Elevator fault intelligent maintenance processing method based on AI large model elevator industry data

Abstract

The invention discloses an elevator fault intelligent maintenance processing method based on AI large model elevator industry data, and relates to the technical field of elevator maintenance, and the method comprises the following steps: constructing a multi-dimensional elevator operation characteristic data set, and establishing an elevator fault prediction model; based on the elevator fault prediction model, the fault type, the fault probability and the fault time of the elevator in a future period of time are predicted; according to the prediction result, a maintenance strategy recommendation model is established based on the elevator use frequency and the importance degree factor, and a personalized intelligent maintenance strategy is formulated; and executing an intelligent maintenance strategy, collecting elevator operation data after maintenance, and feeding back the elevator operation data to the AI large model to optimize and update the elevator fault prediction model. According to the method, elevator faults are predicted through deep learning, personalized maintenance strategies are intelligently recommended, resource allocation is optimized, shutdown and invalid maintenance are reduced, the model is continuously optimized through incremental learning, prediction accuracy is improved, the service life of equipment is prolonged, normal operation is ensured, and potential safety hazards and maintenance cost are reduced.

Description

Technical Field

[0001] This invention relates to the field of elevator maintenance technology, specifically to an intelligent maintenance and repair method for elevator faults based on AI large-scale model elevator industry data. Background Technology

[0002] With the acceleration of modern urbanization, elevators have become an indispensable means of transportation in daily life, especially in high-rise buildings. The frequency of elevator use and operational safety are crucial to people's quality of life. Traditional elevator fault maintenance methods usually rely on manual inspection and regular maintenance. This approach has many shortcomings. For example, manual inspection depends on personnel experience and is difficult to detect potential problems in a timely and accurate manner. Regular maintenance cannot optimize maintenance plans in a targeted manner, which can easily lead to waste of resources or maintenance delays. Therefore, how to improve the accuracy of elevator fault diagnosis and maintenance efficiency has become a technical problem that the elevator industry urgently needs to solve.

[0003] In recent years, with the rapid development of artificial intelligence technology, especially the maturity of deep learning and big data analysis technology, AI-based large models have shown great potential in the field of equipment failure prediction and intelligent maintenance. By using AI large models to analyze elevator operation data in real time, and combining historical failure data and elevator usage, it is possible to accurately predict the time, type and possible causes of elevator failures. This provides a scientific basis for formulating personalized intelligent maintenance strategies. Through such intelligent maintenance methods, not only can the safety of elevator operation be improved, but operating costs can also be significantly reduced and the service efficiency of elevators improved. Summary of the Invention

[0004] To address the aforementioned technical issues, this technical solution provides an intelligent elevator fault maintenance and repair method based on AI-powered large-scale model elevator industry data. This solution resolves the problems mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Intelligent elevator fault maintenance and repair methods based on AI-powered large-scale model elevator industry data include:

[0007] S1: Obtain a multi-dimensional elevator operation feature dataset, perform standardization processing, construct a multi-dimensional elevator operation feature dataset, use deep learning algorithms in the AI ​​large model to train the multi-dimensional elevator operation feature dataset, and establish an elevator fault prediction model.

[0008] S2: Based on the elevator fault prediction model, analyze the elevator operation characteristic dataset in real time to predict the fault type, fault probability and fault time of the elevator in the future.

[0009] S3: Based on the prediction results, and taking into account factors such as elevator usage frequency and importance, establish a maintenance strategy recommendation model. Based on the output of the maintenance strategy recommendation model, formulate personalized intelligent maintenance strategies.

[0010] S4: Execute intelligent maintenance strategies and collect elevator operation data after maintenance, feeding it back to the AI ​​big data model to optimize and update the elevator fault prediction model.

[0011] Preferably, S1 specifically includes:

[0012] Elevator operation data is collected in real time by sensors built into the elevator. Historical fault data is retrieved from the elevator management system database to obtain elevator equipment parameters. Industry statistical data is collected and correlation analysis is used to evaluate the degree of correlation between each feature and elevator faults. Redundant features are removed to obtain multi-dimensional original features. Normalization is used for the multi-dimensional original features to map the data to the [0,1] interval, thus constructing a multi-dimensional elevator operation feature dataset.

[0013] A network combining the Transformer model and the multilayer perceptron was selected to process multi-dimensional time-series data of elevators and output the predicted probability of elevator fault types.

[0014] The collected elevator operation feature data is divided into training set, validation set and test set in a ratio of 7:1:2. The historical fault type is used as the label. The model parameters are adjusted using the Adam optimization algorithm. The training set data is input into the model. The fault type prediction probability is obtained through forward propagation. The difference between the prediction result and the true label is calculated using the cross-entropy loss function. The gradient is obtained through the backpropagation algorithm. The network weights are updated using the Adam algorithm. The parameters are repeatedly adjusted to minimize the loss function value.

[0015] Evaluate the trained model using a validation set to obtain accuracy, recall, and F1 score. Adjust the model based on the evaluation results. If overfitting occurs, simplify the model using Dropout. If underfitting occurs, increase the number of Transformer model layers and heads.

[0016] When the model performs best on the validation set, the test set is used for final testing. Based on the accuracy, recall, and F1 score on the test set, the actual application effect of the model is determined, and the elevator fault prediction model is established.

[0017] Preferably, the process of using a network combining a Transformer model and a multilayer perceptron to process multi-dimensional time-series elevator data and output the predicted probability of elevator fault types specifically includes:

[0018] The Transformer model uses a multi-head attention mechanism to process multi-dimensional time-series elevator data, identify the correlations between data, and then further process the data through a feedforward neural network. The processed data is then input into an MLP, where the features of the input data are transformed and fused at different dimensions and levels. After nonlinear transformation and classification through a fully connected layer, the model outputs the predicted probability of elevator fault types.

[0019] Preferably, the step of updating network weights using the Adam algorithm and repeatedly adjusting parameters to minimize the loss function value specifically includes:

[0020] The Adam optimization algorithm is initialized, the momentum parameter is set, the training set data is input into the built model, the data is passed through the Transformer model and MLP for forward propagation, and finally the elevator fault type probability distribution output by the model is calculated.

[0021] The cross-entropy loss function is used to calculate the gap between the model's prediction and the actual target. The loss value reflects the accuracy of the model's prediction; the smaller the loss value, the more accurate the prediction.

[0022] The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm to obtain the direction of parameter adjustment. Based on the direction of parameter adjustment, the parameters are updated using the Adam algorithm.

[0023] Preferably, S2 specifically includes:

[0024] Real-time collection of multi-dimensional characteristic data of the elevator's current operation, including but not limited to real-time operating speed, car load changes, door operator action time, traction machine current and temperature data;

[0025] The real-time collected data is standardized and mapped to the [[0,1] interval using the same normalization method as in S101 to form a real-time feature dataset;

[0026] The real-time feature dataset is divided into input sequences according to a 10-minute time window and then fed into the elevator fault prediction model.

[0027] The model captures the correlation between real-time data and historical features through the Transformer's multi-head attention mechanism, and outputs the set of fault types, the probability of occurrence of each fault type, and the time interval in which the fault may occur within a preset time period through MLP.

[0028] The output results are filtered by a threshold. When the predicted probability of a fault type exceeds 80%, it is marked as a high-risk fault type and an early warning is triggered.

[0029] Preferably, S3 specifically includes:

[0030] An elevator usage frequency assessment model was constructed, and the usage frequency index was obtained by weighted calculation based on the elevator's historical number of runs, peak-hour operation density, and average daily number of starts and stops.

[0031] An elevator importance assessment model was constructed, and an importance index was obtained by combining the building type where the elevator is located, the number of floors it serves, and historical passenger flow data through the analytic hierarchy process.

[0032] Using fault prediction results, usage frequency index, and importance index as inputs, a maintenance strategy recommendation model is constructed, and a fuzzy decision algorithm is used to output three types of strategies.

[0033] Preferably, the construction of the elevator importance assessment model, which combines the building type, the number of floors served by the elevator, and historical passenger flow data, and obtains the importance index through the analytic hierarchy process, specifically includes:

[0034] An evaluation index system for the importance of elevators was obtained based on correlation analysis, including the weight of building type, the weight of the number of floors served, and the weight of historical passenger flow.

[0035] Building types are classified and assigned values, and an importance index is calculated using the analytic hierarchy process (AHP). The AHP formula is as follows:

[0036] I = W1·c + W2·f + W1·s

[0037] In the formula, I is the importance index, c is the building type score, W1 is the weight coefficient of the building type score, f is the floor score, W2 is the weight coefficient of the floor score, and s is the weight coefficient of the pedestrian flow score, where the value of I ranges from [1,5].

[0038] Preferably, the step of using fault prediction results, usage frequency index, and importance index as inputs to construct a maintenance strategy recommendation model, and employing a fuzzy decision algorithm to output three types of strategies, specifically includes:

[0039] The fault probability, usage frequency index, and importance index in the fault prediction results are used as input variables for fuzzy decision-making, and a fuzzy subset is set.

[0040] Failure probability P: {low, medium, high}, corresponding to intervals [0, 30%], [30%, 70%], [70%, 100%];

[0041] Frequency of use F: {low, medium, high}, corresponding to the intervals [0, 0.4], [0.4, 0.7], and [0.7, 1].

[0042] Importance level I: {low, medium, high}, corresponding to intervals [1,2], [2,4], [4,5];

[0043] Establish a fuzzy rule base, including core rules:

[0044] If P = high and I = high, then output "emergency maintenance".

[0045] If P = Medium and F = High, then output the planned maintenance.

[0046] If P = low or I = low, then output "normal maintenance".

[0047] Through fuzzy reasoning and defuzzification, a maintenance strategy is generated, including maintenance priorities, on-site service time windows, required spare parts lists, and technical personnel level requirements.

[0048] Preferably, S4 specifically includes:

[0049] Work orders are generated according to personalized intelligent maintenance strategies, maintenance tasks, time nodes and responsible persons are obtained, and maintenance personnel are dispatched through the maintenance management system.

[0050] When maintenance personnel perform maintenance operations, they record the fault handling process, the models and parameters of the replaced parts, and the elevator operating status data before and after maintenance.

[0051] After maintenance is completed, sensors are used to collect 48 hours of continuous operation data for the elevator after it resumes operation, including operational stability indicators, temperature changes of key components, and energy consumption data.

[0052] The maintenance record data and the operation data are integrated and marked as maintained samples. They are then added to the original training set at a rate of 10%, while the original ratio of the validation set and the test set is retained.

[0053] The updated dataset is input into the AI ​​model to retrain the elevator fault prediction model. The incremental learning method is used to freeze the parameters of the underlying network and only update the weights of the top classifier, thereby shortening the model iteration cycle and achieving dynamic optimization of the model.

[0054] Preferably, during the incremental learning process, a model performance evaluation threshold is set. When the F1 score of the updated model on the test set is less than 3% higher than that of the original model, a model structure optimization mechanism is triggered. The structure optimization mechanism specifically includes increasing the number of heads in the multi-head attention mechanism of the Transformer model and adding 10% more neurons in the hidden layer of the MLP, retraining the model, and comparing the performance.

[0055] At the same time, stratified sampling was conducted on the maintained samples according to the type of fault.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] This invention proposes to use deep learning algorithms to analyze multi-dimensional elevator operation data in real time, predict potential faults, reduce the probability of downtime, and intelligently recommend personalized maintenance strategies based on factors such as fault prediction, usage frequency, and importance. This optimizes resource allocation, reduces ineffective maintenance, accurately predicts faults, lowers maintenance costs, and reduces downtime losses. Incremental learning and feedback mechanisms continuously optimize the prediction model, improving accuracy and intelligence, thereby ensuring normal elevator operation, extending service life, improving user experience, and reducing safety hazards. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating the steps of the present invention.

[0059] Figure 2 This is a flowchart of steps S1 in this invention;

[0060] Figure 3 This is a flowchart of steps S2 in this invention;

[0061] Figure 4 This is a flowchart of step S3 in this invention;

[0062] Figure 5 This is a flowchart of step S4 in this invention. Detailed Implementation

[0063] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0064] Reference Figure 1 As shown, the intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data includes:

[0065] S1: Obtain a multi-dimensional elevator operation feature dataset, perform standardization processing, construct a multi-dimensional elevator operation feature dataset, use deep learning algorithms in AI large models to train the multi-dimensional elevator operation feature dataset, and establish an elevator fault prediction model.

[0066] S2: Based on the elevator fault prediction model, analyze the elevator operation characteristic dataset in real time to predict the fault type, fault probability and fault time of the elevator in the future.

[0067] S3: Based on the prediction results, and taking into account factors such as elevator usage frequency and importance, establish a maintenance strategy recommendation model. Based on the output of the maintenance strategy recommendation model, formulate personalized intelligent maintenance strategies.

[0068] S4: Execute intelligent maintenance strategies and collect elevator operation data after maintenance, feeding it back to the AI ​​big data model to optimize and update the elevator fault prediction model.

[0069] Reference Figure 2 As shown, S1 specifically includes:

[0070] S101: Real-time elevator operation data is collected through the built-in sensors of the elevator, historical fault data is retrieved from the elevator management system database, elevator equipment parameters are obtained, industry statistical data are collected, correlation analysis is used to evaluate the degree of correlation between each feature and elevator faults, redundant features are removed, multi-dimensional original features are obtained, and a normalization method is used for the multi-dimensional original features to map the data to the [0,1] interval, thus constructing a multi-dimensional elevator operation feature dataset.

[0071] The formula for correlation analysis is as follows:

[0072]

[0073] In the formula, g is the correlation index, and X i This represents the actual value at time i in the elevator feature data. Y is the average value of the elevator feature data. i This represents the actual value of the elevator fault data at time i. is the average value of elevator malfunction data, and n is the total number of time periods;

[0074] The value of g ranges from -1 to 1. When g = 1, it indicates that there is a completely positive linear correlation between variables x and y. When g = -1, it indicates that there is a completely negative linear correlation. When g = 0, it indicates that there is no linear correlation between variables x and y. The closer the absolute value of g is to 1, the stronger the linear correlation between the elevator's operating characteristics and the fault. The closer it is to 0, the weaker the linear correlation.

[0075] S102: A network combining the Transformer model and the multilayer perceptron is used to process multi-dimensional time series data of elevators and output the predicted probability of elevator fault types.

[0076] S103: Divide the collected elevator operation feature data into training set, validation set and test set in a ratio of 7:1:2. Use historical fault types as labels, adjust the model parameters using the Adam optimization algorithm, input the training set data into the model, obtain the fault type prediction probability through forward propagation, calculate the difference between the prediction result and the true label using the cross-entropy loss function, obtain the gradient through the backpropagation algorithm, update the network weights using the Adam algorithm, and repeatedly adjust the parameters to minimize the loss function value.

[0077] S104: Evaluate the trained model using the validation set to obtain accuracy, recall, and F1 score. Adjust the model based on the evaluation results. If overfitting occurs, simplify the model using Dropout. If underfitting occurs, increase the number of Transformer model layers and heads.

[0078] S105: When the model performs best on the validation set, use the test set for final testing. Based on the accuracy, recall, and F1 score on the test set, determine the actual application effect of the model and complete the establishment of the elevator fault prediction model.

[0079] By analyzing elevator operation data in real time and using the multi-head attention mechanism of the Transformer model to capture complex correlations in time-series data, this method can accurately predict the future fault types and their probability of occurrence of elevators in a short period of time. The core of this method lies in providing dynamic and accurate fault prediction based on the combination of real-time data and historical data, which significantly improves the fault early warning capability.

[0080] S102 specifically includes:

[0081] The Transformer model uses a multi-head attention mechanism to process multi-dimensional time-series elevator data, find the correlation between data, and then further process it through a feedforward neural network. The processed data is input into an MLP, which transforms and fuses the features of the input data at different dimensions and levels. After nonlinear transformation and classification through a fully connected layer, the model outputs the predicted probability of elevator fault types.

[0082] The MLP calculation formula is as follows:

[0083] H (i) =σ(W (i) H (i-1) +b (i) )

[0084] In the formula, H (i) The output of the i-th hidden layer in the MLP is the feature vector obtained after computation by the i-th hidden layer, where σ is the ReLU activation function and W is the output of the feature vector. (i) Let H be the weight matrix of the i-th hidden layer. (i-1) For the output of the (i-1)th hidden layer, b (i) Let i be the bias vector of the i-th hidden layer;

[0085] By combining the Transformer model and MLP to process multi-dimensional time-series data of elevators, and using a multi-head attention mechanism to mine long-term dependencies in the data, the accuracy of the model in predicting elevator fault types is further improved. This combination mode enhances the fault prediction model's ability to process complex time-series data.

[0086] S103 specifically includes:

[0087] The Adam optimization algorithm is initialized, the momentum parameter is set, the training set data is input into the built model, the data is passed through the Transformer model and MLP for forward propagation, and finally the elevator fault type probability distribution output by the model is calculated.

[0088] The cross-entropy loss function is used to calculate the gap between the model's prediction and the actual target. The loss value reflects the accuracy of the model's prediction; the smaller the loss value, the more accurate the prediction.

[0089] The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm to obtain the direction of parameter adjustment. Based on the direction of parameter adjustment, the parameters are updated using the Adam algorithm, where the Adam optimization algorithm formula is:

[0090]

[0091] In the formula, m t For first-order moment estimation, m t-1 Let g represent the first-moment estimate at the (t-1)th iteration, β1 be the exponential decay rate of the first-moment estimate, and g be the first-moment estimate. t It is the gradient of the current iteration, v t For second-order moment estimation, v t-1 Let represent the second-order moment estimate at the (t-1)th iteration, and β2 be the exponential decay rate of the second-order moment estimate. This represents the square of the gradient. This is the first-order moment estimate after bias correction. This is the second-order moment estimate after bias correction, where t is the current iteration number and θ is the second-order moment estimate. t Let τ be the model parameters at the t-th iteration, ε be the learning rate, ∈ be a constant, and θ be the model parameters at the t-th iteration. t+1 These are the model parameters at the (t+1)th iteration after the t-th iteration update;

[0092] By employing appropriate dataset partitioning and the Adam optimization algorithm, model training becomes more efficient. In particular, the introduction of the cross-entropy loss function during training to adjust model parameters, and further improvement of prediction accuracy through backpropagation and parameter updates, optimizes the elevator fault prediction training process, ensuring the model's reliability in practical applications.

[0093] Reference Figure 3 As shown, S2 specifically includes:

[0094] S201: Real-time collection of multi-dimensional characteristic data of the elevator's current operation, including but not limited to real-time operating speed, car load changes, door operator action time, traction machine current and temperature data;

[0095] S202: Standardize the real-time collected data and map it to the [[0,1] interval using the same normalization method as in S101 to form a real-time feature dataset;

[0096] S203: Divide the real-time feature dataset into input sequences according to a 10-minute time window and input them into the elevator fault prediction model;

[0097] S204: The model captures the correlation between real-time data and historical features through the multi-head attention mechanism of Transformer, and outputs the set of fault types, the probability of occurrence of each fault type, and the time interval in which the fault may occur within a preset time period through MLP.

[0098] S205: Perform threshold filtering on the output results. When the predicted probability of a fault type exceeds 80%, mark it as a high-risk fault type and trigger an early warning.

[0099] Reference Figure 4 As shown, S3 specifically includes:

[0100] S301: Construct an elevator usage frequency assessment model. Based on the elevator's historical number of runs, peak-hour operating density, and average daily start-stop frequency, a usage frequency index is calculated through weighted averages. The formula for calculating the frequency index is as follows:

[0101] F = α·K J +δ·D L

[0102] In the formula, α and δ are weighting coefficients (α + δ = 1), K J D is the standardized value of the average number of start-stop cycles per day. L Standardized value for operating density during peak hours;

[0103] S302: Construct an elevator importance assessment model, combining the building type where the elevator is located, the number of floors it serves, and historical passenger flow data, and obtain the importance index through the analytic hierarchy process;

[0104] S303: Using fault prediction results, usage frequency index F, and importance index I as inputs, a maintenance strategy recommendation model is constructed, and a fuzzy decision algorithm is used to output three types of strategies.

[0105] S302 specifically includes:

[0106] An evaluation index system for the importance of elevators was obtained based on correlation analysis, including the weight of building type, the weight of the number of floors served, and the weight of historical passenger flow.

[0107] Building types are classified and assigned values, and an importance index is calculated using the analytic hierarchy process (AHP). The AHP formula is as follows:

[0108] I = W1·c + W2·f + W1·s

[0109] In the formula, I is the importance index, c is the building type score, W1 is the weight coefficient of the building type score, f is the floor score, W2 is the weight coefficient of the floor score, and s is the weight coefficient of the pedestrian flow score. The value range of I is [1,5].

[0110] The elevator importance assessment model, built based on building type, number of floors served, and historical passenger flow data, calculates the elevator importance index using the analytic hierarchy process. This innovation combines the environment served by the elevator with actual usage, providing a scientific basis for customizing maintenance strategies.

[0111] S303 specifically includes:

[0112] The fault probability, usage frequency index, and importance index in the fault prediction results are used as input variables for fuzzy decision-making, and a fuzzy subset is set.

[0113] Failure probability P: {low, medium, high}, corresponding to intervals [0, 30%], [30%, 70%], [70%, 100%];

[0114] Frequency of use F: {low, medium, high}, corresponding to the intervals [0, 0.4], [0.4, 0.7], and [0.7, 1].

[0115] Importance level I: {low, medium, high}, corresponding to intervals [1,2], [2,4], [4,5];

[0116] Establish a fuzzy rule base, including core rules:

[0117] If P = high and I = high, then output "emergency maintenance".

[0118] If P = Medium and F = High, then output the planned maintenance.

[0119] If P = low or I = low, then output "normal maintenance".

[0120] Through fuzzy reasoning and defuzzification, a maintenance strategy is generated, including maintenance priority, on-site service time window, required spare parts list and technician level requirements.

[0121] In the maintenance strategy recommendation process, a fuzzy decision algorithm is used to process the probability of fault type, the frequency of use index, and the importance index, and generate corresponding maintenance strategies. This method can accurately output emergency, planned, or routine maintenance solutions according to different situations, thereby improving the automation and intelligence of maintenance decision-making.

[0122] Reference Figure 5 As shown, S4 specifically includes:

[0123] S401: Generates work orders according to personalized intelligent maintenance strategies, obtains maintenance tasks, time nodes and responsible persons, and dispatches maintenance personnel through the maintenance management system;

[0124] S402: When performing maintenance operations, maintenance personnel shall record the fault handling process, the model and parameters of the replaced parts, and the elevator operating status data before and after maintenance.

[0125] S403: After maintenance is completed, sensors are used to collect continuous 48-hour operation data of the elevator after it resumes operation, including operation stability indicators, temperature changes of key components, and energy consumption data;

[0126] S404: Integrate maintenance record data with operational data, mark them as maintained samples, and add them to the original training set at a ratio of 10%, while retaining the original ratio of validation set and test set.

[0127] S405: Input the updated dataset into the AI ​​large model, retrain the elevator fault prediction model, freeze the underlying network parameters using incremental learning, update only the weights of the top classifier, shorten the model iteration cycle, and achieve dynamic model optimization.

[0128] By introducing an incremental learning mechanism, the elevator fault prediction model is updated after maintenance data feedback. This not only ensures real-time performance and efficiency but also avoids the waste of time and resources from retraining each time. Through sampling and dynamic optimization of the maintained samples, the system can continuously evolve and maintain prediction accuracy.

[0129] During incremental learning, a model performance evaluation threshold is set. When the F1 score of the updated model on the test set is less than 3% higher than that of the original model, the model structure optimization mechanism is triggered. The structure optimization mechanism specifically includes increasing the number of heads in the multi-head attention mechanism of the Transformer model and adding 10% more neurons in the hidden layer of the MLP. The model is then retrained and its performance is compared. At the same time, the maintained samples are stratified and sampled according to the fault type.

[0130] In summary, the advantages of this invention are as follows:

[0131] By collecting multi-dimensional elevator operation data and using deep learning algorithms to train an elevator fault prediction model, the elevator's operating status can be analyzed in real time, thereby predicting potential fault types and occurrence times in advance and reducing the probability of elevator shutdown.

[0132] Based on fault prediction results and multiple factors such as elevator usage frequency and importance, intelligent maintenance strategies are recommended. Personalized maintenance plans are formulated according to different fault types and elevator operating conditions, which not only ensures the timeliness of maintenance, but also minimizes ineffective maintenance and optimizes resource allocation.

[0133] By accurately predicting elevator malfunctions, unnecessary periodic inspections and excessive maintenance are avoided, reducing the workload of maintenance personnel and equipment replacement costs. Furthermore, intelligent maintenance can reduce downtime losses caused by elevator malfunctions, further saving on enterprise operating costs.

[0134] Through incremental learning and feedback mechanisms, elevator operation data after maintenance is continuously fed back to the AI ​​big model, continuously optimizing the elevator fault prediction model and improving its accuracy and practicality. In addition, by dynamically optimizing the model structure, the accuracy of elevator fault prediction and the intelligence level of maintenance strategies are continuously improved.

[0135] Intelligent fault prediction and maintenance of elevators can greatly reduce the probability of elevator malfunctions, ensure the normal operation of elevators, and improve the user experience. At the same time, intelligent maintenance can ensure that elevators are in optimal working condition, extend the service life of equipment, and reduce the threat of sudden malfunctions to passenger safety.

[0136] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for intelligent maintenance and repair of elevator faults based on AI large-scale model elevator industry data, characterized in that: include: S1: Obtain a multi-dimensional elevator operation feature dataset, perform standardization processing, construct a multi-dimensional elevator operation feature dataset, use deep learning algorithms in AI large models to train the multi-dimensional elevator operation feature dataset, and establish an elevator fault prediction model. S2: Based on the elevator fault prediction model, analyze the elevator operation characteristic dataset in real time to predict the fault type, fault probability and fault time of the elevator in the future. S3: Based on the prediction results, and taking into account factors such as elevator usage frequency and importance, establish a maintenance strategy recommendation model. Based on the output of the maintenance strategy recommendation model, formulate personalized intelligent maintenance strategies. S4: Execute intelligent maintenance strategies and collect elevator operation data after maintenance, feeding it back to the AI ​​big data model to optimize and update the elevator fault prediction model.

2. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 1, characterized in that, S1 specifically includes: S101: Real-time elevator operation data is collected through the built-in sensors of the elevator, historical fault data is retrieved from the elevator management system database, elevator equipment parameters are obtained, industry statistical data are collected, correlation analysis is used to evaluate the degree of correlation between each feature and elevator faults, redundant features are removed, multi-dimensional original features are obtained, and a normalization method is used for the multi-dimensional original features to map the data to the [0,1] interval, thus constructing a multi-dimensional elevator operation feature dataset. S102: A network combining the Transformer model and the multilayer perceptron is used to process multi-dimensional time series data of elevators and output the predicted probability of elevator fault types. S103: Divide the collected elevator operation feature data into training set, validation set and test set in a ratio of 7:1:

2. Use historical fault types as labels, adjust the model parameters using the Adam optimization algorithm, input the training set data into the model, obtain the fault type prediction probability through forward propagation, calculate the difference between the prediction result and the true label using the cross-entropy loss function, obtain the gradient through the backpropagation algorithm, update the network weights using the Adam algorithm, and repeatedly adjust the parameters to minimize the loss function value. S104: Evaluate the trained model using the validation set to obtain accuracy, recall, and F1 score. Adjust the model based on the evaluation results. If overfitting occurs, simplify the model using Dropout. If underfitting occurs, increase the number of Transformer model layers and heads. S105: When the model performs best on the validation set, use the test set for final testing. Based on the accuracy, recall, and F1 score on the test set, determine the actual application effect of the model and complete the establishment of the elevator fault prediction model.

3. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 2, characterized in that, S102 specifically includes: The Transformer model uses a multi-head attention mechanism to process multi-dimensional time-series elevator data, identify the correlations between data, and then further process the data through a feedforward neural network. The processed data is then input into an MLP, where the features of the input data are transformed and fused at different dimensions and levels. After nonlinear transformation and classification through a fully connected layer, the model outputs the predicted probability of elevator fault types.

4. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 3, characterized in that, S103 specifically includes: The Adam optimization algorithm is initialized, the momentum parameter is set, the training set data is input into the built model, the data is passed through the Transformer model and MLP for forward propagation, and finally the elevator fault type probability distribution output by the model is calculated. The cross-entropy loss function is used to calculate the gap between the model's prediction and the actual target. The loss value reflects the accuracy of the model's prediction; the smaller the loss value, the more accurate the prediction. The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm to obtain the direction of parameter adjustment. Based on the direction of parameter adjustment, the parameters are updated using the Adam algorithm.

5. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 1, characterized in that, S2 specifically includes: S201: Real-time collection of multi-dimensional characteristic data of the elevator's current operation, including but not limited to real-time operating speed, car load changes, door operator action time, traction machine current and temperature data; S202: Standardize the real-time collected data and map it to the [[0,1] interval using the same normalization method as in S101 to form a real-time feature dataset; S203: Divide the real-time feature dataset into input sequences according to a 10-minute time window and input them into the elevator fault prediction model; S204: The model captures the correlation between real-time data and historical features through the multi-head attention mechanism of Transformer, and outputs the set of fault types, the probability of occurrence of each fault type, and the time interval in which the fault may occur within a preset time period through MLP. S205: Perform threshold filtering on the output results. When the predicted probability of a fault type exceeds 80%, mark it as a high-risk fault type and trigger an early warning.

6. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 1, characterized in that, S3 specifically includes: S301: Construct an elevator usage frequency assessment model, and obtain the usage frequency index by weighted calculation based on the elevator's historical number of runs, peak-hour operation density, and average daily number of starts and stops. S302: Construct an elevator importance assessment model, combining the building type where the elevator is located, the number of floors it serves, and historical passenger flow data, and obtain the importance index through the analytic hierarchy process; S303: Using fault prediction results, usage frequency index, and importance index as inputs, a maintenance strategy recommendation model is constructed, and a fuzzy decision algorithm is used to output three types of strategies.

7. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 6, characterized in that, S302 specifically includes: An evaluation index system for the importance of elevators was obtained based on correlation analysis, including the weight of building type, the weight of the number of floors served, and the weight of historical passenger flow. Building types are classified and assigned values, and an importance index is calculated using the analytic hierarchy process (AHP). The AHP formula is as follows: I = W1·c + W2·f + W1·s In the formula, I is the importance index, c is the building type score, W1 is the weight coefficient of the building type score, f is the floor score, W2 is the weight coefficient of the floor score, and s is the weight coefficient of the pedestrian flow score.

8. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data as described in claim 6, characterized in that, Specifically, S303 includes: The fault probability, usage frequency index, and importance index in the fault prediction results are used as input variables for fuzzy decision-making, and a fuzzy subset is set. Through fuzzy reasoning and defuzzification, a maintenance strategy is generated, including maintenance priorities, on-site service time windows, required spare parts lists, and technical personnel level requirements.

9. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data according to claim 1, characterized in that, S4 specifically includes: S401: Generates work orders according to personalized intelligent maintenance strategies, obtains maintenance tasks, time nodes and responsible persons, and dispatches maintenance personnel through the maintenance management system; S402: When performing maintenance operations, maintenance personnel shall record the fault handling process, the model and parameters of the replaced parts, and the elevator operating status data before and after maintenance. S403: After maintenance is completed, sensors are used to collect continuous 48-hour operation data of the elevator after it resumes operation, including operation stability indicators, temperature changes of key components, and energy consumption data; S404: Integrate maintenance record data with operational data, mark them as maintained samples, and add them to the original training set at a ratio of 10%, while retaining the original ratio of validation set and test set. S405: Input the updated dataset into the AI ​​large model, retrain the elevator fault prediction model, freeze the underlying network parameters using incremental learning, update only the weights of the top classifier, shorten the model iteration cycle, and achieve dynamic model optimization.

10. The intelligent elevator fault maintenance and repair method based on AI large-scale model elevator industry data according to claim 9, characterized in that: During the incremental learning process, a model performance evaluation threshold is set. When the F1 score of the updated model on the test set is less than 3% higher than that of the original model, a model structure optimization mechanism is triggered. The structure optimization mechanism specifically includes increasing the number of heads in the multi-head attention mechanism of the Transformer model and adding 10% more neurons in the hidden layer of the MLP, retraining the model, and comparing the performance. At the same time, stratified sampling was conducted on the maintained samples according to the type of fault.

Images