Intelligent detection system for appearance defects of electric energy meter

Abstract

The invention discloses an intelligent detection system for appearance defects of an electric energy meter, and relates to the technical field of power equipment detection. The system comprises an annular multispectral light source module, a clamp mechanism with a pneumatic rotating table, a high-resolution industrial camera, an edge computing terminal and a cloud AI analysis platform, reflective interference is eliminated through multispectral imaging, rapid defect positioning and fine identification are achieved in combination with YOLOv8 and ResNet-50 models, and 30 + type electric energy meter detection is supported. The system detection speed is smaller than or equal to 2 seconds per machine, the defect recognition rate is larger than or equal to 99%, the problems that traditional detection is low in efficiency, insufficient in precision and poor in compatibility are solved, and the method is suitable for quality control scenes of large-scale production of electric energy meters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy meter, and in particular to an electric energy meter appearance defect intelligent detection system. BACKGROUND

[0002] As the core equipment of electric power metering, the appearance quality of electric energy meter directly affects the product reliability and service life. At present, the electric energy meter appearance detection mainly has the following problems: 1. Low efficiency of manual detection: the traditional manual detection relies on naked eye observation, and each person can detect at most 500 meters per day, and is affected by factors such as fatigue and experience, the missed detection rate is more than 5%, which is difficult to meet the demand of large-scale production.

[0003] 2. Limitations of existing machine vision scheme: Ⅰ. Affected by light, the mirror reflection of the metal shell of the electric energy meter is easy to cause defect misjudgment or omission; Ⅱ. The recognition rate is generally less than 90%, and the recognition ability of subtle defects (such as 0.1mm crack, small scratch) is insufficient; Ⅲ. Poor equipment compatibility, different types of electric energy meters need to replace special fixtures and detection programs, and the reuse rate is low.

[0004] 3. Difficulty in balancing detection speed and precision: when pursuing high detection speed, the precision is often sacrificed, and vice versa, which reduces production efficiency, and cannot meet the requirements of large-scale production and quality control. SUMMARY

[0005] In order to make up for the shortcomings of the prior art, the present application provides an electric energy meter appearance defect intelligent detection system to solve the problems in the prior art.

[0006] In order to solve the above technical problems, the present application provides the following technical scheme: An electric energy meter appearance defect intelligent detection system, comprising a ring-shaped multi-spectral light source module, a fixture mechanism with a pneumatic rotary table, a high-resolution industrial camera, an edge computing terminal and a cloud AI analysis platform; the ring-shaped multi-spectral light source module surrounds the shooting path of the high-resolution industrial camera, and is used for providing multi-band illumination; the fixture mechanism with a pneumatic rotary table is used for fixing and rotating the electric energy meter to be tested; the high-resolution industrial camera is used for collecting electric energy meter appearance images; the edge computing terminal is electrically connected with the high-resolution industrial camera, the ring-shaped multi-spectral light source module and the pneumatic rotary table respectively, and is used for receiving image data and performing preliminary defect detection; the cloud AI analysis platform is in communication connection with the edge computing terminal, and is used for fine analysis of the preliminary detection result.

[0007] As a further technical solution of the present application: the annular multi-spectral light source module comprises 8 groups of independently controllable LED light sources, covering visible light band, infrared band and ultraviolet band, and is provided with a polarization filter, which can switch the polarization mode to eliminate the reflection of the metal watchcase, and the light source brightness adjustment range is 50-1000Lux.

[0008] As a further technical solution of the present application: the clamp mechanism with a pneumatic rotary table comprises a three-dimensional adjustable clamp and a pneumatic rotary table; the clamping range is 50mm×80mm-200mm×300mm; the pneumatic rotary table can rotate 0-360°, and the rotation accuracy is ±0.5°, which is used to drive the electric energy meter to collect multi-angle images.

[0009] As a further technical solution of the present application: the resolution of the high-resolution industrial camera is not less than 4K, the frame rate is ≥29fps, the lens focal length can be adjusted, the shooting range covers the overall appearance of the electric energy meter, and supports horizontal 0-90°, vertical 0-45° shooting.

[0010] As a further technical solution of the present application: the edge computing terminal adopts NVIDIA Jetson AGX Orin processor, and is built-in with lightweight YOLOv8 model, which is used for real-time positioning of defect area, and the defect positioning response time is ≤0.5s, and the output defect preliminary judgment result.

[0011] As a further technical solution of the present application: it also comprises a data preprocessing module, which is used for processing the images collected by the high-resolution industrial camera, and the processing steps include image graying, denoising, edge enhancement, rotation, scaling, flipping and illumination equalization, so as to eliminate image interference and highlight defect features.

[0012] As a further technical solution of the present application: the edge computing terminal and the cloud AI analysis platform communicate through 5G or Ethernet, the edge computing terminal uploads the preliminary detection result and the original image data to the cloud AI analysis platform, the cloud AI analysis platform returns the refined detection report, including defect type, position, size and severity level.

[0013] As a further technical solution of the present application: it also comprises an automatic feeding and sorting mechanism, which is electrically connected with the edge computing terminal, and can sort the electric energy meter to qualified or unqualified material channel according to the detection result, and the sorting response time is ≤1s.

[0014] As a further technical solution of the present application: the layout of the annular multi-spectral light source module is annular array, the distance between adjacent LED light sources is 45°, the vertical distance between the light source and the surface of the electric energy meter can be adjusted, and the adjustment range is 100-300mm, so as to adapt to the shooting needs of electric energy meters of different sizes.

[0015] As a further technical solution of the application: the cloud AI analysis platform is deployed with a ResNet-50 multi-task neural network, and the ResNet-50 multi-task neural network is trained in the following way: The data set contains normal samples and defect samples of 30+ models of electric energy meters, and the sample size is ≥100,000; Data enhancement techniques are used, including multi-illumination condition simulation, defect morphology variation, and background interference addition; L1 regularization, L2 regularization, and Dropout algorithm are used to prevent overfitting, and the model verification set accuracy is ≥99.5%.

[0016] One or more technical solutions provided in the embodiments of the application have at least the following technical effects or advantages: 1. The detection efficiency is greatly improved: the single detection time is ≤2 seconds, and 1800 devices can be detected per hour, which is more than 15 times higher than manual detection (500 devices / day), meeting the needs of large-scale production; 2. High defect recognition accuracy: the recognition rate of defects ≥0.1mm is ≥99%, the missed detection rate is ≤0.1%, and the false detection rate is ≤0.5%, which is much higher than traditional machine vision solutions; 3. Strong compatibility: through three-dimensional adjustable clamps and multi-model adaptive algorithms, 30+ electric energy meter models are supported, the equipment reuse rate is 100%, and the enterprise equipment investment cost is reduced; 4. Excellent anti-interference ability: multi-spectrum imaging and polarization filtering technology effectively eliminates metal reflection, data preprocessing further optimizes image quality, and ensures detection stability under complex lighting conditions; 5. High degree of automation: integrated with automatic feeding, detection, and sorting of the whole process, reduces manual intervention, reduces labor costs, and avoids detection errors caused by human factors. BRIEF DESCRIPTION OF DRAWINGS

[0017] Fig. 1 is a schematic diagram of the overall structure of the system; Fig. 2 is a multi-spectrum light source layout section view; Fig. 3 is an AI processing flowchart. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0019] As Figs. 1-3As shown, an electric energy meter appearance defect intelligent detection system is composed of a ring-shaped multi-spectrum light source module, a clamp mechanism with a pneumatic rotary table, a high-resolution industrial camera, an edge computing terminal, a cloud AI analysis platform, a data preprocessing module, and an automatic feeding and sorting mechanism. Each module works together to realize full-automatic detection of electric energy meter appearance defects.

[0020] Ring-shaped multi-spectrum light source module: Structure: 8 groups of LED light sources are uniformly distributed in a ring shape (adjacent interval 45°), surrounding the shooting path, and the distance between the light source and the surface of the electric energy meter can be adjusted (100-300 mm). Spectrum coverage: visible light (400-760 nm), infrared (760-1500 nm), and ultraviolet (200-400 nm). Single or mixed wavebands can be switched according to the material of the electric energy meter. Anti-reflection design: Polarizing filter is configured to eliminate mirror reflection of metal watch case by adjusting polarization angle. Brightness can be automatically adjusted within the range of 50-1000 Lux.

[0021] Clamp mechanism with pneumatic rotary table: Clamp: Three-dimensional adjustable design, clamping range 50mm×80mm-200mm×300mm, compatible with 30+ models of electric energy meters such as 86 type and 118 type. Rotary table: SMCRTQ0720 pneumatic rotary table is adopted, supporting 0-360° continuous rotation, rotation accuracy ±0.5°, which can drive the electric energy meter to realize multi-angle (0°, 90°, 180°, 270°) shooting, ensuring no detection dead angle.

[0022] High-resolution industrial camera: Configuration: Basler acA4024-29um industrial camera is adopted, resolution 4024×3036 pixels, frame rate 29fps, with 8mm fixed focus lens. Shooting mode: Single frame shooting and continuous shooting are supported. Images under different lighting conditions can be synchronously collected when the multi-spectrum light source is switched, ensuring complete capture of defect information.

[0023] Edge computing terminal and cloud AI analysis platform: Edge computing terminal: Based on NVIDIA Jetson AGX Orin processor, lightweight YOLOv8 model is deployed to realize fast positioning of defects (response time ≤0.5 seconds) and screen out suspected defect areas. Cloud AI analysis platform: ResNet-50 multi-task network is adopted to perform fine analysis on suspected defects, output specific parameters of defects such as scratches, missing characters, and structural deformation, and the accuracy meets the ISO2859-1 standard.

[0024] Data preprocessing and model training: Preprocessing steps: including image graying, denoising (Gaussian filter), edge enhancement (Sobel operator), illumination equalization, etc., to eliminate background interference; Model training: trained by 100,000+ sample data set (including normal and defect samples), using data augmentation (rotation, scaling, flipping) and regularization (L1, L2) techniques to prevent overfitting, with strong model generalization ability.

[0025] Workflow: The automatic feeding mechanism sends the electric energy meter to the clamp mechanism with a pneumatic rotary table, and the clamp is self-adaptive clamping; The annular multi-spectral light source module switches different waveband light sources, and the high-resolution industrial camera synchronously collects multi-angle images; The data preprocessing module optimizes the image, and the edge computing terminal runs the YOLOv8 model for preliminary detection; Suspected defect images are uploaded to the cloud AI analysis platform, and the ResNet-50 model is used for fine identification; The cloud returns the detection report, and the edge computing terminal controls the sorting mechanism to classify the electric energy meter into the qualified / unqualified material channel.

[0026] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application.

[0027] In addition, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for the sake of clarity, those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment have been properly combined to form other embodiments easily understood by those skilled in the art.

Claims

1. An appearance defect intelligent detection system for an electric energy meter, characterized in that, It comprises a ring-shaped multi-spectral light source module, a clamp mechanism with a pneumatic rotary table, a high-resolution industrial camera, an edge computing terminal, and a cloud AI analysis platform; the ring-shaped multi-spectral light source module surrounds the shooting path of the high-resolution industrial camera to provide multi-band illumination; the clamp mechanism with a pneumatic rotary table is used to fix and rotate the meter to be tested; the high-resolution industrial camera is used to collect the appearance image of the meter; the edge computing terminal is electrically connected with the high-resolution industrial camera, the ring-shaped multi-spectral light source module, and the pneumatic rotary table, respectively, to receive image data and perform preliminary defect detection; the cloud AI analysis platform is in communication connection with the edge computing terminal to perform fine analysis on the preliminary detection results.

2. The system of claim 1, wherein, The ring-shaped multi-spectral light source module comprises 8 groups of independently controllable LED light sources covering visible light, infrared, and ultraviolet bands, and is configured with a polarization filter to switch the polarization mode to eliminate the reflection of the metal shell, and the light source brightness adjustment range is 50-1000Lux.

3. The system of claim 1, wherein, The clamp mechanism with a pneumatic rotary table comprises a three-dimensional adjustable clamp and a pneumatic rotary table; the clamping range is 50mm×80mm-200mm×300mm; the pneumatic rotary table can rotate 0-360° with a rotation accuracy of ±0.5° to drive the meter to collect multi-angle images.

4. The system of claim 1, wherein, The clamp mechanism with a pneumatic rotary table comprises a three-dimensional adjustable clamp and a pneumatic rotary table; the clamping range is 50mm×80mm-200mm×300mm; the pneumatic rotary table can rotate 0-360° with a rotation accuracy of ±0.5° to drive the meter to collect multi-angle images.

5. The system according to claim 1, wherein: The edge computing terminal adopts NVIDIA Jetson AGX Orin processor and built-in lightweight YOLOv8 model to locate the defect area in real time, the defect positioning response time is ≤0.5s, and the preliminary defect judgment result is output.

6. The system of claim 1, wherein, It also comprises a data preprocessing module for processing the images collected by the high-resolution industrial camera, and the processing steps include image graying, denoising, edge enhancement, rotation, scaling, flipping, and illumination equalization to eliminate image interference and highlight defect features.

7. The system according to claim 1, wherein: The edge computing terminal and the cloud AI analysis platform communicate through 5G or Ethernet, the edge computing terminal uploads the preliminary detection results and original image data to the cloud AI analysis platform, and the cloud AI analysis platform returns a fine detection report including defect type, location, size, and severity level.

8. The system of claim 1, wherein, It also comprises an automatic feeding and sorting mechanism electrically connected with the edge computing terminal to sort the meters into qualified or unqualified channels according to the detection results, and the sorting response time is ≤1s.

9. The system of claim 1, wherein, The layout of the ring-shaped multi-spectral light source module is a ring-shaped array, the distance between adjacent LED light sources is 45°, and the vertical distance between the light source and the surface of the meter can be adjusted in the range of 100-300mm to adapt to the shooting needs of meters of different sizes.

10. The system of claim 1, wherein, The cloud AI analysis platform is deployed with a ResNet-50 multi-task neural network, which is trained in the following way: The data set contains normal samples and defect samples of 30+ models of electric energy meters, and the sample size is greater than 100,000; Data enhancement techniques are used, including multi-illumination condition simulation, defect morphology variation and background interference addition; L1 regularization, L2 regularization and Dropout algorithm are used to prevent overfitting, and the model verification set accuracy is greater than 99.5%.

Images