A digital twin method for self-diagnosis of equipment status in vehicle in-and-out inspection shed

Abstract

The present invention discloses a digital twin method for self-diagnosis of equipment status in a vehicle entry and exit inspection shed, comprising: arranging inspection cameras and sensors in the vehicle entry and exit inspection shed, collecting and processing real-time images; using tools such as SolidWorks and Blender to physically model the equipment in the entry and exit inspection shed (including tracks, wheelset inspection equipment, inspection cameras, pantographs, etc.), and then forming a digital twin model; and transmitting data such as the operating status of each module of the equipment, detection data, operating history, inspection camera video, and video image processing results to the digital twin model through a network to complete the mapping, thereby truly reflecting the full life cycle process and operating status of the entry and exit inspection shed equipment; and embedding algorithms to complete self-diagnosis analysis of the internal and external status of the equipment and fault prediction.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent maintenance of rail transit and relates to a digital twin method for self-diagnosis of equipment status in a vehicle entry and exit inspection shed. Background Art

[0002] The equipment in the vehicle entry and exit inspection shed can detect various parameters of key parts of rail transit vehicles entering and leaving the warehouse, and alarm for excessive data.

[0003] At present, the monitoring of the operating status of the in-and-out inspection shed equipment mainly monitors the internal performance status of the in-and-out inspection shed equipment and presents it to customers in the form of a status list.

[0004] This method can display the internal performance status of the equipment in the warehouse inspection shed, but it cannot achieve the correspondence between the equipment status of the warehouse inspection shed and the actual equipment, and cannot display it visually, that is, the equipment and the status are separated; it is even more impossible to discover potential problems before a failure occurs, and it is impossible to predict potential problems. Maintenance personnel cannot reasonably and effectively plan the inspection and maintenance of the equipment; and only the internal performance status of the equipment is monitored, and the external status of the equipment cannot be monitored. Manual inspections are still required to solve the external status inspection of the equipment, such as possible intrusion, loose bolts, etc.

[0005] Therefore, a digital twin system is needed that can map the equipment in the vehicle entry and exit inspection shed in an all-round 1:1 manner, combine the inspection data, the internal performance status of the equipment, the appearance status of the equipment, the comprehensive status analysis and the maintenance decision analysis, discover potential problems before failures occur, monitor the changes of physical objects in the virtual model, diagnose multi-dimensional complex data processing and abnormal analysis based on artificial intelligence, predict potential failures, and plan maintenance reasonably and effectively. Summary of the invention

[0006] The purpose of the present invention is to provide a digital twin method for self-diagnosis of the equipment status of a vehicle entry and exit inspection shed, which can synchronize the equipment data of the entry and exit inspection shed to the equipment model in real time, and combine the detection data, internal performance status of the equipment, appearance status of the equipment, comprehensive status analysis and maintenance decision analysis together to achieve monitoring, predict equipment failures, and rationally plan maintenance.

[0007] The present invention solves the technical problem, and the technical solution adopted is:

[0008] A digital twin method for self-diagnosis of equipment status in a vehicle in-and-out inspection shed, comprising:

[0009] S1 deploys high-definition infrared inspection cameras and sensors to capture images from all angles and obtain status data of the equipment in the warehouse inspection shed;

[0010] S2 uses deep learning algorithms to train computer vision models and big data models to predict equipment status failures;

[0011] S3 also builds a physical model of the equipment in the warehouse inspection shed to form a digital twin model;

[0012] S4 communicates the status data and inspection images of the in-and-out inspection shed equipment with the digital twin model in real time to complete the mapping; at the same time, the computer vision model and the big data model are used to complete the self-diagnosis analysis of the internal and external status of the equipment and the fault prediction;

[0013] S5 real-time remote monitoring and operation.

[0014] Furthermore, the S1: arranging multiple groups of high-definition infrared inspection cameras to perform regular inspections on the appearance of the in-and-out inspection shed equipment from multiple angles; the sensors monitor the internal data of each module of the in-and-out inspection shed equipment in real time.

[0015] Furthermore, the S4: the status data of the in-and-out inspection shed equipment is packaged into a Json file and transmitted via the TCP / IP protocol;

[0016] The status data is synchronized to the digital twin model and displayed according to the preset animation effect.

[0017] Furthermore, S3: establishing a physical model of the inspection shed and the equipment in the shed through the structural diagram of the in-and-out inspection shed equipment and the CAD drawings of the civil engineering drawings, and forming the digital twin model through processing.

[0018] Furthermore, SolidWorks and Blender tools are used to process the physical model of the device to form the digital twin model.

[0019] Furthermore, in S2: a computer vision model is trained using a deep learning algorithm to identify the appearance status of the device; and a big data model is trained using a deep learning algorithm to predict faults of the internal and external status of the device.

[0020] Furthermore, in S2: a large number of inspection images are collected, and the raw data are cleaned, denoised and normalized to reduce uncertainty and noise interference of the data;

[0021] A convolutional neural network is used to build a computer vision model for fault classification. The inspection images are used to train the fault classification model. The back propagation algorithm is used to continuously adjust the model parameters to improve the model accuracy and robustness.

[0022] Furthermore, in S2: the trained computer vision model and big data model are deployed in the on-site graphics processing industrial computer to perform fault identification and alarm, and the processed video and alarm results can be synchronously transmitted to the digital twin model server.

[0023] Furthermore, the S5: runs in a local area network, and the mobile terminal or host used can receive the alarm information, and the data content accessible to different users is set according to user permissions.

[0024] The present invention establishes a digital twin model of the vehicle entry and exit inspection shed equipment, utilizes multiple sensors and intelligent inspection cameras of each module of the equipment, integrates equipment detection data, equipment internal performance status (including the operating status of each detection module and the probe self-inspection status), equipment appearance status and other data, completes mapping in the digital twin model, realizes real-time synchronization between reality and virtuality, and embeds algorithms in the digital twin model to complete equipment status self-diagnosis analysis and fault prediction.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) With the present invention, maintenance personnel can predict potential problems of the in-and-out inspection shed equipment, find dynamic and visual positioning of fault points, and then determine whether immediate maintenance is needed and reasonably plan the maintenance time;

[0027] (2) By using the present invention, inspection personnel can conduct intelligent inspections by remote login, and only make judgments on the inspection results. There is no need to enter the inspection shed for manual inspection every day. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the modular composition of the wheelset dynamic detection equipment (vehicle entry and exit detection shed equipment).

[0029] Figure 2 Rendering of the wheelset inspection site.

[0030] Figure 3 This is a video surveillance image of the inspection of the wheelset dynamic detection equipment.

[0031] Figure 4 This is the inspection video alarm picture of the wheelset dynamic detection equipment.

[0032] Figure 5 This is the digital twin rendering of the wheelset dynamic detection equipment.

[0033] Figure 6 The figure is a flow chart of the steps of the method of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0035] The principle of a digital twin method for self-diagnosis of the equipment status of a vehicle entry and exit inspection shed in the present embodiment is as follows: inspection cameras and sensors are arranged in the vehicle entry and exit inspection shed, and the real-time equipment appearance images and internal status of the equipment are collected and processed; the equipment physical model of the entry and exit inspection shed equipment (including tracks, wheelset inspection equipment, inspection cameras, pantographs, etc.) is processed using tools such as SolidWorks and Blender to form a digital twin model, and the operating status of each module of the equipment, detection data, operating history, inspection camera video, and video image processing results are transmitted to the digital twin model through the network to complete the mapping, which truly reflects the entire life cycle process and operating status of the entry and exit inspection shed equipment; at the same time, algorithms are embedded in the digital twin model to complete equipment status self-diagnosis analysis and fault prediction. Embodiment 1:

[0036] The process of the digital twin method for self-diagnosis of the equipment status of the vehicle in and out of the warehouse detection shed in this embodiment is as follows: Figure 6 As shown, the following steps are included:

[0037] S1: Deploy high-definition infrared inspection cameras and sensors to capture images without blind spots and obtain status data of the equipment in and out of the warehouse inspection shed;

[0038] S2: Use deep learning algorithms to train computer vision models and big data models to predict equipment status failures;

[0039] S3: Simultaneously build a digital twin model;

[0040] S4: Communicate the status data and inspection images with the digital twin model in real time to complete the mapping; at the same time, embed algorithms (computer vision model and big data model) to complete the self-diagnosis analysis of the internal and external status of the equipment and fault prediction;

[0041] S5: Real-time remote monitoring and operation.

[0042] Among them, in S1: high-definition infrared inspection cameras are arranged to conduct regular inspections of the equipment, rail fasteners and the appearance of the equipment in the shed. These cameras are installed in different positions to capture images from all angles without blind spots. These images can be transmitted to the back-end industrial computer for processing in real time. Sensors monitor the internal data of each module of the equipment in the shed.

[0043] In this embodiment, four groups of inspection cameras are arranged in the inspection room, and the collected real-time images are sent back to the image processing industrial computer for identification and classification to establish an inspection image database. According to the selected inspection camera, the video stream is called and the video data is displayed on the digital twin model, such as Figure 3 ;

[0044] When the inspection camera servo is triggered, a high-resolution picture is taken 10 seconds later. The picture is named according to the shooting time and stored in folders based on the inspection point.

[0045] Among them, in S2: use deep learning algorithms to train computer vision models to identify the appearance status of equipment, such as missing bolts, damaged probes, etc., and use deep learning algorithms to train big data models to predict faults of the internal and external status of the inspection shed equipment. The deep learning algorithm enables it to accurately identify these faults from the inspection images.

[0046] Preferably, these models can be run on a back-end industrial computer and the results can be sent back to the front-end digital twin model server for display.

[0047] In this embodiment, the trained computer vision model and big data model are deployed in the on-site graphics processing industrial computer to perform fault identification and alarm, such as Figure 4 shown.

[0048] In this embodiment, a large number of inspection images are collected, and the original data are cleaned, denoised, and normalized to reduce data uncertainty and noise interference.

[0049] In this embodiment, a convolutional neural network is used to construct a fault classification model (computer vision model), and the inspection images are used to train the fault classification model. Through the back propagation algorithm, the model parameters are continuously adjusted to improve the accuracy and robustness of the model. Preferably, when the data such as the video of the appearance of the equipment obtained by monitoring is large, the above large data is first transmitted to the industrial computer for processing, and then enters the trained computer vision model for prediction after processing. When the data obtained by monitoring is small, such as the internal data of each module of the equipment, the internal data is directly transmitted to the trained big data model for prediction and judgment.

[0050] About the big data model: In this embodiment, some deep learning algorithms are embedded in the digital twin model, and the big data is summarized, aggregated, and compared to achieve the prediction of equipment internal and external state faults. The fault data and abnormal data of more than 100 sets of in-and-out inspection shed equipment in the country that have been running for many years are used to establish a database, annotate these data, and then classify them and place them in a folder respectively as training data, verification data, test data, and fault comparison data.

[0051] Among them, in S3: First, the physical model of the equipment required for the digital twin of the inspection shed and the equipment in the shed is established through CAD drawings such as the inspection shed and trackside equipment structure drawings, civil engineering drawings, etc. The physical model of the equipment includes various parameters and numbers such as structure and function to simulate the operating status of the real equipment.

[0052] Then, the physical model of the device is rendered and processed to form a digital twin model:

[0053] Specifically, yes Figure 1 For the on-site equipment, SolidWorks is used to draw the corresponding sketches, and the 1:1 scale entity construction of each component is realized through stretching, cutting, scanning, lofting, array and other operations. The components are assembled into a complete object model through axis matching and surface matching. After the geometric entity modeling is completed, it is converted into an STL file for export. Then, the STL data of each part is subjected to limited surface merging, simplification, surface reduction and other operations to control the size of a single model file.

[0054] In Blender, shader node tools are used to make maps based on physical rendering, and the maps are combined into materials. Each material includes diffuse maps, metalness maps, normal maps, roughness maps, etc. After the materials are made, they are assigned to the corresponding faces on the above models, and then baking, high-poly conversion to low-poly operations, etc. are performed to control the fineness of the model; finally, the FBX file with embedded materials and textures is exported as an editor asset. Use Blender to draw models of wheel shape detection modules, wheel scratch detection modules, flaw detection modules, rail models, equipment power cord models, data cable models, and various fastener models with anti-loosening marks, etc., to form a series of digital twin models of in-and-out warehouse inspection shed equipment.

[0055] like Figure 2 As shown, two states of animation effects are produced based on the digital twin model, one is a fault state and the other is a non-fault state.

[0056] Among them, in S4: the status data and inspection images of the detection shed equipment are transmitted to the digital twin model in real time through the local area network, and the model updates the status of the equipment in real time according to the received data; at the same time, deep learning algorithms (computer vision models and big data models) are used to make timely warnings for the faults found, realizing automatic inspection and fault diagnosis of the equipment.

[0057] The status data of the equipment in the inspection room is packaged into a Json file and transmitted via the TCP / IP protocol. The status data is synchronized to the digital twin model and displayed according to the preset animation effects, such as Figure 5 .

[0058] The status data includes the operation status and detection data of each detection module, equipment appearance, etc. Specifically, the status data includes: the operation status of the on-site scratch detection module (sensor status, acquisition board status, software operation status), the operation status of the external dimension detection module (internal temperature status, laser status, laser driver board status, camera status, transmission board status, water circuit status, operation software status), the operation status of the flaw detection module (acquisition board status, power board status, control board status, water circuit status, water pressure status, software operation status), probe self-test data, chiller operation status (phase sequence switch, wrong phase, missing phase, compressor 1 low voltage, compressor 1 high voltage / overload, electric auxiliary heating overload, electric heating 1 overload, electric heating overtemperature, insufficient evaporation water flow, air conditioning pump overload, condensing fan 1 overload, compressor current too large, condensing fan 1 current too large, compressor current too small, evaporation temperature probe failure), UPS status ( Output voltage, output frequency, output current, AC input, input voltage, input frequency, battery charge, battery temperature, battery voltage, remaining time), computer room air conditioning status (computer room temperature, computer room humidity, computer room air conditioning status), computer room power distribution cabinet status (power distribution cabinet temperature, humidity, cooling fan status), coupled water system status (water pump status, water level monitoring, water flow monitoring, water temperature), system hard disk status (total capacity, available capacity), system firewall status (switch status, interception status), pantograph detection module operation status (camera status, pressure detection device status, roof monitoring detection status, software operation status), body appearance module operation status (camera status, laser status, motor status, software operation status), appearance status of each detection module (including track and corresponding fasteners) and detection data, etc.

[0059] Among them, in S5: use the local area network within the segment to transmit the collected data to the integrated control room or the mobile device of the inspection personnel, so as to timely discover potential equipment failures, issue early warnings and take corresponding measures.

[0060] S5: Running in the local area network, the mobile terminal or host used can receive alarm information, and set the data content accessible to different users according to user permissions.

[0061] The above description of the embodiments enables professional and technical personnel in the field to implement or use the present invention. Various modifications to the embodiments are obvious to professional and technical personnel in the field. The general principles defined in this document can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in this document. Any technical solution obtained by equivalent replacement or equivalent transformation shall fall within the protection scope of the present invention.

Claims

1. A digital twin method for self-diagnosis of equipment status in a vehicle entry and exit inspection shed, characterized in that: The method comprises: S1 deploys high-definition infrared inspection cameras and sensors to capture images from all angles and obtain status data of the equipment in the warehouse inspection shed; S2 uses deep learning algorithms to train computer vision models and big data models to predict equipment status failures: Use deep learning algorithms to train computer vision models to identify the appearance of equipment. Use deep learning algorithms to train big data models to predict faults based on the internal and external conditions of equipment. Deploy the trained computer vision model and big data model in the on-site graphics processing industrial computer to perform fault identification and alarm. At the same time, the processed video and alarm results can be synchronously transmitted to the server of the digital twin model. S3 simultaneously establishes a physical model of the equipment of the in-and-out inspection shed, thereby forming a digital twin model: the physical model of the equipment of the inspection shed and the equipment in the shed is established through the CAD drawings of the structural diagram and civil engineering diagram of the in-and-out inspection shed, and the digital twin model is formed through processing; S4 communicates the status data and inspection images of the in-and-out inspection shed equipment with the digital twin model in real time to complete the mapping; at the same time, the computer vision model and the big data model are used to complete the self-diagnosis analysis and fault prediction of the internal and external status of the equipment; the status data is synchronized to the digital twin model and displayed according to the preset animation effect; S5 real-time remote monitoring and operation; running in the local area network, the mobile terminal or host used can receive alarm information, and set the data content accessible to different users according to user permissions.

2. A digital twin method for self-diagnosis of equipment status of a vehicle in-and-out inspection shed as claimed in claim 1, characterized in that: S1: multiple groups of high-definition infrared inspection cameras are arranged to perform regular inspections on the appearance of the in-and-out inspection shed equipment from multiple angles; the sensors monitor the internal data of each module of the in-and-out inspection shed equipment in real time.

3. A digital twin method for self-diagnosis of equipment status of a vehicle entry and exit inspection shed as claimed in claim 1 or 2, characterized in that: S4: The status data of the in-and-out inspection shed equipment is packaged into a Json file and transmitted via the TCP / IP protocol.

4. A digital twin method for self-diagnosis of equipment status of a vehicle in-and-out inspection shed as claimed in claim 1, characterized in that: The physical model of the device is processed using SolidWorks and Blender tools to form the digital twin model.

5. A digital twin method for self-diagnosis of equipment status of a vehicle in-and-out inspection shed as claimed in claim 1, characterized in that: S2: collecting a large number of inspection images, cleaning, denoising and normalizing the original data to reduce data uncertainty and noise interference; A convolutional neural network is used to build a computer vision model for fault classification. The inspection images are used to train the fault classification model. The back propagation algorithm is used to continuously adjust the model parameters to improve the model accuracy and robustness.

Images