CN109834050A - A kind of varistor appearance delection device - Google Patents
A kind of varistor appearance delection device Download PDFInfo
- Publication number
- CN109834050A CN109834050A CN201910014598.2A CN201910014598A CN109834050A CN 109834050 A CN109834050 A CN 109834050A CN 201910014598 A CN201910014598 A CN 201910014598A CN 109834050 A CN109834050 A CN 109834050A
- Authority
- CN
- China
- Prior art keywords
- varistor
- detected
- appearance
- conveyer belt
- master controller
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of varistor appearance delection devices, including varistor conveyer belt to be detected, imaging case, master controller, assistant controller and manipulator；Varistor conveyer belt to be detected is for transmitting varistor to be detected；2 industrial cameras inside imaging case are located at the oblique upper and obliquely downward of varistor conveyer belt to be detected, and in the horizontal direction, positioned at the two sides of varistor conveyer belt to be detected, master controller is transferred to for obtaining the photo of varistor to be detected, and by photo to be detected；Master controller is based on trained picture recognition model and determines defective varistor according to the photo of the varistor to be detected of acquisition, and indicates that assistant controller control manipulator grabs defective varistor.The present invention with automatic identification and can sort out that appearance is defective and the flawless varistor of appearance.Improve appearance detection efficiency, reliability, accuracy rate, moreover it is possible to help enterprise that production cost is greatly reduced.
Description
Technical field
The present invention relates to varistor detection field more particularly to a kind of varistor appearance delection devices.
Background technique
At present in varistor production process, varistor appearance substantially has nine class defects: shape is non-round, surface is scarce
Damage, surface is prominent, surface weighs wounded, surface pinholes, edge dew porcelain, disjunctor, torticollis, surface are damaged.Appearance detection includes three kinds of sides
Method:
1. traditional visual inspection method；The low efficiency of visual inspection, poor reliability do not adapt to high-volume
It manufactures.
2. the detection method based on laser measuring technique；Laser measuring technology is higher to hardware requirement, wants to working environment
Ask harsh, maintenance work is many and diverse, and cost is very high.
3. the detection method based on machine vision；With machine vision technique theoretical research and the continuous hair of hardware technology
Exhibition, machine vision technique is also more and more in the application of varistor appearance detection field, and the research of this respect is also more and more.
Foreign countries are more early in the application study starting of varistor appearance detection field to machine vision, and existing Related product surveys device now
It is limited to technical level at that time, practical application effect is simultaneously bad, and major defect is that generalization ability is poor, and accuracy rate is low, often judges by accident good
Product are substandard products.
Summary of the invention
It is an object of the present invention in place of solving the above deficiencies in the existing technologies.
To achieve the above object, the present invention provides a kind of varistor appearance delection device, including varistor to be detected
Conveyer belt, imaging case, master controller, assistant controller and manipulator；Varistor conveyer belt to be detected is for transmitting pressure to be detected
Quick resistance；Imaging case is internally provided with the controllable secondary light source and 2 industrial cameras for being located at imaging box top, wherein 2
Industrial camera is located at the oblique upper and obliquely downward of varistor conveyer belt to be detected, and in the horizontal direction, 2 industry
Camera is located at the two sides of varistor conveyer belt to be detected, for obtaining the photo of varistor to be detected, and by pressure to be detected
The photo of quick resistance is transferred to master controller；Master controller is based on trained picture recognition model, according to the to be checked of acquisition
The photo for surveying varistor, determines defective varistor, and sends fetching instruction to assistant controller, and assistant controller is according to grabbing
Instruction fetch controls manipulator and grabs defective varistor.
Preferably, each industrial camera includes at least an optical lens, polarization camera lens, a zoom lens
With a camera support.
Preferably, industrial camera is ZW-H2000 industrial camera.
Preferably, varistor appearance delection device further includes one group to laser sensor is penetrated, and is located at imaging case
Two sides.
Preferably, master controller includes at least PC machine and GPU card.
Preferably, assistant controller is programmable logic controller (PLC) PLC.
Preferably, picture recognition model is to utilize varistor appearance image data collection, and pressure based on deep neural network
The manual identified of quick resistance appearance image data collection is as a result, what training obtained.
The present invention with automatic identification and can sort out that appearance is defective and the flawless varistor of appearance.Improve appearance inspection
Survey efficiency, reliability, accuracy rate, moreover it is possible to help enterprise that production cost is greatly reduced.
Detailed description of the invention
Fig. 1 is a kind of composition schematic diagram of varistor appearance delection device provided in an embodiment of the present invention；
Fig. 2 is a kind of position view of industrial camera and varistor provided in an embodiment of the present invention；
Fig. 3 is a kind of deep learning network structure provided in an embodiment of the present invention；
Description of symbols:
1- varistor conveyer belt to be detected, 2- imaging case, 3- controllable secondary light source, 4- industrial camera, 5- master controller,
6- assistant controller, 7- manipulator, 8- is to penetrating laser sensor, 9- varistor.
Specific embodiment
With reference to the accompanying drawings and examples, technical scheme of the present invention will be described in further detail.
Such as Fig. 1, the embodiment of the present invention provides a kind of varistor appearance delection device, including varistor to be detected transmission
Band 1, imaging case 2, master controller 5, assistant controller 6 and manipulator 7.
Varistor conveyer belt 1 to be detected is for transmitting varistor 9 to be detected.
Imaging case 2 is internally provided with controllable secondary light source 3 and 2 industrial camera 4 for being located at 2 top of imaging case,
In, 2 industrial cameras 4 are located at the oblique upper and obliquely downward of varistor conveyer belt to be detected, obtain pressure-sensitive electricity to be detected
The photo of resistance, and the photo of varistor to be detected is transferred to master controller.Case 2 is imaged and controllable secondary light source 3 is industry
Camera 4 provides a metastable imaging circumstances, avoids occurring the influence image quality such as shade in photo.
Wherein, in order to find that non-round varistor shape, surface damage, surface is prominent, surface weighs wounded, surface pinholes, side
The defect that edge dew porcelain, disjunctor, torticollis, surface are damaged, an industrial camera 4 are taken pictures downwards from the oblique upper position of varistor 9, separately
One industrial camera 4 is taken pictures upwards from the position obliquely downward of varistor 9.If Fig. 2 is the pressure-sensitive electricity that the direction A from Fig. 1 is seen
The positional diagram of resistance and industrial camera, varistor approximation regard a flat cylindrical body, including upper and lower surface+side as
The appearance in face, two sides will detect, and side is also required to detect, so two industrial camera one is taken pictures from oblique upper, another
It takes pictures from obliquely downward, and is located at the two sides of varistor in the horizontal direction, specific location is as shown in Figure 2.
Master controller 5 is based on picture recognition model and is determined defective according to the photo of the varistor to be detected of acquisition
Then the defective varistor marking signal of appearance is sent to assistant controller 6 by varistor, assistant controller 6 controls machinery
Defective varistor is isolated varistor conveyer belt 1 to be detected by hand 7.
In one example, each industrial camera 4 includes at least an optical lens, a polarization camera lens, a change
Zoom lens and a camera support.
In one example, industrial camera 4 selects ZW-H2000 industrial camera.
In one example, master controller 5 includes at least PC machine and GPU card.
In one example, assistant controller 6 is programmable logic controller (PLC) PLC.
In one example, varistor appearance delection device further includes to penetrating laser sensor 8, to penetrating laser sensor
8 for capture varistor by signal, and assistant controller 6 will be sent to by signal, 6 pairs of disengaging imaging casees of assistant controller
2 varistor 9 counts, and will count and be sent to master controller 5, and master controller 5 is according to the varistor in imaging case
Quantity determines defective varistor in the position of transmission belt.
Picture recognition model based on deep learning, mode input are varistor appearance image data collection, the core of model
The heart is deep neural network, automatically extracts multi-level characteristic of division, and model output is classification, that is, has zero defect.
After deep neural network has been trained to, it is to have existing for some redundancies or unessential neuron, passes through pressure
Contracting accelerated method Optimized model, it is possible to reduce calculation amount improves classification speed under computing resource limited situation.
It wherein, include trimming and sharing, low-rank based on parameter by deep neural network compression acceleration method Optimized model
Decompose the accelerated method with the disaggregated model compression of knowledge distillation.
The present invention selects the method based on convolutional neural networks to carry out appearance detection to varistor.First pass through identification library
Imageai locks the effective coverage of varistor, and then model is realized based on Keras.Input color picture, dimension 268*
298*3, after step-length by three non-filling be 1 two-dimensional convolution neural network.Each convolutional neural networks include a volume
Lamination and a maximum pond layer, activation primitive ' relu '.First convolutional neural networks includes the convolution kernel of 32 3*3, pond
Change layer is 2*2, and output dimension is 133*148*32；Second convolutional neural networks includes the convolution kernel of 32 3*3, and pond layer is
2*2, output dimension are 65*73*32；Third convolutional neural networks include the convolution kernel of 64 3*3, and pond layer is 2*2, output
Dimension is 31*35*64；Afterwards by a smooth layer, exporting dimension is 69440；By a full articulamentum, activation primitive is
' relu ', dimension 64；Using a full articulamentum after Dropout layers, exporting dimension is 2 dimensions, into after crossing softmax
Obtain final result.The structure of deep neural network and the dimension variation of data are as shown in Figure 3.Wherein, Keras is a high level
Neural network API, softmax are that will export the function for being converted into probability.
Wherein, convolutional layer+activation primitive+pond layer is to carry out feature extraction to image, and smooth layer is by convolutional layer and Quan Lian
Layer smooth transition is connect, prevents over-fitting with Dropout layers.
In the training process of model, the loss function of network uses ' binary_crossentropy ', and optimizer makes
With ' rmsprop '.The pseudocode of model training process is described as shown in Algorithm1:
Algorithm1
The frequency of training that batchsize is selected as 40, epoch is 50 times.
Workflow:
Firstly, picture recognition models coupling constructs data set with automatic method by hand, model is trained；Then,
The varistor photo of industrial camera shooting is pre-processed, the varistor in picture is detected, zooms in or out picture extremely
Uniform sizes；Finally, being entered into picture recognition model, classification results are obtained, i.e., defective or zero defect.
Wherein, automatic method building data set includes the varistor appearance based on self-encoding encoder and confrontation neural network
Image data collection automatic generation method.Training deep neural network model needs a large amount of training sets, due to manually acquiring all pressures
Quick resistance appearance picture be it is unpractical, need to automatically generate varistor appearance by self-encoding encoder and confrontation neural network
Image data collection.It is as picture that the varistor appearance image data of self-encoding encoder and confrontation neural network, which integrates automatic generation method,
Identification model provides training set and test set.
In practical work process, identified using the picture recognition model based on deep learning method defective and flawless
Varistor, marking defective varistor is 0, and zero defect varistor is 1.Finally, the transmission of 0,1 flag bit to be given to and can compile
Journey logic controller PLC, PLC control manipulator by number or analog pattern input/output and separate defective varistor
Varistor conveyer belt out.
Above specific embodiment has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Illustrate, it should be understood that the above is only a specific embodiment of the invention, the protection model that is not intended to limit the present invention
It encloses, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should be included in the present invention
Protection scope within.
Claims (7)
1. a kind of varistor appearance delection device, which is characterized in that including varistor conveyer belt to be detected, imaging case, master
Controller, assistant controller and manipulator；
The varistor conveyer belt to be detected is for transmitting varistor to be detected；
The imaging case is internally provided with the controllable secondary light source and 2 industrial cameras for being located at the imaging box top,
In, 2 industrial cameras are located at the oblique upper and obliquely downward of the varistor conveyer belt to be detected, and in the horizontal direction
On, 2 industrial cameras are located at the two sides of varistor conveyer belt to be detected, for obtaining the photograph of the varistor to be detected
Piece, and the photo of the varistor to be detected is transferred to the master controller；
The master controller is based on trained picture recognition model, according to the photograph of the varistor to be detected of acquisition
Piece determines defective varistor, and sends fetching instruction to the assistant controller, and the assistant controller is according to fetching instruction
It controls the manipulator and grabs defective varistor.
2. varistor appearance delection device according to claim 1, which is characterized in that each industrial camera,
Including at least an optical lens, a polarization camera lens, a zoom lens and a camera support.
3. varistor appearance delection device according to claim 1, which is characterized in that the industrial camera is ZW-
H2000 industrial camera.
4. varistor appearance delection device according to claim 1, which is characterized in that the varistor appearance detection
Device further includes one group to laser sensor is penetrated, and is located at the two sides of the imaging case.
5. varistor appearance delection device according to claim 1, which is characterized in that the master controller includes at least
PC machine and GPU card.
6. varistor appearance delection device according to claim 1, which is characterized in that the assistant controller is programmable
Logic controller PLC.
7. varistor appearance delection device according to claim 1, which is characterized in that the picture recognition model is base
In deep neural network, the manual identified of varistor appearance image data collection and varistor appearance image data collection is utilized
As a result, what training obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910014598.2A CN109834050A (en) | 2019-01-08 | 2019-01-08 | A kind of varistor appearance delection device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910014598.2A CN109834050A (en) | 2019-01-08 | 2019-01-08 | A kind of varistor appearance delection device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109834050A true CN109834050A (en) | 2019-06-04 |
Family
ID=66883721
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910014598.2A Pending CN109834050A (en) | 2019-01-08 | 2019-01-08 | A kind of varistor appearance delection device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109834050A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113504263A (en) * | 2021-06-24 | 2021-10-15 | 海南电网有限责任公司电力科学研究院 | Insulator abnormal heating defect detection device and application method thereof |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101341815B1 (en) * | 2012-12-04 | 2014-01-06 | 대한민국 | Screening devices of seeds using hyperspectral image processing |
| CN204448619U (en) * | 2014-12-23 | 2015-07-08 | 中国农业大学 | A kind of seed Real-time online sorting device |
| CN206489090U (en) * | 2016-12-16 | 2017-09-12 | 惠州市德赛电池有限公司 | A kind of battery appearance detecting device |
| CN108144872A (en) * | 2017-12-25 | 2018-06-12 | 东莞辰达电器有限公司 | A kind of PCBA automatic shape inspection systems using CCD vision techniques |
| CN108897925A (en) * | 2018-06-11 | 2018-11-27 | 华中科技大学 | A kind of casting technological parameter optimization method based on casting defect prediction model |
| CN208194986U (en) * | 2017-12-18 | 2018-12-07 | 深圳市巨领智能装备有限公司 | A kind of patch capacitor appearance intelligence sorting machine |
-
2019
- 2019-01-08 CN CN201910014598.2A patent/CN109834050A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101341815B1 (en) * | 2012-12-04 | 2014-01-06 | 대한민국 | Screening devices of seeds using hyperspectral image processing |
| CN204448619U (en) * | 2014-12-23 | 2015-07-08 | 中国农业大学 | A kind of seed Real-time online sorting device |
| CN206489090U (en) * | 2016-12-16 | 2017-09-12 | 惠州市德赛电池有限公司 | A kind of battery appearance detecting device |
| CN208194986U (en) * | 2017-12-18 | 2018-12-07 | 深圳市巨领智能装备有限公司 | A kind of patch capacitor appearance intelligence sorting machine |
| CN108144872A (en) * | 2017-12-25 | 2018-06-12 | 东莞辰达电器有限公司 | A kind of PCBA automatic shape inspection systems using CCD vision techniques |
| CN108897925A (en) * | 2018-06-11 | 2018-11-27 | 华中科技大学 | A kind of casting technological parameter optimization method based on casting defect prediction model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113504263A (en) * | 2021-06-24 | 2021-10-15 | 海南电网有限责任公司电力科学研究院 | Insulator abnormal heating defect detection device and application method thereof |
| CN113504263B (en) * | 2021-06-24 | 2022-12-27 | 海南电网有限责任公司电力科学研究院 | Insulator abnormal heating defect detection device and application method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110543878B (en) | Pointer instrument reading identification method based on neural network | |
| CN111507976B (en) | Defect detection method and system based on multi-angle imaging | |
| CN107944504B (en) | Board recognition and machine learning method and device for board recognition and electronic equipment | |
| CN104063873B (en) | A kind of Model For The Bush-axle Type Parts surface defect online test method based on compressed sensing | |
| CN110473173A (en) | A kind of defect inspection method based on deep learning semantic segmentation | |
| CN109724984A (en) | A kind of defects detection identification device and method based on deep learning algorithm | |
| CN109840900A (en) | A kind of line detection system for failure and detection method applied to intelligence manufacture workshop | |
| CN107052086A (en) | Stamping parts surface defect detection apparatus and detection method based on 3D vision | |
| CN108090896B (en) | Wood board flatness detection and machine learning method and device and electronic equipment | |
| CN108108679B (en) | Full-automatic tungsten ore concentrating machine | |
| CN104914111A (en) | Strip steel surface defect on-line intelligent identification and detection system and detection method | |
| CN102654464A (en) | Copper strip surface defect detection system based on multi-feature fuzzy recognition | |
| CN114529510B (en) | Automatic detection and classification method for cathode copper on-line quality | |
| CN112233067A (en) | Hot rolled steel coil end face quality detection method and system | |
| CN112893159B (en) | Coal gangue sorting method based on image recognition | |
| CN103644852A (en) | Lens film thickness automatic detection system | |
| CN206132657U (en) | Gilt quality intelligent detecting system based on machine vision | |
| CN106248680A (en) | A kind of engine commutator quality detecting system based on machine vision and detection method | |
| CN110554052A (en) | artificial board surface defect detection method and system | |
| CN111558542A (en) | Ceramic tile surface quality online detection sorting system and method | |
| CN102663340A (en) | Device and method for classifying plates | |
| CN114549391A (en) | Circuit board surface defect detection method based on polarization prior | |
| CN110044921A (en) | Lithium battery open defect detection system and method | |
| CN109834050A (en) | A kind of varistor appearance delection device | |
| CN103984967B (en) | A kind of automatic checkout system being applied to Commercial goods labels detection and automatic testing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190604 |
|
| RJ01 | Rejection of invention patent application after publication |