Open elevator fault identification system

Abstract

The present invention provides an open elevator fault identification system, which relates to the elevator Internet of Things and computer software technology fields. The system includes an elevator fault identification front end, a server end, and a client end. The data acquisition device of the elevator fault identification front end collects video stream data and sends it to an edge computing device. The fault algorithm model library of the edge computing device includes a fault algorithm model, and the fault identification module uses the fault algorithm model in the fault algorithm model library to perform fault identification on the video stream data. When the maintenance engineer obtains the video stream data and captures the elevator operation fault characteristics, the fault characteristics are input into the model training server through the model training APP of the client. The model training server uses the input fault characteristics to train a new fault algorithm model and updates it to the corresponding fault algorithm model library. This solution can open the elevator fault model training service to the maintenance engineers, continuously update the model rule library, and thus improve the accuracy of fault identification.

Description

Technical Field

[0001] The present invention relates to the field of elevator Internet of Things and computer software technology, and in particular to an open elevator fault identification system. Background Art

[0002] With the rapid increase of high-rise buildings in various places, elevators have played an increasingly important role in cross-floor dispatching in high-rise buildings, greatly improving the travel efficiency of personnel, while also providing convenient and comfortable travel conditions for specific groups such as the disabled and the elderly.

[0003] The operation mechanism of the elevator includes power system, control system, sensor and safety protection device. The overall process includes passenger calling, elevator response, passenger entry, target floor selection, operation process, arrival floor and cyclic operation. The elevator is a highly integrated product of machinery, elevator and electronics, consisting of mechanical system and electrical system. The elevator is powered on for a long time, and the structure of the electrical control system is very complex. There are high-voltage devices. Due to electronic devices, electrical components are installed in the machine room, shaft, pit, car, floor station, floor door and car door. Under the influence of many factors such as high-frequency use of elevators, uncivilized elevator riding behavior of elevator passengers, wear of elevator equipment and complex environment, emergencies caused by elevator failures occur from time to time, posing a serious threat to the life and property safety of elevator passengers. Therefore, monitoring and fault diagnosis of the elevator status can be used to know the fault and fault type in advance when the elevator fails, thereby shortening the time for fault handling, improving the maintenance efficiency of staff, reducing unnecessary personal injuries and losses, and providing strong technical support for ensuring the smooth operation of the elevator.

[0004] With the development of AI technology, machine vision has been used to diagnose and alarm elevator faults in real-time monitoring and identification. However, the existing elevator fault identification solutions usually fix a number of limited fault models in the machine vision equipment. For elevators, the corresponding fault types are complex, and it is difficult to cover various fault types through limited elevator fault models. Moreover, as the elevator is used for a longer time, the fault types of the elevator will become more complex. Therefore, the existing elevator fault identification solutions can only cover a limited number of elevator fault types, resulting in low accuracy and precision in fault identification. Summary of the invention

[0005] In view of this, in order to address the above shortcomings, it is necessary to propose an open elevator fault identification system that can open elevator fault model training services to elevator maintenance engineers, so as to continuously update the rule base of the elevator fault identification model and thus improve the accuracy of fault identification.

[0006] The present invention provides an open elevator fault identification system, which includes: an elevator fault identification front end, a server and a client; the elevator fault identification front end is connected to the server via 5G RedCap communication, and the server and the client are connected via Ethernet communication;

[0007] The elevator fault identification front end includes a data acquisition device and an edge computing device, which are connected via power carrier or Ethernet communication; the data acquisition device is used to collect video stream data of N areas of the elevator and send it to the edge computing device; the edge computing device includes a fault identification module and N fault algorithm model libraries, each of which includes at least one fault algorithm model, and the i-th fault algorithm model library corresponds to the acquisition device of the video stream data of the i-th area, and the fault identification module uses the fault algorithm model in the i-th fault algorithm model library to perform fault identification on the video stream data of the i-th area, N≥1, 1≤i≤N;

[0008] The client includes a model training APP, and the server includes a model training server; when the elevator maintenance engineer obtains the video stream data collected by the data acquisition device and captures the elevator operation fault characteristics, the fault characteristics are input into the model training server of the server through the model training APP of the client. The model training server uses the input fault characteristics to train a new fault algorithm model, and notifies the edge computing device to download the new fault algorithm model to the corresponding fault algorithm model library.

[0009] Preferably, the data acquisition device includes a machine room camera, a hoistway camera, a pit camera and a car camera, which are respectively used to collect video stream data of four areas: the machine room, the hoistway, the pit and the car; the edge computing device includes a machine room fault algorithm model library, a hoistway fault algorithm model library, a pit fault algorithm model library and a car fault algorithm model library, which are respectively used to identify faults in the machine room, the hoistway, the pit and the car.

[0010] Preferably, for each video stream data collected by a camera, the fault identification module uses each fault algorithm model in the fault algorithm model library corresponding to the camera to perform identification in sequence.

[0011] Preferably, the server also includes a fault alarm server, and the client also includes an elevator fault alarm handling APP; when the fault identification module identifies an elevator fault, the fault information is sent to the server; after receiving the fault information, the fault alarm server of the server pushes the fault information to the elevator fault alarm handling APP of the corresponding elevator maintenance engineer according to the management relationship of the elevator and the type of elevator fault, so that the elevator maintenance engineer can perform maintenance work on the elevator.

[0012] Preferably, the client also includes a data processing APP, which is used to receive video stream data collected by a data acquisition device imported by an elevator maintenance engineer, and is used by the elevator maintenance engineer to find fault points; when the elevator maintenance engineer finds an elevator fault in the video stream data based on experience, the data processing APP intercepts pictures containing fault features according to the trigger command of the elevator maintenance engineer, and inputs the picture data containing the fault features into the model training server on the server side through the model training APP for model training.

[0013] Preferably, the edge computing device also includes a data storage device, which is used to store the video stream data collected by the data acquisition device; the way in which the elevator maintenance engineer obtains the video stream data includes: exporting the video stream data from the data storage device of the on-site edge computing device.

[0014] Preferably, the server also includes a database server; the edge computing device sends the video stream data to the server while performing fault identification on the video stream data; the server receives the video stream data and stores it in the database server; the data processing APP is also used to send a video stream data acquisition request to the server; after receiving the video stream data acquisition request sent by the data processing APP and passing the identity verification, the server obtains the corresponding video stream data from the database server and sends it to the client's data processing APP.

[0015] Preferably, the edge computing device is an edge computing gateway, which adopts an ARM CPU+NPU architecture, the business logic is calculated by the CPU, and the reasoning of elevator fault identification is calculated by the NPU.

[0016] Preferably, the N fault algorithm model libraries can at least cover the identification of abnormal elevator door switch, abnormal hydraulic buffer, abnormal cable hooking, foreign matter in the pit, and loose speed limiter rope on the tensioner pulley.

[0017] Preferably, the method in which the fault identification module uses the fault algorithm model to perform fault identification on the video stream data includes:

[0018] Get 640×640×3 image data from the video stream data;

[0019] Use the focus network to compress the image data and obtain a primary feature layer;

[0020] The primary feature layer is processed at different levels using a convolutional neural network, normalization, and activation functions to obtain three secondary feature layers of 80×80×256, 40×40×512, and 20×20×1024 respectively;

[0021] The three secondary feature layers are passed into the feature pyramid network, and enhanced feature extraction is performed through upsampling feature fusion and downsampling feature fusion to obtain fused features;

[0022] The fusion features are used to perform regression prediction and classification prediction to obtain a fault identification result.

[0023] It can be seen from the above technical scheme that the open elevator fault identification system provided by the embodiment of the present invention includes a fault identification front end, a server and a client. The data acquisition device in the elevator fault identification front end can collect video stream data of multiple areas and send it to the edge computing device. The edge computing device stores a fault algorithm model library for fault identification in each area, so that the fault identification module uses the fault identification algorithm model in the corresponding fault algorithm model library to identify the video stream data, thereby realizing the identification of elevator faults. Moreover, the client includes a model training APP, and the server includes a model training server. When the elevator maintenance engineer obtains the video stream data collected by the data acquisition device and captures the fault characteristics of the elevator operation, he can input the fault characteristics into the model training server on the server through the model training APP on the client. The model training server uses the input fault characteristics to train a new fault algorithm model. The trained new fault algorithm model will be downloaded by the edge computing device to the corresponding fault algorithm model library. In this way, the training of the fault algorithm model is open to elevator maintenance engineers. Elevator maintenance engineers can continuously update the fault algorithm model library according to actual data, so that the fault algorithm model library grows dynamically and becomes more and more intelligent, thereby making the accuracy of fault identification higher and higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of an open elevator fault identification system provided by an embodiment of the present invention.

[0025] Figure 2 A schematic diagram of an elevator fault identification front end provided by an embodiment of the present invention.

[0026] Figure 3 A schematic diagram of another open elevator fault identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] See also Figure 1-3 , the present invention provides an open elevator fault identification system, the system comprising: an elevator fault identification front end, a server and a client; the elevator fault identification front end is connected to the server via 5G RedCap communication, and the server and the client are connected via Ethernet communication;

[0029] The elevator fault identification front end includes a data acquisition device and an edge computing device, which are connected via power carrier or Ethernet communication; the data acquisition device is used to collect video stream data of N areas of the elevator and send it to the edge computing device; the edge computing device includes a fault identification module and N fault algorithm model libraries, each of which includes at least one fault algorithm model, and the i-th fault algorithm model library corresponds to the acquisition device of the video stream data of the i-th area, and the fault identification module uses the fault algorithm model in the i-th fault algorithm model library to perform fault identification on the video stream data of the i-th area, N≥1, 1≤i≤N;

[0030] The client includes a model training APP, and the server includes a model training server; when the elevator maintenance engineer obtains the video stream data collected by the data acquisition device and captures the elevator operation fault characteristics, the fault characteristics are input into the model training server on the server through the client's model training APP. The model training server uses the input fault characteristics to train a new fault algorithm model, and notifies the edge computing device to download the new fault algorithm model to the corresponding fault algorithm model library.

[0031] In view of the current situation where elevator faults are identified, only a limited number of fault models are fixed in machine vision equipment, which can only identify limited elevator faults, resulting in low accuracy in fault identification. This solution considers building an open elevator fault identification platform, mobilizing and relying on the elevator maintenance unit's deep understanding of elevator operation faults, combined with machine vision, and the elevator maintenance engineer selects the elevator fault video stream information on the AI ​​open elevator fault identification platform established by the present invention, autonomously trains the fault model algorithm on the model training server on the service side, and seamlessly upgrades the newly generated fault algorithm model to the fault algorithm model library of the edge computing device through the system, so that the elevator fault identification type of the elevator fault identification front-end equipment can grow dynamically and become more and more intelligent, thereby making the accuracy of fault identification higher and higher.

[0032] Elevator failures usually occur in areas such as the machine room, hoistway, pit, and car. Therefore, in one embodiment, the data acquisition device may include a machine room camera, a hoistway camera, a pit camera, and a car camera, which are respectively used to collect video stream data in the four areas of the machine room, hoistway, pit, and car; the edge computing device includes a machine room fault algorithm model library, a hoistway fault algorithm model library, a pit fault algorithm model library, and a car fault algorithm model library, which are respectively used to identify faults in the machine room, hoistway, pit, and car.

[0033] In this embodiment, it is considered to install a high-definition camera in the machine room, shaft, pit and car of the elevator. The camera transmits the RTSP video stream data of the elevator operation to the edge computing device through Ethernet or power carrier. The edge computing device also includes a fault algorithm model library for these four areas, namely, the machine room fault algorithm model library, the shaft fault algorithm model library, the pit fault algorithm model library and the car fault algorithm model library. In this way, the fault algorithm model library is divided according to the area where the video data stream is collected. The video stream data collected by the camera in the corresponding area can be used for fault identification using the fault algorithm model library in the corresponding area, which can not only improve the accuracy of fault identification, but also greatly improve the efficiency of fault identification.

[0034] Specifically, each fault algorithm model library includes several fault algorithm models. Therefore, for each video stream data collected by a camera, the fault identification module uses each fault algorithm model in the fault algorithm model library corresponding to the camera to identify the video stream data in turn when performing fault identification. For example, the shaft fault algorithm model library has 5 fault algorithm models, namely A1, A2, A3, A4, and A5. When performing fault identification on the video stream data collected by the shaft camera, the fault identification module uses fault algorithm models A1-A5 to identify the video stream data in turn, and the results of each fault algorithm model identification are the fault identification results corresponding to the video stream data.

[0035] Of course, it is easy to understand that the data collection areas may include more than the four areas listed above, and the number of cameras installed in each location may not be limited to one.

[0036] In this embodiment, the edge computing device can be an edge computing gateway, which adopts the architecture of ARM CPU+NPU, the business logic is calculated by the CPU, and the reasoning of elevator fault identification is calculated by the embedded neural network processor NPU, so that the business logic and the reasoning of elevator fault identification are processed by different processors, and the NPU adopts the architecture of "data-driven parallel computing", which is particularly good at processing massive multimedia data such as videos and images, and can solve the problem of low computing efficiency of the CPU when processing neural network operations. The CPU starts four processes to collect video stream data from four cameras respectively, and then the fault identification module sends the video stream data to four corresponding fault algorithm model libraries respectively, and each fault algorithm model in each fault algorithm model library identifies the input video stream data in turn.

[0037] Of course, it should be pointed out that the number of fault algorithm models contained in each fault algorithm model library may be the same or different, that is, n1, n2, n3, and n4 in the accompanying drawings may be completely the same or different, and can be set according to specific circumstances.

[0038] In one embodiment, the server also includes a fault alarm server, and the client also includes an elevator fault alarm handling APP; when the fault identification module identifies an elevator fault, the fault information is sent to the server; after receiving the fault information, the server pushes the fault information to the elevator fault alarm handling APP of the corresponding elevator maintenance engineer based on the management relationship of the elevator and the type of elevator fault, so that the elevator maintenance engineer can perform maintenance work on the elevator.

[0039] In this embodiment, when the fault identification module in the edge computing device identifies an elevator fault, the edge computing device reports the fault information to the cloud elevator fault alarm server through the communication protocol interacting with the server, and then pushes the elevator fault information to the corresponding elevator maintenance engineer's elevator fault alarm handling APP according to the management relationship and the elevator fault type, so as to perform elevator maintenance operations. It can be seen that in this embodiment, when the elevator fault information is pushed, the fault alarm server pushes it according to the management relationship and the elevator fault type. For example, the elevator maintenance engineer who manages the machine room, the elevator maintenance engineer who manages the well, the elevator maintenance engineer who manages the pit, and the elevator maintenance engineer who manages the car. Moreover, for the elevator maintenance engineer, the types of elevator faults that each elevator maintenance engineer is good at handling are not exactly the same, so it is considered to push the elevator fault information according to the elevator fault type, so that the elevator fault can be solved with higher quality and faster speed.

[0040] In order to facilitate elevator maintenance engineers to process video stream data, the processed data can be used to train fault algorithm models. In one embodiment, the client can also include a data processing APP, which is used to receive video stream data collected by a data acquisition device imported by the elevator maintenance engineer, and is used by the elevator maintenance engineer to find fault points; when the elevator maintenance engineer finds that there is an elevator fault in the video stream data based on experience, the data processing APP intercepts the picture containing the fault characteristics according to the trigger command of the elevator maintenance engineer, so as to input the picture data containing the fault characteristics into the model training server of the server through the model training APP for model training.

[0041] In this embodiment, the elevator maintenance engineer can search and analyze the fault characteristics through the client's data processing APP, and process the video stream data at the same time, which is more convenient for the operation of the elevator maintenance project and improves the video analysis efficiency of the elevator maintenance engineer. Furthermore, if the elevator maintenance engineer captures the fault characteristics of the elevator operation, the image data of the elevator fault characteristics can be input into the model training server on the server side through the model training APP to train the fault algorithm model, thereby forming a new elevator fault algorithm model, and then notifying the edge computing device to download the newly added algorithm model through the communication protocol. It can be seen that this solution provides an open elevator fault model training platform, with a large number of elevator maintenance units providing professional elevator fault information extraction and training, and the elevator edge computing gateway dynamically adding fault algorithm models, forming a dynamically growing and increasingly intelligent elevator fault identification mode, thereby greatly improving the richness, accuracy and precision of elevator fault alarms.

[0042] The video stream data imported into the data processing APP by the elevator maintenance engineer can come directly from the edge computing device or be obtained from the server. For example, in one embodiment, the edge computing device also includes a data storage device, which is used to store the video stream data collected by the data acquisition device; the elevator maintenance engineer obtains the video stream data by exporting the video stream data from the data storage device of the edge computing device on site.

[0043] Of course, in another embodiment, the server also includes a database server; the edge computing device sends the video stream data to the server while performing fault identification on the video stream data; the server receives the video stream data and stores it in the database server; the data processing APP is also used to send a video stream data acquisition request to the server; after receiving the video stream data acquisition request sent by the data processing APP and the identity verification is passed, the server obtains the corresponding video stream data from the database server and sends it to the client's data processing APP.

[0044] It can be seen that elevator maintenance engineers can export the video stream data in the edge computing device through the storage device when on site, or they can request the video stream data from the server by communicating with the server, and then use the obtained video stream data to find and analyze fault characteristics, and use the obtained fault characteristics to train the fault algorithm model.

[0045] Of course, the data transmission considers using the Http standard hypertext transfer protocol, and in order to ensure the security of data transmission, consider adding fields to the Http protocol Header, such as adding Bclient to represent AI-BOX, a fixed item; adding Fault-code to represent the fault code; adding Fault-time to represent the fault timestamp; adding File-name to represent the image or file name; adding Check-hash to represent the check hash code, which is calculated using a fixed addition code and variables. The specific calculation method is Md5, and the format is "AI-BOX+File-name+Fault-time".

[0046] In the actual application of elevators, the most common faults include abnormal opening and closing of elevator floor doors, abnormal hydraulic buffers, abnormal cable hooking, foreign objects in the pit, loosening of the speed limiter rope on the tensioner pulley, etc. For example, for abnormal opening and closing of elevator floor doors, the elevator floor doors may not be opened or closed, or the elevator may be running during operation, while the floor doors are open. This may be caused by a sensor failure of the floor door, a problem with the floor door track, or a failure of the floor door lock; the closing speed of the elevator floor door should be appropriate, neither too fast nor too slow. If the elevator floor door closes too quickly, there may be safety hazards and passengers may be pinched. If the elevator floor door closes too slowly, the elevator may not operate normally or malfunction. The elevator floor door should have a suitable time when it is opened so that passengers can enter and exit the elevator. If the elevator floor door is open for too long, the elevator may not operate normally or malfunction.

[0047] For another example, in the case of an abnormal elevator hydraulic buffer, when the elevator rapidly descends to the bottom pit buffer, the buffer can slow down the impact of the elevator and protect the safety of passengers and the elevator itself. Common elevator buffer failures include: lack of oil in the hydraulic buffer, failure to reset the hydraulic buffer after actuation, or failure to meet the reset time requirements. When these failures occur, timely maintenance or replacement of the hydraulic buffer is required to ensure the normal operation of the elevator and the safety of passengers. In addition to routine maintenance and inspection, fault analysis of elevator buffers also requires professional technical support. Only on the basis of understanding the cause of the failure can correct maintenance measures be taken to avoid safety accidents.

[0048] For another example, in the case of cable hooking anomalies, the elevator's accompanying cable is an important part of the elevator's operation. If there is a problem, it may cause the elevator to lose control, causing serious safety risks, such as falling, unexpected elevator stops, etc.; it may cause the elevator to fail to operate normally or fail to stop at the required floor; in more serious cases, the accompanying cable may be pulled off. This may cause surprises to passengers and may also cause the elevator to interrupt operation; it may cause damage to other elevator equipment, such as cables, rails, and cars, which may cause the overall performance of the elevator to decline; the elevator cable hooking failure requires professional maintenance personnel to handle, and may require replacement of parts or maintenance work. This may increase maintenance costs and downtime.

[0049] For another example, the loose speed limiter rope on the tensioner pulley means that the speed limiter rope on the elevator tensioner pulley is loose. A severely loose speed limiter rope will trigger the elevator safety switch, and then the elevator will stop running. This failure may be due to rope aging. The elevator speed limiter rope will be worn, corroded, fatigued, etc. after long-term use, causing the wire rope to become loose; speed limiter rope system failure. The speed limiter rope system of the elevator includes components such as the tensioner pulley and the speed limiter rope. If these components are damaged or malfunction, the elevator speed limiter rope will become loose; elevator control system failure. If the sensors, rope residual detectors and other components in the elevator control system are damaged or malfunction, the elevator will misjudge the tension state of the speed limiter rope, thereby loosening the speed limiter rope. It should be noted that the loose speed limiter rope of the elevator is a serious fault. If it is not repaired in time, it may cause dangerous situations such as the elevator losing control and suddenly falling.

[0050] For example, if there are foreign objects in the pit, this fault may cause the elevator to be unable to close the pit door normally, the elevator to be obstructed or stuck when going up and down, and other problems.

[0051] That is to say, the above five types of elevator failures are the most common elevator failures, and the hazards when they occur are also relatively high. Therefore, when arranging the fault algorithm model library, the fault algorithm models used to identify these five types of failures are first arranged in the fault algorithm model library to ensure the safe operation of the elevator to the greatest extent. Subsequently, the fault algorithm model library is continuously updated by elevator maintenance engineers, thereby continuously improving the accuracy and reliability of fault identification.

[0052] In one embodiment, when the fault identification module uses the fault algorithm model to identify the fault of the video stream data, it can be implemented in the following manner:

[0053] Get 640×640×3 image data from the video stream data;

[0054] Use the focus network to compress the image data and obtain a primary feature layer;

[0055] The primary feature layer is processed at different levels using convolutional neural networks, normalization, and activation functions to obtain three secondary feature layers of 80×80×256, 40×40×512, and 20×20×1024 respectively;

[0056] The three secondary feature layers are passed into the feature pyramid network, and enhanced feature extraction is performed through upsampling feature fusion and downsampling feature fusion to obtain fused features;

[0057] The fusion features are used for regression prediction and classification prediction to obtain the fault identification results.

[0058] In this embodiment, when performing elevator fault identification, image data is obtained from the video stream data, and a 640×640×3 image is input into the fault algorithm model. The backbone feature extraction network of the fault algorithm model will enhance the features of this image. First, the image is compressed in height and width through the focus network to obtain a feature layer, and then continuously through convolution, standardization and activation functions to obtain three effective feature layers of 80×80×256, 40×40×512 and 20×20×1024 respectively, which are passed into the feature pyramid network FPN to enhance feature extraction through upsampling feature fusion and downsampling feature fusion, and finally regression prediction and classification prediction are performed in YoloHead to obtain the fault identification result. It can be seen that this solution extracts features of different levels when performing elevator fault identification, so that monitoring can be performed at different scales, which can greatly improve the accuracy of elevator fault identification.

[0059] The modules or units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. The above disclosure is only the preferred embodiment of the present invention, and of course it cannot be used to limit the scope of the rights of the present invention. Those skilled in the art can understand that all or part of the processes of the above embodiment are implemented, and the equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. An open elevator fault identification system, characterized in that: The system includes: an elevator fault identification front end, a server and a client; the elevator fault identification front end is connected to the server via 5G RedCap communication, and the server and the client are connected via Ethernet communication; The elevator fault identification front end includes a data acquisition device and an edge computing device, which are connected via power carrier or Ethernet communication; the data acquisition device is used to collect video stream data of N areas of the elevator and send it to the edge computing device; the edge computing device includes a fault identification module and N fault algorithm model libraries, each of which includes at least one fault algorithm model, and the i-th fault algorithm model library corresponds to the acquisition device of the video stream data of the i-th area, and the fault identification module uses the fault algorithm model in the i-th fault algorithm model library to perform fault identification on the video stream data of the i-th area, N≥1, 1≤i≤N; The client includes a model training APP, and the server includes a model training server; when the elevator maintenance engineer obtains the video stream data collected by the data acquisition device and captures the elevator operation fault characteristics, the fault characteristics are input into the model training server of the server through the model training APP of the client, and the model training server uses the input fault characteristics to train a new fault algorithm model, and notifies the edge computing device to download the new fault algorithm model to the corresponding fault algorithm model library; The client also includes a data processing APP, which is used to receive video stream data collected by a data collection device imported by an elevator maintenance engineer, and is used by the elevator maintenance engineer to find fault points; when the elevator maintenance engineer finds that there is an elevator fault in the video stream data based on experience, the data processing APP intercepts a picture containing fault features according to a trigger command of the elevator maintenance engineer, and inputs the picture data containing the fault features into a model training server on the server side through a model training APP for model training; The method in which the fault identification module uses the fault algorithm model to identify the fault of the video stream data includes: Get 640×640×3 image data from the video stream data; Use the focus network to compress the image data and obtain a primary feature layer; The primary feature layer is processed at different levels using a convolutional neural network, normalization, and activation functions to obtain three secondary feature layers of 80×80×256, 40×40×512, and 20×20×1024 respectively; The three secondary feature layers are passed into the feature pyramid network, and enhanced feature extraction is performed through upsampling feature fusion and downsampling feature fusion to obtain fused features; The fusion features are used to perform regression prediction and classification prediction to obtain a fault identification result.

2. The open elevator fault identification system according to claim 1, characterized in that: The data acquisition device includes a machine room camera, a shaft camera, a pit camera and a car camera, which are respectively used to collect video stream data of four areas: the machine room, the shaft, the pit and the car; the edge computing device includes a machine room fault algorithm model library, a shaft fault algorithm model library, a pit fault algorithm model library and a car fault algorithm model library, which are respectively used to identify faults occurring in the machine room, the shaft, the pit and the car.

3. The open elevator fault identification system according to claim 2, characterized in that: For each video stream data collected by a camera, the fault identification module uses each fault algorithm model in the fault algorithm model library corresponding to the camera to perform identification in sequence.

4. The open elevator fault identification system according to claim 1, characterized in that: The server also includes a fault alarm server, and the client also includes an elevator fault alarm handling APP; When the fault identification module identifies an elevator fault, the fault information is sent to the server; after receiving the fault information, the fault alarm server of the server pushes the fault information to the elevator fault alarm handling APP of the corresponding elevator maintenance engineer according to the management relationship of the elevator and the type of elevator fault, so that the elevator maintenance engineer can perform maintenance work on the elevator.

5. The open elevator fault identification system according to claim 1, characterized in that: The edge computing device also includes a data storage device, which is used to store the video stream data collected by the data collection device; The method for elevator maintenance engineers to obtain video stream data includes: exporting the video stream data from the data storage device of the edge computing device on site.

6. The open elevator fault identification system according to claim 1, characterized in that: The server also includes a database server; the edge computing device sends the video stream data to the server while performing fault identification on the video stream data; the server receives the video stream data and stores it in the database server; the data processing APP is also used to send a video stream data acquisition request to the server; after receiving the video stream data acquisition request sent by the data processing APP and passing the identity verification, the server obtains the corresponding video stream data from the database server and sends it to the client's data processing APP.

7. The open elevator fault identification system according to claim 1, characterized in that: The edge computing device is an edge computing gateway, which adopts the ARM CPU+NPU architecture. The business logic is calculated by the CPU, and the reasoning of elevator fault identification is calculated by the NPU.

8. The open elevator fault identification system according to claim 1, characterized in that: The N fault algorithm model libraries can at least cover the identification of abnormal elevator door switches, hydraulic buffer abnormalities, cable hook abnormalities, foreign objects in the pit, and loose speed limiter rope on the tensioner pulley.

Images