maintenance in the past, and combining the current use frequency.

[0129] Through real-time data monitoring and dynamic fault diagnosis, the traditional fixed cycle maintenance mode is replaced, the actual running state and the degree of wear of the elevator can be targeted, the problem that the traditional mode lacks pertinence and easily leads to excessive maintenance or hidden danger is effectively solved, the deep learning model is used to improve the fault recognition accuracy, the accuracy of elevator maintenance is further improved, the fault risk caused by improper maintenance is reduced, and finally the safety of the elevator running is ensured.

[0130] Reference Figure 2 In one of the embodiments of the present embodiment, step S104 performs correlation analysis on the abnormal label to obtain the core feature set of the elevator fault, including steps S201 to S208:

[0131] Step S201: extracting the abnormal type, abnormal position, and abnormal duration corresponding to different abnormal labels.

[0132] Specifically, in the embodiment, all abnormal labels generated within the last 24 hours are filtered out from the abnormal label library stored in the system (e.g., label 1: "Abnormal type: car vibration exceeds the standard; location: top of the car; duration: 15 minutes"; label 2: "Abnormal type: excessive noise of the traction machine; location: traction machine room A; duration: 20 minutes"), and are arranged in a table in the format of "label ID-Abnormal type-occurrence location-duration" to ensure that no key information of each label is missed.

[0133] Step S202: Calculate the coincidence degree of different abnormal labels in terms of abnormal type, the proximity degree of abnormal occurrence location, and the correlation degree of abnormal duration.

[0134] Specifically, in the embodiment, the coincidence degree refers to an index for measuring the overlapping degree of different abnormal labels in terms of abnormal type, the higher the coincidence degree, the more likely that the abnormality is caused by the same reason; the proximity degree refers to an index for measuring the spatial distance between abnormal occurrence locations, the higher the proximity degree (the closer the distance), the more likely that the abnormality is physically related; and the correlation degree refers to an index for measuring the overlapping degree of abnormal duration, the higher the correlation degree, the more likely that the abnormality occurs synchronously, and there is a cause-and-effect relationship.

[0135] Taking label 1 (type: car vibration exceeds the standard, door machine response delay) and label 2 (type: car vibration exceeds the standard, traction machine current fluctuation) as examples, the total number of abnormal types of the two labels (4 items) and the number of the same types (1 item "car vibration exceeds the standard") are counted, and the coincidence degree = the number of the same types / total number = 1 / 4 = 25%.

[0136] The actual distance between the location of label 1 (top of the car) and the location of label 2 (traction machine room A) is obtained (e.g., 5 meters), and the maximum distance between components of the elevator (e.g., the total height of the elevator shaft is 18 meters), and the proximity degree = 1-(actual distance / maximum distance between components) = 1-(5 / 18) ≈ 72.2%.

[0137] The duration of label 1 is 14:00-14:15 (15 minutes), and the duration of label 2 is 14:10-14:25 (15 minutes), the intersection duration is 14:10-14:15 (5 minutes), and the union duration is 14:00-14:25 (25 minutes), and the correlation degree = intersection duration / union duration = 5 / 25 = 20%.

[0138] Step S203: Assign weight values corresponding to the coincidence degree, the proximity degree, and the correlation degree, respectively.

[0139] Specifically, in this embodiment, the weight values are preset based on the influence degree of the elevator component failure, the abnormal type coincidence degree weight is 0.4 (because the failure type directly reflects the nature of the failure, the relevance is the strongest), the abnormal occurrence location proximity weight is 0.3 (the location correlation is an important reference for failure positioning, but some failures may have a "remote effect", the weight is second), and the abnormal duration correlation weight is 0.3 (time correlation can assist in determining causality, but single time overlap may be a coincidence, and the weight is consistent with the proximity).

[0140] Step S204: Based on the coincidence degree, proximity, correlation degree, and corresponding weight values, the comprehensive correlation coefficient between different abnormal labels is calculated.

[0141] Specifically, the comprehensive correlation coefficient is a comprehensive index integrating the coincidence degree, proximity, and correlation degree, and the value range is 0-100%. The higher the value, the stronger the relevance of the two abnormal labels, and the more likely they belong to the same failure-induced abnormality. In this embodiment, the comprehensive correlation coefficient is calculated using a weighted summation formula, which is: comprehensive correlation coefficient = (coincidence degree x coincidence degree weight) + (proximity x proximity weight) + (correlation degree x correlation degree weight). Taking the data in S202 as an example, the comprehensive correlation coefficient = (25% x 0.4) + (72.2% x 0.3) + (20% x 0.3) = 10% + 21.66% + 6% = 37.66%.

[0142] Step S205: Determine whether the comprehensive correlation coefficient exceeds the preset correlation coefficient threshold.

[0143] Specifically, the correlation coefficient threshold is a critical value for dividing the strength of the relevance of abnormal labels. In this embodiment, the correlation coefficient threshold can be set to 60%.

[0144] Step S206: If the correlation coefficient threshold is exceeded, the corresponding abnormal labels are grouped into the same failure correlation group.

[0145] Specifically, in this embodiment, the failure correlation group refers to a set composed of abnormal labels with strong relevance. Abnormalities within the same group are usually caused by the same failure root cause (such as bearing wear of the traction machine, which may cause three abnormal labels of high current, excessive vibration, and excessive noise, forming a group). Grouping can reduce the complexity of failure diagnosis. For example, if the comprehensive correlation coefficient of label A and label B is 65% (> 60%), and the comprehensive correlation coefficient of label B and label C is 72% (> 60%), then labels A, B, and C are grouped into the same failure correlation group. If the comprehensive correlation coefficient of label D and any label in the group is 55% (< 60%), then label D is listed separately as another group.

[0146] Step S207: Extract the abnormal type with the highest occurrence frequency, the most concentrated abnormal occurrence position, and the longest abnormal duration in the fault association group as the fault key information.

[0147] Specifically, the fault key information refers to the core data that best reflects the fault characteristics extracted from the fault association group. In this embodiment, the fault key information includes the abnormal type with the highest occurrence frequency, the most concentrated abnormal occurrence position, and the longest abnormal duration in the fault association group.

[0148] Step S208: Based on the fault key information, filter out parameters that can represent the nature of the fault to form a core feature set of the elevator fault.

[0149] Specifically, in this embodiment, the parameters representing the nature of the fault refer to the operating parameters that can directly reflect the root cause of the fault, which are distinguished from indirect influence parameters (such as door machine timeout may cause passenger waiting time to increase, but this parameter does not reflect the nature of the fault, so it is not included). The core feature set must be composed of essential parameters to ensure the accuracy of the data input into the model.

[0150] The elevator maintenance method provided by the present embodiment can effectively mine the potential association between scattered abnormal labels, avoid fragmented analysis of multiple abnormalities caused by the same fault, accurately locate the root cause of the fault, and reduce the fault misjudgment or omission caused by isolated judgment of abnormalities. At the same time, by extracting fault key information to form a core feature set, redundant data interference can be removed, providing accurate and effective data basis for subsequent deep learning model input, further improving the accuracy and efficiency of elevator fault diagnosis, providing a reliable basis for subsequent development of targeted maintenance solutions, and helping to solve the problem of poor maintenance effect caused by fuzzy fault positioning in traditional maintenance.

[0151] Reference Figure 3 In one of the embodiments of the present embodiment, step S202 of calculating the coincidence degree of abnormal types, the proximity degree of abnormal occurrence positions, and the correlation degree of abnormal durations between different abnormal labels includes steps S301 to S306:

[0152] Step S301: Count the total number of abnormalities of the abnormal types corresponding to the first abnormal label and the second abnormal label and the number of single abnormalities of the same type in the abnormal types.

[0153] Specifically, in this embodiment, the total number of abnormalities refers to the sum of the number of abnormal types contained in the two abnormal labels; the number of single abnormalities refers to the number of completely consistent abnormal types in the two labels. Complete consistency must meet the industry-unified abnormal type naming standard (such as car vertical vibration exceeding the standard and car vibration exceeding the standard, which are not the same type, and must be accurate to the vibration direction).

[0154] Step S302: Obtain the abnormal type coincidence degree based on the total number of abnormalities and the number of single abnormalities.

[0155] Specifically, the abnormal type coincidence degree reflects the degree of overlap of the two abnormal labels in the fault manifestation type. The higher the coincidence degree, the more similar the external manifestations of the two abnormalities. In this embodiment, the abnormal type coincidence degree is calculated according to the formula: abnormal type coincidence degree = number of single abnormalities / total number of abnormalities x 100%. For example, if the two labels have the same abnormal type (e.g., both are "high current" and "excessive temperature"), the coincidence degree is 2 / 4 x 100% = 50%. If they are completely different (e.g., label X is "vibration" and label Y is "door machine timeout"), the coincidence degree is 0%.

[0156] Step S303: Obtain the actual distance between the abnormal occurrence positions of the first abnormal label and the second abnormal label and the maximum component distance of the elevator.

[0157] Specifically, in this embodiment, the actual distance between the abnormal occurrence positions refers to the straight-line distance between the two abnormal positions in the elevator shaft space, rather than the polyline distance along the components. It needs to be calculated through three-dimensional coordinates, and the precision needs to be controlled within ±0.1 m. The maximum component distance of the elevator refers to the maximum straight-line distance between any two key components (such as the traction machine, control cabinet, car, buffer, etc.) in the overall structure of the elevator. It is the reference value for calculating the proximity, and the maximum component distance of different types of elevators is different (e.g., about 25 m for a 10-story residential elevator and about 80 m for a 30-story office building elevator). It needs to be input into the system in advance.

[0158] Step S304: Obtain the abnormal occurrence position proximity based on the actual distance and the maximum component distance of the elevator.

[0159] Specifically, the abnormal occurrence position proximity refers to the relative closeness of the spatial distance between the two abnormal positions, rather than the absolute distance. By comparing with the maximum component distance, the influence of the size difference of different elevators can be eliminated (e.g., a distance of 0.3 m has a proximity of 98.8% in a 25 m elevator and a proximity of 99.6% in an 80 m elevator, both of which belong to the extremely close category). In this embodiment, the calculation formula of the abnormal occurrence position proximity is: abnormal occurrence position proximity = (1-actual distance / maximum component distance of elevator) x 100%.

[0160] Step S305: Calculate the intersection time length and union time length of the abnormal duration of the first abnormal label and the second abnormal label.

[0161] Specifically, in this embodiment, the intersection time length refers to the time length of the overlapping part of the two abnormal durations. The longer the overlapping time, the higher the possibility of synchronous occurrence of the two abnormalities, and there may be a causal relationship (e.g., after the occurrence of A abnormality, B abnormality immediately appears synchronously). The union time length refers to the total time length covered by the two abnormal durations (from the earliest start time to the latest end time).

[0162] Step S306: Based on the intersection duration and the union duration, an abnormal duration correlation degree is obtained.

[0163] Specifically, the abnormal duration correlation degree reflects the synchronization degree of two abnormalities in the time dimension. The higher the correlation degree, the closer the occurrence and end time of the two abnormalities, which further supports the judgment of the same origin fault (such as abnormalities caused by the same fault, which usually appear and disappear at similar times). In this embodiment, the calculation formula of the abnormal duration correlation degree is: abnormal duration correlation degree = (intersection duration / union duration) x 100%.

[0164] The elevator maintenance method provided by the embodiment can accurately measure the correlation strength of different abnormalities in the type, position, and time dimension by calculating the three types of correlation indicators through a clear formula, ensuring that the calculation result of the subsequent comprehensive correlation coefficient is objective and reliable, thereby laying a solid foundation for accurately dividing the fault correlation group and extracting the core feature set, effectively reducing the fault misjudgment caused by inaccurate correlation analysis, and further improving the refinement degree of elevator fault diagnosis, and providing a more accurate basis for targeted maintenance.

[0165] Reference Figure 4 In one of the embodiments of the present embodiment, step S107 of judging whether the preliminary maintenance scheme meets the preset maintenance effect standard includes steps S401 to S408:

[0166] Step S401: From the elevator historical maintenance database, the historical maintenance records of the same fault type as the preliminary maintenance scheme are retrieved.

[0167] Specifically, the same fault type refers to the consistency of the fault position, fault cause, and fault degree, such as light wear of the traction machine deep groove ball bearing and moderate wear of the traction machine needle bearing, which are not the same type, so as to avoid the reduction of the reference value of historical data due to fault differences; the historical maintenance database refers to a structured database that stores the maintenance data of the entire life cycle of the elevator, supports multi-dimensional retrieval according to fault types, components, time, etc., and the data needs to be updated regularly (within 24 hours after maintenance is completed) to ensure timeliness; in this embodiment, the fault type corresponding to the preliminary maintenance scheme (such as wear of the traction machine deep groove ball bearing) is first determined, and the "fault type keyword + component name" combination is searched in the historical maintenance database (search keywords: "wear of the traction machine bearing + deep groove ball bearing"), all matching records within the last 3 years are retrieved, and the number of records is required to be not less than 10 (if there are less than 10, the search range can be expanded to the historical records of the same brand and model of elevator), each record needs to include the maintenance date, the maintenance scheme, the operation data (such as whether the fault recurs, the stability score) within 1-3 months after maintenance, and the maintenance cost details.

[0168] Step S402: Extract the post-maintenance failure recurrence rate, post-maintenance elevator operation stability score and maintenance cost data in the historical maintenance record.

[0169] Specifically, the post-maintenance failure recurrence rate is an index for measuring the long-term effectiveness of the maintenance plan. The lower the recurrence rate, the more complete the solution to the fault, effectively extending the service life of the component. In this embodiment, for a single record, the number of recurrences of the same fault within 3 months after maintenance is counted (for example, 1 recurrence in 1 month after maintenance, and 1 recurrence within 3 months), and the recurrence rate is calculated as (recurrence number / 3 months) x 100%. The post-maintenance elevator operation stability score is a comprehensive index for measuring the overall operation state of the elevator after maintenance. The higher the score, the better the overall performance of the elevator, indicating that the maintenance not only solved the target fault but also did not cause negative effects on other components. In this embodiment, according to the elevator operation data (such as vibration value, current fluctuation, door machine switching time) within 1 month after maintenance, the score is calculated according to the industry standard "Elevator Operation Stability Evaluation Specification" (full score 10 points, vibration value ≤0.1 m / s 2 2 points, current fluctuation ≤5% 2 points, door machine switching time 3-6 seconds 2 points, no fault shutdown 4 points, total score 8 points, i.e. stability score 8 points). The maintenance cost data includes all costs required to complete the maintenance, including direct and indirect costs. In this embodiment, the total cost of a single record is extracted, including component procurement cost (such as deep groove ball bearing 800 yuan), labor service cost (2 people x 2 hours x 150 yuan / hour = 600 yuan), and auxiliary material cost (lubricating oil 50 yuan), total cost = 800 + 600 + 50 = 1450 yuan.

[0170] Step S403: Calculate the average recurrence rate corresponding to the recurrence rate, the average operation stability score corresponding to the operation stability score, and the average maintenance cost corresponding to the maintenance cost.

[0171] Specifically, in this embodiment, the average recurrence rate is calculated by selecting all valid records from the elevator historical maintenance database that are related to the same fault type as the current preliminary maintenance plan, calculating the post-maintenance failure recurrence rate for each record, and then taking the arithmetic mean of these single recurrence rates. The average operation stability score is calculated by extracting the elevator operation data (such as vibration value, current fluctuation, door machine switching time) within 1 month after maintenance for each record of the historical maintenance record related to the same fault type, converting it into an operation stability score according to the industry standard (full score 10 points), and then taking the arithmetic mean of all single scores. The average maintenance cost is calculated by extracting the total cost (including component procurement cost, labor service cost, and auxiliary material cost) of each record of the historical maintenance record related to the same fault type, and then taking the arithmetic mean of all single total costs.

[0172] Step S404: Compare the average recurrence rate with the preset recurrence rate threshold, compare the average running stability score with the preset stability threshold, and compare the average maintenance cost with the preset cost threshold.

[0173] Specifically, in this embodiment, the preset recurrence rate threshold refers to the maximum probability value of allowing the fault to occur again after maintenance, which is determined by elevator safety specifications and operation and maintenance experience (for example, the threshold value of the fault related to the safety circuit is set to 5%, and the threshold value of the ordinary mechanical fault is set to 10%), and a value lower than the threshold value indicates that the maintenance effect is durable; the preset stability threshold refers to the minimum qualified score of the elevator running parameters after maintenance, which is converted based on data such as vibration and speed fluctuation (a full score of 10 points), and a value higher than the threshold value indicates that the elevator running state is good; and the preset cost threshold refers to the maximum cost upper limit allowed by the maintenance scheme, which comprehensively considers the market price of components, labor cost and budget of the elevator using unit, and a value lower than the threshold value indicates that the economic efficiency of the scheme meets the standard.

[0174] Step S405: If the average recurrence rate is lower than the preset recurrence rate threshold, the average running stability score is higher than the preset stability threshold, and the average maintenance cost is lower than the preset cost threshold, it is determined that the preliminary maintenance scheme meets the maintenance effect standard.

[0175] Step S406: If any one of them does not meet the standard, mark the preliminary maintenance scheme as a to-be-optimized scheme, and record the index that does not meet the threshold value.

[0176] Specifically, in this embodiment, the to-be-optimized scheme refers to the preliminary maintenance scheme that needs to be adjusted and improved because one or more indicators do not meet the standard.

[0177] Step S407: For the to-be-optimized scheme, adjust the component replacement frequency or maintenance process parameters in the maintenance steps based on the index that does not meet the threshold value.

[0178] Step S408: Recalculate the recurrence rate, stability score and cost data corresponding to the adjusted scheme, and compare them with the preset threshold value again until all the standards are met.

[0179] Specifically, in this embodiment, the adjusted maintenance scheme (such as replacing the bearing with high durability) is input into the historical maintenance database, the historical records of similar adjustment schemes (such as the last three maintenance cases using bearings with high durability) are matched, the average values of the three indicators (such as the recurrence rate is reduced to 9%, the cost is increased to 1100 yuan, and the stability score is maintained at 8.5 points) are recalculated; and the threshold value is compared again, if there is still an unqualified item (such as no suitable historical record leading to data missing), the adjustment process of S407 (such as further optimizing the bearing brand) is repeated, until all the three indicators meet the threshold value. Finally, the qualified scheme is marked as optimization completed and is transferred to step S108.

[0180] The elevator maintenance method provided by the embodiment quantitatively evaluates the long-term effectiveness, stability and economy of the scheme through historical data, ensures that the final maintenance scheme can effectively reduce the risk of fault recurrence, guarantee the stable operation of the elevator, and control the maintenance cost, solves the problem that the traditional maintenance scheme is dependent on experience and the effect is difficult to guarantee, and further improves the reliability and rationality of the maintenance plan, and provides strong support for the safe and efficient operation of the elevator.

[0181] With reference to Figure 5 In one of the embodiments of the present embodiment, step S407 adjusts the component replacement frequency or maintenance process parameters in the maintenance step based on the indicators that do not meet the threshold for the to-be-optimized scheme, including steps S501 to S507:

[0182] Step S501: If only the post-maintenance fault recurrence rate is higher than the preset recurrence rate threshold, check the vulnerable components corresponding to the fault type in the historical maintenance record.

[0183] Specifically, the vulnerable component refers to a component that is more prone to wear and tear and leads to fault recurrence due to frequent stress, environmental impact or material characteristics in a specific fault type, and is an object that needs to be focused on to reduce the recurrence rate; in the present embodiment, all records of the same fault type (such as “door machine jam”) are screened from the historical maintenance database, the most commonly damaged components (such as “door machine belt” appearing in 80% of the recurrence cases) are counted when the fault recurs, and it is determined that “door machine belt” is a vulnerable component, and the average service life (such as 6 months) and common damage reasons (such as material aging and insufficient tension) of the component are recorded to provide a direction for adjustment.

[0184] Step S502: Increase the replacement frequency of the vulnerable component, or replace it with a higher durability of the same type of component.

[0185] Specifically, for the determined vulnerable component (such as the door machine belt), if the original replacement frequency is 6 months / time, it can be adjusted to 4 months / time (increased frequency); or the 6-month replacement cycle is retained, but the ordinary rubber belt is replaced with a wear-resistant polyurethane belt (improved durability).

[0186] Step S503: If only the post-maintenance elevator operation stability score is lower than the preset stability threshold, analyze the component calibration link involved in the maintenance step.

[0187] Specifically, the component calibration link refers to adjusting the component parameters (such as position, angle, tension) to make the elevator operation meet the standard requirements of the operation process (such as the leveling accuracy needs to be controlled within ±5mm); in this embodiment, the calibration steps in the preliminary maintenance scheme (such as "car leveling accuracy calibration" and "door machine opening and closing position calibration") are called, combined with historical cases with low scores, to analyze the potential problems in the calibration link: such as leveling calibration is only performed once (standard requires twice), or the calibration tool precision is insufficient (such as using a common level instead of a laser calibration instrument).

[0188] Step S504: Optimize calibration parameters, extend the duration of calibration operation, or increase the number of calibrations.

[0189] Specifically, in this embodiment, if the parameter setting is incorrect, the leveling calibration parameters are tightened from "±10mm" to "±5mm" (optimized parameters); if the calibration time is too short (originally 5 minutes), it is extended to 10 minutes (extended time); if there is an error in single calibration, the number of calibrations is increased from 1 to 2 (30 minutes of rest after the first calibration for re-calibration); after adjustment, the stability improvement effect needs to be verified through simulation operation (such as continuous operation for 100 times, and the leveling deviation value is counted).

[0190] Step S505: If only the average maintenance cost is higher than the preset cost threshold, the link with the highest cost proportion in the maintenance step is selected.

[0191] Specifically, in this embodiment, the cost composition of the disassembled preliminary maintenance scheme is analyzed (such as component procurement accounting for 70%, labor accounting for 20%, and auxiliary materials accounting for 10%), and the link with the highest cost proportion is located (such as "imported traction machine bearing procurement" accounting for 60% of the total cost); the cost composition of the link is analyzed (such as customs duty accounting for 20%, and brand premium accounting for 30%), and the optimization space is identified (such as whether there is a same performance domestic replacement component).

[0192] Step S506: Replace the maintenance material with lower cost and the same function, or simplify the non-critical maintenance step under the premise of ensuring the maintenance effect.

[0193] Specifically, in this embodiment, for high-cost links (such as imported bearings), on the premise of ensuring the service life and performance parameters, replace them with domestic bearings of the same model with a 30% lower price (after testing, the service life is the same as that of imported parts); if the labor cost is too high (such as including the "comprehensive disassembly and inspection" step), it can be simplified to "key part disassembly and inspection" (non-key parts are changed to visual inspection), reducing 2 hours of labor cost; after adjustment, the total cost needs to be recalculated to ensure that it does not exceed the preset threshold, and the maintenance effect is verified through historical data to ensure that there is no decline (such as the failure recurrence rate of domestic bearings is consistent with that of imported parts).

[0194] Step S507: If the multiple indicators do not meet the requirements, the parameter that has the greatest impact on the failure recurrence rate is adjusted first, and then the stability and cost-related parameters are adjusted in turn.

[0195] Specifically, in the embodiment, when the recurrence rate, stability, and cost do not meet the requirements, the adjustment priority is determined by the weight analysis method: the weight of the recurrence rate is set to 0.5 (directly affecting safety), the weight of the stability is 0.3 (affecting the operation experience), and the weight of the cost is 0.2 (affecting the economy). For example, a certain scheme has a recurrence rate of 18% (exceeding the threshold of 8%), a stability of 7 points (lower than the threshold of 1 point), and a cost of 1500 yuan (exceeding the threshold of 300 yuan). The recurrence rate is adjusted first (for example, replacing the vulnerable parts), so that the recurrence rate is reduced to 9%. Then, the calibration step is optimized to improve the stability to 8.2 points. Finally, the low-cost material is replaced to reduce the cost to 1100 yuan. After each adjustment, the three indicators are re-evaluated until they all meet the requirements.

[0196] The elevator maintenance method provided by the embodiment can target a single indicator defect and determine the adjustment priority when multiple indicators conflict. This can improve the long-term effectiveness, stability, and economy of the maintenance scheme, balance the indicators while ensuring the core effect (such as reducing the recurrence rate), effectively compensate for the deficiencies of the initial scheme, ensure that the final maintenance plan addresses the fault while considering operation safety and cost control, and further improve the scientificity and practicality of elevator maintenance.

[0197] Reference Figure 6 In one of the embodiments of the present embodiment, step S109 predicts the time interval at which the elevator needs to be maintained next based on the final maintenance plan and the historical maintenance records of the elevator, including steps S601 to S608:

[0198] Step S601: From the historical maintenance records of the elevator, extract multiple sets of maintenance records in the near future and the current final maintenance plan for the same fault type.

[0199] Specifically, in the present embodiment, taking the "door machine jam" fault addressed by the current final maintenance plan as an example, the maintenance records in the past two years with a similarity of ≥80% (such as both including the steps of "door machine belt replacement + guide rail lubrication") to the current plan are selected from the historical maintenance database. The number of extracted sets is required to be ≥3 (to ensure data representativeness). If there are less than 3 sets, the time range is expanded to 3 years or the similarity is reduced to 70%. If there are still less than 3 sets, the backup scheme is executed.

[0200] Step S602: Calculate the length of time during which the elevator operates normally after each maintenance, and obtain multiple sets of normal operation time data.

[0201] Specifically, in the embodiment, for each group of maintenance records, the interval between the maintenance completion time and the time when the same fault occurs again is calculated. If the elevator does not appear the fault again in a certain group of records, the interval between the maintenance completion time and the current time is taken as the normal operation time, and finally a plurality of groups of data are obtained.

[0202] Step S603: removing extreme values in the normal operation time data, calculating the average value of the remaining data as the reference operation time.

[0203] Specifically, the extreme value refers to abnormal data caused by accidental factors (such as human damage, sudden power failure), which does not reflect the operation rule after normal maintenance, and needs to be removed to avoid affecting the accuracy of the reference value. The reference operation time refers to the average fault-free operation time of the elevator under normal use conditions after the same maintenance, which is the basic reference value for predicting the next maintenance interval. In the embodiment, the quartile method can be used to identify the extreme value. After sorting the normal operation time data (such as 6, 5.5, 7, 3, 6.5), the lower quartile Q1=5.5 and the upper quartile Q3=6.5 are calculated, the interquartile range IQR=Q3-Q1=1, and the extreme value determination standard is less than Q1-1.5×IQR (5.5-1.5=4) or greater than Q3+1.5×IQR (6.5+1.5=8). The data "3 months" is less than 4, which is determined as an extreme value (may be caused by sudden external force leading to early failure), and is removed. The average value of the remaining data (6, 5.5, 7, 6.5) is (6+5.5+7+6.5) / 4=6.25 months, that is, the reference operation time is 6.25 months.

[0204] Step S604: obtaining the operation time of the current elevator, subtracting the operation time from the reference operation time to obtain the preliminary estimated interval.

[0205] Specifically, in the embodiment, the operation time refers to the interval from the completion of the current maintenance to the current time, reflecting the degree of wear of the elevator after the current maintenance (the longer the time, the greater the wear, the closer to the next maintenance); the preliminary estimated interval refers to the theoretical maintenance interval without considering the difference in use frequency, which is calculated based on the historical average level. The cumulative operation time after the current maintenance is recorded by the elevator control system (for example, the maintenance is completed on June 1, 2025, and the current time is August 1, 2025, so the operation time is 2 months). Subtracting the operation time (2 months) from the reference operation time (6.25 months) gives the preliminary estimated interval = 6.25-2=4.25 months.

[0206] Step S605: judging whether the use frequency of the current elevator is higher than the average use frequency of a plurality of groups of maintenance in recent period.

[0207] Specifically, in the embodiment, the current use frequency is counted through the elevator operation log (e.g., 200 times per day in the last 30 days, and the average value is 200 times / day); the average number of daily operations in the 30 days after each maintenance is obtained from the historical maintenance records (e.g., 180 times / day, 190 times / day, and 210 times / day), and the historical average use frequency is calculated as (180+190+210) / 3=193.3 times / day. By comparing the current use frequency (200 times / day) with the historical average value (193.3 times / day), it is determined that the current use frequency is higher.

[0208] Step S606: If higher than the average use frequency, multiply the preliminary estimated interval by a preset down adjustment coefficient to obtain the final next maintenance time interval.

[0209] Specifically, in the embodiment, the preset down adjustment coefficient can be set to 0.9 (the use frequency is higher than the historical average value by 10%, and the interval is shortened by 10%); the current use frequency (200 times / day) is 3.5% higher than the historical average value (193.3 times / day), which does not reach 10%, and the final interval is calculated according to the basic down adjustment coefficient: final interval=4.25x0.9≈3.8 months; if the use frequency is 20% higher (e.g., 232 times / day), the down adjustment coefficient can be set to 0.8, and the interval is further shortened.

[0210] Step S607: If not higher than the average use frequency, multiply the preliminary estimated interval by a preset up adjustment coefficient to obtain the final next maintenance time interval.

[0211] Specifically, the preset up adjustment coefficient can be set to 1.1 (the use frequency is lower than the historical average value by 10%, and the interval is extended by 10%); if the current use frequency is 170 times / day (lower than the historical average value 193.3 times / day), the final interval is 4.25x1.1≈4.7 months; in the embodiment, the upper limit of the up adjustment coefficient is 1.5 (even if the use frequency is extremely low, the interval will not exceed 1.5 times of the baseline value, avoiding excessive extension leading to hidden dangers).

[0212] Step S608: If sufficient historical records cannot be extracted, the next maintenance time interval is estimated in proportion based on the rated service life of the elevator and the used service life.

[0213] Specifically, in the embodiment, if the elevator is newly installed (no historical record) or the historical record is less than 3 groups, the rated service life of the elevator (such as 15 years) and the used life (such as 3 years) are called. Assuming that the elevator of this model needs to be maintained 30 times (once every 6 months on average) within the rated life, the remaining maintenance times = 30 x (15-3) / 15 = 24 times, and the next maintenance interval = (15-3) x 12 / 24 = 6 months. If it has been used for 10 years (close to retirement), the remaining maintenance times = 30 x (15-10) / 15 = 10 times, and the interval = (15-10) x 12 / 10 = 6 months (old elevators need to be maintained at a fixed short interval).

[0214] The elevator maintenance method provided by the embodiment not only ensures the rationality of interval prediction by referring to historical data, but also dynamically adjusts and adapts to the actual wear speed of the elevator by using the frequency, avoids the hidden dangers of high-load elevators caused by too long intervals or the over-maintenance of low-load elevators caused by too short intervals, and provides a reliable alternative estimation method for the case of insufficient historical data. Ultimately, the maintenance interval is accurately matched with the actual state of the elevator, further improving the economy of maintenance and the safety of elevator operation.

[0215] Reference Figure 7 In one of the embodiments of the embodiment, S605 determines whether the use frequency of the current elevator is higher than the average use frequency of the recent multiple maintenance groups, including steps S701 to S706:

[0216] Step S701: Count the number of daily operations of the current elevator in the recent fixed period, calculate the average value of the number of daily operations as the current use frequency.

[0217] Specifically, the current use frequency refers to an index reflecting the current actual load intensity of the elevator. The more the number of operations, the faster the wear of the components, and the maintenance interval needs to be shortened. In the embodiment, the fixed period is set to 30 days, and the number of daily operations in the recent 30 days is extracted from the elevator control system. After removing the abnormal values due to special events, the average value obtained is the current use frequency.

[0218] Step S702: Extract the number of daily operations in the fixed period after the recent multiple maintenance groups from the historical maintenance records, and calculate the fixed period average number of operations for each group.

[0219] Specifically, in the embodiment, the number of daily operations in the fixed period after the last 3 maintenance groups can be extracted.

[0220] Step S703: Add the fixed period average number of operations of the multiple groups and take the average value to obtain the average use frequency of the recent multiple maintenance groups.

[0221] Step S704: compare the current use frequency with the average use frequency in the recent multiple maintenance.

[0222] Step S705: if the current use frequency is greater than the average use frequency in the recent multiple maintenance, it is determined that the use frequency of the current elevator is higher than the average use frequency in the recent multiple maintenance.

[0223] Step S706: if the current use frequency is less than or equal to the average use frequency in the recent multiple maintenance, it is determined that the use frequency of the current elevator is not higher than the average use frequency in the recent multiple maintenance.

[0224] Specifically, if there is data missing in a fixed period after a certain group of historical maintenance, the average number of running times in each half period before and after the maintenance is used to supplement, and then the average use frequency is calculated.

[0225] The elevator maintenance method provided by the embodiment ensures the comparability of the current and historical use frequencies by uniformly calculating the average number of running times in a period (such as 30 days), so that the judgment result of the use frequency accurately reflects the difference in actual load strength of the elevator, thereby providing reliable support for subsequent adjustment of the maintenance interval through the coefficient, further ensuring the matching degree of the maintenance interval and the actual wear rate of the elevator, and effectively improving the scientificity and pertinence of the maintenance plan.

[0226] Reference Figure 8 In one of the embodiments of the present embodiment, after step S101, steps S801 to S807 are further included:

[0227] Step S801: pre-process the real-time running data, compare the pre-processed real-time running data with the standard running data of the elevator at the time of leaving the factory, and calculate the data deviation value.

[0228] Specifically, the data deviation value is a quantitative index for measuring the difference between the real-time running data and the standard data at the time of leaving the factory, which can intuitively reflect the percentage of the current running state of the elevator deviating from the design standard; in the present embodiment, the abnormal values and missing values in the data are removed, the abnormal values are identified by a pre-set statistical principle and replaced by the average value of the adjacent data, and the missing values are supplemented by the interpolation method.

[0229] Step S802: determine whether the data deviation value exceeds the pre-set deviation warning threshold.

[0230] Specifically, in the present embodiment, the deviation warning threshold is a pre-set percentage standard, which is used to determine whether the deviation of the pre-processed real-time running data of the elevator and the standard data at the time of leaving the factory exceeds the normal range, and then determines whether to trigger a warning.

[0231] Step S803: if the deviation warning threshold is not exceeded, continue to monitor the real-time running data of the elevator.

[0232] Step S804: If the deviation exceeds the deviation warning threshold, generate data deviation warning information and send it to the maintenance personnel terminal.

[0233] Specifically, the warning information includes the deviation parameter, the real-time value, the standard value, the deviation rate, the occurrence time, and the recommended inspection items.

[0234] Step S805: After receiving the warning information, the maintenance personnel performs a preliminary inspection on the elevator within a preset time and records the inspection results.

[0235] Step S806: If the inspection results show no abnormalities, update the deviation warning threshold.

[0236] Specifically, in this embodiment, if no physical failure is found during the inspection (such as high vibration value caused by floor structure resonance, not a problem with the elevator itself), the warning threshold for this parameter is increased from 15% to 25% (dynamically adapted based on actual operation), to avoid subsequent false positives. After updating, record the adjustment reason (“normal deviation caused by resonance”) and the approver.

[0237] Step S807: If the inspection results show that there is a potential failure, start the fault diagnosis process in advance.

[0238] The elevator maintenance method provided by the present embodiment not only avoids neglecting minor hidden dangers through deviation value monitoring, but also balances the accuracy and flexibility of the warning through on-site verification by maintenance personnel and dynamic threshold updating, reducing ineffective maintenance caused by data false positives, and enabling early intervention in diagnosis when potential failures are confirmed, effectively addressing the lag in traditional maintenance that relies on post-failure processing, further improving the timeliness of elevator hidden danger investigation and the safety of operation.

[0239] Reference Figure 9 In one embodiment of the present embodiment, step S105 builds an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm, including steps S901 to S907:

[0240] Step S901: Collect running data samples corresponding to different fault types of the elevator.

[0241] Specifically, in this embodiment, the fault data within the last 5 years is collected through the elevator Internet of Things platform, covering 10 common faults (such as traction machine failure, door machine failure, safety circuit failure, etc.), with a sample size of ≥1000 for each fault type, and each sample containing 1 hour of running data before the fault occurs and a fault type label.

[0242] Step S902: Divide the sample data into a training set and a test set according to a preset ratio.

[0243] Specifically, the preset ratio can be set to 7:3; the training set is used for model training, and the test set is used for model verification.

[0244] Step S903: a preset deep learning model structure is selected, training set data is input, a preset learning rate and iteration number are set for model training.

[0245] Step S904: after each iteration, the model diagnosis accuracy is verified using test set data, and if the accuracy does not improve for a preset number of iterations in succession, the training is stopped.

[0246] Specifically, in this embodiment, the diagnosis accuracy (the number of correctly predicted samples / the total number of test set samples) is calculated using test set data after each iteration (model parameter update once); if the accuracy remains at 88% (no improvement) for 5 iterations in succession (such as the 10th to 15th), the early stopping mechanism is triggered, the training is stopped, and invalid iterations are avoided.

[0247] Step S905: the trained model is saved, and the average diagnosis accuracy of the model on the test set is calculated.

[0248] Step S906: if the average diagnosis accuracy is higher than a preset accuracy threshold, the model is used as an elevator fault diagnosis and maintenance recommendation model.

[0249] Specifically, in this embodiment, the preset accuracy threshold is a minimum accuracy standard set in advance before the elevator fault diagnosis model is put into use, which is used to judge whether the trained model is qualified and can meet the actual fault diagnosis requirements.

[0250] Step S907: if the average diagnosis accuracy is lower than the preset accuracy threshold, the number of sample data is increased, the model learning rate is adjusted, and the training is performed again until the accuracy meets the standard.

[0251] The elevator maintenance method provided by the embodiment provides comprehensive data support for model training through sufficient samples and scientific data division; avoids model overfitting by means of the early stopping mechanism, ensures that the model performance meets the standard through the accuracy threshold and parameter adjustment, effectively improves the accuracy of model diagnosis and the rationality of output maintenance scheme, avoids fault misjudgment or improper maintenance suggestion caused by insufficient model precision, provides core technical support for subsequent accurate fault diagnosis and development of targeted maintenance scheme, and further guarantees the scientificity and efficiency of elevator maintenance.

[0252] In a second aspect, the application also discloses an elevator maintenance system.

[0253] Referring to Figure 10 An elevator maintenance system comprises:

[0254] The first acquisition module is configured to acquire real-time running data of the elevator.

[0255] The first judgment module is configured to judge whether the real-time running data is in a normal range based on a preset normal running parameter range of the elevator.

[0256] The label generation module is configured to generate a corresponding abnormal label if the real-time running data is out of the normal range.

[0257] The second acquisition module is configured to perform correlation analysis on the abnormal label and acquire a core feature set of the elevator fault.

[0258] The model construction module is configured to construct an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm.

[0259] The scheme output module is configured to input the core feature set into the fault diagnosis and maintenance recommendation model and output a fault diagnosis result and a preliminary maintenance scheme.

[0260] The second judgment module is configured to judge whether the preliminary maintenance scheme meets a preset maintenance effect standard.

[0261] The plan generation module is configured to generate a final maintenance plan based on the preliminary maintenance scheme if the maintenance effect standard is met.

[0262] The interval prediction module is configured to predict a time interval at which the elevator needs to be maintained next based on the final maintenance plan and a historical maintenance record of the elevator.

[0263] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, any equivalent changes made in the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. An elevator maintenance method characterized by, The method comprises the following steps: acquiring real-time operation data of an elevator; judging whether the real-time operation data is within a normal range based on preset normal operation parameter ranges of the elevator; generating corresponding abnormal labels if the real-time operation data is outside the normal range; performing correlation analysis on the abnormal labels to obtain a core feature set of elevator faults; constructing an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm; inputting the core feature set into the fault diagnosis and maintenance recommendation model to output fault diagnosis results and a preliminary maintenance plan; judging whether the preliminary maintenance plan meets preset maintenance effect standards; generating a final maintenance plan based on the preliminary maintenance plan if the maintenance effect standards are met; predicting a time interval at which the elevator needs to be maintained next based on the final maintenance plan and historical maintenance records of the elevator; wherein the correlation analysis on the abnormal labels to obtain the core feature set of elevator faults comprises the following steps: extracting abnormal types, abnormal occurrence positions and abnormal durations corresponding to different abnormal labels; calculating coincidence degrees of abnormal types, proximity degrees of abnormal occurrence positions and correlation degrees of abnormal durations between different abnormal labels; assigning weight values corresponding to the coincidence degrees, proximity degrees and correlation degrees respectively; calculating comprehensive correlation coefficients between different abnormal labels based on the coincidence degrees, proximity degrees, correlation degrees and corresponding weight values; judging whether the comprehensive correlation coefficients exceed a preset correlation coefficient threshold; grouping corresponding abnormal labels into the same fault correlation group if the comprehensive correlation coefficients exceed the preset correlation coefficient threshold; extracting the abnormal type with the highest occurrence frequency, the most concentrated abnormal occurrence position and the longest abnormal duration in the fault correlation group as fault key information; filtering out parameters representing the essence of the fault based on the fault key information to form the core feature set of elevator faults; the judgment of whether the preliminary maintenance plan meets the preset maintenance effect standards comprises the following steps: obtaining a fault type to which the preliminary maintenance plan is directed, and calling historical maintenance records of the same fault type from an elevator historical maintenance database, wherein the same fault type refers to a fault type that meets the consistency of fault position, fault cause and fault degree; extracting fault recurrence rates after maintenance, elevator operation stability scores after maintenance and maintenance cost data in the historical maintenance records; calculating average values of the recurrence rates, average values of the operation stability scores and average values of the maintenance costs; comparing the average values of the recurrence rates with a preset recurrence rate threshold, comparing the average values of the operation stability scores with a preset stability threshold, and comparing the average values of the maintenance costs with a preset cost threshold; determining that the preliminary maintenance plan meets the maintenance effect standards if the average values of the recurrence rates are lower than the preset recurrence rate threshold, the average values of the operation stability scores are higher than the preset stability threshold and the average values of the maintenance costs are lower than the preset cost threshold; if any of the above conditions is not met, marking the preliminary maintenance plan as a to-be-optimized plan and recording the index that does not meet the threshold; adjusting the component replacement frequency or maintenance process parameters in the maintenance steps based on the index that does not meet the threshold for the to-be-optimized plan. Recalculate the recurrence rate, stability score and cost data corresponding to the adjusted scheme, and compare them with the preset threshold again until all criteria are met.

2. The method of claim 1, wherein, The calculation of the coincidence degree of abnormal types, the proximity of abnormal occurrence positions, and the correlation of abnormal durations between different abnormal labels comprises: Count the total number of abnormalities of the abnormal types corresponding to the first abnormal label and the second abnormal label, and the number of single abnormalities of the same type in the abnormal types, the total number of abnormalities being the sum of the number of abnormal types contained in the two abnormal labels, and the number of single abnormalities being the number of completely consistent abnormal types in the two labels; Obtain the abnormal type coincidence degree based on the total number of abnormalities and the number of single abnormalities; Obtain the actual distance between the abnormal occurrence positions of the first abnormal label and the second abnormal label and the maximum component spacing of the elevator; Obtain the abnormal occurrence position proximity based on the actual distance and the maximum component spacing of the elevator; Calculate the intersection duration and union duration of the abnormal duration of the first abnormal label and the second abnormal label; Obtain the abnormal duration correlation degree based on the intersection duration and the union duration.

3. The method of claim 1, wherein, The adjustment of the component replacement frequency or maintenance process parameters in the maintenance step based on the indicators that do not meet the threshold for the to-be-optimized scheme comprises: If only the post-maintenance failure recurrence rate is higher than the preset recurrence rate threshold, check the vulnerable components corresponding to the failure type in the historical maintenance records; Increase the replacement frequency of the vulnerable components, or replace the vulnerable components with higher durability of the same model; If only the post-maintenance elevator operation stability score is lower than the preset stability threshold, analyze the component calibration link involved in the maintenance step; Optimize the calibration parameters, extend the duration of the calibration operation, or increase the number of calibrations; If only the average maintenance cost is higher than the preset cost threshold, filter the link with the highest cost proportion in the maintenance step; Replace the maintenance materials with lower cost and the same function, or simplify the non-critical maintenance steps under the premise of ensuring the maintenance effect; If multiple indicators do not meet the requirements, prioritize adjusting the parameter that has the greatest impact on the failure recurrence rate, and then adjust the stability and cost-related parameters in turn.

4. The method of claim 1, wherein, The prediction of the time interval for the next maintenance of the elevator based on the final maintenance plan and the historical maintenance records of the elevator comprises: Obtain the failure type to which the current final maintenance plan is directed, and extract multiple sets of maintenance records for the same failure type from the historical maintenance records of the elevator; Calculate the duration of normal operation of the elevator after each maintenance to obtain multiple sets of normal operation duration data; Remove the extreme values from the normal operation duration data, calculate the average of the remaining data as the baseline operation duration; Obtain the running duration of the current elevator, subtract the running duration from the baseline operation duration to obtain a preliminary estimated interval, the running duration being the interval from the completion of the current maintenance to the current time; Determine whether the usage frequency of the current elevator is higher than the average usage frequency during the recent multiple maintenance; If it is higher than the average usage frequency, multiply the preliminary estimated interval by a preset downscaling coefficient to obtain the final next maintenance time interval; If it is not higher than the average usage frequency, multiply the preliminary estimated interval by a preset upscaling coefficient to obtain the final next maintenance time interval; If the number of historical records cannot be extracted, the next maintenance time interval is estimated proportionally based on the rated service life and the used service life of the elevator.

5. The method of claim 4, wherein, The method comprises the following steps: The average value of the daily operation times in the fixed period is calculated as the current use frequency. The average use frequency of the multiple sets of maintenance is obtained by adding the fixed period average operation times of the multiple sets and taking the average value. The current use frequency is compared with the average use frequency of the multiple sets of maintenance. If the current use frequency is greater than the average use frequency of the multiple sets of maintenance, it is determined that the use frequency of the current elevator is higher than the average use frequency of the multiple sets of maintenance. If the current use frequency is less than or equal to the average use frequency of the multiple sets of maintenance, it is determined that the use frequency of the current elevator is not higher than the average use frequency of the multiple sets of maintenance. After the real-time operation data of the elevator is obtained, the following steps are further included:

6. The method of claim 1, wherein, The preprocessed real-time operation data is compared with the standard operation data of the elevator at the time of leaving the factory, and a data deviation value is calculated. It is judged whether the data deviation value exceeds a preset deviation warning threshold. If the deviation warning threshold is not exceeded, the real-time operation data of the elevator is continuously monitored. If the deviation warning threshold is exceeded, a data deviation warning information is generated and sent to the maintenance personnel terminal. After receiving the warning information, the maintenance personnel performs a preliminary inspection on the elevator within a preset time and records the inspection results. If the inspection results show no abnormalities, the deviation warning threshold is updated. If the inspection results show potential faults, the fault diagnosis process is started in advance. The method comprises the following steps:

7. The method of claim 1, wherein, Collect operation data samples corresponding to different fault types of the elevator. The sample data is divided into a training set and a test set according to a preset proportion. A preset deep learning model structure is selected, the training set data is input, a preset learning rate and iteration number are set for model training. After each iteration, the test set data is used to verify the model diagnosis accuracy, and if the accuracy does not improve continuously for a preset number of iterations, the training is stopped. The trained model is saved, and the average diagnosis accuracy of the model on the test set is calculated. If the average diagnosis accuracy is higher than a preset accuracy threshold, the model is used as the elevator fault diagnosis and maintenance recommendation model. If the average diagnosis accuracy is lower than the preset accuracy threshold, the number of sample data is increased, the model learning rate is adjusted, and the training is restarted until the accuracy meets the standard. The method comprises the following steps:

8. An elevator maintenance system for performing the method of any one of claims 1 to 7, characterized by A first acquisition module is configured to acquire real-time operation data of an elevator. A first judgment module is configured to judge whether the real-time operation data is in a normal range based on a preset normal operation parameter range of the elevator. A label generation module is configured to generate a corresponding abnormal label if the real-time operation data exceeds the normal range. ​ The second acquisition module is configured to perform correlation analysis on the abnormal label and acquire a core feature set of the elevator fault; The model construction module is configured to construct an elevator fault diagnosis and maintenance recommendation model based on a deep learning algorithm; The scheme output module is configured to input the core feature set into the fault diagnosis and maintenance recommendation model and output a fault diagnosis result and a preliminary maintenance scheme; The second judgment module is configured to judge whether the preliminary maintenance scheme meets a preset maintenance effect standard; The plan generation module is configured to generate a final maintenance plan based on the preliminary maintenance scheme if the maintenance effect standard is met; The interval prediction module is configured to predict a time interval at which the elevator needs to be maintained next based on the final maintenance plan and a historical maintenance record of the elevator.

Images