CN105229548B - The method and apparatus that assembling failure is detected using the process interface marking after assembling - Google Patents
The method and apparatus that assembling failure is detected using the process interface marking after assembling Download PDFInfo
- Publication number
- CN105229548B CN105229548B CN201480028747.6A CN201480028747A CN105229548B CN 105229548 B CN105229548 B CN 105229548B CN 201480028747 A CN201480028747 A CN 201480028747A CN 105229548 B CN105229548 B CN 105229548B
- Authority
- CN
- China
- Prior art keywords
- product
- marking
- failure
- success
- interaction
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000008569 process Effects 0.000 title claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 63
- 230000003993 interaction Effects 0.000 claims abstract description 48
- 238000004519 manufacturing process Methods 0.000 claims abstract description 21
- 230000002452 interceptive effect Effects 0.000 claims abstract description 20
- 238000005259 measurement Methods 0.000 claims abstract description 9
- 230000008859 change Effects 0.000 claims description 11
- 238000000513 principal component analysis Methods 0.000 claims description 8
- 238000006073 displacement reaction Methods 0.000 claims description 7
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000007514 turning Methods 0.000 description 10
- 238000012549 training Methods 0.000 description 9
- 239000011159 matrix material Substances 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 6
- 238000005070 sampling Methods 0.000 description 5
- 238000007689 inspection Methods 0.000 description 4
- 238000012805 post-processing Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000009434 installation Methods 0.000 description 3
- 238000003825 pressing Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 2
- 229910052782 aluminium Inorganic materials 0.000 description 2
- 239000004411 aluminium Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000001066 destructive effect Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004888 barrier function Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000994 depressogenic effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000006260 foam Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0221—Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32201—Build statistical model of past normal proces, compare with actual process
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- General Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- General Factory Administration (AREA)
- Manipulator (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Automatic Assembly (AREA)
Abstract
The present disclosure describes a kind of successful technology for being used to detect the automation process of article of manufacture.The product of success statistically in a large number and failure is produced by automation process.Each for making in these products is interacted with test platform, and successfully product and unsuccessfully the interaction marking of product indicates with measurement.Calculate the correlation of the difference between the interaction marking.Then obtain for passing through the interaction marking of the product manufactured by the process after the product manufactured previously.The new interaction marking is analyzed relative to the difference in correlation calculated, so that additional article automatically is categorized as into success or failure.A kind of technology for being used to optimize the motion for being used to test article is also described, to improve the correlation of the difference between the interactive signal of success product and the interactive signal of failure product.
Description
Technical field
The present invention relates to automatic assembling, and relate more particularly to the detection of assembling failure after assembly is completed.
Background technology
Automatic assembling (such as but not limiting, robot assembling) is in the various industries such as automobile, electronic installation
Used.The robot assembling for electronic installation was described in WO2011153156 disclosed in 8 days December in 2011
One example.
Fault detect during there are, such as entitled Failure Detection in Rodriguez et al.
Assembly:Described in Force Signature Analysis (the IEEE CASE meetings for coming from 2010) paper
Those methods.Such method uses the data from sensor, and the dynamic characteristic that the data are usually assembled process is damaged.
If it would be of interest to be assembled the mechanical strength of part, can be held with automatic visual inspection and/or destructive tension force
Checked after row assembling.
The content of the invention
A kind of successful method for being used to detect the automation process of article of manufacture, includes but is not limited to:
Successful product statistically in a large number and failure product are produced using automation process;Make the success product
Interact with test platform with each product in the failure product, described make with the measurement instruction successfully product and unsuccessfully
The interaction marking of product;
Calculate the related of the difference between the interaction marking and the interaction marking of the failure product of the success product
Property;
Obtain the interaction for being directed to the additional article produced after the success product and the failure product are by production
The marking;And
Relative to the success product and the correlation of the failure product interaction marking calculated, it is directed to analyze
The interaction marking that the success product and the failure product are obtained by the additional article produced after production, with automatically
By in the success product and the failure product by production after the additional article that is produced be categorized as success or failure.
A kind of successful method for being used to detect the automation process of article of manufacture, includes but is not limited to:
Calculate the statistically interaction marking of successful product in a large number and failure product statistically in a large number
The correlation of difference between the interaction marking;
Obtain the interaction marking and the difference interacted between the marking of the failure product being directed in the success product
The interaction marking for the product that correlation is produced after calculating;And
Relative to the success product and the correlation of the failure product interaction marking calculated, it is directed to analyze
The correlation of difference between the interaction marking of the interaction marking of the success product and the failure product is by institute after calculating
The product of production and the interaction marking obtained, with automatically by the interaction marking of the success product and the failure product
The correlation of difference between the interaction marking is categorized as success or failure by the product produced after calculating.
It is a kind of to be used to use the successful product statistically in a large number of automation process production by test and unsuccessfully make
Product detect the successful method of the automation process of article of manufacture, include but is not limited to:
Make it is described success product and it is described failure product in it is each interacted with test platform, with measurement instruction successfully
The interaction marking of product and the product of failure;
Calculate the correlation of the difference between the interaction marking of the success product and the interaction marking of the failure product;
And
By changing the interacting and with interactive step of each success product and failure product and the test platform
In the case of change, the repeatedly the step of production, the interaction and calculating, to cause the interaction in the success product
The correlation of difference between the interaction marking of the marking and the failure product minimizes.
Brief description of the drawings
Fig. 1 shows the robot that hold completed knocked down products.
Fig. 2 shows that when testing the assembling failure of completed knocked down products the test used in test system after assembling is put down
Platform.
Fig. 3 and Fig. 4, which is shown, to be assembled into by an example of the part of the product of test system and test after assembling.
Fig. 5 a to Fig. 5 c show the example of the failure assembling of the part for showing in figures 3 and 4, and Fig. 5 d are shown
It is directed to the correct assembling of those parts.
Fig. 6 a and Fig. 6 b are shown to be used for for the part detection assembling event shown in Fig. 3 and 4 by test system after assembling
The motion of barrier.
Fig. 7 a and 7b show the flow in two stages during for SVMs (SVM) to be used for assembling fault detect
Figure.
Fig. 8 shows the flow chart of the post processing for the collected stress marking.
Fig. 9 shows the hyperplane that can be used for being classified to the collected stress marking.
Figure 10 a and 10b show that the motion that test article is used to for optimizing is believed with the interaction improved in success product
Number failure product interactive signal between difference correlation flow chart.
Embodiment
With reference now to Fig. 1 and 2, one embodiment for test system after assembling is shown.As shown in Figure 1 should
In embodiment, completed knocked down products 16 are gripped by the clamper 14 on the tip of robot 10.Robot 10 can be such as
It is to be hinged 6 axle robots, Descartes gantry robot, the robot having less than 6 axles of such as SCARA robots etc,
Or the robot having more than 6 axles of such as multi-arm robot etc.The motion of robot 10 is controlled by controller 12.Such as figure
Shown in 1, the test platform 18 contacted with completed knocked down products 16 is installed on workbench 20.
Test platform 18 can be anything to be interacted with completed knocked down products i.e. product 16.Although Fig. 1 show by
Test platform 18 on workbench 20, it is known that, test platform 18 can be by the machine not shown in Fig. 1
People keeps, and robot 10 can take completed knocked down products 16 in the robot that remain test platform 18, or remains
The robot of test platform 18 can be such that test platform 18 is contacted with completed knocked down products 16.
As this field technicians can recognize, controller 12 of the invention can include computer-readable Jie
Matter, it has the computer-readable instruction in the face that is stored thereon, and it performs operations described herein when being executed by processor.Meter
Calculation machine computer-readable recording medium can be any tangible medium, and it can include, stores, transmits, propagates or transmit user interface program and refer to
Order for instruction execution system, device or equipment use or it is in conjunction use, and limiting for example but not
In the case of, can be electronics, magnetic, optics, electromagnetism, infrared or semiconductor system, device, equipment or propagate it is tangible
Medium.The more specifically example (non exhaustive list) of computer-readable tangible medium includes:Portable computer diskette, hard disk,
It is random access memory (RAM), read-only storage (ROM), erasable programmable read only memory (EPROM or flash memory), portable
Formula compact disk read-only storage (CD-ROM), light storage device, magnetic memory device.It can be compiled using any properly programmed language
Computer program code or the instruction of the operation for performing the present invention are write, as long as it allows for the technical result.Although
Controller 12 can perform the operation shown in Fig. 7, Fig. 8 and Figure 10 flow chart, but the embodiment shown in Fig. 1 can also wrap
Include and communicated with controller 12 to perform the independent computing device of those operations.
As shown in Figure 2, test platform 18 is configured with least bottom 26 and top layer 27.Top layer 27 is by hard material system
Into, and bottom 26 is made up of the conformable material of such as rubber and foam etc.The purpose of such design is to provide test platform 18
Contacted with the submissive and non-destructive of completed knocked down products 16.In an alternate embodiment, due to other places in systems exist it is submissive
Property or the property due to assembly operation, it may not be necessary to compliant layers.
An example of completed knocked down products 16 is illustrated as printed circuit board (PCB) 24, the and of general rectangular socket 30 in figures 3 and 4
The combination of aluminium cask flask 36, wherein having various circuit elements printed circuit board (PCB) 24 is installed above, in general rectangular socket 30
In there are other circuit elements, and aluminium cask flask 36 covers general rectangular socket 30.Socket 30 includes forming four turnings
34a to 34d four raised side wall 32a to 32d.
As seen from Figure 3, cask flask 28 includes the tabular surface 36 of general rectangular shape.Side wall 38a to 38d from
Each edge projects upwards and forms turning 40a to 40d.Cask flask 28 is dimensioned so as to snap on socket 30.Can
In a manner of the technical staff with robot assembling field is well-known, pass through one or more machines in addition to robot 10
People's (not shown) come perform cask flask 28 arrive socket 30 assembling.
Fig. 5 a to 5d show that the good and failure of cask flask 28 to socket 30 assembles.More particularly, Fig. 5 a, which are shown, lacks
Cask flask 28, that is, cask flask 28 is not assembled into socket 30.
When cask flask 28 is assembled into socket 30, the major failure of the assembling be four turning 40a of cask flask 28 extremely
One or two in 40d is not depressed sufficiently in socket 30.Fig. 5 b are shown for a turning not in correct position
On such failure, and Fig. 5 c are shown for this failure of two turnings not on correct position.
Fig. 5 d show good assembling.As shown in this figure, cask flask 28 covers socket 30, and the whole of cask flask 28
Four turning 40a to 40d are in correct position on socket 30.
Fig. 6 a and Fig. 6 b are shown to be detected using the setting in Fig. 1 and Fig. 2 for the part shown in Fig. 3 and 4
Assembling failure test campaign.As shown in Fig. 6 a and Fig. 6 b, the test campaign is to abut against the top layer 27 of test platform 18
Press the oscillating motion of each in four turning 40a-40d of socket 30.It is the one or more of cask flask 28 in failure
Turning when being assembled into socket 30 not on correct position in the case of, it is all as shown in figures 5 b and 5 c shown in, four of socket 30
Each pressing for abutting against layer 27 in turning 40a to 40d can cause the reparation that the failure assembles.
Test motion is performed by robot 10, and is programmed during system is set.Because the height of robot 10 can
Repeatability, this test motion is identical for each completed knocked down products 16.If each clamping of completed knocked down products 16
It is repeatable, then the test condition for assembling failure has almost no change.As a result, what is induced during waving and pressing motion connects
Touch other side effects of the stress without the dynamic characteristic during such as actual assembled.
Institute can be moved from test by using force snesor to handle using many existing algorithms for fault detect
The contact marking (force signature) of acquisition.Exemplary algorithm is professional standard statistical classification instrument, supporting vector
Machine (SVM).Contact blotting data can be categorized into two classes by SVM:Success and unsuccessful failure.
Respectively as shown in Fig. 7 a and 7b flow chart 710 and 722, there is two when SVM is used for into assembling fault detect
Individual stage, i.e., the test rank that the training stage and its flow chart 722 that its flow chart 710 is shown in figure 7 a show in fig.7b
Section.The two stages share the common action that will be described in more detail below, and such as perform test motion, collect the stress marking,
The stress marking is post-processed and taken characteristic vector.
The step of SVM training continues shown in Fig. 7 a as follows:
1. at step 712 and 714, as follows to the statistically huge amount with the success of generally half and half failure
N number of completed knocked down products 16 in each completed knocked down products 16 perform test motion:
A. test motion is performed at step 712 and is directed to each completed knocked down products 16 in the series and collects the stress marking
Information;And
B. the classification of each in a series of completed knocked down products 16 that will be tested at the step 714 be recorded as successfully or
Failure.
2. perform step 716 and 718 for a series of each completed knocked down products 16 in N number of completed knocked down products 16.
The stress marking collected at step 712 is post-processed at step 716.Figure 8 illustrates for after
The flow chart 800 of processing and its have steps of:
A. it is if the sampling time is uneven, then smooth to the stress marking re-sampling and progress that are recorded at step 802
Change to remove noise (such as average etc the various widely-known techniques of boxcar can be used to perform this function);
B. each collected stress marking is made to be aligned with reference to the marking at step 804, to remove because of the different time starteds
Caused by skew;And
C. stress blotting data is normalized at step 806 so that across all markings, peak is 1, and
Minimum is 0.
Fig. 7 a are now turned to, perform step 718 after processing step 716 after execution.In this step, extraction is located after
The characteristic vector of the marking of reason.If L.Smith is in " A tutorial on principal components analysis "
(2002, the U.S., Cornell University) described in, first several points from principal component analysis (PCA) can be used
Amount is used as characteristic vector.
3. after each in for N number of completed knocked down products 16 performs step 716 and 718, at step 720, base
SVM is trained in characteristic vector, if C.Burges is in " A Tutorial on Support Vector Machines for
Pattern Recognition " (Data Mining and Knowledge Discovery 2,121-167,1998
(" Burges ")) described in.
SVM has been trained to be saved and for each using in the test phase of completed knocked down products.The test phase
Set using the system of the system setting identical (or similar) with being used during the training stage.
Herein in use, statistically N number of completed knocked down products 16 in a large number mean to obtain enough samples to cause
The stress marking can be classified with the degree of accuracy of predeterminated level.For example, as described in the bibliography in C.Burges, can
To calculate the error rate or " practical risk " (page 156) of having trained SVM, if based on predetermined threshold, SVM " practical risk " value
It is too high, then more multisample can be obtained, and re -training SVM is untill value-at-risk is acceptable.Alternatively, can be with
Determine whether the number of sample is statistically very big using other SVM attributes.
Show the flow chart 722 for test phase in fig.7b, wherein step 724,726 and 728 respectively with instruction
It is identical to practice step 716,718 and 720 in stage, and is therefore not described any further.Contact print during test phase
Note is recorded and is fed in SVM.After using identical post processing and feature extraction, characteristic vector is then input into
SVM, it runs at step 730 and has trained SVM.SVM output has been trained to predict that the assembling of each tested product is success
Or failure.At step 732, when having trained SVM output to predict that tested product is not assembled correctly, abandon tested
Trial product.In an alternate embodiment, different actions can be performed to product, number of assembling steps is such as retried or is placed on product
While for doing over again manually later.
It note that and be directed to each step, especially post processing and feature extraction, many algorithms of different can be used.For example,
The re-sampling of stress blotting data can use simple linear interpolation technique:
Wherein, F'(t) be new sample point stress data, F (ta) and F (tb) it is in crude sampling time taAnd tbPlace
The stress data of record.
The example of noise remove algorithm includes low pass filter (formula 2) and weighted moving average (formula 3).In two formulas,It is the stress data after noise is removed.
Re-sampling and noise remove can be in a single step handled as shown in the step 802 in Fig. 8.Above
The interpolating method referred to is good candidate for this purpose.
The stress marking is set to be aligned and can be realized by cross-correlation method with reference to the marking:
Wherein, fkIt is the stress marking of record, and gkIt is to refer to the marking.Cross-correlation function (f*g)kMaximum at
Parameter k is stress marking fkRelative to reference to marking gkMisalignment.By simply being shifted markers with this misalignment, and obtain
The stress marking that must be aligned:
Wherein, q is misalignment.
The normalization of the stress marking, step 806 can be completed using formula 6:
Wherein, μ and σ is stress marking fkAverage and standard deviation.
Note that the post processing of the stress marking can have the step different from those steps shown in Fig. 8, and this depends on
In the quality of the stress marking recorded.
After post-treatment, feature extraction will be carried out to the stress marking at step 718.Good time for feature extraction
It is principal component analysis (PCA) to select algorithm.PCA is for being asked by the correlation between heuristic data or redundancy to reduce study
The common tool of the size of data set in topic.PCA becomes to bring by the linear coordinate of initial data realizes this target.It is referred to as main
Try to make the change of data to maximize in the new basis of component.The direction pair of the maximum change of the first factor and raw data set
Standard, therefore include most information on data set.Each continuous principal component with not by previous component catch it is maximum remaining
Variational direction alignment.Mathematically, it is assumed that data set includes N number of measurement resultiX, i=1 ... N.EachiX be m dimension to
Amount, so thatiX=[ix1,ix2...ixm]′.For stress assembling fault detect,iX is the post-treated stress marking.Make P
For the linear coordinate transformation matrix of m*m dimensions, after transformation, initial dataiX becomes new dataiy:
iY=PiX, i=1 ... N (7)
Write using the Simple form of N number of measurement result including whole, then formula (7) becomes
Y=PX
X=[1x 2x…N-1x NX], Y=[1y 2y…N-1y Ny] (8)
Make the maximized transformation matrix P of change relevant with the eigenvector of the following covariance matrix of raw data set:
The eigenvector and characteristic value for calculating covariance matrix C provide:
V-1CV=D (10)
Wherein, V is the matrix of eigenvector, and D is the diagonal matrix by the C for order arrangement of successively decreasing characteristic value.In formula (7)
Transformation matrix P then be equal to matrix V.WithiY provides data pointiX principal component.
For assembling fault detect, preceding several principal components of the stress marking recorded, which can include, to be enough to be selected as feature
The information of vector.For example, test has shown that first five (5) individual principal component is the good candidate for characteristic vector.
After obtaining for each recording the characteristic vector of the stress marking, grader is ready to train.Training data
Including the classification belonging to a series of characteristic vector and each characteristic vector.In the case of assembling fault detect, classification is into
Work(and failure.Grader has been trained to predict new feature vector in which classification.One good grader is linearly to prop up
Hold vector machine (SVM).Linear SVM trial divides feature space with hyperplane, so that two classifications fall in hyperplane
Opposite side.
As used in fig.9 shown in 2D feature spaces, many hyperplane that possibility is classified to data be present.As optimal
One reasonable selection of hyperplane is to represent that selection of largest interval or nargin between two classifications.Referred to as maximum nargin
This hyperplane of hyperplane has optimum stabilization relative to noise.Linear SVM algorithm will find such super in feature space
Plane.Mathematically, it is assumed that training set includes N number of data set:
Wherein,iX is dimension p characteristic vector,iY is 1 or -1, instruction this feature vectoriClassification belonging to x.Can be by spy
Any hyperplane in sign space is written as the set for the point for meeting following formula:
Wx-b=0 (12)
Wherein, dot product is represented, and w represents the normal vector of hyperplane.
Described two classifications are on the opposite side of hyperplane, therefore hyperplane meets
w·ix-b≥1 for iY=1wix-b≤-1 for iY=-1 (13)
Or simplyiy(w·i) >=1, x-b for alliX, i=1 ... N.
Thus the nargin between two classifications of hyperplane production is 2/ | | w | |.Therefore, can be by asking following optimization
Topic solves, to find maximum nargin hyperplane:
Minimize | | w | |, constraintsiy(w·iX-b) >=1, i=1 ... N (14)
Many software programs can be used for training SVM;These software programs are market either open-sources on sale.
With w and b SVM will be trained to be parameterized as shown in formula 12.The prediction of SVM during test phase can be with
The classification belonging to new test is predicted using following decision logic:
If wiX-b >=1 itemiY=1
If wiX-b≤- 1 itemiY=-1 (15)
Otherwise, SVM can not predict output.
It is to be appreciated that completed knocked down products 16 be only can be used together with method described herein and system with
An example of the completed knocked down products of the assembling failure in completed knocked down products is detected after product is assembled.For those products
Speech is that successfully assembling or manufacture or the measurement of unsuccessful assembling or manufacture will depend on product.
Although described above is obtained using force snesor between completed knocked down products 16 and the top layer 27 of test platform 18
The interaction marking, it will be appreciated that can be measured using other types of sensor between product and its surrounding environment
Interaction.One sensors with auxiliary electrode is displacement transducer.In cask flask example described above, unsuccessfully assemble generally with convex
The turning or edge risen.When it is abutted against submissive subject presses, the submissive object and those the assembling feelings successfully assembled
Condition is more compared to that will deform.Therefore the marking can be interacted between product and submissive object to obtain using displacement transducer.When
During using displacement transducer, program described above and algorithm can also be followed to train SVM and detect product using SVM
Success or failure.The interaction marking from displacement transducer can be along one or more shaft positions measurement result and/or
Around redirecting for those axles.
It is also to be appreciated that the interaction marking can carry out the group of the interaction marking of force sensor and displacement transducer
Close.
With reference now to Figure 10 a and 10b, it illustrated therein is and the motion of test article is used to for optimizing to improve in success
The flow chart 1000 of the correlation of difference between the interactive signal of product and the interactive signal of failure product.
Square frame 1002 and 1004 shown in Figure 10 a is two operations that each being directed in N number of product is repeated.
At square frame 1002, in N number of product each perform test motion, and collect from perform the motion obtain stress print
Note.At square frame 1004, the success or failure classification for the assembling of each in N number of product is recorded.
When completing all operations in square frame 1002 and 1004 for N number of product, flow is advanced to in N number of product
Each two of second group operation performed.At square frame 1006, to the stress marking for each in N number of product
Post-processed.At square frame 1008, for each of each in N number of product from for N number of product after from
The stress marking extraction characteristic vector of reason.
When each in for N number of product completes all operations in square frame 1006 and 1008, the flow side of advancing to
Frame 1010, wherein training SVM for N number of product.
After SVM is trained at square frame 1010, routine advances to the square frame 1012 shown in Figure 10 b, there pin
Change test motion to N number of product.The change of test motion can be relative to Fig. 6 a for example and without limitation
It is related to the test campaign shown in 6b, the different positions of the pressing socket 30 of test platform top layer 27 are abutted against by oscillating motion
Put.Alternatively, the test kinematic parameter such as speed, angle is changed, or from the diverse location and orientation of test platform 18
The test kinematic parameter such as speed, angle is read in installation different force snesors in place.
Flow then continues to square frame 1014 and 1016, and the square frame is to be repeated for each in N number of product
Two operation.What is performed at square frame 1014 and 1016 operates the operation phase with being performed at square frame 1002 and 1004 respectively
Together, simply these operations are to be directed to the change of each the test motion being used in N number of product.
When completing all operations in square frame 1002 and 1004 for N number of product, flow is advanced to in N number of product
Each square frame 1018 and 1020 performed at two of second group operations.The operation performed at square frame 1018 and 1020
Identical with the operation performed at square frame 1006 and 1008 respectively, simply these operations are for in N number of product
Extracted characteristic vector is moved in the test that changed of each.
In the completion of operation 1020 for N number of product, flow advances to square frame 1022, and wherein SVM is trained to.Such as can be with
Recognize, it is for having the N number of product for having changed test motion at square frame 1012 that this, which has trained SVM,.
Flow then continues to square frame 1024, in square frame, it is determined whether must optimal inspection motion to improve correlation.Stream
The target of journey 1000 be make success product interactive signal and failure product the interaction marking between difference correlation most
Smallization.If correlation must be improved, necessary optimal inspection motion, and flow returns to square frame 1012, in the square frame, surveys
Test run is dynamic to be changed again.If need not improve difference in correlation, flow advances to square frame 1026, and SVM training is complete
Into.When difference in correlation meets the predetermined criterion for the difference, the poor minimum terminates.
The predetermined criterion for example and can be without limitation improve correlation value be less than preset threshold value (this
Mean that optimal inspection motion can be generated on successful and the unsuccessfully force snesor marking between assembling clear and definite difference);Or
Person is after all different test campaigns are performed, and on the reduction of value of correlation, (this means optimal inspection motion
Generating among all test campaigns on the difference of the force snesor marking between successfully assembling and unsuccessfully assembling
It is optimal);Or the overall test number of product exceedes preset number;Or to produce N number of product for optimization
Total time exceedes preset time.
It should be understood that (one or more) exemplary embodiment above description be intended to be merely illustrative the present invention rather than
The limit present invention.In the case where not departing from the spirit or its scope of the invention being defined by the appended claims, technology people
Member will can be added, delete and/or change to (one or more) embodiment of open theme.
Claims (9)
1. a kind of successful method for being used to detect the automation process of article of manufacture, including:
Successful product statistically in a large number and failure product are produced using the automation process;
Make it is described success product and it is described failure product in each product interacted with test platform, with measurement indicate it is described into
The interaction marking of work(product and the failure product;
Calculate and interact the difference between the marking with the described of the failure product in the interactive marking of the success product
Correlation;
Obtain the interaction marking for being directed to the additional article produced after the success product and the failure product are by production;
And
Relative to the success product and the correlation of the failure product interaction marking calculated, it is directed to analyze
The interactive marking that the success product and the failure product are obtained by the additional article produced after production,
With automatically by the success product and the failure product by production after the additional article that is produced be categorized as into
Work(or failure.
2. according to the method for claim 1, wherein the interaction makes the product with being arranged on above the test platform
Force snesor contact, to obtain the stress marking.
3. according to the method for claim 1, wherein measuring the displacement of the product by using displacement transducer to obtain
Obtain the interactive marking.
4. according to the method for claim 1, wherein the correlation calculations use SVMs (SVM).
5. according to the method for claim 1, wherein calculating the interactive marking and the mistake in the success product
Before the correlation for losing the difference between the interactive marking of product, described make the success product and unsuccessfully
The interactive marking of product is normalized.
6. according to the method for claim 1, wherein calculating the interactive marking and the mistake in the success product
Before the correlation for losing the difference between the interactive marking of product, use principal component analysis (PCA).
A kind of 7. successful method for being used to detect the automation process of article of manufacture:
Calculate interacting for the statistically interaction marking of successful product in a large number and statistically failure product in a large number
The correlation of difference between the marking;
Obtain for the described related of the difference between the interaction marking of the interaction marking of success product and failure product
The interaction marking for the product that property is produced after calculating;And
Relative to the success product and the correlation of the failure product interaction marking calculated, it is directed to analyze
The correlation of difference between the interaction marking of success product and the interaction marking of failure product is produced after calculating
The product and the interactive marking that obtains, with automatically by interacting in the interaction marking of success product and unsuccessfully product
The correlation of difference between the marking is categorized as success or failure by the product produced after calculating.
8. a kind of be used for by test using the automation process of article of manufacture the successful system statistically in a large number that produces
Product detect the successful method of the automation process with failure product, and methods described includes:
Make it is described success product and it is described failure product in each product interacted with test platform, with measurement indicate it is described into
The interaction marking of work(product and the failure product;
Calculate and interact the difference between the marking with the described of the failure product in the interactive marking of the success product
Correlation;And
Interacted by changing each success product and the failure product with the described of the test platform and in the friendship
Have in mutual step in the case of changing, the step of repeating the production, the interaction and the calculating, with cause it is described into
The correlation of the difference between the interactive marking of the interactive marking of work(product and the failure product is most
Smallization.
9. according to the method for claim 8, wherein repeating the life in the case that there is change in the interactive step
The step of production, the interaction and calculating, is used for the phase so that the difference in correlation minimizes in the satisfaction that minimizes
Terminate during the predetermined criterion for closing sex differernce.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US201361805864P | 2013-03-27 | 2013-03-27 | |
US61/805,864 | 2013-03-27 | ||
PCT/US2014/031837 WO2014160760A2 (en) | 2013-03-27 | 2014-03-26 | Method and apparatus for using post assembly process interaction signatures to detect assembly failures |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105229548A CN105229548A (en) | 2016-01-06 |
CN105229548B true CN105229548B (en) | 2018-04-06 |
Family
ID=50694025
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201480028747.6A Active CN105229548B (en) | 2013-03-27 | 2014-03-26 | The method and apparatus that assembling failure is detected using the process interface marking after assembling |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN105229548B (en) |
WO (1) | WO2014160760A2 (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117280286A (en) * | 2022-03-30 | 2023-12-22 | 京东方科技集团股份有限公司 | Method, apparatus and storage medium for analyzing correlation of production process data |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1506802A (en) * | 2002-12-10 | 2004-06-23 | 安捷伦科技有限公司 | System and method for classification of cost efficiency to detected objects |
CN102879881A (en) * | 2012-10-31 | 2013-01-16 | 中国科学院自动化研究所 | Element holding device |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101128830A (en) * | 2005-02-25 | 2008-02-20 | 比奥根艾迪克Ma公司 | Process equipment monitoring |
US8090676B2 (en) * | 2008-09-11 | 2012-01-03 | Honeywell International Inc. | Systems and methods for real time classification and performance monitoring of batch processes |
JP4730451B2 (en) * | 2009-03-16 | 2011-07-20 | 富士ゼロックス株式会社 | Detection data processing apparatus and program |
WO2011153156A2 (en) | 2010-06-02 | 2011-12-08 | Abb Research Ltd | Robotic assembly with partial fixturing |
-
2014
- 2014-03-26 CN CN201480028747.6A patent/CN105229548B/en active Active
- 2014-03-26 WO PCT/US2014/031837 patent/WO2014160760A2/en active Application Filing
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1506802A (en) * | 2002-12-10 | 2004-06-23 | 安捷伦科技有限公司 | System and method for classification of cost efficiency to detected objects |
CN102879881A (en) * | 2012-10-31 | 2013-01-16 | 中国科学院自动化研究所 | Element holding device |
Non-Patent Citations (1)
Title |
---|
Failure detection in assembly:Force signature analysis;RODRIGUEZ A et al.;《IEEE CONFERENCE ON AUTOMATION SCIENCE AND ENGINEERING(CASE)》;20100821;第210-215页 * |
Also Published As
Publication number | Publication date |
---|---|
WO2014160760A2 (en) | 2014-10-02 |
WO2014160760A3 (en) | 2014-11-13 |
CN105229548A (en) | 2016-01-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110930347B (en) | Convolutional neural network training method, and method and device for detecting welding spot defects | |
US11010838B2 (en) | System and method for optimizing damage detection results | |
Tibaduiza et al. | Structural damage detection using principal component analysis and damage indices | |
Khodayari-Rostamabad et al. | Machine learning techniques for the analysis of magnetic flux leakage images in pipeline inspection | |
USRE47557E1 (en) | Method and computer system for characterizing a sheet metal part | |
JP2020126601A (en) | Fault detecting method and fault detecting device | |
US20200320402A1 (en) | Novelty detection using deep learning neural network | |
CN110648305B (en) | Industrial image detection method, system and computer readable recording medium | |
JP6749655B1 (en) | Inspection device, abnormality detection method, computer program, learning model generation method, and learning model | |
US11461584B2 (en) | Discrimination device and machine learning method | |
EP4068140A1 (en) | Method and system for optimizing a simulation model using machine learning | |
JP2020198092A (en) | Method and system for unsupervised anomaly detection and cause explanation with majority voting for high-dimensional sensor data | |
CN113487533A (en) | Part assembly quality digital detection system and method based on machine learning | |
Wu et al. | An end-to-end learning method for industrial defect detection | |
Adam et al. | Multiple faults diagnosis for an industrial robot fuse quality test bench using deep-learning | |
Surovi et al. | A study on the acoustic signal based frameworks for the real-time identification of geometrically defective wire arc bead | |
CN105229548B (en) | The method and apparatus that assembling failure is detected using the process interface marking after assembling | |
Degadwala et al. | DeepTread: Exploring Transfer Learning in Tyre Quality Classification | |
Sauter et al. | Defect detection of metal nuts applying convolutional neural networks | |
CN109992106B (en) | Gesture track recognition method, electronic device and storage medium | |
JP7322560B2 (en) | Program, information processing method and information processing apparatus | |
CN114091347A (en) | Electronic component service life prediction method | |
Sahoo et al. | A robotic assistance machine vision technique for an effective inspection and analysis | |
CN113128158A (en) | Coupling coefficient identification method of wireless power transmission system | |
CN117589109B (en) | Quality detection and control method for quartz center tray manufacturing process |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20180516 Address after: Baden, Switzerland Patentee after: ABB TECHNOLOGY LTD. Address before: Zurich Patentee before: ABB T & D Technology Ltd. |