Fault Prediction and Processing Method Based on AI Large Model Elevator Industry Data

Abstract

This invention discloses a fault prediction and processing method based on AI large-scale model elevator industry data. This invention relates to the field of elevator fault prediction technology. First, it extracts elevator maintenance information from a cloud database where the faulty component is the traction machine and performs cluster analysis to determine the fault type set and the optimal handling solution sequence. Next, it constructs a fault precursor feature group and builds an elevator fault prediction model based on a convolutional neural network. The model is trained and validated using a corresponding proportion of the fault precursor feature group, fault type sequence, and optimal handling solution sequence as samples. Then, it acquires the elevator to be monitored, constructs a digital twin model of its traction machine, and inputs the real-time operating parameter sequence into the fault prediction model. The model outputs the fault-related type, the optimal handling solution, and the fault occurrence time, and provides alarm reminders to operators. The core of this method is to achieve accurate prediction and early warning of elevator traction machine faults, ensuring elevator operational safety.

Description

Technical Field

[0001] This invention belongs to the field of elevator fault prediction technology, specifically, it relates to a fault prediction and processing method based on AI large model elevator industry data. Background Technology

[0002] With sustained economic growth and accelerated urbanization, elevators, as an indispensable vertical transportation tool in urban buildings, have seen explosive growth in number. With the advancement of information technology and the increasing popularity of cloud databases, centralized storage and management of maintenance data have been achieved, laying the foundation for subsequent data analysis.

[0003] Existing technologies for processing elevator maintenance data typically rely on manual statistical analysis, which is inefficient, error-prone, and makes it difficult to achieve rapid and accurate clustering analysis and fault type matching. Secondly, in terms of extracting early warning features, they often depend on limited operating parameters and simple statistical analysis, failing to fully capture the complex changing trends of the traction machine before a fault occurs. Furthermore, in terms of real-time monitoring and early warning, existing technologies can usually only achieve passive response after a fault occurs, lacking the ability to predict and accurately locate the time of fault occurrence. Moreover, due to insufficient in-depth mining and analysis of historical maintenance data, it is difficult to determine the optimal handling strategy.

[0004] To address the aforementioned issues, this invention proposes a fault prediction and processing method based on AI large-scale model elevator industry data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a fault prediction and processing method based on AI large-scale model elevator industry data, which solves the problems of inaccurate prediction and passive response in existing technologies for elevator fault prediction.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A fault prediction and processing method based on AI large-scale model elevator industry data, which includes the following:

[0008] Step 1: Extract elevator maintenance information of several faulty components associated with the traction machine of any type of elevator from the cloud database, perform cluster analysis, and determine the set of fault types associated with the traction machine of that type of elevator and the corresponding sequence of optimal handling solutions.

[0009] Step 2: Based on the elevator maintenance information of the faulty component being the traction machine, extract the operating parameters of the traction machine within a retrospective period before the fault occurred from the cloud database, construct the operating parameter sequence associated with each operating parameter, and generate the fault precursor feature group associated with all fault types of the traction machine.

[0010] Step 3: Construct an elevator fault prediction model associated with the traction machine. Divide the fault precursor feature group, fault type sequence, and optimal handling solution sequence into training samples and validation samples respectively to train and validate the elevator fault prediction model.

[0011] Step 4: Obtain the elevator to be monitored, construct a digital twin model of the traction machine associated with the traction machine of the elevator to be monitored, obtain the simulated operating parameters, and construct a fault precursor feature group based on the operating parameter sequence as input data. Input the model into the elevator fault prediction model, output the fault type associated with the traction machine of the elevator to be monitored, the best handling solution, and the fault occurrence time, and provide corresponding alarm reminders to the operators.

[0012] As a further aspect of the present invention, in step one, the elevator maintenance information includes the elevator number, the time of the fault occurrence, the time of the fault repair, the type of fault, the faulty component, and the handling plan.

[0013] As a further aspect of the present invention, the specific method for performing cluster analysis on elevator maintenance information in step one is as follows:

[0014] Using the faulty component as the traction machine as the search condition, elevator maintenance information for elevators with the faulty component being the traction machine is extracted. The total number is denoted as j, resulting in the elevator maintenance information set. ;

[0015] Get Let m be the total number of all different fault types, and let m be the fault type set. ;

[0016] Similarly, determine All different processing schemes are used to obtain the processing scheme set. Where n represents the total number of different processing schemes, ;

[0017] Will and By performing cross combinations, we obtain The combined results are denoted as the clustering condition sequence. ,in, ;

[0018] Build There are 1 cache pool, denoted as the cache pool sequence: ,in, correspond , u is the counting index, ;

[0019] Traversal Sequences that meet the clustering criteria Elevator maintenance information corresponding to the clustering conditions is included in the corresponding cache pool;

[0020] extract The elevator maintenance information cached in each cache pool is arranged according to the cache pool sequence. The order is denoted as the elevator maintenance information clustering sequence. In any given elevator maintenance information cluster, there shall be no less than one elevator maintenance information and no more than j elevator maintenance information.

[0021] As a further aspect of the present invention, in step one, the specific method for determining the set of fault types associated with the traction machine of this type of elevator and the corresponding sequence of optimal handling solutions is as follows:

[0022] S41. Extract the fault type set ;

[0023] S42, from Extract any fault type Where k is the counting index, ;

[0024] S43, Clustering sequence from elevator maintenance information Extract the fault type as The total number of elevator maintenance information is denoted as x.

[0025] Let the x elevator maintenance information records be extracted in the order listed, and denote this as the elevator maintenance information filtering sequence. ,in, ;

[0026] S44, from Extract all types of treatment schemes, denoted as y, and denote the extraction order as the treatment scheme selection sequence. ;

[0027] S45, from Extract x elevator numbers from the list, and denote the extraction order as the elevator number filtering sequence. ;

[0028] S46, From Extract x maintenance times, and denote the maintenance time screening sequence in the order of extraction. ;

[0029] S47, Continuous Monitoring The corresponding x elevators, until the traction machine of the x elevators experiences another failure of type 0. The malfunction;

[0030] Extract x fault occurrence times, and sort them by The order is denoted as the fault occurrence time filtering sequence. ;

[0031] S48, Adopt Subtract accordingly The maintenance and upkeep time screening sequence was obtained. ;

[0032] S49, Extraction Any processing scheme and from Extract all processing schemes as The maintenance and upkeep time was averaged to obtain The associated mean maintenance time is denoted as Where z is the counting index, ;

[0033] Similarly, determine The average maintenance uptime associated with all treatment options, and according to The order of these sequences is denoted as the average maintenance and upkeep time series. ;

[0034] S410, from The longest average maintenance duration is extracted and denoted as . ;

[0035] Get Corresponding solutions ,Will As The associated best-case scenario is renamed ;

[0036] S411. Repeat steps S42 to S410 to confirm. The optimal handling solutions associated with all fault types are summarized to form a sequence of optimal handling solutions. .

[0037] As a further aspect of the present invention, in step two, the specific method for constructing the sequence of operating parameters associated with each operating parameter is as follows:

[0038] S51. Determine the fault type The associated elevator maintenance information filtering sequence And elevator number filtering sequence ;

[0039] S52, from Determine the fault occurrence time associated with x elevator numbers, and then... The order of the fault occurrences is sorted to obtain the fault occurrence time series. ;

[0040] S53. Obtain the backtracking period and duration preset by the operator. ;

[0041] S54, Confirm Any elevator number and elevator number The associated failure occurrence time Where v is the counting index, ;

[0042] S55, with As the end time of the backtracking, we look back one backtracking cycle. The time must be traced back to the start time, recorded as... ;

[0043] S56, with Extract as query criteria to The elevator number inside is The operating parameters of the elevator's traction machine;

[0044] The operating parameters include the three-phase current of the traction machine, the bearing temperature of the traction machine, the vibration amplitude of the motor, and the rotational speed of the traction sheave.

[0045] S57, will The associated operating parameters were serialized in chronological order to obtain the traction machine three-phase current sequence, traction machine bearing temperature sequence, motor vibration amplitude sequence, and traction sheave speed sequence, which were then summarized as follows: The associated sequence of operating parameters;

[0046] S58. Repeat steps S54 to S57 to confirm. The sequence of operating parameters associated with all elevator numbers.

[0047] As a further aspect of the present invention, the specific method for generating the fault precursor feature group associated with all fault types of the traction machine in step two is as follows:

[0048] S61. Obtain elevator number The associated sequence of operating parameters;

[0049] S62. Construct a two-dimensional coordinate system with the timeline as the horizontal axis and the values ​​of the running parameters as the vertical axis;

[0050] The three-phase current sequence of the traction machine is plotted on a two-dimensional coordinate system to obtain several three-phase current data points of the traction machine;

[0051] S63. Connect two adjacent three-phase current data points of the traction machine with short lines to obtain the three-phase current broken line of the traction machine;

[0052] S63. Determine the slope between all pairs of adjacent data points in the three-phase current piecewise linear curve of the traction machine, and average them. The average value is recorded as the slope of the three-phase current change of the traction machine. ;

[0053] S64. Similarly, determine the slope of the temperature change of the traction machine bearing. Slope of motor vibration amplitude change and the slope of the traction sheave speed change ;

[0054] S65, will as well as The combination is recorded as the elevator number. The associated fault precursor feature vector ;

[0055] S66. Repeat steps S61 to S65 to determine the elevator number filtering sequence. The fault precursor feature vectors associated with all elevator numbers are collected and summarized, and denoted as fault types. The associated fault precursor feature group , represented as: ;

[0056] S67. By analogy, determine the set of fault types. The fault precursor characteristic groups associated with all fault types are denoted in sequence as follows: .

[0057] As a further aspect of the present invention, the specific method for training and validating the elevator fault prediction model in step three is as follows:

[0058] S71. Construct an elevator fault prediction model for traction machine fault prediction;

[0059] S72. Extract the fault type set and its related:

[0060] Fault precursor characteristic group Optimal processing sequence This is summarized and recorded as the overall dataset;

[0061] S73. Randomly and evenly select 70% of the data from the overall dataset as training samples and the remaining 30% as validation samples.

[0062] S74. Input the training samples into the elevator fault prediction model to perform training operations and adjust the network parameters of the elevator fault prediction model.

[0063] S75. Use the verification sample to input the elevator fault prediction model to verify the prediction accuracy, and compare it with the prediction accuracy threshold preset by the operator.

[0064] If the prediction accuracy exceeds the prediction accuracy threshold, the verification is successful, and the elevator fault prediction model is output.

[0065] If the prediction accuracy does not exceed the prediction accuracy threshold, the verification fails. Continue the training operation until the verification is successful.

[0066] As a further aspect of the present invention, the specific method for constructing the digital twin model of the traction machine associated with the traction machine of the elevator to be monitored in step four is as follows:

[0067] Obtain the physical structural parameters associated with the traction machine of the elevator to be monitored, including model, size, and material properties, all of which are considered known values;

[0068] Extract the operating parameters associated with the traction machine of the elevator to be monitored from the cloud database;

[0069] A digital twin model of the traction machine is constructed by combining the physical structural parameters and operating parameters of the traction machine of the elevator to be monitored, denoted as . ;

[0070] Real-time extraction of operating parameters associated with the traction machine of the elevator to be monitored from the cloud database. Synchronize the mapping in the middle.

[0071] As a further aspect of the present invention, the specific method for outputting the fault type, optimal handling solution, and fault occurrence time associated with the traction machine of the elevator to be monitored in step four is as follows:

[0072] Using digital twin model of traction machine The simulation runs for a future time period and extracts the simulation parameters. The duration of the simulation period is determined by the operator.

[0073] The simulated operating parameters are serialized to obtain an operating parameter sequence. Based on the operating parameter sequence, a fault precursor feature group is constructed and input into the elevator fault prediction model.

[0074] If no fault occurs, continue the simulation;

[0075] If a fault occurs, the system will output the fault type associated with the traction machine of the elevator under monitoring, the best handling solution, and the time of the fault occurrence.

[0076] The elevator number of the elevator to be monitored is then obtained and combined with the obtained fault type, optimal handling solution and fault occurrence time to provide corresponding alarm reminders to the operators as alarm information.

[0077] The beneficial effects of this invention are:

[0078] (1) This invention extracts elevator maintenance information from cloud database and performs cluster analysis to obtain operating parameters to construct fault precursor feature groups. Based on the above data, it provides dynamic and accurate data for training the elevator fault prediction model. Then, it uses the traction machine digital twin model to map the elevator traction machine operating parameters in real time. The simulated operating parameters are obtained through the traction machine digital twin model and input into the constructed elevator fault prediction model to achieve accurate elevator fault location prediction and real-time monitoring. This helps operators respond quickly and make scientific decisions, reduce elevator downtime, and ensure reliable elevator operation.

[0079] (2) This invention achieves refined classification of massive maintenance information by cross-combination clustering of fault types and treatment schemes, ensuring that subsequent analysis focuses on specific fault scenarios and avoids decision bias caused by mixed data; secondly, it introduces maintenance time as a key indicator to quantify the actual effect of different treatment schemes, and takes the one with the longest average maintenance time as the best treatment scheme, abandoning subjective experience dependence and ensuring that the best treatment scheme can maximize the extension of the equipment's fault-free operation cycle; finally, it independently calculates the best treatment scheme for each fault type to form a continuously optimized knowledge base, which is convenient for determining the best treatment scheme after subsequent fault prediction.

[0080] (3) This invention uses the time of failure as a benchmark to dynamically backtrack the operating parameters, ensuring that the collected data is closely related to the cause of the failure and avoiding the lag of traditional fixed-cycle monitoring; furthermore, the original operating parameter sequence is transformed into a variable slope feature vector to extract the essential law of the deterioration of the traction machine's operating status, which both compresses the data volume and enhances the interpretability of the features; finally, by fusing multiple variable slope feature vectors, a fault precursor feature group is constructed, which avoids misjudgment of single parameters while improving the robustness of prediction, realizing the transformation from "post-event maintenance" to "pre-event intervention", and significantly reducing the risk of downtime and safety costs caused by sudden failures;

[0081] (4) This invention uses a convolutional neural network to construct an elevator fault prediction model to process the fault precursor feature group, divides the training set and the validation set, and sets a threshold to dynamically optimize the model to ensure that the prediction results have strong robustness. Secondly, it constructs a digital twin model associated with the traction machine of the elevator to be monitored, realizes the dynamic mapping of the entire life cycle of the traction machine of the elevator to be monitored, and obtains the simulated operating parameters by simulating the future cycle operating state. Combined with the constructed elevator fault prediction model, it pre-simulates the fault scenario and outputs the three-in-one early warning information of fault type, best handling solution and fault occurrence time, transforming post-event maintenance into pre-event precise intervention, helping operators to deal with elevator faults in a timely manner and extend the service life of the elevator. Attached Figure Description

[0082] The invention will now be further described with reference to the accompanying drawings.

[0083] Figure 1 This is a flowchart illustrating the method described in this invention;

[0084] Figure 2 This is a flowchart illustrating the method described in Embodiment 2 of the present invention;

[0085] Figure 3 This is a flowchart illustrating the method described in Embodiment 3 of the present invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] Example 1

[0088] Fault prediction and processing methods based on AI large-scale model elevator industry data, such as Figure 1 As shown, this method includes the following:

[0089] First, before implementing this method, a cloud database with a certain storage capacity is required (generally, elevator-related data will be persistently stored, such as maintenance information or operating parameters). The data stored in the cloud database includes elevator maintenance information and elevator operating parameters.

[0090] This method mainly focuses on the analysis of the traction machine, an important component of elevators. The traction machine is the main equipment for controlling the elevator's up and down movement, and its operating parameters can directly reflect the overall operating status of the elevator.

[0091] Next, extract several elevator maintenance information records associated with any type of elevator from the cloud database (the elevator maintenance information refers to the record information that is persistently stored after the operator performs maintenance on the elevator after a malfunction, and the elevator maintenance information includes: elevator number, malfunction time, malfunction repair time, malfunction type, malfunctioning component, and handling plan).

[0092] It should be noted that "any type of elevator" refers to only one type of elevator, with identical specifications, operating mode, manufacturer, and material composition. This avoids practical errors caused by mixing different types of elevators in the analysis. Although this method uses only a single type of elevator as an example, it is also applicable to other types of elevators with traction machines.

[0093] Among the extracted elevator maintenance information, the traction machine is used as the query condition for the faulty component. The elevator maintenance information is retrieved from the extracted elevator maintenance information where the faulty component is the traction machine. The extracted elevator maintenance information is then subjected to cluster analysis (as mentioned above, this method mainly targets the fault handling of elevator traction machines). The set of fault types associated with the traction machine of this type of elevator is determined, as well as the best handling scheme that corresponds one-to-one with the fault types in the fault type set. These are then summarized into a sequence of best handling schemes, where the fault type set also corresponds one-to-one with the sequence of best handling schemes.

[0094] Next, the elevator number of each elevator is determined from the elevator maintenance information of the elevator with the identified faulty component being the traction machine. Then, the determined elevator number is used as the query condition to query the cloud database for the operating parameters associated with the traction machine of the elevator with the determined elevator number in the previous backtracking period when the traction machine of the elevator with the identified elevator number experienced the fault corresponding to the above-mentioned elevator maintenance information.

[0095] In other words, the operating parameters of the traction machine are determined within a retrospective period before the elevator malfunctions, wherein the elevator is the elevator in the elevator maintenance information where the malfunctioning component is determined to be the traction machine.

[0096] It should be explained here that the operating parameters include the three-phase current of the traction machine, the temperature of the traction machine bearings, the vibration amplitude of the motor, and the speed of the traction sheave when the elevator is in operation.

[0097] Next, construct the operating parameter sequence associated with each operating parameter, and generate the fault precursor feature group associated with all fault types of the traction machine based on the constructed operating parameter sequence;

[0098] When a traction machine malfunctions, its operating parameters will fluctuate or change drastically. Constructing a sequence of operating parameters can intuitively show the trend of the traction machine's operating status changes before the malfunction. Secondly, by calculating the slope of adjacent data points in each parameter sequence and taking the average, the slope of each parameter's change can be obtained, which can capture the abnormal change trend of the traction machine before the malfunction. Then, the change slopes are combined to form a fault precursor feature vector. By summarizing the feature vectors corresponding to all fault types, a fault precursor feature group is generated (the fault precursor feature group will be used as important training data, with the aim of training a high-precision fault prediction model to accurately predict faults in advance).

[0099] The following steps illustrate how to build and train an elevator failure prediction model, specifically:

[0100] Obtain the fault precursor feature group, fault type sequence, and optimal handling solution sequence, which correspond one-to-one with the fault type.

[0101] Next, a deep learning model based on convolutional neural networks is designed to extract key local patterns and hierarchical features from the fault precursor feature group, and it is named the elevator fault prediction model (this step can be implemented by existing technology, and will not be elaborated in this solution; for ease of reference, the term "model" will refer to "elevator fault prediction model" in the following sections).

[0102] Next, the fault precursor feature group, fault type sequence, and optimal handling solution sequence are randomly and evenly divided into two sets of mutually exclusive data according to a ratio of 70% and 30%. One set consists of 70% fault precursor feature group, 70% fault type sequence, and 70% optimal handling solution sequence, while the other set consists of 30% fault precursor feature group, 30% fault type sequence, and 30% optimal handling solution sequence.

[0103] A set of data comprising 70% is used as training samples (to train the model, which learns how to map from the set of fault precursor features to fault types and optimal solutions by optimizing its internal network parameters), while another set of data comprising 30% is used as validation samples (to evaluate the model's performance at different stages of training, monitor whether the model is overfitting, and decide when to stop training).

[0104] When the prediction progress of the trained model exceeds the prediction accuracy threshold preset by the operator, it means that the model can be used in the actual production environment. If the prediction progress of the model does not exceed the prediction accuracy threshold preset by the operator, it means that training needs to continue (if the training samples or validation samples are exhausted, new elevator maintenance information and operating parameters will be obtained from the cloud database to generate new training samples and validation samples).

[0105] Once the training is complete, an elevator fault prediction model can be obtained. However, this model alone is not enough, because by the time the elevator traction machine's operating parameters are collected, a fault may have already occurred (there is a delay in data collection and transmission). Even if real-time collection is achieved, delays may still occur due to subjective factors such as operators, which could lead to dangerous situations. Therefore, it is still impossible to predict faults and solve problems before they occur.

[0106] Based on this, the following content presents a method for predicting faults, specifically:

[0107] Obtain any elevator to be monitored, and the type of elevator to be monitored must be consistent with the elevators described above (if they are inconsistent, the results and models described above cannot be used for elevators of that type).

[0108] Further determine the physical structural parameters associated with the traction machine of the elevator to be monitored (the physical structural parameters can be obtained from the manufacturer and equipment nameplate, which are all public information and can be regarded as known values). The physical structural parameters include model, size and material properties.

[0109] Next, obtain the elevator number of the elevator to be monitored, and use the determined elevator number as a query condition to retrieve and extract the operating parameters associated with the elevator to be monitored.

[0110] Using digital twin technology, a digital twin model of the traction machine associated with the traction machine of the elevator to be monitored is constructed by combining the physical structural parameters associated with the traction machine of the elevator to be monitored (the content covered by the existing technology will not be elaborated in this solution). For ease of reference, the "digital twin model" in the following text refers to the "traction machine digital twin model".

[0111] Then, the operating parameters associated with the traction machine of the elevator to be monitored are extracted from the cloud database in real time and input into the constructed digital twin model of the traction machine for mapping and interconnection, so that the digital twin model of the traction machine can be synchronized with the real traction machine.

[0112] Next, the digital twin model of the traction machine associated with the elevator to be monitored is simulated for a future simulation cycle (the duration of the simulation cycle is preset by the operator), and the operating parameters simulated within this simulation cycle are obtained based on the digital twin model of the traction machine.

[0113] Repeat the method described above to serialize the simulated operating parameters to obtain the operating parameter sequence associated with each operating parameter. Then, construct a fault precursor feature group based on the operating parameter sequence (the fault precursor feature group mentioned here is not only associated with faults and fault types, but also represents the changing trend of the operating parameter sequence; that is, the fault precursor feature group can also represent the changing trend of normal operating parameters).

[0114] The constructed fault precursor feature set is input into the trained elevator fault prediction model. The elevator fault prediction model outputs the fault type, optimal handling solution, and fault occurrence time associated with the traction machine of the elevator to be monitored (if there is no fault, the fault type, optimal handling solution, and fault occurrence time are not output, and a fault-free reminder is output).

[0115] If a fault occurs, the elevator number of the elevator to be monitored will be combined with the obtained fault type, optimal handling solution and fault occurrence time to provide corresponding alarm reminders to the operators as alarm information.

[0116] Example 2

[0117] This embodiment, based on embodiment 1, further discloses a method for performing cluster analysis on elevator maintenance information to determine the optimal processing sequence, such as... Figure 2 As shown, it specifically includes the following:

[0118] As described in Example 1, several elevator maintenance information entries for elevators with traction machines as the faulty components were identified. The total number of these elevator maintenance information entries was counted and denoted as j, meaning a total of j elevator maintenance information entries were obtained. These j elevator maintenance information entries were then sorted according to the order in which they were obtained. The sorted result is denoted as the elevator maintenance information set, and is represented as follows: ;

[0119] Then from the elevator maintenance information set The system retrieves the different associated fault types (one elevator maintenance information corresponds to one fault type, but duplicate fault types exist; duplicate fault types are not counted repeatedly), and the total number is denoted as m. The m fault types are then grouped into a fault type set according to the order in which they were retrieved, represented as: m does not exceed j;

[0120] Following the method described above for determining the fault type set, then from the elevator maintenance information set... The system retrieves different handling solutions associated with different fault types (one elevator maintenance information corresponds to one handling solution, but duplicate handling solutions exist and are not counted repeatedly), and sorts them according to the retrieval order to obtain the set of handling solutions represented as follows: , where n represents the total number of different processing schemes, and n does not exceed j;

[0121] Then the determined set of processing schemes With fault type set Perform cross-combination (the set of post-processing schemes for cross-combination) Each processing solution in the set is related to the fault type. Each fault type in the algorithm has combinations, provided that combinations are possible. If a solution cannot be applied to the corresponding fault type, the combination is canceled (this algorithm assumes that all solutions can be combined with all fault types). The combination results, in the order of combination, are denoted as the clustering condition sequence, as follows: ,in, equal ;

[0122] Next, build There are several cache pools, and they are recorded as a cache pool sequence in the order they were built, as follows: ,in, correspond , indicating cache pool Used to store clustering conditions The associated elevator maintenance information, u is the counting index, and its value ranges from 1 to... .

[0123] Then extract the elevator maintenance information set determined in the above content. The system then iterates through the sequences (extracting fault types and processing solutions for verification) and identifies those that meet the clustering criteria. Elevator maintenance information corresponding to the clustering conditions is included in the cache pool associated with the corresponding clustering conditions. For example, if a piece of elevator maintenance information meets the clustering conditions... Then the elevator maintenance information will be included in the clustering criteria. The associated cache pool middle.

[0124] extract The elevator maintenance information cached in each cache pool is arranged according to the cache pool sequence. The order of the elevator maintenance information clustering sequence is denoted as: .

[0125] Then extract the determined fault type set and from the fault type set Extract any fault type and denote it as... , where k is the counting index, with a value ranging from 1 to m.

[0126] Then extract the determined elevator maintenance information cluster sequence. The fault type was determined to be... We collect elevator maintenance information, count the total number, and denote it as x. The x extracted elevator maintenance information entries are then categorized into an elevator maintenance information filtering sequence based on the order in which they were extracted, as follows: , where x does not exceed n.

[0127] Next, the elevator maintenance information is filtered from the determined sequence. Extract all types of processing schemes (duplicate processing schemes are counted as one processing scheme), then count the total number of determined processing scheme types, denoted as y. The y processing schemes are arranged in the extraction order and denoted as the processing scheme selection sequence, as follows: .

[0128] Then filter the sequence from elevator maintenance information. Extract the elevator number associated with each elevator maintenance information, totaling x elevator numbers. Record these x elevator numbers in the order they were extracted as the elevator number filtering sequence, represented as: .

[0129] Then filter the sequence from elevator maintenance information. Extract the maintenance times associated with all elevator maintenance information, totaling x maintenance times. These x maintenance times are then categorized into a maintenance time filtering sequence based on the order in which they were extracted, as follows: .

[0130] At this point, the basic data has been prepared. Based on the above, we can see the associated elevator numbers and the elevator number filtering sequence. Continuously monitor elevator number screening sequence The corresponding x elevators, and obtain the next fault type of x elevators. The fault occurrence time was determined, and x fault occurrence times were filtered according to elevator number sequence. Sort the results in the order of occurrence, and denote this as the fault occurrence time filtering sequence, represented as: .

[0131] Filtering sequences using failure occurrence time Corresponding screening sequence minus maintenance time ( minus (and so on), resulting in x time intervals, denoted as maintenance and upkeep times. These x maintenance and upkeep times are then filtered according to the fault occurrence time sequence. and maintenance time screening sequence The items are sorted in order, and the sorted result is denoted as the maintenance and upkeep time screening sequence, as follows: .

[0132] Next, the processing scheme screening sequence is extracted. Any of the processing schemes And filter sequences based on maintenance and upkeep time. Extract all processing schemes as The average maintenance time is used to obtain the solution. The associated mean maintenance time is denoted as Where z is the counting index, with a value ranging from 1 to y (if the maintenance time screening sequence is used). It does not contain any processing solution. If the current processing method fails, skip that processing method and continue to verify the next processing method.

[0133] Repeat the above steps to determine the processing scheme and filter the sequence. The average maintenance uptime associated with all treatment options (if there is no corresponding treatment option, it will not participate in this step or subsequent steps), and the sequence will be selected according to the treatment options. The order is denoted as the average maintenance maintenance time series, and is expressed as: ;

[0134] Then, from the determined average maintenance time series The longest average maintenance maintenance time is determined and denoted as the average maintenance maintenance time. Then obtain the mean maintenance time. Corresponding processing solutions Because of the solution After maintenance, it will remain in operation until the next occurrence of the corresponding fault type. The longest time was observed, therefore this solution is considered suitable for the corresponding fault type. There is a certain priority among them, which are considered to be fault types. The associated best processing solution will be the processing solution. Renamed , indicating the solution Fault type The associated best processing solution.

[0135] Using the method described above, the set of fault types can be determined. The best handling solution is associated with all fault types, and then all best handling solutions are sorted according to the fault type set. The optimal processing solution sequence is obtained by summarizing the sorting order, and is represented as follows: .

[0136] Example 3

[0137] This embodiment, based on Embodiments 1 and 2, further discloses a method for constructing a fault precursor feature set, such as... Figure 3 As shown, it specifically includes the following:

[0138] As can be seen from the content described in Example 2, the fault type The associated elevator maintenance information filtering sequence And elevator number filtering sequence The following solutions are based on fault type. As an example, construct the fault type. The associated fault precursor characteristic group, and the remaining fault types are all handled according to the fault type. The same methods and steps are applied.

[0139] Filtering sequence from elevator maintenance information The process further determines the occurrence time of faults associated with x elevators, and then filters these x fault occurrence times according to elevator maintenance information. The sequence is serialized (sorted) to obtain the fault occurrence time series, represented as: .

[0140] Next, the operator's preset backtracking period and the duration of the backtracking period are obtained. .

[0141] Then obtain the elevator number filtering sequence Any elevator number Filtering sequences from elevator maintenance information Determine the elevator number The associated fault occurrence time is marked as , where v is the counting index, with a value ranging from 1 to x.

[0142] Next, based on the time of the failure The end time of a backtracking cycle is used to rewind to a past time for one backtracking cycle. The time is used to obtain the start time of the backtracking cycle, and it is marked as... .

[0143] Then retrieve the elevator number from the cloud database. Extract the backtracking start time as a query condition. Time until the fault occurred During this period, the elevator number The operating parameters associated with the traction machine.

[0144] As described in Example 1, elevator numbering The operating parameters associated with the traction machine include the three-phase current of the traction machine, the temperature of the traction machine bearings, the vibration amplitude of the motor, and the speed of the traction sheave when the elevator is in operation.

[0145] Then extract the elevator number The associated operating parameters are serialized according to the timeline (sorted in chronological order and divided by time point), resulting in the serialized traction machine three-phase current sequence, traction machine bearing temperature sequence, motor vibration amplitude sequence, and traction sheave speed sequence, which are then summarized and recorded as the elevator number. The associated sequence of operating parameters.

[0146] By analogy, the elevator numbering selection sequence is further determined. The sequence of operating parameters associated with all elevator numbers.

[0147] Extract the elevator number again The associated operating parameter sequences are processed as examples, including the traction machine three-phase current sequence, the traction machine bearing temperature sequence, the motor vibration amplitude sequence, and the traction sheave speed sequence.

[0148] Next, a two-dimensional coordinate system is constructed with the timeline as the horizontal axis and the values ​​of the operating parameters as the vertical axis. The three-phase current sequence of the traction machine is then plotted on the constructed two-dimensional coordinate system to obtain the three-phase current data points of the traction machine associated with the three-phase current sequence (one data point corresponds to each moment). The total number of three-phase current data points of the traction machine is the same as the total number of three-phase currents of the traction machine in the three-phase current sequence. Then, adjacent three-phase current data points of the traction machine are obtained and connected with short lines to finally obtain the three-phase current broken line of the traction machine associated with the three-phase current sequence.

[0149] Next, extract all pairwise adjacent three-phase current data points of the traction machine from the three-phase current polygonal line, and calculate the slope between any two adjacent three-phase current data points. Finally, average all the obtained slopes to obtain an average slope, and record this average slope as the slope of the three-phase current change of the traction machine, expressed as: ;

[0150] It is necessary to explain here the slope of the three-phase current change in the traction machine. This indicates that the traction machine has experienced a fault type of [missing information]. At that time, it is a quantitative indicator of the overall trend of the three-phase current of the traction machine.

[0151] By analogy, the slope of the traction machine bearing temperature change associated with the traction machine bearing temperature sequence can be obtained. The slope of the change in motor vibration amplitude associated with the motor vibration amplitude sequence and the slope of the traction sheave speed change associated with the traction sheave speed sequence .

[0152] Then, the slope of the three-phase current change of the traction machine. traction machine bearing temperature change slope Slope of motor vibration amplitude change and the slope of the traction sheave speed change Combine them and record them as elevator numbers. The elevator traction machine experienced a malfunction of type 1. The fault precursor feature vector associated with the time is labeled as .

[0153] Based on the above, continue to determine the elevator numbering filter sequence. The traction machine of all elevators with the specified elevator numbers is in the fault type. The fault precursor feature vectors associated with the fault are identified, and all obtained fault precursor feature vectors are summarized as the fault type. The associated set of fault precursor characteristics is represented as follows: .

[0154] At this point, the type of fault has been determined. Similarly, the set of fault types can be determined from the associated set of fault precursor features. The set of pre-fault characteristic groups associated with all fault types, and arranged according to the fault type set. The order is represented as follows: .

[0155] All data in the formulas described above are numerical calculations performed with dimensions removed. Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0156] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0157] It should be stated that all user data collected in this application was collected with the user's consent and authorization. Furthermore, the uses of user data are legal and compliant, and the use and processing of user data comply with the relevant laws, regulations, and standards of the relevant regions.

Claims

1. A fault prediction and processing method based on AI large-scale model elevator industry data, characterized in that, This method includes the following: Step 1: Extract elevator maintenance information for several faulty components associated with the traction machine of any type of elevator from the cloud database, and perform cluster analysis to determine the set of fault types associated with the traction machine of that type of elevator and its corresponding optimal handling solution sequence. Specifically: Using the faulty component as the traction machine as the search condition, elevator maintenance information for elevators with the faulty component being the traction machine is extracted. The total number is denoted as j, resulting in the elevator maintenance information set. ; Get Let m be the total number of all different fault types, and let m be the fault type set. ; Obtain and determine the elevator maintenance information set Let m be the total number of all different fault types, and let m be the fault type set. One elevator maintenance information corresponds to one fault type. If there are duplicate fault types, the duplicate fault types will not be counted again. Similarly, determine All different processing schemes are used to obtain the processing scheme set. Where n represents the total number of different processing schemes, ; Set of processing solutions With fault type set Cross-combinations are performed to obtain o combinations, denoted as the clustering condition sequence. ,in, ; Construct o cache pools, denoted as the cache pool sequence: ,in, correspond , u is the counting index, ; Traversal Sequences that meet the clustering criteria Elevator maintenance information corresponding to the clustering conditions is included in the corresponding cache pool; Extract elevator maintenance information cached in each of the o cache pools, according to the cache pool sequence. The order is denoted as the elevator maintenance information clustering sequence. Among them, any cluster of elevator maintenance information contains no less than 1 elevator maintenance information and no more than j elevator maintenance information; S11. Extract the fault type set ; S12, from Extract any fault type Where k is the counting index, ; S13, Clustering sequence from elevator maintenance information Extract the fault type as The total number of elevator maintenance information is denoted as x. Let the x elevator maintenance information records be extracted in the order listed, and denote this as the elevator maintenance information filtering sequence. ,in, ; S14, from Extract all types of treatment schemes, denoted as y, and denote the extraction order as the treatment scheme selection sequence. ; S15, from Extract x elevator numbers from the list, and denote the extraction order as the elevator number filtering sequence. ; S16, from Extract x maintenance times, and denote the maintenance time screening sequence in the order of extraction. ; S17, Continuous Monitoring The corresponding x elevators, until the traction machine of the x elevators experiences another failure of type 0. The malfunction; Extract x fault occurrence times, and sort them by The order is denoted as the fault occurrence time filtering sequence. ; S18, Adopt Subtract accordingly The maintenance and upkeep time screening sequence was obtained. ; S19, Extraction Any processing scheme and from Extract all processing schemes as The maintenance and upkeep time was averaged to obtain The associated mean maintenance time is denoted as Where z is the counting index, ; Similarly, determine The average maintenance uptime associated with all treatment options, and according to The order of these sequences is denoted as the average maintenance and upkeep time series. ; S110, from The longest average maintenance duration is extracted and denoted as . ; Get Corresponding solutions ,Will As The associated best-case scenario is renamed ; S111, Repeat steps S12 to S110, and confirm. The optimal handling solutions associated with all fault types are summarized to form a sequence of optimal handling solutions. ; Step 2: Based on the elevator maintenance information of the faulty component being the traction machine, extract the operating parameters of the traction machine within a retrospective period before the fault occurred from the cloud database, construct the operating parameter sequence associated with each operating parameter, and generate the fault precursor feature group associated with all fault types of the traction machine. Step 3: Construct an elevator fault prediction model associated with the traction machine. Divide the fault precursor feature group, fault type sequence, and optimal handling solution sequence into training samples and validation samples respectively to train and validate the elevator fault prediction model. Step 4: Obtain the elevator to be monitored, construct a digital twin model of the traction machine associated with the traction machine of the elevator to be monitored, obtain the simulated operating parameters, and construct a fault precursor feature group based on the operating parameter sequence as input data. Input the model into the elevator fault prediction model, output the fault type associated with the traction machine of the elevator to be monitored, the best handling solution, and the fault occurrence time, and provide corresponding alarm reminders to the operators.

2. The fault prediction and processing method based on AI large-scale model elevator industry data according to claim 1, characterized in that, In step one, the elevator maintenance information includes the elevator number, the time of the fault occurrence, the time of the fault repair, the type of fault, the faulty component, and the handling plan.

3. The fault prediction and processing method based on AI large-scale model elevator industry data according to claim 1, characterized in that, In step two, the specific method for constructing the sequence of operating parameters associated with each operating parameter is as follows: S31. Determine the fault type The associated elevator maintenance information filtering sequence And elevator number filtering sequence ; S32, from Determine the fault occurrence time associated with x elevator numbers, and then... The order of the fault occurrences is sorted to obtain the fault occurrence time series. ; S33. Obtain the backtracking period and duration preset by the operator. ; S34, Confirm Any elevator number and elevator number The associated failure occurrence time Where v is the counting index, ; S35, with As the end time of the backtracking, we look back one backtracking cycle. The time must be traced back to the start time, recorded as... ; S36, with Extract as query criteria to The elevator number inside is The operating parameters of the elevator's traction machine; The operating parameters include the three-phase current of the traction machine, the bearing temperature of the traction machine, the vibration amplitude of the motor, and the rotational speed of the traction sheave. S37, will The associated operating parameters were serialized in chronological order to obtain the traction machine three-phase current sequence, traction machine bearing temperature sequence, motor vibration amplitude sequence, and traction sheave speed sequence, which were then summarized as follows: The associated sequence of operating parameters; S38. Repeat steps S34 to S37 to confirm. The sequence of operating parameters associated with all elevator numbers.

4. The fault prediction and processing method based on AI large-scale model elevator industry data according to claim 3, characterized in that, In step two, the specific method for generating the fault precursor feature group associated with all fault types of the traction machine is as follows: S41. Obtain elevator number The associated sequence of operating parameters; S42. Construct a two-dimensional coordinate system with the timeline as the horizontal axis and the values ​​of the running parameters as the vertical axis; The three-phase current sequence of the traction machine is plotted on a two-dimensional coordinate system to obtain several three-phase current data points of the traction machine; S43. Connect two adjacent three-phase current data points of the traction machine with short lines to obtain the three-phase current broken line of the traction machine; S44. Determine the slope between all pairs of adjacent data points in the three-phase current piecewise linear curve of the traction machine, and average them. The average value is recorded as the slope of the three-phase current change of the traction machine. ; S45. Similarly, determine the slope of the temperature change of the traction machine bearing. Slope of motor vibration amplitude change and the slope of the traction sheave speed change ; S46, will as well as The combination is recorded as the elevator number. The associated fault precursor feature vector ; S47. Repeat steps S41 to S46 to determine the elevator number filtering sequence. The fault precursor feature vectors associated with all elevator numbers are collected and summarized, and denoted as fault types. The associated fault precursor feature group , represented as: ; S48. By analogy, determine the set of fault types. The fault precursor characteristic groups associated with all fault types are denoted in sequence as follows: .

5. The fault prediction and processing method based on AI large-scale model elevator industry data according to claim 1, characterized in that, In step three, the specific method for training and validating the elevator fault prediction model is as follows: S51. Construct an elevator fault prediction model for traction machine fault prediction; S52. Extract the fault type set and its related: Fault precursor characteristic group Optimal processing sequence This is summarized and recorded as the overall dataset; S53. Randomly and evenly select 70% of the data from the overall dataset as training samples and the remaining 30% as validation samples. S54. Input the training samples into the elevator fault prediction model to perform training operations and adjust the network parameters of the elevator fault prediction model. S55. Use the verification sample to input the elevator fault prediction model to verify the prediction accuracy, and compare it with the prediction accuracy threshold preset by the operator. If the prediction accuracy exceeds the prediction accuracy threshold, the verification is successful, and the elevator fault prediction model is output. If the prediction accuracy does not exceed the prediction accuracy threshold, the verification fails. Continue the training operation until the verification is successful.

6. The fault prediction and processing method based on AI large model elevator industry data according to claim 5, characterized in that, In step four, the specific method for constructing the digital twin model of the traction machine associated with the traction machine of the elevator to be monitored is as follows: Obtain the physical structural parameters associated with the traction machine of the elevator to be monitored, including model, size, and material properties, all of which are considered known values; Extract the operating parameters associated with the traction machine of the elevator to be monitored from the cloud database; A digital twin model of the traction machine is constructed by combining the physical structural parameters and operating parameters of the traction machine of the elevator to be monitored, denoted as . ; Real-time extraction of operating parameters associated with the traction machine of the elevator to be monitored from the cloud database. Synchronize the mapping in the middle.

7. The fault prediction and processing method based on AI large model elevator industry data according to claim 6, characterized in that, In step four, the specific method for outputting the fault type, optimal handling solution, and fault occurrence time associated with the traction machine of the elevator to be monitored is as follows: Using digital twin model of traction machine The simulation runs for a future time period and extracts the simulation parameters. The duration of the simulation period is determined by the operator. The simulated operating parameters are serialized to obtain an operating parameter sequence. Based on the operating parameter sequence, a fault precursor feature group is constructed and input into the elevator fault prediction model. If no fault occurs, continue the simulation; If a fault occurs, the system will output the fault type associated with the traction machine of the elevator under monitoring, the best handling solution, and the time of the fault occurrence. The elevator number of the elevator to be monitored is then obtained and combined with the obtained fault type, optimal handling solution and fault occurrence time to provide corresponding alarm reminders to the operators as alarm information.

Images