Defects detection training machine and its application method based on deep learning
Technical field
The invention belongs to industry spot defect detecting technique field, in particular to a kind of defects detection based on deep learning
Training machine and its application method.
Background technique
In industrial scene, there is very strict requirement to the factory index of product parts, these requirements are product zero
The effect of functioning property provides guarantee after part factory.Under industrial scene, various machinery, sound, light, the complex environment of electricity and crowd
Multi-process may be damaged to the appearance of product parts, become the product parts with defect.It is largely directed at present
The defects detection of product parts is all highly dependent on manually to be detected one by one, occupies a large amount of human resources, is consumed a large amount of
Human cost.Least a portion of defects detection is to be solved by traditional image processing algorithm, but have the defects that such as to detect kind
The deficiencies of class is very single, and the early period of Processing Algorithm, verifying was extremely complex, and then influence the efficiency of detection.
Summary of the invention
For the technical problems in the prior art, the present invention provides a kind of defects detection training based on deep learning
Machine and its application method.
In order to solve the above technical problems, the technical solution adopted by the present invention is that:
A kind of defects detection training machine based on deep learning, including industrial personal computer, bottom light source, turntable, distance are adjusted
Bogie, side light source, programmable light source and camera lens assembly are equipped with deep learning software and equipment pipe in the industrial personal computer
Reason system, the bottom light source are located at the lower section of turntable, and the distance adjusts rotating turret and is located above the turntable, the programmable
Light source is connect by light source shelf with apart from adjusting bogie, and the camera lens assembly is connect with apart from adjusting bogie, and institute
It states the lower end camera lens of camera lens assembly downward and is directed at the upper end outlet of programmable light source.
Preferably, it is linear mould group that the distance, which adjusts bogie,.
Preferably, the bottom light source and side light source are luminescent panel.
A kind of application method of the defects detection training machine based on deep learning, includes the following steps:
1) light emitting region and the brightness of programmable light source are adjusted, by the equipment management system of industrial personal computer to obtain picture number
According to;
2) image data that will acquire carries out algorithm training by the deep learning software of industrial personal computer, and training is generated
Algorithm model detects workpiece for measurement;
3) classification and statistics of defects count, defect classification are carried out to workpiece for measurement, and qualified product is grabbed by manipulator
With non-qualified product.
5. the application method of the defects detection training machine according to claim 4 based on deep learning, feature
It is, the deep learning software includes:
Labeling module: for realizing the labeling of the image data to camera and camera lens acquisition, and defect image is carried out
Type mark, is the data source of post deep learning training;
Training module: learning the data marked, to obtain optimal disaggregated model;
Cloud storage module: the image at scene is returned in Cloud Server, the data-optimized of later period is carried out;
Test module: trained model is loaded into, and the image acquired by camera and lens assembly is lacked in real time
Fall into detection.
Preferably, the equipment management system comprises the following modules:
Light source camera control module: the periodic light on and off of light source, stroboscopic are realized by software or hardware trigger interface.
Machine control modules: for controlling angle direction, the distance of light source to realize automatic movement;
Scheme memory module: it is used for the different types of polishing mode of record storage;
External communication module: there is the interface externally communicated with PLC and manipulator and the integral machine of entire combination;
The automatic categorization module of product: for variety classes product classification;
Product quality warning module: it is used for real-time statistics defects count;
Real-time online rejects module: for rejecting faulty goods/mixing by the device for eliminating outside docking
Compared with prior art, the present invention has the beneficial effects that the present invention utilizes adjustable programmable light source
And equipment management system, there can be a variety of environment of simulation, and fast implement trained detection, improve the efficiency and accuracy of detection.
Detailed description of the invention
Fig. 1 is the schematic view of the front view of the defects detection training machine based on deep learning in the present invention;
Fig. 2 is the side structure schematic view of the defects detection training machine based on deep learning in the present invention;
Fig. 3 is the sectional structure signal of the programmable light source of the defects detection training machine based on deep learning in the present invention
Figure;
Fig. 4 is that the programmable light source of defects detection training machine based on deep learning in the present invention looks up structural representation
Figure;
Fig. 5 is the test specimen to be measured of the defects detection training machine based on deep learning and the relationship of bottom light source in the present invention
Schematic diagram;
Fig. 6 is the flow diagram of the defect application method based on deep learning in the present invention.
In figure, 1. bottom light sources;2. turntable;3. distance adjusts bogie;4. side light source;5. programmable light source;51. bowl-shape
Body;52. light source up;Qu Guangyuan in 53.;54. lower area's light source;6. camera lens assembly;7. workpiece for measurement;8. light source shelf;9.
Screw.
Specific embodiment
Technical solution in order to enable those skilled in the art to better understand the present invention, in the following with reference to the drawings and specific embodiments
It elaborates to the present invention.
As shown in Figures 1 to 6, embodiment of the invention discloses a kind of defects detection training airplane based on deep learning
Device, including industrial personal computer (not shown), bottom light source 1, turntable 2, distance adjust bogie 3, side light source 4, programmable light source 5
With camera lens assembly 6, bottom light source 1 and side light source 4 are luminescent panel.Deep learning software is installed in industrial personal computer and is set
Standby management system bottom light source 1 is located at the lower section of turntable 2, and distance adjusts the top that bogie 3 is located at turntable 2, and programmable light
Source 5 is connect by light source shelf 8 with apart from adjusting bogie 3, and camera lens assembly 6 is connect with apart from adjusting bogie 3, and camera
The lower end camera lens of lens assembly 6 downward and is directed at the upper end outlet of programmable light source 5.
In the present embodiment, it is linear mould group that distance, which adjusts bogie 3,.
In the present embodiment, the rotating manner of turntable 2 is the prior art, such as can be motor and drive turntable by gear pair
Rotation etc..
Programmable light source is connect by light source shelf 8 and screw 9 with apart from adjusting bogie 3, can journey by process control
The circumferential direction of formula light source is divided into eight sections, and is axially divided into area's light source 51, middle area's light source 52 and lower area light source 53, and logical
The brightness in each section and upper area's light source 51, middle area's light source 52 and lower area light source 53 can individually be controlled by crossing program.
The application method of the invention also discloses a kind of defects detection training machine based on deep learning, including walk as follows
It is rapid:
1) light emitting region and the brightness of programmable light source 5 are adjusted by the equipment management system of industrial personal computer;
2) algorithm training is carried out by the deep learning software of industrial personal computer, and the algorithm model that training is generated is to work to be measured
Part is detected;
3) image data of acquisition is uploaded in the Cloud Server for the deep learning software being pre-installed, and passes through equipment pipe
Reason system to workpiece for measurement 10 carry out defects count, defect classification classification and statistics, and by manipulator grab qualified product with
Non- qualified product.
Acquire the photo of workpiece for measurement (workpiece for measurement is placed on turntable 2) by camera and lens assembly 6, and by acquisition
Photographic intelligence uploads to the Cloud Server in deep learning software, for improving detection model.
Loading and unloading mechanical device is accessed by equipment management system, part is grabbed by the mechanical arm of loading and unloading mechanical device
To detection platform, by detection platform output as a result, and grabbing rejected product to not by the mechanical arm of loading and unloading mechanical device
Qualified product area, qualified product are grabbed to qualified product area.
In the present invention, deep learning software includes:
Labeling module: for realizing the labeling of the image data to camera and camera lens acquisition, and defect image is carried out
Type mark is the data source of deep learning later period training;
Training module: learning the data marked, to obtain optimal disaggregated model;
Cloud storage module: the image at scene is returned in Cloud Server, the data-optimized of later period is carried out;
Test module: trained model is loaded into, and the image acquired by camera and lens assembly is lacked in real time
Fall into detection.
In the present embodiment, equipment management system can connect host computer, slave computer, be collected simultaneously arrangement production, detection data,
Equipment management system comprises the following modules:
Light source camera control module: the periodic light on and off of light source, stroboscopic are realized by software or hardware trigger interface.
Machine control modules: automatic movement is realized for angle direction, the distance controlling of light source;
Scheme memory module: the different types of polishing mode of record storage;
External communication module: there is the interface externally communicated with PLC and manipulator and the integral machine of entire combination;
Product defects categorization module: for realizing the differentiation for defect classification;
The automatic categorization module of product: for realizing the classification for variety classes product;
Product quality warning module: real-time statistics are carried out for the defects count to product;
Real-time online rejects module: by the device for eliminating (such as manipulator) outside docking and by faulty goods/mixing
It rejects in time.
The detection machine of hundreds of polishing scheme of analog of the present invention, configured with manual zoom (12-36mm) camera lens+program-controlled
Light source+multifunctional workstation, while can recorde hundreds of camera position and corresponding polishing scheme, it can call cut at any time
It changes.
Above embodiments are only exemplary embodiment of the present invention, are not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.