Elevator dynamic maintenance period optimization method and system combined with Internet of Things analysis

Abstract

The invention relates to the field of elevator maintenance, and discloses an elevator dynamic maintenance period optimization method and system combined with Internet of Things analysis, and the method comprises the following steps: collecting multi-modal data of an elevator in a manner of deploying a heterogeneous sensor array, and determining elevator part damage root cause associated nodes; and the aim of constructing an elevator fault prediction model is fulfilled by combining a federal learning mode. And through the elevator fault prediction model, risk simulation analysis is carried out in combination with a parallel Monte Carlo simulation method, and a periodic optimization optimal solution of elevator dynamic maintenance is output according to an analysis result. According to the method, the continuity and intelligence of data inspection can be realized in combination with an advanced data analysis algorithm, the operation and maintenance cost is remarkably reduced, and the resource utilization efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of elevator maintenance, and in particular to a method and system for optimizing the dynamic maintenance cycle of elevators by incorporating Internet of Things (IoT) analysis. Background Technology

[0002] Elevators are maintained at fixed intervals regardless of their actual operating condition. However, over-maintenance, such as frequent disassembly and inspection of healthy components, not only wastes manpower, resources, and spare parts but may also introduce new malfunction risks due to improper operation. Conversely, under-maintenance can occur. For example, elevators operating in high-intensity, harsh environments may not have their fixed maintenance cycles cover rapid component wear, leading to malfunctions within the maintenance intervals and significant safety hazards. In subsequent repairs, malfunctions can occur suddenly during operation, potentially causing entrapment or even more serious accidents. Small problems can escalate into major malfunctions, drastically increasing repair costs and extending downtime, disrupting normal user experience. Therefore, the core problems of the traditional elevator maintenance model are unscientific resource allocation, an inability to balance safety and economy, and a lack of data-driven decision-making.

[0003] Optimizing elevator dynamic maintenance cycles using the Internet of Things (IoT) essentially drives the transformation of elevator maintenance towards predictive maintenance. Timely maintenance before components enter a period of severe wear or failure prevents malfunctions and avoids unnecessary, premature intervention, thus saving costs. Predictive maintenance can also eliminate faults in their early stages, significantly reducing the probability of sudden elevator stops and ensuring smooth building traffic. The IoT can continuously collect elevator operation status data and, combined with advanced data analysis algorithms, achieve continuous and intelligent data inspection, significantly reducing operation and maintenance costs and improving resource utilization efficiency. Summary of the Invention

[0004] This invention overcomes the shortcomings of existing technologies and provides a method and system for optimizing the dynamic maintenance cycle of elevators by combining Internet of Things (IoT) analysis.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for optimizing the dynamic maintenance cycle of elevators by incorporating Internet of Things (IoT) analysis, comprising the following steps: A heterogeneous sensor array is deployed in the elevator to collect and synchronize multimodal data of the elevator, and the multimodal data of the elevator is analyzed by a causal discovery algorithm to trace the key root cause nodes of the elevator. By combining the federated learning model, the target elevator and the data stored in the elevator database are jointly trained to build an elevator fault prediction model. By combining an elevator failure prediction model with a digital twin sand table simulation method and a parallel Monte Carlo simulation method, risk simulation analysis is conducted, and different Monte Carlo simulation results are output. Based on different Monte Carlo simulation results, the optimal solution for the cycle optimization of elevator dynamic maintenance is output.

[0006] Furthermore, in a preferred embodiment of the present invention, the deployment of a heterogeneous sensor array in the elevator for collecting and synchronizing multimodal elevator data, and analyzing the multimodal elevator data using a causal discovery algorithm to trace the key root cause nodes of the elevator, specifically includes: The elevators that need maintenance are identified and designated as target elevators. A heterogeneous sensor array is then deployed on the target elevators. The heterogeneous sensor array includes sensors that collect elevator data in different modes, including sensors corresponding to vibration mode, acoustic mode, visual mode, and electrical power mode. Time synchronization is performed on elevator data of all modes collected by the heterogeneous sensor array, and timestamps are constructed. The elevator data is mapped to the timestamps. At the same time, the spatial coordinates of elevator data of different modes are determined to generate multimodal spatiotemporal synchronized elevator data. High-dimensional spatiotemporal features are extracted from multimodal spatiotemporal synchronous elevator data, including time-domain features, frequency-domain features, visual features, acoustic features, and electrical parameters. At the same time, data preprocessing of high-dimensional spatiotemporal features is required, and the preprocessed high-dimensional spatiotemporal features are combined to construct a high-dimensional spatiotemporal feature pool. A causal discovery algorithm engine is introduced, wherein the causal discovery algorithm engine is a dual engine, including a conditional independence test engine and a continuous optimization algorithm engine based on NOTEARS. Within a high-dimensional spatiotemporal feature pool, a causal discovery algorithm engine is used to construct the causal graph of the target elevator and trace the key root cause nodes of the elevator.

[0007] Furthermore, in a preferred embodiment of the present invention, the step of constructing a causal graph of the target elevator's operation within a high-dimensional spatiotemporal feature pool based on a causal discovery algorithm engine, and tracing the elevator's key root cause nodes, specifically involves: Within the high-dimensional spatiotemporal feature pool, the conditional independence test engine is used to test whether different high-dimensional spatiotemporal features are independent. If they are, then there is no direct causal relationship between the corresponding high-dimensional spatiotemporal features. If not, then check whether the other high-dimensional spatiotemporal features are independent of the two previously checked high-dimensional spatiotemporal features. After all high-dimensional spatiotemporal features have been checked, generate a network structure that describes the direct causal relationship between all high-dimensional spatiotemporal features and label it as a type of running causal graph. By using the NOTEARS-based continuous optimization algorithm engine and combining it with a high-dimensional spatiotemporal feature pool to construct a weighted adjacency matrix, the causal influence strength between different high-dimensional spatiotemporal features is iteratively calculated in the weighted adjacency matrix using a loss function combined with gradient descent until the number of iterations equals the preset number of iterations. The iterated weighted adjacency matrix is ​​then labeled as a binary running causal graph. By fusing the first-class and second-class operational causal graphs, a fused operational causal graph of the target elevator is output and labeled as the target operational causal graph. The target operational causal graph describes the causal relationships between different high-dimensional spatiotemporal features. Within the target causal graph, high-dimensional spatiotemporal features are treated as nodes. The directed edges connected to different nodes are analyzed, and nodes with more than a preset number of directed edges are marked as key root cause nodes. The key root cause node is the high-dimensional spatiotemporal feature corresponding to the probability of failure in the target elevator being higher than a predetermined value.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of combining federated learning mode to jointly train the target elevator and the data stored in the elevator database to construct an elevator fault prediction model specifically involves: A database for data connection and data storage with the target elevator is obtained and labeled as an elevator database. The elevator database not only stores elevator data of all modes collected by the heterogeneous sensor array on the target elevator, but also stores elevator data of all modes of other elevators. An LSTM network model is introduced and combined with an elevator database to perform lateral federated learning on elevator data of all modalities of the target elevator in the elevator database, thereby constructing a federated learning system. The horizontal federated learning method involves randomly sampling data from an elevator database and importing it into the input layer of the LSTM network model. The random sampling targets elevator data of all modes of other elevators stored in the elevator database. In the training layer, the input layer data is weighted and trained locally. The local training is performed by constructing a loss function and simultaneously iterating the mean squared error and cross-entropy of the input layer data. An L2 regularization term is introduced into the loss function to prevent the LSTM network model from training bias during the local training process. When the number of synchronous iterations equals a preset value, the horizontal federated learning is stopped, and the LSTM network model after horizontal federated learning is obtained and labeled as an elevator fault prediction model. The elevator fault prediction model can predict the remaining lifespan of different components of the target elevator and the probability of failure of different components by analyzing elevator data of all modes of the target elevator, that is, analyzing the key root cause nodes in the target elevator.

[0009] Furthermore, in a preferred embodiment of the present invention, the step of combining the elevator fault prediction model with risk simulation analysis through digital twin sand table simulation and parallel Monte Carlo simulation method, and outputting different Monte Carlo simulation results, specifically includes: Obtain the direct costs of different components of the target elevator, and calculate the indirect costs during the failure of different components of the target elevator; Direct costs include component prices and labor costs, while indirect costs include the value of energy resources lost during periods of no failure. Obtain resource data corresponding to the target elevator, including human resources data, spare parts resource data, and tool resource data; Based on the cost and corresponding resource data of different components of the target elevator, an economic-resource correlation model is constructed, and an elevator failure prediction model is simultaneously correlated within the economic-resource correlation model. The output is a simulation model that can be used for risk simulation analysis through digital twin sand table simulation, which is calibrated as a risk simulation analysis model. When the risk simulation analysis model is run, and the elevator failure prediction model in the risk simulation analysis model predicts that there is a faulty elevator component, the corresponding cost and resource consumption data when the component is damaged are generated, the component loss correlation data are combined, and all maintenance strategies for the corresponding component loss are retrieved in the big data network, and a maintenance strategy space is generated. Parallel Monte Carlo simulations are performed within the maintenance strategy space. These simulations simulate the execution of all maintenance strategies for component loss and, in conjunction with component loss correlation data, output different Monte Carlo simulation results. These results include strategy costs, comprehensive risk indicators, and resource utilization rates.

[0010] Furthermore, in a preferred embodiment of the present invention, the step of outputting the optimal solution for the cycle optimization of elevator dynamic maintenance based on different Monte Carlo simulation results specifically includes: Preset the cost threshold, comprehensive risk index threshold, and resource utilization threshold for maintenance of components that are predicted to fail; Among all Monte Carlo simulation results, the Monte Carlo simulation results in which the strategy cost, comprehensive risk index and resource utilization rate are all maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold are selected and marked as candidate Monte Carlo simulation results. If the number of candidate Monte Carlo simulation results is 1, then the corresponding maintenance strategy output from the candidate Monte Carlo simulation results is selected and marked as the optimal solution for the dynamic maintenance cycle optimization of the corresponding component. If the quantity is 0, select the candidate Monte Carlo simulation result with the highest strategy cost, comprehensive risk index and resource utilization rate that are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and output its corresponding maintenance strategy. If there are more than one candidate Monte Carlo simulation result whose strategy cost, comprehensive risk index and resource utilization rate are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and the number is equal, then the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost shall be selected first. If the number is greater than 1, the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost will be selected first.

[0011] The second aspect of this invention also provides an elevator dynamic maintenance cycle optimization system incorporating Internet of Things (IoT) analysis. This system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory consisting of a DDR4 RDIMM memory module with ECC verification and an NVMe solid-state storage array using 3D NAND flash memory, as well as a multi-core processor based on the Zen4 microarchitecture. The memory contains a program for optimizing the elevator dynamic maintenance cycle, which has an optimization engine. When this program is executed in parallel via a superscalar pipeline execution unit within the processor, the following steps are implemented: A heterogeneous sensor array is deployed in the elevator to collect and synchronize multimodal data of the elevator, and the multimodal data of the elevator is analyzed by a causal discovery algorithm to trace the key root cause nodes of the elevator. By combining the federated learning model, the target elevator and the data stored in the elevator database are jointly trained to build an elevator fault prediction model. By combining an elevator failure prediction model with a digital twin sand table simulation method and a parallel Monte Carlo simulation method, risk simulation analysis is conducted, and different Monte Carlo simulation results are output. Based on different Monte Carlo simulation results, the optimal solution for the cycle optimization of elevator dynamic maintenance is output.

[0012] This invention addresses the technical deficiencies in the prior art and offers the following advantages: It collects multimodal data about elevators by deploying a heterogeneous sensor array, identifies the root cause correlation nodes of elevator component damage, and then combines this with a federated learning model to construct an elevator fault prediction model. Using this elevator fault prediction model, it performs risk simulation analysis using a parallel Monte Carlo simulation method and outputs the optimal solution for the cycle optimization of elevator dynamic maintenance based on the analysis results. This invention can combine advanced data analysis algorithms to achieve continuous and intelligent data inspection, significantly reducing operation and maintenance costs and improving resource utilization efficiency. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0014] Figure 1 A flowchart is shown for a method to optimize the dynamic maintenance cycle of elevators by incorporating Internet of Things (IoT) analytics. Figure 2 The flowchart of the method for tracing the key root cause nodes of the elevator is shown; Figure 3 This paper presents a program view of an elevator dynamic maintenance cycle optimization system that incorporates Internet of Things (IoT) analytics. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flowchart illustrating a method for optimizing elevator dynamic maintenance cycles using IoT analytics is shown, including the following steps: S102: Deploy a heterogeneous sensor array in the elevator to collect and synchronize multimodal data of the elevator, and analyze the multimodal data of the elevator through a causal discovery algorithm to trace the key root cause nodes of the elevator. S104: Combining federated learning, the target elevator and the data stored in the elevator database are jointly trained to build an elevator fault prediction model. S106: Combining the elevator failure prediction model, risk simulation analysis is conducted through digital twin sand table simulation and parallel Monte Carlo simulation method, and different Monte Carlo simulation results are output. S108: Based on different Monte Carlo simulation results, output the optimal solution for the cycle optimization of elevator dynamic maintenance.

[0018] Furthermore, in a preferred embodiment of the present invention, the step of combining federated learning mode to jointly train the target elevator and the data stored in the elevator database to construct an elevator fault prediction model specifically involves: A database for data connection and data storage with the target elevator is obtained and labeled as an elevator database. The elevator database not only stores elevator data of all modes collected by the heterogeneous sensor array on the target elevator, but also stores elevator data of all modes of other elevators. An LSTM network model is introduced and combined with an elevator database to perform lateral federated learning on elevator data of all modalities of the target elevator in the elevator database, thereby constructing a federated learning system. The horizontal federated learning method involves randomly sampling data from an elevator database and importing it into the input layer of the LSTM network model. The random sampling targets elevator data of all modes of other elevators stored in the elevator database. In the training layer, the input layer data is weighted and trained locally. The local training is performed by constructing a loss function and simultaneously iterating the mean squared error and cross-entropy of the input layer data. An L2 regularization term is introduced into the loss function to prevent the LSTM network model from training bias during the local training process. When the number of synchronous iterations equals a preset value, the horizontal federated learning is stopped, and the LSTM network model after horizontal federated learning is obtained and labeled as an elevator fault prediction model. The elevator fault prediction model can predict the remaining lifespan of different components of the target elevator and the probability of failure of different components by analyzing elevator data of all modes of the target elevator, that is, analyzing the key root cause nodes in the target elevator.

[0019] It should be noted that federated learning models by aggregating operational data from other elevators to jointly train a high-precision fault prediction model, achieving the goal of knowledge sharing. The LSTM model is used to handle root cause features strongly correlated with time series data, enabling fault prediction. The model is trained by inputting all data from the elevator database into the input layer and training it in the training layer. The mean squared error and cross-entropy calculations are performed synchronously on the input layer data to calculate component lifespan and fault severity, respectively. An L2 regularization term is introduced into the loss function to prevent the local model from deviating excessively from the global model, i.e., to prevent training bias in the LSTM network model during local training.

[0020] Furthermore, in a preferred embodiment of the present invention, the step of combining the elevator fault prediction model with risk simulation analysis through digital twin sand table simulation and parallel Monte Carlo simulation method, and outputting different Monte Carlo simulation results, specifically includes: Obtain the direct costs of different components of the target elevator, and calculate the indirect costs during the failure of different components of the target elevator; Direct costs include component prices and labor costs, while indirect costs include the value of energy resources lost during periods of no failure. Obtain resource data corresponding to the target elevator, including human resources data, spare parts resource data, and tool resource data; Based on the cost and corresponding resource data of different components of the target elevator, an economic-resource correlation model is constructed, and an elevator failure prediction model is simultaneously correlated within the economic-resource correlation model. The output is a simulation model that can be used for risk simulation analysis through digital twin sand table simulation, which is calibrated as a risk simulation analysis model. When the risk simulation analysis model is run, and the elevator failure prediction model in the risk simulation analysis model predicts that there is a faulty elevator component, the corresponding cost and resource consumption data when the component is damaged are generated, the component loss correlation data are combined, and all maintenance strategies for the corresponding component loss are retrieved in the big data network, and a maintenance strategy space is generated. Parallel Monte Carlo simulations are performed within the maintenance strategy space. These simulations simulate the execution of all maintenance strategies for component loss and, in conjunction with component loss correlation data, output different Monte Carlo simulation results. These results include strategy costs, comprehensive risk indicators, and resource utilization rates.

[0021] It's important to note that by extrapolating various maintenance strategies, the potential consequences of each strategy over a future period can be precisely quantified, providing a solid data foundation for optimized decision-making. During the simulation, it's necessary to determine the cost and resource data related to the components. Cost data includes spare parts prices, labor costs, travel expenses, downtime losses, and energy waste, categorized as direct and indirect costs. Resource data includes human resource data, spare parts resource data, and tool resource data, encompassing the maintenance team's skill matrix, geographical location, work schedule, efficiency curves, warehouse inventory levels, supplier replenishment lead times, and procurement price fluctuations. Different data sets necessitate different maintenance strategies, resulting in varying outcomes. Parallel Monte Carlo simulations enable quantitative analysis of these consequences, outputting different simulation results. These results include strategy costs, comprehensive risk indicators, and resource utilization. The Monte Carlo simulation results need to be filtered, and the selected maintenance strategy is based on the filtering results.

[0022] Furthermore, in a preferred embodiment of the present invention, the step of outputting the optimal solution for the cycle optimization of elevator dynamic maintenance based on different Monte Carlo simulation results specifically includes: Preset the cost threshold, comprehensive risk index threshold, and resource utilization threshold for maintenance of components that are predicted to fail; Among all Monte Carlo simulation results, the Monte Carlo simulation results in which the strategy cost, comprehensive risk index and resource utilization rate are all maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold are selected and marked as candidate Monte Carlo simulation results. If the number of candidate Monte Carlo simulation results is 1, then the corresponding maintenance strategy output from the candidate Monte Carlo simulation results is selected and marked as the optimal solution for the dynamic maintenance cycle optimization of the corresponding component. If the quantity is 0, select the candidate Monte Carlo simulation result with the highest strategy cost, comprehensive risk index and resource utilization rate that are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and output its corresponding maintenance strategy. If there are more than one candidate Monte Carlo simulation result whose strategy cost, comprehensive risk index and resource utilization rate are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and the number is equal, then the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost shall be selected first. If the number is greater than 1, the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost will be selected first.

[0023] It should be noted that the selection of maintenance strategies is based on the number of Monte Carlo simulation results maintained within the corresponding cost threshold, comprehensive risk index threshold, and resource utilization rate threshold. Priority is given to strategy cost, which must meet both economic efficiency and safety requirements.

[0024] Figure 2 The flowchart illustrating the method for tracing the key root cause nodes of an elevator accident includes the following steps: S202: Deploy a heterogeneous sensor array in the elevator to collect and synchronize multimodal data of the elevator, and analyze the multimodal data of the elevator through a causal discovery algorithm to trace the key root cause nodes of the elevator. S204: Within a high-dimensional spatiotemporal feature pool, based on a causal discovery algorithm engine, construct the causal graph of the target elevator's operation and trace the elevator's key root cause nodes.

[0025] Furthermore, in a preferred embodiment of the present invention, the deployment of a heterogeneous sensor array in the elevator for collecting and synchronizing multimodal elevator data, and analyzing the multimodal elevator data using a causal discovery algorithm to trace the key root cause nodes of the elevator, specifically includes: The elevators that need maintenance are identified and designated as target elevators. A heterogeneous sensor array is then deployed on the target elevators. The heterogeneous sensor array includes sensors that collect elevator data in different modes, including sensors corresponding to vibration mode, acoustic mode, visual mode, and electrical power mode. Time synchronization is performed on elevator data of all modes collected by the heterogeneous sensor array, and timestamps are constructed. The elevator data is mapped to the timestamps. At the same time, the spatial coordinates of elevator data of different modes are determined to generate multimodal spatiotemporal synchronized elevator data. High-dimensional spatiotemporal features are extracted from multimodal spatiotemporal synchronous elevator data, including time-domain features, frequency-domain features, visual features, acoustic features, and electrical parameters. At the same time, data preprocessing of high-dimensional spatiotemporal features is required, and the preprocessed high-dimensional spatiotemporal features are combined to construct a high-dimensional spatiotemporal feature pool. A causal discovery algorithm engine is introduced, wherein the causal discovery algorithm engine is a dual engine, including a conditional independence test engine and a continuous optimization algorithm engine based on NOTEARS. Within a high-dimensional spatiotemporal feature pool, a causal discovery algorithm engine is used to construct the causal graph of the target elevator and trace the key root cause nodes of the elevator.

[0026] It should be noted that elevators requiring maintenance are those that may have component failures or have reached their maintenance due date. The purpose of deploying a heterogeneous sensor array is to monitor different data on the elevator to analyze potential faults and maintenance needs within the elevator. Vibration modes, acoustic modes, visual modes, and electrical power modes are all modal data detected by the sensors. Time synchronization of data from different modes ensures that vibration impacts, sound events, visual frames, and current transients are accurately correlated on the time axis. Spatial synchronization strongly correlates the collected data with the elevator's operating stage and location. Subsequently, high-dimensional spatiotemporal features need to be extracted. Time-domain features include, but are not limited to, mean, variance, peak value, kurtosis, waveform factor, and impulse factor; frequency-domain features include, but are not limited to, spectral centroid and frequency variance; visual features include, but are not limited to, component size and angle; acoustic features include, but are not limited to, Mel-frequency cepstral coefficients and signal-to-noise ratio in a specific frequency band; and electrical parameters include total harmonic distortion, specific harmonic amplitude, and instantaneous power fluctuation variance.

[0027] Furthermore, in a preferred embodiment of the present invention, the step of constructing a causal graph of the target elevator's operation within a high-dimensional spatiotemporal feature pool based on a causal discovery algorithm engine, and tracing the elevator's key root cause nodes, specifically involves: Within the high-dimensional spatiotemporal feature pool, the conditional independence test engine is used to test whether different high-dimensional spatiotemporal features are independent. If they are, then there is no direct causal relationship between the corresponding high-dimensional spatiotemporal features. If not, then check whether the other high-dimensional spatiotemporal features are independent of the two previously checked high-dimensional spatiotemporal features. After all high-dimensional spatiotemporal features have been checked, generate a network structure that describes the direct causal relationship between all high-dimensional spatiotemporal features and label it as a type of running causal graph. By using the NOTEARS-based continuous optimization algorithm engine and combining it with a high-dimensional spatiotemporal feature pool to construct a weighted adjacency matrix, the causal influence strength between different high-dimensional spatiotemporal features is iteratively calculated in the weighted adjacency matrix using a loss function combined with gradient descent until the number of iterations equals the preset number of iterations. The iterated weighted adjacency matrix is ​​then labeled as a binary running causal graph. By fusing the first-class and second-class operational causal graphs, a fused operational causal graph of the target elevator is output and labeled as the target operational causal graph. The target operational causal graph describes the causal relationships between different high-dimensional spatiotemporal features. Within the target causal graph, high-dimensional spatiotemporal features are treated as nodes. The directed edges connected to different nodes are analyzed, and nodes with more than a preset number of directed edges are marked as key root cause nodes. The key root cause node is the high-dimensional spatiotemporal feature corresponding to the probability of failure in the target elevator being higher than a predetermined value.

[0028] It's important to note that the parallel dual-causal discovery algorithm engine can clearly reveal the root causes of elevator component damage. This includes a conditional independence testing engine and a NOTEARS-based continuous optimization algorithm engine. The conditional independence testing engine constructs a causal graph by testing conditional independence, for example, whether "vibration peak value" and "abnormal noise energy" are independent. If independent, there is no direct causal relationship between them; if not independent, it further tests whether they become independent after considering the condition "guide rail perpendicularity." If so, "guide rail perpendicularity" is their common cause. Through a series of such statistical tests, unrelated edges are gradually eliminated, leaving a network structure that depicts direct causal relationships—a type I running causal graph. The NOTEARS-based continuous optimization algorithm engine transforms the discrete problem of edge presence into a continuous optimization problem. It iteratively calculates the causal influence strength between different high-dimensional spatiotemporal features using a loss function combined with gradient descent. Higher strength indicates a stronger causal relationship, ultimately generating a type II running causal graph. By combining the first and second types of causal graphs, and identifying key root cause nodes in the causal graph, the more directed edges there are, the higher the degree of association, and the node is the key root cause node.

[0029] like Figure 3As shown, the second aspect of the present invention also provides an elevator dynamic maintenance cycle optimization system combining Internet of Things (IoT) analysis. This system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory consisting of a DDR4 RDIMM memory module with ECC verification and an NVMe solid-state storage array using 3D NAND flash memory, and a multi-core processor based on the Zen4 microarchitecture. The memory contains a program for optimizing the elevator dynamic maintenance cycle with an optimization engine. When the program is executed in parallel through a superscalar pipeline execution unit within the processor, the following steps are implemented: A heterogeneous sensor array is deployed in the elevator to collect and synchronize multimodal data of the elevator, and the multimodal data of the elevator is analyzed by a causal discovery algorithm to trace the key root cause nodes of the elevator. By combining the federated learning model, the target elevator and the data stored in the elevator database are jointly trained to build an elevator fault prediction model. By combining an elevator failure prediction model with a digital twin sand table simulation method and a parallel Monte Carlo simulation method, risk simulation analysis is conducted, and different Monte Carlo simulation results are output. Based on different Monte Carlo simulation results, the optimal solution for the cycle optimization of elevator dynamic maintenance is output.

[0030] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing the dynamic maintenance cycle of elevators by combining Internet of Things (IoT) analysis, characterized in that: Includes the following steps: A heterogeneous sensor array is deployed in the elevator to collect and synchronize multimodal data of the elevator, and the multimodal data of the elevator is analyzed by a causal discovery algorithm to trace the key root cause nodes of the elevator. By combining the federated learning model, the target elevator and the data stored in the elevator database are jointly trained to build an elevator fault prediction model. By combining an elevator failure prediction model with a digital twin sand table simulation method and a parallel Monte Carlo simulation method, risk simulation analysis is conducted, and different Monte Carlo simulation results are output. Based on different Monte Carlo simulation results, the optimal solution for the cycle optimization of elevator dynamic maintenance is output.

2. The elevator dynamic maintenance cycle optimization method combining IoT analysis as described in claim 1, characterized in that, The deployment of a heterogeneous sensor array in the elevator is used to collect and synchronize multimodal elevator data. A causal discovery algorithm is then used to analyze this multimodal data to trace the key root causes of elevator problems. Specifically: The elevators that need maintenance are identified and designated as target elevators. A heterogeneous sensor array is then deployed on the target elevators. The heterogeneous sensor array includes sensors that collect elevator data in different modes, including sensors corresponding to vibration mode, acoustic mode, visual mode, and electrical power mode. Time synchronization is performed on elevator data of all modes collected by the heterogeneous sensor array, and timestamps are constructed. The elevator data is mapped to the timestamps. At the same time, the spatial coordinates of elevator data of different modes are determined to generate multimodal spatiotemporal synchronized elevator data. High-dimensional spatiotemporal features are extracted from multimodal spatiotemporal synchronous elevator data, including time-domain features, frequency-domain features, visual features, acoustic features, and electrical parameters. At the same time, data preprocessing of high-dimensional spatiotemporal features is required, and the preprocessed high-dimensional spatiotemporal features are combined to construct a high-dimensional spatiotemporal feature pool. A causal discovery algorithm engine is introduced, wherein the causal discovery algorithm engine is a dual engine, including a conditional independence test engine and a continuous optimization algorithm engine based on NOTEARS. Within a high-dimensional spatiotemporal feature pool, a causal discovery algorithm engine is used to construct the causal graph of the target elevator and trace the key root cause nodes of the elevator.

3. The elevator dynamic maintenance cycle optimization method combining IoT analysis as described in claim 2, characterized in that, Within a high-dimensional spatiotemporal feature pool, based on a causal discovery algorithm engine, a causal graph of the target elevator's operation is constructed, and the key root cause nodes of the elevator are traced. Specifically: Within the high-dimensional spatiotemporal feature pool, the conditional independence test engine is used to test whether different high-dimensional spatiotemporal features are independent. If they are, then there is no direct causal relationship between the corresponding high-dimensional spatiotemporal features. If not, then check whether the other high-dimensional spatiotemporal features are independent of the two previously checked high-dimensional spatiotemporal features. After all high-dimensional spatiotemporal features have been checked, generate a network structure that describes the direct causal relationship between all high-dimensional spatiotemporal features and label it as a type of running causal graph. By using the NOTEARS-based continuous optimization algorithm engine and combining it with a high-dimensional spatiotemporal feature pool to construct a weighted adjacency matrix, the causal influence strength between different high-dimensional spatiotemporal features is iteratively calculated in the weighted adjacency matrix using a loss function combined with gradient descent until the number of iterations equals the preset number of iterations. The iterated weighted adjacency matrix is ​​then labeled as a binary running causal graph. By fusing the first-class and second-class operational causal graphs, a fused operational causal graph of the target elevator is output and labeled as the target operational causal graph. The target operational causal graph describes the causal relationships between different high-dimensional spatiotemporal features. Within the target causal graph, high-dimensional spatiotemporal features are treated as nodes. The directed edges connected to different nodes are analyzed, and nodes with more than a preset number of directed edges are marked as key root cause nodes. The key root cause node is the high-dimensional spatiotemporal feature corresponding to the probability of failure in the target elevator being higher than a predetermined value.

4. The elevator dynamic maintenance cycle optimization method combining IoT analysis as described in claim 1, characterized in that, The method of combining federated learning with data stored in the elevator database to jointly train an elevator fault prediction model is as follows: A database for data connection and data storage with the target elevator is obtained and labeled as an elevator database. The elevator database not only stores elevator data of all modes collected by the heterogeneous sensor array on the target elevator, but also stores elevator data of all modes of other elevators. An LSTM network model is introduced and combined with an elevator database to perform lateral federated learning on elevator data of all modalities of the target elevator in the elevator database, thereby constructing a federated learning system. The horizontal federated learning method involves randomly sampling data from an elevator database and importing it into the input layer of the LSTM network model. The random sampling targets elevator data of all modes of other elevators stored in the elevator database. In the training layer, the input layer data is weighted and trained locally. The local training is performed by constructing a loss function and simultaneously iterating the mean squared error and cross-entropy of the input layer data. An L2 regularization term is introduced into the loss function to prevent the LSTM network model from training bias during the local training process. When the number of synchronous iterations equals a preset value, the horizontal federated learning is stopped, and the LSTM network model after horizontal federated learning is obtained and labeled as an elevator fault prediction model. The elevator fault prediction model can predict the remaining lifespan of different components of the target elevator and the probability of failure of different components by analyzing elevator data of all modes of the target elevator, that is, analyzing the key root cause nodes in the target elevator.

5. The elevator dynamic maintenance cycle optimization method combining IoT analysis as described in claim 1, characterized in that, The method combines an elevator failure prediction model with a digital twin sand table simulation and a parallel Monte Carlo simulation method to perform risk simulation analysis, outputting different Monte Carlo simulation results, specifically: Obtain the direct costs of different components of the target elevator, and calculate the indirect costs during the failure of different components of the target elevator; Direct costs include component prices and labor costs, while indirect costs include the value of energy resources lost during periods of no failure. Obtain resource data corresponding to the target elevator, including human resources data, spare parts resource data, and tool resource data; Based on the cost and corresponding resource data of different components of the target elevator, an economic-resource correlation model is constructed, and an elevator failure prediction model is simultaneously correlated within the economic-resource correlation model. The output is a simulation model that can be used for risk simulation analysis through digital twin sand table simulation, which is calibrated as a risk simulation analysis model. When the risk simulation analysis model is run, and the elevator failure prediction model in the risk simulation analysis model predicts that there is a faulty elevator component, the corresponding cost and resource consumption data when the component is damaged are generated, the component loss correlation data are combined, and all maintenance strategies for the corresponding component loss are retrieved in the big data network, and a maintenance strategy space is generated. Parallel Monte Carlo simulations are performed within the maintenance strategy space. These simulations simulate the execution of all maintenance strategies for component loss and, in conjunction with component loss correlation data, output different Monte Carlo simulation results. These results include strategy costs, comprehensive risk indicators, and resource utilization rates.

6. The elevator dynamic maintenance cycle optimization method combining IoT analysis as described in claim 1, characterized in that, The optimal solution for the cycle optimization of elevator dynamic maintenance is output based on different Monte Carlo simulation results, specifically as follows: Preset the cost threshold, comprehensive risk index threshold, and resource utilization threshold for maintenance of components that are predicted to fail; Among all Monte Carlo simulation results, the Monte Carlo simulation results in which the strategy cost, comprehensive risk index and resource utilization rate are all maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold are selected and marked as candidate Monte Carlo simulation results. If the number of candidate Monte Carlo simulation results is 1, then the corresponding maintenance strategy output from the candidate Monte Carlo simulation results is selected and marked as the optimal solution for the dynamic maintenance cycle optimization of the corresponding component. If the quantity is 0, select the candidate Monte Carlo simulation result with the highest strategy cost, comprehensive risk index and resource utilization rate that are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and output its corresponding maintenance strategy. If there are more than one candidate Monte Carlo simulation result whose strategy cost, comprehensive risk index and resource utilization rate are maintained within the corresponding cost threshold, comprehensive risk index threshold and resource utilization rate threshold, and the number is equal, then the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost shall be selected first. If the number is greater than 1, the maintenance strategy output corresponding to the candidate Monte Carlo simulation result with the lowest strategy cost will be selected first.

7. An elevator dynamic maintenance cycle optimization system combining IoT analysis, characterized in that, The elevator dynamic maintenance cycle optimization system integrates a high-performance computing architecture and a data storage module, including a non-volatile memory consisting of a DDR4 RDIMM memory module with ECC verification and an NVMe solid-state storage array using 3D NAND flash memory, as well as a multi-core processor based on the Zen4 microarchitecture. The memory contains a program for optimizing the elevator dynamic maintenance cycle with an optimization engine. When the program is decoded and executed in parallel by the superscalar pipeline execution unit in the processor, the elevator dynamic maintenance cycle optimization steps as described in any one of claims 1-6 are implemented.

Images