CN109840900A - A kind of line detection system for failure and detection method applied to intelligence manufacture workshop - Google Patents
A kind of line detection system for failure and detection method applied to intelligence manufacture workshop Download PDFInfo
- Publication number
- CN109840900A CN109840900A CN201811651018.2A CN201811651018A CN109840900A CN 109840900 A CN109840900 A CN 109840900A CN 201811651018 A CN201811651018 A CN 201811651018A CN 109840900 A CN109840900 A CN 109840900A
- Authority
- CN
- China
- Prior art keywords
- image
- unit
- manufacturing defect
- neural network
- computer control
- 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.)
- Granted
Links
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 119
- 238000001514 detection method Methods 0.000 title claims abstract description 68
- 230000007547 defect Effects 0.000 claims abstract description 63
- 238000012545 processing Methods 0.000 claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 42
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000012423 maintenance Methods 0.000 claims abstract description 24
- 230000001235 sensitizing effect Effects 0.000 claims description 49
- 239000013598 vector Substances 0.000 claims description 42
- 238000005259 measurement Methods 0.000 claims description 28
- 238000006243 chemical reaction Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000005452 bending Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 11
- 238000003708 edge detection Methods 0.000 claims description 10
- 239000013307 optical fiber Substances 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000003745 diagnosis Methods 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 5
- 210000004218 nerve net Anatomy 0.000 claims description 4
- 238000005286 illumination Methods 0.000 claims description 3
- 230000002950 deficient Effects 0.000 claims description 2
- 239000000284 extract Substances 0.000 claims description 2
- 238000012797 qualification Methods 0.000 claims description 2
- 230000035807 sensation Effects 0.000 claims description 2
- 238000004891 communication Methods 0.000 claims 1
- 239000000835 fiber Substances 0.000 claims 1
- 230000035945 sensitivity Effects 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 4
- 238000010276 construction Methods 0.000 abstract description 2
- 238000003062 neural network model Methods 0.000 abstract description 2
- 238000012394 real-time manufacturing Methods 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 description 5
- 238000007689 inspection Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000005622 photoelectricity Effects 0.000 description 1
- 238000010223 real-time analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Abstract
The present invention relates to a kind of line detection system for failure and method applied to intelligence manufacture workshop, it is based on deep neural network and constructs manufacturing defect prediction model, efficiently use the typical manufacturing defect obtained in the manufacturing process of a large amount of practical intelligence manufacture workshops, in conjunction with Image Acquisition and image processing techniques, study is trained to the master sample image library of typical manufacturing defect image construction, so that the deep neural network model can be used for real-time manufacturing defect identification and classification, it can be in the manufacturing process of product, pass through real time image collection, processing dynamically obtains the manufacture information of product with analysis, and combine PLC controller and history Maintenance plan, provide and execute automatically the maintenance policy of manufacturing defect, it can be the product manufacturing precision in intelligence manufacture workshop, it manufactures information and collects the reference more accurate with analysis offer Information.
Description
Technical field
The invention belongs to intelligent Manufacturing Technology fields, and in particular, to a kind of failure applied to intelligence manufacture workshop exists
Line detection system and its method.
Background technique
Current intelligence manufacture workshop, since the high speed of use, high-precision numerically-controlled machine tool are more, the product of processing and manufacturing
Allowable error is small and intelligence manufacture workshop is to the requirement all with higher of personnel's overall qualities, environment, so that manufacture manufacture vehicle
Between manufacture course of products in manufacturing defect be difficult to sometimes by timely artificial discovery, typically only by the artificial of finishing operations
Inspection just found, thus cause the waste product of intelligence manufacture to increase, productivity decline the problems such as, when serious, or even will cause
What postorder produced lathes scraps, this has been occurred in intelligence manufacture assembly interval, therefore, to the product of manufacture manufacturing shop
Dynamic manufacturing defect inspection is carried out in the fabrication process after diagnosing, is the hindering factor for influencing intelligence manufacture and further developing
One of, nowadays also obtain the extensive research of scholar in the industry.
Summary of the invention
The object of the present invention is to provide a kind of line detection system for failure and detection method applied to intelligence manufacture workshop.
Above-mentioned purpose according to the present invention proposes that a kind of line detection system for failure, including workpiece sensing platform, image are adopted
Collect unit, image processing unit, characteristic vector pickup unit, deep neural network unit and computer control unit, wherein
The workpiece sensing platform includes detection station, the area array cameras acquisition testing station in described image acquisition unit
The image of upper workpiece sensing platform, and it is sent to image processing unit；
Described image processing unit carries out resolution scan to the image received, obtains the sensitizing range of current detection station
Area image, and sensitizing range image is denoised, then the sensitizing range image after denoising is sent to characteristic vector pickup list
Member；
Characteristic vector pickup unit carries out edge detection to sensitizing range image, forms target area, and pass through public affairs respectively
Formula (1) to (3) calculates edge area, the edge shape factor and the target area mean radius for obtaining target area, adds
Bending moment, composition do not have the feature vector of the sensitizing range there are four characteristic variable to the Hu of preceding 3 dimension, to reflect that current workpiece sensing is flat
The workpiece quality information of platform, feature vector are sent to deep neural network unit as input layer；
In above formula, parameter M and N are the marginal point number of target area,Wherein t (x,
It y) is the gray value of each marginal point；Parameter L is the perimeter of target area, and the Chain-Code-Method that can be used in image processing techniques is counted
It calculates and obtains, be the marginal point number on target area boundaries, (x with reference to Kk,yk) indicate that the pixel being located on target area boundaries is sat
Mark,The center-of-mass coordinate for indicating target area, can be calculated by the following formula:
Wherein, parameter A indicates the area of sensitizing range, and is suitable for obtaining it when recognizing sensitizing range in image procossing
Size；
Deep neural network unit is based on neural network algorithm and constructs manufacturing defect prediction model, to workpiece sensing platform
Image feature vector is trained, learns and classifies, and identifies the manufacturing defect class of the workpiece for measurement on current workpiece sensing platform
Type, and classification results are fed back into computer control unit.
Another aspect, the present invention also provides a kind of failure on-line detecting methods applied to intelligence manufacture workshop, comprising:
Step 1: manufacturing defect prediction model of the building based on deep neural network, by sample image to depth nerve net
Network is trained study；
Step 2: workpiece sensing platform drives workpiece for measurement in detection station under the control instruction of computer control module
Upper movement makes it reach preset detection position, triggers close to switch, and send trigger signal to computer control unit；
Step 3: computer control unit sends instruction, the LED to LED area light source matrix and image acquisition units respectively
Area source matrix opens illumination, and described image acquisition unit shoots workpiece for measurement, and the image of generation is sent to figure
As processing unit；
Step 4: image processing unit identifies and is partitioned into the sensitizing range of current detection station, for the regional area into
The processing of row image denoising, by the sensitizing range, image is sent to characteristic vector pickup unit after denoising；
Step 5: characteristic vector pickup unit carries out edge detection to sensitizing range, forms target area, and calculate acquisition
Edge area, the edge shape factor and the target area mean radius of target area are constituted in conjunction with the Hu not bending moment of preceding 3 dimension
There are four the feature vectors of the sensitizing range of characteristic variable for tool；
Step 6: carrying out the diagnosis of manufacture information to feature vector based on trained deep neural network, predict and classify
The manufacturing defect of workpiece for measurement, and classification results are fed back into computer control unit.
The beneficial effects of the present invention are:
(1) Image Acquisition and image processing techniques are applied to the product manufacturing defect in intelligence manufacture workshop by the present invention for the first time
In information identification and diagnosis, by the manufacturing defect prediction model constructed based on deep neural network, to intelligence manufacture workshop
Product carries out real time image collection, processing and feature extraction, and the feature vector for obtaining Product Manufacturing Information indicates, to realize
Product Manufacturing Information during intelligence manufacture can be shown, defect is diagnosable, maintenance policy can automatically generate, and form product system
The real-time monitoring made, the real-time analysis of defect conclusion, maintenance policy automatic implementation organic integral system, thus effectively
Solves one of widely applied factor of existing restriction intelligent Manufacturing Technology；
(2) present invention is in order to obtain accurately processing result image, at several optical fiber slip rings of application and image
Application resolution scanning technique during reason efficiently solves the lathe and detection station being likely to occur in intelligence manufacture workshop
The deflection lens problem of microvibration bring Image Acquisition precision and area array cameras improves product manufacturing of the invention and exists
The precision of line accident defect diagnosis and identification；
(3) time-consuming of the invention in order to reduce image procossing, is likely to occur on current detection station based on workpiece for measurement
Typical manufacturing defect, pointedly be arranged background target mark point position so that can directly determine in postorder image procossing
Position goes out the sensitizing range image of current detection station, for this regional area carry out the extraction of image procossing and characteristic variable with
It calculates, image processing efficiency is greatly improved, so that On-line Product failure prediction system of the invention and method are compared to biography
System carries out processing and the extraction of characteristic variable to whole image, it appears more in real time and efficiently, greatly improves the present invention
Application value in practical intelligence manufacture workshop.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is the schematic diagram of line detection system for failure of the invention.
Specific embodiment
In conjunction with the accompanying drawings, the present invention is further explained in detail.These attached drawings are simplified schematic diagram, only with
Illustration illustrates basic structure of the invention, therefore it only shows the composition relevant to the invention.
The present invention provides a kind of line detection system for failure and its method applied to intelligence manufacture workshop, based on deep
It spends neural network and constructs manufacturing defect prediction model, efficiently use the allusion quotation obtained in the manufacturing process of a large amount of practical intelligence manufacture workshops
Type manufacturing defect, in conjunction with Image Acquisition and image processing techniques, to the master sample image library of typical manufacturing defect image construction
It is trained study, so that the deep neural network model can be used for real-time manufacturing defect identification and classification, can produced
In the manufacturing process of product, the manufacture information of product is dynamically obtained by real time image collection, processing and analysis, and combines PLC
Controller and history Maintenance plan provide and execute automatically the maintenance policy of manufacturing defect, can be intelligence manufacture workshop
Product manufacturing precision, manufacture information collects and provides more accurate reference information with analysis.
Realization process of the invention is discussed in detail by following examples.
As shown in Figure 1, reference attached drawing 1 of the invention, the present invention provides a kind of failures applied to intelligence manufacture workshop
On-line detecting system, including product testing platform 1, image acquisition units 2, image processing unit 3, characteristic vector pickup unit 4,
Deep neural network unit 5 and computer control unit 6.
In a preferred embodiment, the quantity of product testing platform 1 can be several, and each product testing platform 1 is right
Answer an image acquisition units 2, image processing unit 3, characteristic vector pickup unit 4 and deep neural network unit 5 can be with
As software system integration in computer control unit 6.
Wherein, product testing platform 1 includes detection station 11, close switch 12 and intelligent positioning and releasing mechanism 13,
In a preferred embodiment, it is communicated to connect close to switch 12, intelligent positioning and releasing mechanism 13 with computer control unit 6,
Wherein intelligent positioning and releasing mechanism 13 receive the control instruction of computer control unit 6, drive workpiece for measurement in detection station
Movement makes its reach preset detection position on 11, triggers close to switch 12 and sends trigger signal to computer control unit 6,
Image acquisition units 2 are started by computer control unit 6 and carry out Image Acquisition.
On the other hand, the background target 14 continuously arranged along flexible production line is provided in the side of detection station 11,
The mark point of multiple corresponding current detection stations 11 is provided on background target 14, in a preferred embodiment, based on current inspection
The typical manufacturing defect information that the workpiece for measurement of station is likely to occur is surveyed, the mark point position on background target 14 is set, so as to
In identification and segmentation of the subsequent image processing to sensitizing range.In addition, the other side of detection station 11 is provided with LED area light source square
Battle array 15, the LED area light source matrix 15 and computer control unit 6 communicate to connect, and intelligent control may be implemented in computer control unit 6
Lighting and extinguishing for LED area light source matrix 15 in the entire production line processed, moves to current detection station 11 in workpiece for measurement
When predeterminated position, computer control unit 6 sends instructions to the LED area light source matrix 15 of the corresponding detection station 11, makes its point
Bright, when workpiece for measurement is transported to next detection station, the LED area light source matrix 15 of control current detection station 11 is put out
Go out, background target 14 and LED area light source matrix 15 collectively form back lighting environment, be conducive to workpiece for measurement on the image with back
The intensity contrast of scape target 14, edge detection when reducing image procossing are time-consuming.
Image acquisition units 2 include area array cameras 21, the first photo-electric conversion element 22, optical fiber slip ring successively communicated to connect
23 and second photo-electric conversion element 24, wherein when workpiece for measurement moves to preset detection position simultaneously in detection station 11
Close to area array cameras 21 shooting central region when, trigger close to switch 12 simultaneously send trigger signal to computer control unit 6；
The sensation lens axis of area array cameras 21 is vertically arranged with detection station circulation direction, it is contemplated that in intelligence manufacture workshop, equipment
Operating may be detected station and generate certain vibration, to influence the area array cameras 21 being fixedly connected with detection station
Precision is shot, in order to obtain clear reliable shooting effect, the image that area array cameras 21 of the invention is shot is being sent to image
Before processing unit 3, carried out respectively by the first photo-electric conversion element 22, optical fiber slip ring 23 and the second photo-electric conversion element 24
Photoelectric conversion, so that the received electric image signal of image processing unit 3 tends to be complete and reliable.
Specifically, the first photo-electric conversion element 22 and the second photo-electric conversion element 24 are provided with the input terminal of identical quantity
Be corresponding to it with the quantity of multiple output ends, optical fiber slip ring, each input terminal of the first photo-electric conversion element 22 respectively with face
Array camera connection, each output end passes through an optical fiber slip ring 23 respectively and the input terminal of the second photo-electric conversion element 24 connects
It connects, each output end of the second photo-electric conversion element 24 is connect with image processing unit 3.
Image processing unit 3 receives after the image of photoelectric conversion, and in order to improve image processing efficiency, the present invention is first
It first passes through resolution scan to obtain the sensitizing range image of current detection station and carry out image segmentation, then only to acquisition
Sensitizing range carries out denoising and edge detection, traditional image procossing, is all to be denoised to carry out image again to whole image
Dividing processing, the present invention is partitioned into sensitizing range first, then carries out local denoising, can effectively improve image processing efficiency.
Specifically, image processing unit 3 is automatically positioned out the center of multiple mark points on image, determines background
The angle of target 14 and horizontal direction, so that the deflection angle between area array cameras 21 and background target 14 is calculated, at image
It manages unit 3 and controls the image along deflection angle progress resolution scan, to complete the sensitizing range to current detection station
Identification, after identifying the sensitizing range, image processing unit 3 carries out image segmentation operations to the image, to obtain new
Image to be processed, in a preferred embodiment, it is contemplated that the back lighting that background target 14 and LED area light source matrix 15 are constituted
Environment, the sensitizing range of image and the contrast of background are higher, and gray value difference is larger, and the present invention uses the image based on region
Partitioning algorithm is partitioned into the sensitizing range image.
When being denoised to the sensitizing range image being partitioned into, in order to enable the sensitizing range image has more natural put down
Sliding effect enhances the treatment effect to the random noise of sensitizing range image, and the present invention is using gaussian filtering method to the sensitizing range
Area image is denoised；After obtaining smooth sensitizing range image, characteristic vector pickup unit is to treated sensitizing range
Image further processes.
Specifically, the purpose of characteristic vector pickup unit be the dimensional images information characterized by pixel set is reduced to
Vector set is combined into the low-dimensional image information of feature, in order to computer processing and guarantee deep neural network unit classification
Accuracy；In the present invention, the Flexible Production feature based on intelligence manufacture, using this bullet image feature of shape feature come
Obtain workpiece for measurement intelligence manufacture information, be it is proper, in order to capture the system of workpiece for measurement sensitizing range image comprehensively
Information is made, the present invention considers the external margin information and interior zone information of target area simultaneously, by the edge of target area
Bending moment is not used as characterization sensitizing range image to the Hu of area, the edge shape factor and target area mean radius and preceding 3 dimension
Characteristic variable be input to based on deep neural network unit 5 and in this, as the feature vector of sensitizing range image.
In a preferred embodiment, the edge attributes based on sensitizing range image, characteristic vector pickup unit is first to quick
Sensillary area area image carries out edge detection, obtains target area, and calculates the side for obtaining target area by formula (1)-(3) respectively
Edge area, the edge shape factor and target area mean radius, along with the Hu not bending moment of preceding 3 dimension, constituting tool, there are four special
The feature vector of the sensitizing range of variable is levied, with manufacturing quality informations such as the processing or the assembly that reflect current production detection platform,
Feature vector is sent to deep neural network unit 5 as input layer；
In above formula, parameter M and N are the marginal point number of target area,Wherein t (x,
It y) is the gray value of each marginal point；Parameter L is the perimeter of target area, and the Chain-Code-Method that can be used in image processing techniques is counted
It calculates and obtains, be the marginal point number on target area boundaries, (x with reference to Kk,yk) indicate that the pixel being located on target area boundaries is sat
Mark,The center-of-mass coordinate for indicating target area, can be calculated by the following formula:
Wherein, parameter A indicates the area of sensitizing range, and it is big that it is obtained when can recognize sensitizing range in image procossing
It is small.
In addition, Hu not bending moment global characteristics important as image, are not influenced by light, noise, have good several
Why not deform, can effectively describe the more complicated subject image of shape, it is contemplated that the typical manufacturing defect of intelligence manufacture
The efficiency of matter and image procossing chooses preceding 3 dimension Hu not characteristic variable one of of the bending moment as workpiece for measurement sensitizing range image,
Be it is effective, specific calculation can refer to the Normal practice in image processing techniques, will not repeat them here.
In a preferred embodiment, deep neural network unit 5 of the invention, specifically based on the system of deep neural network
Make bug prediction model, and including three-layer neural network, be input layer, hidden layer and output layer respectively, wherein input layer with it is defeated
Layer scale having the same out, input interface of the input layer as manufacturing defect prediction model receive the spy of workpiece for measurement image
Vector is levied, by information coding, reaches hidden layer, using information decoded transform to output layer, the present invention is using classical volume
Code and decoding formula model, will not repeat them here；The manufacturing defect of workpiece for measurement is formally carried out in deep neural network unit 5
It before classification, needs first to carry out learning training, specifically, be lacked for the typical manufacture that current detection station workpiece for measurement is likely to occur
It falls into, establishes master sample image library, in a preferred embodiment, master sample image library may include that manufacture is qualified, manufacture is scarce
Four kinds of sample database types such as I, manufacturing defect II, manufacturing defect III are fallen into, in this, as the training sample database of deep neural network；
Similar to extraction workpiece for measurement feature vector above-mentioned, the image in master sample image library is equally carried out edge inspection by the present invention
It surveys, and successively edge area, edge standard deviation, form factor and the Hu of extraction image not instruct by the characteristic variables such as bending moment, composition
Practice the feature vector of sample database；Finally, the input layer of deep neural network unit 5 reads the feature vector in training sample database,
Coding and decoding based on deep neural network unit 5, to the corresponding manufacture information of image in each master sample image library
Deep learning is carried out, to obtain the manufacturing defect prediction model of current detection station.
After deep neural network unit 5 carries out learning training to training sample database, current detection station 11 can be used to
Workpiece for measurement carries out the classification prediction of manufacturing defect information, to identify the workpiece for measurement in current production detection platform 1
Manufacturing defect type, and the manufacturing defect information of workpiece for measurement is further transmitted to computer control unit 6, by computer control
Unit 6 processed carries out the maintenance and processing of manufacturing defect.
In a preferred embodiment, computer control unit 6 include PLC controller 61, and with product testing platform 1, figure
As acquisition unit 2 and deep neural network unit 5 communicate to connect respectively, computer control unit 6 is according to deep neural network list
Manufacturing defect information indicated by the classification results of member 5 is determined needed for repairing the manufacturing defect in conjunction with history Maintenance plan
Resource, and for the defect type, defective locations, defect level, maintenance personnel selection etc. automatically generate maintenance this lack
Sunken maintenance policy, and form corresponding work order and be sent to PLC controller 61, specific dimension is executed by PLC controller 61
Operation is repaired, and updates addition one maintenance record for being directed to the executive condition in history Maintenance plan.
Embodiment two
Invention further provides a kind of failure on-line detecting methods applied to intelligence manufacture workshop, use this hair
Bright aforementioned intelligent manufacturing shop line detection system for failure, and include the following steps:
Step 1: manufacturing defect prediction model of the building based on deep neural network, by sample image to depth nerve net
Network is trained study；
Step 2: intelligent positioning and releasing mechanism drive workpiece for measurement examining under the control instruction of computer control module
Surveying movement on station makes its preset detection position of arrival, triggers close to switch, and sends trigger signal to computer and control list
Member；
Step 3: computer control unit sends instruction, the face LED light to LED area light source matrix and image acquisition units respectively
Source matrix opens illumination, shoots according to a pre-set procedure with parameter, image acquisition units to workpiece for measurement, and photoelectricity turns
After changing, the image of generation is sent to image processing unit；
Step 4: image processing unit identifies and is partitioned into the sensitizing range of current detection station, for the regional area into
The processing of row image denoising, by the sensitizing range, image is sent to characteristic vector pickup unit after denoising；
Step 5: characteristic vector pickup unit carries out edge detection to sensitizing range, forms target area, and pass through respectively
Formula (1)-(3) calculate edge area, the edge shape factor and the target area mean radius for obtaining target area, in conjunction with preceding
Bending moment, composition do not have the feature vector of the sensitizing range there are four characteristic variable to the Hu of 3 dimensions；
Step 6: carrying out the diagnosis of manufacture information to feature vector based on trained deep neural network, predict and classify
The manufacturing defect of workpiece for measurement, and classification results are fed back into computer control unit；
Step 7: computer control unit is based on classification results and history Maintenance plan information, determines to repair the system
The maintenance policy of defect is made, work order is sent to PLC controller, specific maintenance operation is executed by PLC controller.
In a preferred embodiment, above-mentioned steps 1 specifically include:
Step 1.1: manufacturing defect prediction model of the building based on deep neural network, the deep neural network are 3 layers
Neural network, including input layer, output layer, hidden layer, wherein input layer and output layer scale having the same；
Step 1.2: the typical manufacturing defect being likely to occur for current detection station establishes master sample image library, packet
Four kinds of sample database types such as manufacture qualification, manufacturing defect I, manufacturing defect II, manufacturing defect III are included, as deep neural network
Training sample database；
Step 1.3: the image in master sample image library being subjected to edge detection, and successively extracts the edge surface of image
Product, edge standard deviation, form factor and Hu not characteristic variables such as bending moment, the feature vector of composing training sample database；
Step 1.4: the input layer of deep learning network unit reads the feature vector in the training sample database, and to every
The corresponding manufacture information of image in a master sample image library carries out deep learning, to obtain the manufacture of current detection station
Bug prediction model.
Taking the above-mentioned ideal embodiment according to the present invention as inspiration, through the above description, relevant staff is complete
Various changes and amendments can be carried out without departing from the scope of the technological thought of the present invention' entirely.The technology of this invention
Property range is not limited to the contents of the specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (9)
1. a kind of line detection system for failure, which is characterized in that including workpiece sensing platform, image acquisition units, image procossing
Unit, characteristic vector pickup unit, deep neural network unit and computer control unit, wherein
The workpiece sensing platform includes detection station, work on the area array cameras acquisition testing station in described image acquisition unit
The image of part detection platform, and it is sent to image processing unit；
Described image processing unit carries out resolution scan to the image received, obtains the sensitizing range figure of current detection station
Picture, and sensitizing range image is denoised, then the sensitizing range image after denoising is sent to characteristic vector pickup unit；
Characteristic vector pickup unit carries out edge detection to sensitizing range image, forms target area, and passes through formula (1) respectively
Edge area, the edge shape factor and the target area mean radius for obtaining target area are calculated to (3), along with preceding 3 dimension
Hu not bending moment, constituting tool, there are four the feature vectors of the sensitizing range of characteristic variable, to reflect current workpiece sensing platform
Workpiece quality information, feature vector are sent to deep neural network unit as input layer；
In above formula, parameter M and N are the marginal point number of target area,Wherein t (x, y) is each
The gray value of marginal point；Parameter L be target area perimeter, can be used the Chain-Code-Method in image processing techniques carry out calculate obtain
, it is the marginal point number on target area boundaries, (x with reference to Kk,yk) indicate the pixel coordinate being located on target area boundaries,The center-of-mass coordinate for indicating target area, can be calculated by the following formula:
Wherein, parameter A indicates the area of sensitizing range, and is suitable for obtaining it when recognizing sensitizing range in image procossing big
It is small；
Deep neural network unit is based on neural network algorithm and constructs manufacturing defect prediction model, to the image of workpiece sensing platform
Feature vector is trained, learns and classifies, and identifies the manufacturing defect type of the workpiece for measurement on current workpiece sensing platform, and
Classification results are fed back into computer control unit.
2. line detection system for failure according to claim 1, which is characterized in that
The workpiece sensing platform further includes being provided with close to switch, intelligent positioning and releasing mechanism in the side of detection station
The background target continuously arranged along flexible production line is provided with the label of multiple corresponding current detection stations on background target
Point, the other side of detection station are provided with LED area light source matrix, the LED area light source matrix and computer control unit communication link
It connects, background target and LED area light source matrix collectively form back lighting environment；And
The workpiece sensing platform is communicated to connect with computer control unit close to switch, intelligent positioning and releasing mechanism,
Wherein the intelligent positioning and releasing mechanism receive the control instruction of computer control unit, drive workpiece for measurement in detection station
Upper movement makes it reach preset detection position and the shooting central region close to area array cameras, triggers close to switch and sends touching
It signals to the computer control unit, image is carried out by computer control unit starting described image acquisition unit and is adopted
Collection.
3. line detection system for failure according to claim 1, which is characterized in that
Described image acquisition unit includes the first photo-electric conversion element successively communicated to connect with area array cameras, optical fiber slip ring,
Two photo-electric conversion elements, wherein the sensation lens axis of area array cameras is vertically arranged with detection station circulation direction, area array cameras
The electric signal of image is sent to image processing unit after photoelectric conversion, by the second photo-electric conversion element by the image of acquisition；
The first photo-electric conversion element and the second photo-electric conversion element of described image acquisition unit are provided with the defeated of identical quantity
Enter end and multiple output ends, the quantity of the optical fiber slip ring is corresponding to it, each input of first photo-electric conversion element
End is connect with the area array cameras respectively, each output end of first photo-electric conversion element passes through a fiber slide respectively
Ring is connect with the input terminal of second photo-electric conversion element, each output end of second photo-electric conversion element is and institute
State image processing unit connection.
4. line detection system for failure according to claim 1, which is characterized in that
Described image processing unit is automatically positioned out the center of multiple mark points on image, determines the background target
With the angle of horizontal direction, to calculate the deflection angle between the area array cameras and the background target, described image
Processing unit controls the image and carries out resolution scan along the deflection angle, to complete the sensitivity to current detection station
The identification in region.
5. line detection system for failure according to claim 1, which is characterized in that
The computer control unit includes PLC controller, and with workpiece sensing platform, image acquisition units and depth nerve net
Network unit communicates to connect respectively, after the computer control unit receives the manufacturing defect information of workpiece sensing platform, automatically
Maintenance policy is generated, and issues work order to PLC controller, specific maintenance operation is executed by PLC controller.
6. line detection system for failure according to claim 1-5, which is characterized in that
Computer control unit manufacturing defect information according to indicated by the classification results of the deep neural network unit,
In conjunction with history Maintenance plan, determine resource needed for repairing the manufacturing defect, and for the defect type, defective locations,
Defect level, maintenance personnel's selection automatically generate the maintenance policy for repairing the defect, and form corresponding work order and be sent to
The PLC controller executes specific maintenance operation by the PLC controller, and updates in the history Maintenance plan
Addition one is directed to the maintenance record of the executive condition.
7. a kind of failure on-line detecting method applied to intelligence manufacture workshop characterized by comprising
Step 1: building the manufacturing defect prediction model based on deep neural network, by sample image to deep neural network into
Row training study；
Step 2: workpiece sensing platform drives workpiece for measurement to transport in detection station under the control instruction of computer control module
It is dynamic that it is made to reach preset detection position, it triggers close to switch, and send trigger signal to computer control unit；
Step 3: computer control unit sends instruction, the face LED light to LED area light source matrix and image acquisition units respectively
Source matrix opens illumination, and described image acquisition unit shoots workpiece for measurement, and the image of generation is sent at image
Manage unit；
Step 4: image processing unit identifies and is partitioned into the sensitizing range of current detection station, carries out figure for the regional area
As denoising, by the sensitizing range, image is sent to characteristic vector pickup unit after denoising；
Step 5: characteristic vector pickup unit carries out edge detection to sensitizing range, forms target area, and calculate acquisition target
Edge area, the edge shape factor and the target area mean radius in region, bending moment, composition do not have the Hu tieed up in conjunction with preceding 3
The feature vector of the sensitizing range of four characteristic variables；
Step 6: carrying out the diagnosis of manufacture information to feature vector based on trained deep neural network, predict and classify to be measured
The manufacturing defect of workpiece, and classification results are fed back into computer control unit.
8. failure on-line detecting method according to claim 7, which is characterized in that
The failure on-line detecting method is suitable for the system using line detection system for failure as described in claim 1 detection workpiece
Make defect.
9. failure on-line detecting method according to claim 8, which is characterized in that
The step 1: manufacturing defect prediction model of the building based on deep neural network, by sample image to depth nerve net
The method that network is trained study includes:
Step 1.1: manufacturing defect prediction model of the building based on deep neural network, the mind that the deep neural network is 3 layers
Through network, including input layer, output layer, hidden layer, wherein input layer and output layer scale having the same；
Step 1.2: the typical manufacturing defect being likely to occur for current detection station establishes master sample image library, including system
Make four kinds of sample database types such as qualification, manufacturing defect I, manufacturing defect II, manufacturing defect III, the instruction as deep neural network
Practice sample database；
Step 1.3: the image in master sample image library being subjected to edge detection, and successively extracts edge area, the side of image
Edge standard deviation, form factor and Hu not characteristic variables such as bending moment, the feature vector of composing training sample database；
Step 1.4: the input layer of deep learning network unit reads the feature vector in the training sample database, and to each mark
The corresponding manufacture information of image in quasi- sample image library carries out deep learning, to obtain the manufacturing defect of current detection station
Prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811651018.2A CN109840900B (en) | 2018-12-31 | 2018-12-31 | Fault online detection system and detection method applied to intelligent manufacturing workshop |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811651018.2A CN109840900B (en) | 2018-12-31 | 2018-12-31 | Fault online detection system and detection method applied to intelligent manufacturing workshop |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109840900A true CN109840900A (en) | 2019-06-04 |
| CN109840900B CN109840900B (en) | 2023-12-19 |
Family
ID=66883679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811651018.2A Active CN109840900B (en) | 2018-12-31 | 2018-12-31 | Fault online detection system and detection method applied to intelligent manufacturing workshop |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109840900B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110989507A (en) * | 2019-11-02 | 2020-04-10 | 温州大学 | Automatic production digital twin workshop generation device that detects of circuit breaker |
| CN111272775A (en) * | 2020-02-24 | 2020-06-12 | 上海感图网络科技有限公司 | Device and method for detecting defects of heat exchanger by using artificial intelligence |
| CN111583190A (en) * | 2020-04-16 | 2020-08-25 | 浙江浙能技术研究院有限公司 | Automatic identification method for hidden crack defect of internal cascade structure component |
| CN111721728A (en) * | 2020-07-16 | 2020-09-29 | 陈皓 | Fruit online detection device and use method thereof |
| CN112270284A (en) * | 2020-11-06 | 2021-01-26 | 南京斌之志网络科技有限公司 | Lighting facility monitoring method and system and electronic equipment |
| CN113172422A (en) * | 2021-05-18 | 2021-07-27 | 中品智能机械有限公司 | Woodworking mechanical equipment assembly production line and assembly process thereof |
| CN113217374A (en) * | 2020-02-05 | 2021-08-06 | 中国石油天然气股份有限公司 | Operation maintenance method and system for vertical well screw pump |
| CN113781448A (en) * | 2021-09-14 | 2021-12-10 | 国电四子王旗光伏发电有限公司 | Intelligent photovoltaic power station assembly defect identification method based on infrared image analysis |
| CN116805204A (en) * | 2023-08-24 | 2023-09-26 | 超网实业(成都)股份有限公司 | Intelligent plant monitoring method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016201947A1 (en) * | 2015-06-16 | 2016-12-22 | 华南理工大学 | Method for automated detection of defects in cast wheel products |
| US20180211373A1 (en) * | 2017-01-20 | 2018-07-26 | Aquifi, Inc. | Systems and methods for defect detection |
-
2018
- 2018-12-31 CN CN201811651018.2A patent/CN109840900B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016201947A1 (en) * | 2015-06-16 | 2016-12-22 | 华南理工大学 | Method for automated detection of defects in cast wheel products |
| US20180211373A1 (en) * | 2017-01-20 | 2018-07-26 | Aquifi, Inc. | Systems and methods for defect detection |
Non-Patent Citations (3)
| Title |
|---|
| 万书亭等: "基于BP神经网络的GIS缺陷图像识别系统的研究", 《电力科学与工程》 * |
| 孙创开: ""基于机器视觉的彩钢板缺陷检测和智能分类研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 戴舒文: ""触点零件形貌在线自学习视觉检测系统研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110989507A (en) * | 2019-11-02 | 2020-04-10 | 温州大学 | Automatic production digital twin workshop generation device that detects of circuit breaker |
| CN113217374A (en) * | 2020-02-05 | 2021-08-06 | 中国石油天然气股份有限公司 | Operation maintenance method and system for vertical well screw pump |
| CN111272775A (en) * | 2020-02-24 | 2020-06-12 | 上海感图网络科技有限公司 | Device and method for detecting defects of heat exchanger by using artificial intelligence |
| CN111583190B (en) * | 2020-04-16 | 2022-07-22 | 浙江浙能技术研究院有限公司 | Automatic identification method for hidden crack defect of internal cascade structure component |
| CN111583190A (en) * | 2020-04-16 | 2020-08-25 | 浙江浙能技术研究院有限公司 | Automatic identification method for hidden crack defect of internal cascade structure component |
| CN111721728A (en) * | 2020-07-16 | 2020-09-29 | 陈皓 | Fruit online detection device and use method thereof |
| CN111721728B (en) * | 2020-07-16 | 2023-02-21 | 陈皓 | Fruit online detection device and use method thereof |
| CN112270284A (en) * | 2020-11-06 | 2021-01-26 | 南京斌之志网络科技有限公司 | Lighting facility monitoring method and system and electronic equipment |
| CN113172422A (en) * | 2021-05-18 | 2021-07-27 | 中品智能机械有限公司 | Woodworking mechanical equipment assembly production line and assembly process thereof |
| CN113172422B (en) * | 2021-05-18 | 2022-04-12 | 中品智能机械有限公司 | Woodworking mechanical equipment assembly production line and assembly process thereof |
| CN113781448A (en) * | 2021-09-14 | 2021-12-10 | 国电四子王旗光伏发电有限公司 | Intelligent photovoltaic power station assembly defect identification method based on infrared image analysis |
| CN113781448B (en) * | 2021-09-14 | 2024-01-23 | 国电四子王旗光伏发电有限公司 | Intelligent defect identification method for photovoltaic power station assembly based on infrared image analysis |
| CN116805204A (en) * | 2023-08-24 | 2023-09-26 | 超网实业(成都)股份有限公司 | Intelligent plant monitoring method and system |
| CN116805204B (en) * | 2023-08-24 | 2023-12-01 | 超网实业(成都)股份有限公司 | Intelligent plant monitoring method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109840900B (en) | 2023-12-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109840900A (en) | A kind of line detection system for failure and detection method applied to intelligence manufacture workshop | |
| US10803573B2 (en) | Method for automated detection of defects in cast wheel products | |
| CN104992449B (en) | Information identification and surface defect online test method based on machine vision | |
| CN111325713A (en) | Wood defect detection method, system and storage medium based on neural network | |
| CN111189837B (en) | Cigarette appearance online detection method and device | |
| CN109461149A (en) | The intelligent checking system and method for lacquered surface defect | |
| CN111612737B (en) | Artificial board surface flaw detection device and detection method | |
| CN109959661A (en) | Automatic optical detection method, equipment and its deep learning system | |
| CN109724984A (en) | A kind of defects detection identification device and method based on deep learning algorithm | |
| CN109693140A (en) | A kind of intelligent flexible production line and its working method | |
| CN111507976B (en) | Defect detection method and system based on multi-angle imaging | |
| CN111667455A (en) | AI detection method for various defects of brush | |
| JP2021515885A (en) | Methods, devices, systems and programs for setting lighting conditions and storage media | |
| CN110186375A (en) | Intelligent high-speed rail white body assemble welding feature detection device and detection method | |
| Yang et al. | An automatic aperture detection system for LED cup based on machine vision | |
| KR20200099977A (en) | Image generating apparatus, inspection apparatus, and image generating method | |
| CN114280075A (en) | Online visual inspection system and method for surface defects of pipe parts | |
| CN114136975A (en) | Intelligent detection system and method for surface defects of microwave bare chip | |
| CN110412055A (en) | A kind of lens white haze defect inspection method based on multiple light courcess dark-ground illumination | |
| CN116978834A (en) | Intelligent monitoring and early warning system for wafer production | |
| CN212646436U (en) | Artificial board surface flaw detection device | |
| CN113267506A (en) | Wood board AI visual defect detection device, method, equipment and medium | |
| CN113176281A (en) | System and method for detecting surface defects based on artificial intelligence drawing line | |
| CN113706496B (en) | Aircraft structure crack detection method based on deep learning model | |
| CN116843615B (en) | Lead frame intelligent total inspection method based on flexible light path |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 213100 No.28, Mingxin Middle Road, Wujin District, Changzhou City, Jiangsu Province Applicant after: Changzhou Polytechnic Address before: 213100 No.28, Mingxin Middle Road, Wujin District, Changzhou City, Jiangsu Province Applicant before: Changzhou Institute of Industry Technology |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |