An elevator fault prediction and health management system and method based on artificial intelligence and internet of things

Abstract

This invention discloses an elevator fault prediction and health management system and method based on artificial intelligence and the Internet of Things. The system includes a perception layer, a transmission layer, and a platform layer connected in sequence. The perception layer collects data on elevator equipment, electrical circuits, and passenger scenarios through a multimodal acquisition module. The transmission layer employs a hybrid wired and wireless encrypted transmission architecture to ensure data security. The platform layer achieves early and accurate prediction of elevator faults through a multimodal fusion AI model, and is equipped with quantitative health assessment, model self-optimization, full lifecycle health records, and multi-terminal collaboration functions. This invention transforms elevator operation and maintenance from passive repair to proactive prevention, significantly improving elevator operational safety, reducing operation and maintenance costs, enabling precise supervision, and is applicable to the full lifecycle safety management of various types of elevators.

Description

Technical Field

[0001] This invention belongs to the technical field of elevator safety, specifically relating to an elevator fault prediction and health management system and method based on artificial intelligence and the Internet of Things. Background Technology

[0002] With the acceleration of urbanization, elevators have become an indispensable vertical transportation tool in high-rise buildings, and their operational safety is directly related to the safety of public life and property. Currently, elevator safety management mainly adopts the traditional "time-based maintenance" model, relying on the experience of maintenance personnel for regular inspections and repairs, which has several inherent defects: First, it is passive and lacks early warning, only able to handle repairs after a fault occurs, unable to identify progressive faults such as traction machine bearing wear and wire rope fatigue in advance, easily leading to sudden shutdowns, entrapment, and other safety accidents; second, data collection is limited and analysis capabilities are insufficient, with existing monitoring systems mostly collecting only a few parameters such as speed and door opening / closing status, lacking comprehensive monitoring of electrical circuits and passenger scenarios, and outdated data processing methods, resulting in high rates of false alarms and missed alarms; third, maintenance resources are allocated irrationally, failing to consider the actual operating load and health status of the elevator, resulting in both over-maintenance and under-maintenance; fourth, full life-cycle management is lacking, making it impossible to achieve full-process data traceability from elevator installation to scrapping, and difficult to support accurate supervision and lifespan prediction.

[0003] In recent years, IoT and big data technologies have been gradually applied in the field of elevator monitoring. However, existing related patent technologies have obvious limitations. For example, the prior art document "Elevator Health Management and Maintenance System and Data Acquisition and Evaluation Method Based on IoT" (CN201210234167) discloses an elevator health management system based on IoT, which uses Euclidean distance algorithm for fault matching and FMEA analysis. It can only achieve post-fault diagnosis and cannot predict early faults. The prior art document "An Elevator Health Monitoring System and Monitoring Method Based on IoT" (CN201510138001) only collects data through acceleration and vibration sensors. The data collection dimension is single and lacks monitoring of electrical circuits and passenger scenarios, so it cannot comprehensively reflect the health status of the elevator. The prior art document "An Elevator Health Value Calculation System and Method Based on IoT and Big Data" (CN202011608298) only constructs a quantitative calculation model for health values. It lacks artificial intelligence prediction capabilities, cannot capture early weak abnormal signals of components, and has not formed a complete business closed loop of "monitoring-early warning-maintenance-evaluation".

[0004] In summary, existing technologies have failed to achieve deep integration of artificial intelligence and Internet of Things technologies, resulting in problems such as insufficient early warning of faults, poor comprehensiveness of data collection, weak robustness of AI models, security risks in data transmission, and management functions limited to equipment-level monitoring. These issues cannot meet the intelligent and precise requirements of elevator safety management. Summary of the Invention

[0005] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide an elevator fault prediction and health management system and method based on artificial intelligence and the Internet of Things. This system enables early and accurate warning of elevator faults, dynamic quantitative assessment of health status, intelligent optimization of maintenance strategies, and closed-loop management throughout the entire life cycle. It also improves elevator operation safety, reduces operation and maintenance costs, and empowers regulatory authorities to conduct precise supervision.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an elevator fault prediction and health management system based on artificial intelligence and the Internet of Things, comprising a perception layer, a transmission layer and a platform layer that are connected in sequence.

[0008] The perception layer includes a multimodal acquisition unit and a field edge processing unit; the multimodal acquisition unit includes a vibration monitoring module, a current monitoring module, and a visual monitoring module, used to acquire elevator operation data in real time; the field edge processing unit is used to preprocess and extract features from the acquired raw data, mark fault data and upload it, and at the same time realize rapid response to local anomalies.

[0009] The transmission layer adopts a hybrid wired and wireless transmission architecture and has a built-in data encryption module to match the corresponding transmission mode according to the deployment characteristics of the acquisition nodes.

[0010] The platform layer includes a data storage module, a fault sample library, a model training module, an automatic judgment module, an elevator health assessment module, and a model self-optimization module. The data storage module stores elevator lifecycle operation data. The fault sample library stores tagged fault feature data, establishing a unique sample set for each elevator. The model training module trains a multimodal fusion fault prediction and anomaly detection model based on the fault sample library and normal operation data. The automatic judgment module analyzes real-time collected feature data based on the trained model, identifying early abnormal signals and predicting potential faults. The elevator health assessment module outputs quantitative health results for the elevator as a whole and its core components. The model self-optimization module performs incremental updates and parameter self-adjustments based on newly added operation data and fault samples.

[0011] As a preferred technical solution, the vibration monitoring module includes several triaxial vibration sensors, which are respectively deployed at the core monitoring points of the elevator traction machine, brake, door operator, car, and guide rail, to collect the vibration time-domain and frequency-domain signals of the core moving parts of the elevator; the current monitoring module includes several Hall current sensors, which are respectively connected to the elevator power supply circuit, safety circuit, door lock circuit, and brake circuit, to collect real-time current timing data of key electrical circuits; the visual monitoring module includes a high-definition network camera deployed in the car and an industrial camera deployed in the hoistway, to collect video image data of the car passenger scene and the hoistway environment.

[0012] As a preferred technical solution, the visual monitoring module includes a YOLOv8-based target detection submodule, an OpenPose-based pose estimation submodule, and a FERNet-based facial expression recognition submodule. The target detection submodule is used to identify the status of objects and hardware facilities inside the elevator car. The pose estimation submodule is used to extract human limb movement features. The facial expression recognition submodule is used to extract the facial expression features of passengers and, in conjunction with the elevator's operating status, assist in determining emergency faults.

[0013] As a preferred technical solution, the on-site edge processing unit includes a feature engineering unit and an on-site response unit. The feature engineering module is used to preprocess and extract features from the collected raw data. The preprocessing includes data filtering, denoising, and normalization. The feature extraction includes: extracting time-domain-frequency domain fusion features from vibration signals through wavelet packet decomposition, extracting time-series statistical features from current signals through a sliding window, and extracting deep semantic features from visual signals through a deep learning network.

[0014] The on-site response unit is used to make preliminary anomaly judgments on the extracted features. When an emergency fault is detected, it directly triggers the local emergency response and simultaneously marks the fault feature data and uploads it to the platform-level fault sample library.

[0015] As a preferred technical solution, the wired and wireless hybrid transmission architecture transmits fixed sensor data deployed in the elevator machine room via Ethernet wired transmission with IPSec VPN tunnel encryption, while mobile / distributed sensors deployed in the hoistway and car transmit data via one or more wireless methods, including 5G, Wi-Fi, LoRa, and Bluetooth, and are encrypted using the DTLS protocol. The data encryption module adopts an end-to-end national cryptographic encryption architecture to encrypt data and generate data verification codes.

[0016] As a preferred technical solution, the model training module includes a fault prediction model and a model training module. The fault prediction model adopts a CNN-LSTM hybrid neural network to fuse three types of features: vibration, current, and vision across modes. It is trained under supervision with fault type, fault severity, and fault occurrence window as labels, and outputs the type, probability, and occurrence time of potential faults.

[0017] The anomaly detection model uses an isolated forest for unsupervised training based on elevator normal operation data. An anomaly score threshold is set according to the 3σ criterion. When the anomaly score of the real-time data exceeds the threshold for three consecutive sampling periods, it is determined to be an abnormal state.

[0018] As a preferred technical solution, the elevator health assessment module is used to construct a quantitative assessment system and use a large language model to generate on-demand maintenance decision suggestions based on health scores and fault prediction results.

[0019] The quantitative evaluation system includes primary indicators and several secondary indicators. The primary indicators include component health, operational stability, safety redundancy, and environmental compliance. The weights of each indicator are determined by the analytic hierarchy process (AHP), and the overall health score of the elevator is calculated from 0 to 100 using the fuzzy comprehensive evaluation method, which is then divided into 5 health levels.

[0020] As a preferred technical solution, the model self-optimization module adopts the deep reinforcement learning (DQN) framework, with fault prediction accuracy and false alarm rate as the core reward functions. It continuously introduces new fault samples and operating data, completes an incremental model update at a certain period, and automatically adjusts the model weights and feature weights.

[0021] As a preferred technical solution, the platform layer also includes a health record management module and a multi-terminal collaborative interaction module;

[0022] The health record management module establishes a unique full life cycle health record for each elevator, recording data from the entire process of elevator installation and acceptance, daily operation, fault handling, maintenance work, health assessment and life prediction. Based on a finely tuned large language model, it automatically generates periodic health reports and archives them.

[0023] The multi-terminal collaborative interaction module includes a Web management terminal, a mobile operation and maintenance terminal, an elevator car interaction terminal, and a monitoring terminal, which are respectively matched with the permissions and functional requirements of property management personnel, maintenance personnel, elevator passengers, and regulatory departments to achieve full-process collaboration in emergency response, maintenance scheduling, and off-site supervision.

[0024] Secondly, the present invention also provides an elevator fault prediction and health management method based on artificial intelligence and the Internet of Things, applied to the aforementioned elevator fault prediction and health management system based on artificial intelligence and the Internet of Things, comprising the following steps:

[0025] S1. Multimodal data acquisition: Raw data is acquired in real time through the multimodal acquisition unit of the perception layer. The raw data includes data on elevator core components, electrical circuits, and passenger scenarios.

[0026] S2. Edge data processing: The on-site edge processing unit preprocesses and extracts features from the raw data, marks fault data, and completes rapid response to local emergency anomalies.

[0027] S3. Encrypted data transmission: The transmission layer matches wired and wireless transmission modes according to the characteristics of the acquisition node, and transmits the data to the platform layer after end-to-end encryption through the built-in encryption module.

[0028] S4. Automatic analysis: After decrypting and verifying the integrity of the received data, the platform layer analyzes the real-time feature data through a pre-trained multimodal fusion model to identify early abnormal signals and predict the type, probability, and occurrence window of potential faults.

[0029] S5. Health Status Assessment: Based on real-time elevator operation data, fault prediction results and historical full life cycle data, output the overall health quantitative score, health level and degradation trend of the elevator and its core components.

[0030] S6. Decision Output and Model Optimization: Based on the results of fault prediction and health assessment, generate decision instructions for on-demand maintenance, emergency response, and safety intervention. At the same time, store the newly added fault samples and operational data into the fault sample library to complete the incremental update and optimization of the model.

[0031] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0032] (1) This invention enables accurate early warning of faults, which greatly improves the safety of elevator operation. Through multimodal data acquisition and deep integration of AI, this invention can capture early weak abnormal signals of elevator components 7-14 days in advance, accurately predict potential faults such as traction machine bearing wear and door machine jamming, transform passive maintenance into active prevention, and effectively avoid safety accidents such as sudden shutdown and entrapment.

[0033] (2) This invention enables quantitative assessment of health status, optimizes maintenance strategies, and reduces operation and maintenance costs. This invention constructs a multi-dimensional quantitative health assessment system, which can intuitively reflect the elevator's operating status and degradation trend, providing a scientific basis for operation and maintenance decisions, realizing "on-demand maintenance," and avoiding over-maintenance and under-maintenance.

[0034] (3) This invention realizes closed-loop management of the entire life cycle and empowers precise and intelligent supervision. This invention establishes a full life cycle health record for each elevator, realizes full-process data traceability from installation, operation to maintenance and scrapping, and provides a multi-terminal collaborative interactive interface to provide regulatory departments with off-site inspection capabilities. It can accurately grasp the operating status of high-risk elevators, improve regulatory efficiency, and reduce interference with the normal operation of enterprises.

[0035] (4) This invention has the ability to self-optimize the model and has excellent adaptability and robustness. This invention achieves incremental updates and self-optimization of the model through a deep reinforcement learning framework. The prediction accuracy continues to improve as the system runs. It can adapt to elevators of different brands, service years and working conditions. Tests on publicly available elevator fault datasets show that... Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the elevator fault prediction and health management system based on artificial intelligence and the Internet of Things according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the sensing layer structure according to an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the platform layer structure according to an embodiment of the present invention;

[0040] Figure 4 This is a flowchart of an elevator fault prediction and health management method based on artificial intelligence and the Internet of Things, according to an embodiment of the present invention. Detailed Implementation

[0041] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0042] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0043] Please see Figure 1 This embodiment provides an elevator fault prediction and health management system based on artificial intelligence and the Internet of Things. The core architecture is divided into three layers, from bottom to top: perception layer 1, transmission layer 2 and platform layer 3.

[0044] Example 1 - Specific implementation of the perception layer.

[0045] like Figure 2 As shown, the perception layer 1 includes a multimodal acquisition unit 110 and a field edge processing unit 120, which are deployed at the elevator field end and serve as the data foundation of the system.

[0046] The multimodal acquisition unit 110 includes a vibration monitoring module 111, a current monitoring module 112, and a visual monitoring module 113, enabling elevator data acquisition.

[0047] The vibration monitoring module 111 uses eight high-precision triaxial MEMS vibration sensors with a sampling frequency of 10kHz. These sensors are deployed at eight core points: the traction machine input end, the traction machine output end, the brake, the door motor, the top of the car, the bottom of the car, the upper support of the guide rail, and the lower support of the guide rail. This comprehensively collects vibration signals from the core moving parts of the elevator, covering the characteristic signal acquisition of faults such as bearing wear, abnormal gear meshing, guide rail deformation, and car vibration.

[0048] The current monitoring module 112 uses four high-precision Hall current sensors, which are connected in series in the elevator main power supply circuit, safety circuit, door lock circuit, and brake circuit respectively. The sampling frequency is set to 1kHz to collect the current timing data of each circuit in real time, capture early signals of electrical faults such as short circuit, open circuit, poor contact, and abnormal brake, and make up for the lack of electrical circuit monitoring in traditional monitoring systems.

[0049] The visual monitoring module 113 includes two 2-megapixel high-definition network cameras and one shaft industrial camera. One camera is deployed on the top of the car to collect data on the car's passenger experience; the other camera is deployed at the car door to monitor the door operator's operation and door opening / closing status; and the shaft industrial camera is deployed on the top of the car to collect image data of the shaft guide rails and limit switches as the car moves.

[0050] The visual monitoring module 113 incorporates an edge computing chip and features a lightweight target detection submodule optimized based on YOLOv8n. This submodule can identify prohibited items such as electric vehicles and flammable or explosive materials inside the elevator car in real time, as well as abnormalities such as malfunctioning buttons, jammed door operators, and deformed guide rails. It also features an OpenPose lightweight posture estimation submodule, which extracts limb features from 17 key points on the human body to identify abnormal behaviors such as climbing, falling, and prolonged lingering. Furthermore, it features a FERNet facial expression recognition submodule, which extracts facial features of passengers. When it detects persistent fear, anxiety, or other abnormal facial expressions, combined with the elevator stopping and doors not opening, it automatically determines that a passenger is trapped and triggers a local emergency response.

[0051] If the detected signals from the body language or facial expressions of passengers conflict with the elevator's operational status, the assessment of body language will have the highest priority during emergency response, followed by the elevator's operational status, and then facial expressions. For example, even if the elevator is operating normally, if abnormal body language is detected, such as climbing, falling, or prolonged lingering, the local emergency response will be triggered.

[0052] Meanwhile, the visual monitoring module can use historical visual data to count elevator usage frequency, peak hours, and passenger distribution, build an elevator usage profile, and generate personalized maintenance suggestions based on the equipment's operating status.

[0053] The field edge processing unit 120, which includes a feature engineering unit 121 and a field response unit 122, adopts an ARM architecture edge computing host, is deployed in the elevator machine room, and communicates with each acquisition module via a fieldbus. The core steps include:

[0054] 1. Data preprocessing: Butterworth filtering, wavelet denoising, and minimum-maximum normalization are performed on the raw collected data to remove environmental noise and outliers, thereby improving data quality;

[0055] 2. Feature Extraction: The vibration signal is decomposed into three layers of wavelet packets to extract time-domain and frequency-domain features of eight frequency bands, including energy entropy, root mean square, peak factor, and kurtosis. The 15 core features with the highest correlation to the fault are selected using the mutual information method. The current signal is processed through a 500ms sliding time window to extract 25-dimensional time-series features, including mean, variance, peak value, and harmonic distortion rate. The visual signal is processed through a deep learning network to extract deep semantic features, completing the initial fusion of multimodal features.

[0056] 3. Fault Marking and Local Response: The extracted features are used for preliminary anomaly judgment. When emergency faults such as entrapment or risk of overshooting are detected, the local emergency response is triggered directly. At the same time, the fault feature data is labeled with fault type, fault time, and elevator number, and then uploaded to the platform-level fault sample library. The features of normal operation data are compressed and then uploaded to the platform level.

[0057] Example 2 - Specific implementation of the transport layer.

[0058] The transmission layer 2 serves as a data transmission bridge between the perception layer 1 and the platform layer 3. It adopts a hybrid wired and wireless transmission architecture and has a built-in data encryption module 201, which balances transmission stability, deployment flexibility, and data security.

[0059] The wired and wireless hybrid transmission architecture includes a wired transmission channel and a wireless transmission channel.

[0060] The wired transmission channel, deployed in the elevator machine room, includes edge processing units, fixed vibration sensors, and Hall current sensors. It uses industrial Ethernet for wired transmission, coupled with IPSec VPN tunnel encryption, ensuring the stability and reliability of core data transmission within the machine room, with a transmission rate of up to 100Mbps. The wireless transmission channel, deployed in the hoistway and car, includes mobile / distributed sensors and visual monitoring equipment. The wireless transmission method is matched to the deployment environment: high-definition video data in the car is transmitted via 5G / Wi-Fi 6 to ensure bandwidth requirements; low-speed sensor data in the hoistway is transmitted via LoRa to reduce wiring complexity and improve wall penetration; and short-range debugging data is transmitted via Bluetooth to meet on-site maintenance needs. All wireless transmissions use DTLS encryption to adapt to the transmission characteristics of the wireless channel.

[0061] The data encryption module 201 adopts an end-to-end national cryptographic encryption architecture, specifically:

[0062] Sending end: The edge processing unit of the perception layer uses the SM2 national cryptographic asymmetric encryption algorithm to encrypt the core feature data and fault marking data, and generates a data check code through the SM3 national cryptographic hash algorithm, which is appended to the end of the data packet;

[0063] Transmission process: The entire link adopts the TLS 1.3 secure transmission protocol to establish an encrypted transmission tunnel and prevent data from being stolen midway;

[0064] Receiver: After receiving data, the platform layer first verifies the data integrity using the SM3 algorithm. If the verification fails, the data packet is discarded and a retransmission is requested. If the verification passes, the data is decrypted using the SM2 private key to ensure that the data is not leaked or tampered with throughout the entire data chain.

[0065] Example 3 - Specific implementation of the platform layer.

[0066] like Figure 3 As shown, platform layer 3 is a cloud platform deployed on a cloud server, which is the core of the system's intelligent decision-making. It includes a data storage module 301, a fault sample library 302, a model training module 303, an AI automatic judgment module 304, an elevator health assessment module 305, a model self-optimization module 306, a health record management module 307, and a multi-terminal collaborative interaction module 308.

[0067] The data storage module 301 adopts a hybrid storage architecture of distributed time-series database and relational database. The time-series database is used to store time-series data of high-frequency elevator operation, while the relational database is used to store structured data such as elevator basic information, fault records, maintenance records, and user information. At the same time, a cold and hot data separation storage strategy is adopted to reduce storage costs and improve data retrieval efficiency.

[0068] Fault Sample Library 302 is used to store labeled elevator fault feature data. It is classified and managed according to elevator number, fault type and fault severity. A unique sample set is established for each elevator. At the same time, an industry-wide general fault sample library is built, covering 12 major categories and 36 subcategories of common elevator faults, providing a data foundation for model training.

[0069] Model training module 303 is used to build and train elevator fault prediction and anomaly detection models. The core model includes:

[0070] A multimodal fusion CNN-LSTM fault prediction model: The input consists of three types of features: vibration, current, and visual. Spatial features are extracted through a CNN network, and temporal features are extracted through an LSTM network. Cross-modal feature fusion is achieved through an attention mechanism. The output includes fault type, fault severity, and fault occurrence window (7 days or 14 days). The model uses a cross-entropy loss function, an Adam optimizer, an initial learning rate of 1e-4, a batch size of 32, and 100 iterations of training. Overfitting is prevented through an early stopping mechanism and a Dropout layer.

[0071] Anomaly detection model: An isolated forest is used for unsupervised training based on elevator normal operation data. Anomaly score threshold is set according to the 3σ criterion. When the anomaly score of real-time data exceeds the threshold for three consecutive sampling periods, it is judged as an abnormal state, thus solving the problem of identifying unknown faults.

[0072] The automatic judgment module 304 loads the trained model and synchronously analyzes the real-time feature data received by the platform layer. It identifies early abnormal signals of the elevator in real time and outputs the type, probability of occurrence, and expected occurrence time of potential faults. When the probability of fault occurrence exceeds 80%, it automatically triggers a fault warning and pushes it to the corresponding maintenance personnel and property management personnel. For predicted high-risk serious faults, it automatically sends safety intervention instructions to the elevator control system, including slowing down the door operator, reducing the traction machine speed, and activating emergency protection, to prevent the fault from escalating and causing a safety accident.

[0073] The elevator health assessment module 305 constructs a quantitative health assessment system, including four primary indicators: component health, operational stability, safety redundancy, and environmental compliance, and 28 secondary indicators. The weight of each indicator is determined by industry experts through the Analytic Hierarchy Process (AHP), and an overall health score of 0-100 is calculated using fuzzy comprehensive evaluation, categorized into five health levels: Excellent (90-100), Good (80-89), Caution (70-79), Warning (60-69), and Dangerous (below 60). Simultaneously, it outputs the health scores and degradation trends of each core component. Based on the health level and fault prediction results, it generates targeted on-demand maintenance decision-making suggestions, including maintenance priorities, maintenance locations, and maintenance process recommendations.

[0074] To better explain the fuzzy comprehensive evaluation method, the following description is provided in this embodiment, as shown in Table 1:

[0075] Table 1 Definition of Basic Parameters

[0076]

[0077] The specific steps are as follows:

[0078] 1. Construct a single-index fuzzy membership matrix .

[0079] For each secondary indicator, we distinguish between positive indicators (higher values ​​indicate better health, such as insulation resistance) and negative indicators (higher values ​​indicate worse health, such as vibration amplitude and wear rate). Using the engineering-standard trapezoidal membership function, we calculate the membership degree of each indicator to the five health levels, resulting in a membership degree row vector. .

[0080] Finally, the number was obtained Fuzzy relation matrix of each primary indicator:

[0081] ,

[0082] in, The matrix represents the number of secondary indicators under this primary indicator. Rows 5 columns.

[0083] 2. First-level fuzzy comprehensive evaluation (second-level indicator level).

[0084] use A weighted average operator (fully retaining all indicator weight information, avoiding information loss, and best suited for multi-dimensional device health evaluation) calculates the evaluation result of a single primary indicator:

[0085] ,

[0086] The formula for calculating the membership degree of a single level is as follows:

[0087] ,

[0088] For the first The first-level indicator for the first The membership degree of each health level satisfies .

[0089] 3. Second-level fuzzy comprehensive evaluation (overall health level).

[0090] The evaluation results of the four primary indicators are combined into a fuzzy relation matrix:

[0091] ,

[0092] By combining the first-level weight set with the overall matrix, a second-level fuzzy evaluation result of the overall health of the elevator is obtained. :

[0093] ,

[0094] The formula for calculating the overall membership degree is as follows:

[0095] ,

[0096] For the elevator as a whole to the first The membership degree of each health level satisfies .

[0097] The final formula for calculating the overall health score of the elevator.

[0098] The final health score H, ranging from 0 to 100, is obtained by weighted summation of the fuzzy evaluation results and the grade score vector, as shown in the following formula:

[0099] ,

[0100] in, Assess the overall health of the elevator. Let be the membership degree of the elevator as a whole to the k-th health level. This represents the score for the k-th health level. Let be the transpose of the score vector C.

[0101] The model self-optimization module 306 adopts the deep reinforcement learning DQN framework, with fault prediction accuracy and false alarm rate as the core reward functions. It continuously introduces new fault samples and operational data, completes an incremental model update every 7 days, automatically adjusts model weights and feature weights, and continuously optimizes the model's adaptability to different elevators and different working conditions, so that the model's prediction accuracy continues to improve as the system operates.

[0102] The Health Record Management Module 307 assigns a unique identification code to each elevator, establishing a corresponding full lifecycle health record. This record comprehensively documents elevator installation and acceptance data, daily operation data, fault handling records, maintenance operation records, health assessment reports, and lifespan prediction data, enabling traceability of elevator lifecycle data. Simultaneously, based on large language models, such as finely tuned Qwen-7B and Deepseek V4.0, it automatically extracts key node data from the health record, generating standardized elevator health reports quarterly and annually. These reports include health trend analysis, fault risk warnings, maintenance optimization suggestions, and remaining service life predictions, which are automatically archived into the health record and simultaneously pushed to users with corresponding permissions.

[0103] The multi-terminal collaborative interaction module 308 includes a web management terminal, a mobile operation and maintenance terminal, an in-car interaction terminal, and a monitoring terminal, to meet the needs of different roles.

[0104] Web management interface: Designed for property management personnel, providing functions such as real-time elevator status monitoring, fault warning reception, maintenance work order management, and health report viewing.

[0105] Mobile maintenance terminal: Designed for maintenance personnel, it provides functions such as receiving and closing maintenance work orders, navigating fault locations, viewing elevator drawings, and uploading maintenance records.

[0106] Car interaction terminal: For elevator passengers, it provides functions such as floor selection, emergency call, voice reassurance, and elevator safety prompts. In case of malfunction, it automatically plays reassurance voice and displays the rescue progress.

[0107] For regulatory authorities: It provides functions such as overall elevator safety status within the jurisdiction, a list of high-risk elevators, fault statistics and analysis, and compliance analysis of maintenance units, enabling precise off-site supervision and key supervision of high-risk elevators.

[0108] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0109] Furthermore, in the above embodiments of the elevator fault prediction and health management system based on artificial intelligence and the Internet of Things, the logical division of each program module is only an example. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the elevator fault prediction and health management system based on artificial intelligence and the Internet of Things can be divided into different program modules to complete all or part of the functions described above.

[0110] Please see Figure 4 In one embodiment, a method for predicting elevator malfunctions and managing elevator health based on artificial intelligence and the Internet of Things is provided, including:

[0111] S1. Multimodal data acquisition: Raw data is acquired in real time through the multimodal acquisition unit of the perception layer. The raw data includes data on elevator core components, electrical circuits, and passenger scenarios.

[0112] S2. Edge data processing: The on-site edge processing unit preprocesses and extracts features from the raw data, marks fault data, and completes rapid response to local emergency anomalies.

[0113] S3. Encrypted data transmission: The transmission layer matches wired and wireless transmission modes according to the characteristics of the acquisition node, and transmits the data to the platform layer after end-to-end encryption through the built-in encryption module.

[0114] S4. Automatic analysis: After decrypting and verifying the integrity of the received data, the platform layer analyzes the real-time feature data through a pre-trained multimodal fusion model to identify early abnormal signals and predict the type, probability, and occurrence window of potential faults.

[0115] S5. Health Status Assessment: Based on real-time elevator operation data, fault prediction results and historical full life cycle data, output the overall health quantitative score, health level and degradation trend of the elevator and its core components.

[0116] S6. Decision Output and Model Optimization: Based on the results of fault prediction and health assessment, generate decision instructions for on-demand maintenance, emergency response, and safety intervention. At the same time, store the newly added fault samples and operational data into the fault sample library to complete the incremental update and optimization of the model.

[0117] Meanwhile, the platform's health record management module updates elevator health records in real time and generates health briefings periodically; regulatory authorities can view the elevator safety status within their jurisdiction in real time through the regulatory terminal, achieving precise off-site supervision.

[0118] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An elevator fault prediction and health management system based on artificial intelligence and the Internet of Things, characterized in that, This includes the perception layer, transmission layer, and platform layer, which are connected in sequence. The perception layer includes a multimodal acquisition unit and a field edge processing unit; the multimodal acquisition unit includes a vibration monitoring module, a current monitoring module, and a visual monitoring module, used to acquire elevator operation data in real time; the field edge processing unit is used to preprocess and extract features from the acquired raw data, mark fault data and upload it, and at the same time realize rapid response to local anomalies. The transmission layer adopts a hybrid wired and wireless transmission architecture and has a built-in data encryption module to match the corresponding transmission mode according to the deployment characteristics of the acquisition nodes. The platform layer includes a data storage module, a fault sample library, a model training module, an automatic judgment module, an elevator health assessment module, and a model self-optimization module. The data storage module stores elevator lifecycle operation data. The fault sample library stores tagged fault feature data, establishing a unique sample set for each elevator. The model training module trains a multimodal fusion fault prediction and anomaly detection model based on the fault sample library and normal operation data. The automatic judgment module analyzes real-time collected feature data based on the trained model, identifying early abnormal signals and predicting potential faults. The elevator health assessment module outputs quantitative health results for the elevator as a whole and its core components. The model self-optimization module analyzes newly added operation data and fault samples to achieve incremental model updates and parameter self-adjustment.

2. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The vibration monitoring module includes several triaxial vibration sensors, which are deployed at key monitoring points of the elevator traction machine, brake, door operator, car, and guide rails to collect vibration time-domain and frequency-domain signals of the elevator's core moving components. The current monitoring module includes several Hall current sensors, which are connected to the elevator power supply circuit, safety circuit, door lock circuit, and brake circuit to collect real-time current timing data of key electrical circuits. The visual monitoring module includes a high-definition network camera deployed in the car and an industrial camera deployed in the hoistway to collect video image data of the car's passenger experience and the hoistway environment.

3. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The visual monitoring module includes a YOLOv8-based target detection submodule, an OpenPose-based pose estimation submodule, and a FERNet-based facial expression recognition submodule. The target detection submodule is used to identify objects and hardware facilities inside the car; the pose estimation submodule is used to extract human limb movement features. The facial expression recognition submodule is used to extract the facial expression features of elevator passengers and combine them with the elevator's operating status to help determine emergency malfunctions.

4. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The on-site edge processing unit includes a feature engineering unit and an on-site response unit. The feature engineering module is used to preprocess and extract features from the collected raw data. The preprocessing includes data filtering, denoising, and normalization. The feature extraction includes: extracting time-domain-frequency domain fusion features from vibration signals through wavelet packet decomposition, extracting time-series statistical features from current signals through a sliding window, and extracting deep semantic features from visual signals through a deep learning network. The on-site response unit is used to make preliminary anomaly judgments on the extracted features. When an emergency fault is detected, it directly triggers the local emergency response and simultaneously marks the fault feature data and uploads it to the platform-level fault sample library.

5. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The wired and wireless hybrid transmission architecture uses wired Ethernet transmission for fixed sensors deployed in the elevator machine room, encrypted with an IPSec VPN tunnel. Mobile or distributed sensors deployed in the hoistway and car transmit wirelessly via one or more of 5G, Wi-Fi, LoRa, and Bluetooth, encrypted using the DTLS protocol. The data encryption module adopts an end-to-end national cryptographic encryption architecture to encrypt data and generate data verification codes.

6. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The model training module includes a fault prediction model and a model training module. The fault prediction model uses a CNN-LSTM hybrid neural network to fuse three types of features: vibration, current and vision across modes. It is trained under supervision with fault type, fault severity and fault occurrence window as labels, and outputs the type, probability and occurrence time of potential faults. The anomaly detection model uses an isolated forest for unsupervised training based on elevator normal operation data. An anomaly score threshold is set according to the 3σ criterion. When the anomaly score of the real-time data exceeds the threshold for three consecutive sampling periods, it is determined to be an abnormal state.

7. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The elevator health assessment module is used to construct a quantitative assessment system and use a large language model to generate on-demand maintenance decision recommendations based on health scores and fault prediction results. The quantitative evaluation system includes primary indicators and several secondary indicators. The primary indicators include component health, operational stability, safety redundancy, and environmental compliance. The weights of each indicator are determined by the analytic hierarchy process (AHP), and the overall health score of the elevator is calculated from 0 to 100 using the fuzzy comprehensive evaluation method, which is then divided into 5 health levels.

8. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The model self-optimization module adopts the deep reinforcement learning (DQN) framework, with fault prediction accuracy and false alarm rate as the core reward functions. It continuously introduces new fault samples and operational data, completes an incremental model update at a certain period, and automatically adjusts the model weights and feature weights.

9. The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things as described in claim 1, characterized in that, The platform layer also includes a health record management module and a multi-terminal collaborative interaction module; The health record management module establishes a unique full life cycle health record for each elevator, recording data from the entire process of elevator installation and acceptance, daily operation, fault handling, maintenance work, health assessment and life prediction. Based on a finely tuned large language model, it automatically generates periodic health reports and archives them. The multi-terminal collaborative interaction module includes a Web management terminal, a mobile operation and maintenance terminal, an elevator car interaction terminal, and a monitoring terminal, which are respectively matched with the permissions and functional requirements of property management personnel, maintenance personnel, elevator passengers, and regulatory departments to achieve full-process collaboration in emergency response, maintenance scheduling, and off-site supervision.

10. A method for elevator fault prediction and health management based on artificial intelligence and the Internet of Things, characterized in that, The elevator fault prediction and health management system based on artificial intelligence and the Internet of Things, as described in any one of claims 1-9, includes the following steps: S1. Multimodal data acquisition: Raw data is acquired in real time through the multimodal acquisition unit of the perception layer. The raw data includes data on elevator core components, electrical circuits, and passenger scenarios. S2. Edge data processing: The on-site edge processing unit preprocesses and extracts features from the raw data, marks fault data, and completes rapid response to local emergency anomalies. S3. Encrypted data transmission: The transmission layer matches wired and wireless transmission modes according to the characteristics of the acquisition node, and transmits the data to the platform layer after end-to-end encryption through the built-in encryption module. S4. Automatic analysis: After decrypting and verifying the integrity of the received data, the platform layer analyzes the real-time feature data through a pre-trained multimodal fusion model to identify early abnormal signals and predict the type, probability, and occurrence window of potential faults. S5. Health Status Assessment: Based on real-time elevator operation data, fault prediction results and historical full life cycle data, output the overall health quantitative score, health level and degradation trend of the elevator and its core components. S6. Decision Output and Model Optimization: Based on the results of fault prediction and health assessment, generate decision instructions for on-demand maintenance, emergency response, and safety intervention. At the same time, store the newly added fault samples and operational data into the fault sample library to complete the incremental update and optimization of the model.

Images