CN110428408A

CN110428408A - Flaw detection method based on ELM-in-ELM

Info

Publication number: CN110428408A
Application number: CN201910698736.3A
Authority: CN
Inventors: 马宏宾; 宋利
Original assignee: Gongji Intelligent Technology (suzhou) Co Ltd
Current assignee: Gongji Intelligent Technology (suzhou) Co Ltd
Priority date: 2019-07-31
Filing date: 2019-07-31
Publication date: 2019-11-08
Anticipated expiration: 2039-07-31
Also published as: CN110428408B

Abstract

The present invention is based on the flaw detection methods of ELM-in-ELM, and steps are as follows: 1) imaging device carries out workpiece image molding, and the rear extraction and matching that characteristic point is carried out by ORB algorithm further extracts the area-of-interest in present image, the i.e. image of workpiece；2) Automatic Feature Extraction is carried out to the workpiece image extracted using ELM-in-ELM and differentiates whether workpiece has flaw.

Description

Flaw detection method based on ELM-in-ELM

Technical field

The present invention relates to intelligent detecting methods, specifically, it shows a kind of flaw detection method based on ELM-in-ELM.

Background technique

Defect Detection is exactly to judge according to the appearance of product or using the product internal structure that special sensor obtains Whether the product meets certain production standard, that is, judges whether the product is substandard products.Currently, Defect Detection has had many realities The application on border, such as the detection of surface of workpiece Defect Detection, vehicle leather and the detection of capacitor bubble etc..

Defect Detection at this stage is mainly to pass through the Defect Detection of magnetic field, laser, sound wave and infrared thermal imaging, detection Object be mostly in object fault of construction or be not easy the flaw of the body surface directly contacted or by measurement object week The magnetic field or electromagnetic wave enclosed come whether judgment object has flaw.

Specific cocoa are as follows:

It realizes by the variation of alternating current magnetic field around measurement object and its with the difference of the Distribution of Magnetic Field of normal structure for water The Defect Detection of flowering structure solves the problems, such as directly to carry out Defect Detection for submerged structure relatively difficult.

It is real using the means of finite element analysis using the electromagnetic acoustic creeping wave characteristic sensitive to surface and near surface flaw The Defect Detection to cylindrical body cavity is showed.

Defect areas information is reinforced using laser irradiation target object and the method for collecting reflected light, is convenient for flaw Positioning and measurement.

The inner ring and roller defect of rolling bearing are detected using Measurement Technology of Acoustic Emission, but it is lesser to be only limitted to size Flaw, this method will fail after the size of flaw is more than certain range.

It a kind of is solved the above problems based on the flaw detection method of ELM-in-ELM therefore, it is necessary to provide.

Summary of the invention

The object of the present invention is to provide a kind of flaw detection methods based on ELM-in-ELM, are with ELM-in-ELM network Core carries out the solution of on-line study, is trained using small sample, in actual use for identification mistake Workpiece image carries out re -training, so that network performance is continuously improved, and then reaches higher detection accuracy；It is calculated simultaneously with ORB Method is attitude updating and the region of interesting extraction that core realizes workpiece, is conducive to improve Defect Detection precision.

Technical solution is as follows:

A kind of flaw detection method based on ELM-in-ELM, steps are as follows:

1) imaging device carries out workpiece image molding, and the rear extraction and matching that characteristic point is carried out by ORB algorithm further mentions Take out the area-of-interest in present image, the i.e. image of workpiece；

2) Automatic Feature Extraction is carried out to the workpiece image extracted using ELM-in-ELM and differentiates whether workpiece has flaw.

Further, imaging device can be high definition CDD camera.

Further, the depth convolutional neural networks that ELM-in-ELM includes are judged by importing a variety of different flaws Data carry out deep learning, to adapt to different detection occasion detections.

Further, flaw determines that data include penetrating through the flaw inspection of magnetic field, laser, sound wave and infrared thermal imaging Measured data.

Further, flaw determines that data include the Defect Detection number that different objects are carried out using infrared thermal imaging as tool The Defect Detection data etc. of different objects are carried out according to, pulse thermal imaging.

Further, ORB algorithm has many advantages, such as that calculation amount is small, amount of storage is small, be highly suitable in real-time system into Row Feature Points Matching；According to the coordinate of characteristic point after matching, homography matrix is calculated using RANSAC algorithm, is effectively reduced and makes an uproar Sound and Feature Points Matching mistake bring influence.

Further, ELM-in-ELM remain its input weight at random give feature on the basis of, it is multiple using internal layer Multiple ELM networks are combined by the method that ELM network is respectively trained and outer layer ELM network is individually trained, and improve classification Stability and veracity.

Further, ELM-in-ELM is on-line study network, has preferable real-time, can be used for real-time flaw inspection It surveys.

Compared with prior art, the present invention carries out the solution of on-line study using ELM-in-ELM network as core, adopts It is trained with small sample, re -training is carried out for the workpiece image of identification mistake in actual use, so that network Performance is continuously improved, and then reaches higher detection accuracy；Simultaneously using ORB algorithm as core realize the attitude updating of workpiece with And region of interesting extraction, be conducive to improve Defect Detection precision.

Specific embodiment

Embodiment:

The present embodiment shows a kind of flaw detection method based on ELM-in-ELM, and steps are as follows:

Imaging device can be high definition CDD camera.

The depth convolutional neural networks that ELM-in-ELM includes judge that data carry out deeply by importing a variety of different flaws Degree study, to adapt to different detection occasion detections.

Flaw determines that data include penetrating through the Defect Detection data of magnetic field, laser, sound wave and infrared thermal imaging, packet It includes:

It realizes by the variation of alternating current magnetic field around measurement object and its with the difference of the Distribution of Magnetic Field of normal structure for water The Defect Detection of flowering structure.

The inner ring and roller defect of rolling bearing are detected using Measurement Technology of Acoustic Emission, but it is lesser to be only limitted to size Flaw.

Flaw determine data include using infrared thermal imaging as tool carry out the Defect Detection data of different objects, pulse heat at Defect Detection data etc. as carrying out different objects.

Include: using the Defect Detection data that infrared thermal imaging carries out different objects as tool

To the Defect Detection with low thermal diffusivity material, this particular problem proposes a kind of simple active infrared thermal imaging Method, by count detectivity obtain this material conclusion whether defective.

The element after Electronic Packaging is heated using steady-state DC electroheat technology, and passes through the statistics in infrared image Information obtains the temperature defective effect factor information of element, so that whether the element after further being encapsulated has flaw.

The improvement to infrared thermoviewer is crossed so that the distance between sensor and testee are constant, and in the flaw of pipeline The validity of its method is demonstrated in Detection task.

Infrared temperature sensor is improved, the temperature point on cylinder metal surface can be more accurately measured Cloth to measure the Temperature Distribution variation of object after a heating, and from which further follows that whether object has flaw.

Pulse thermal imaging carry out different objects Defect Detection data include:

The pulse graphic images of carbon fibre reinforced plastic are carried out by the method that image segmentation and Laplce map dimensionality reduction Processing, to judge whether the material has flaw.

ORB algorithm has many advantages, such as that calculation amount is small, amount of storage is small, is highly suitable for carrying out characteristic point in real-time system Match；According to the coordinate of characteristic point after matching, homography matrix is calculated using RANSAC algorithm, noise and characteristic point is effectively reduced Matching error bring influences.

ELM-in-ELM is remained on the basis of its input weight gives feature at random, using multiple ELM networks of internal layer point Not Xun Lian and the method individually trained of outer layer ELM network multiple ELM networks are combined, improve classification accuracy and Stability.

ELM-in-ELM is on-line study network, has preferable real-time, can be used for real-time Defect Detection.

Above-described is only some embodiments of the present invention.For those of ordinary skill in the art, not Under the premise of being detached from the invention design, various modifications and improvements can be made, these belong to protection model of the invention It encloses.

Claims

1. a kind of flaw detection method based on ELM-in-ELM, it is characterised in that: steps are as follows:

2. a kind of flaw detection method based on ELM-in-ELM according to claim 1, it is characterised in that: imaging device It can be high definition CDD camera.

3. a kind of flaw detection method based on ELM-in-ELM according to claim 2, it is characterised in that: ELM-in- The depth convolutional neural networks that ELM includes judge that data carry out deep learning by importing a variety of different flaws, to adapt to not With detection occasion detection.

4. a kind of flaw detection method based on ELM-in-ELM according to claim 3, it is characterised in that: flaw determines Data include penetrating through the Defect Detection data of magnetic field, laser, sound wave and infrared thermal imaging.

5. -4 any a kind of flaw detection method based on ELM-in-ELM according to claim 1, it is characterised in that: the flaw Defect determines that data include that Defect Detection data, the pulse thermal imaging progress difference of different objects are carried out using infrared thermal imaging as tool Defect Detection data of object etc..

6. a kind of flaw detection method based on ELM-in-ELM according to claim 5, it is characterised in that: ORB algorithm Have many advantages, such as that calculation amount is small, amount of storage is small, is highly suitable for carrying out Feature Points Matching in real-time system；According to special after matching The coordinate for levying point calculates homography matrix using RANSAC algorithm, noise and Feature Points Matching mistake bring shadow is effectively reduced It rings.

7. a kind of flaw detection method based on ELM-in-ELM according to claim 6, it is characterised in that: ELM-in- ELM is remained on the basis of its input weight gives feature at random, is respectively trained using the multiple ELM networks of internal layer and outer layer Multiple ELM networks are combined by the method that ELM network is individually trained, and improve the Stability and veracity of classification.

8. a kind of flaw detection method based on ELM-in-ELM according to claim 6 or 7, it is characterised in that: ELM- In-ELM is on-line study network, has preferable real-time, can be used for real-time Defect Detection.