CN107696034A - A kind of wrong autonomous restoration methods for industrial robot - Google Patents
A kind of wrong autonomous restoration methods for industrial robot Download PDFInfo
- Publication number
- CN107696034A CN107696034A CN201710922428.5A CN201710922428A CN107696034A CN 107696034 A CN107696034 A CN 107696034A CN 201710922428 A CN201710922428 A CN 201710922428A CN 107696034 A CN107696034 A CN 107696034A
- Authority
- CN
- China
- Prior art keywords
- data signal
- wrong data
- robot
- wrong
- elm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/1653—Programme controls characterised by the control loop parameters identification, estimation, stiffness, accuracy, error analysis
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1602—Programme controls characterised by the control system, structure, architecture
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Automation & Control Theory (AREA)
- Numerical Control (AREA)
Abstract
The present invention, which provides a kind of autonomous restoration methods of mistake for industrial robot and device, methods described, to be included：Obtain the processing information of wrong data signal and all wrong data signals corresponding to robot；The wrong data signal is pre-processed using PCA algorithms, obtains pretreated wrong data signal；Quantitative analysis mathematical modeling is established using ELM algorithms and handles the pretreated wrong data signal, classification processing is carried out to the wrong data signal, and determines the processing information of corresponding wrong data signal, robot is able to respond the mistake.The above method can quickly and accurately realize that the mistake of industrial robot is autonomous and recover, to improve industrial machine human reriability.
Description
Technical field
The present invention relates to data processing technique, and in particular to a kind of autonomous restoration methods of mistake for industrial robot and
Device.
Background technology
World-leading, six axles crawl robot is for the industrial robot product of ABB AB and software kit solution
Relatively common industrial robot, the series of IRB 6640 is one of which.It has different mechanical arm lengths optional and can be with
Different instruments is equipped according to specific industrial environment, to reach highest operating efficiency.For example, machine mentioned below
Device people is equipped with tape clamp instrument on the robotic arm.
However, when we perform task in industrial environment using ABB crawls robot, due to diversified original
Cause, some system mistakes will be quoted inevitably.At present, the ABB IRB systems of Robotware machine handing softwares are carried
Row robot provides the user pre-defined its diagnosis processing method：User can program is pre-defined to work as robot by writing
Specific processing scheme when encountering such wrong.This abnormality eliminating method is by the self-defined mistake based on type of error
Manage program.But the drawbacks of this error handling method, is also apparent from：Robot can only make to pre-defined type of error
Reaction, when a kind of new mistake is quoted, robot can stop and wait for operating personnel's processing immediately, due to industry spot feelings
Condition is more complicated, by gather data, it is found that this situation often occurs.It will be apparent that for customer perspective, this is can not be by
Receive.In real commercial Application, client needs more solutions.Flexibility and reliability just seem very heavy herein
Will.
Therefore, need a kind of autonomous sorting technique of the wrong data that can realize robot badly.
The content of the invention
To solve the problems of the prior art, the present invention provides a kind of wrong autonomous restoration methods for industrial robot
And device, this method can quickly and accurately realize the mistake of industrial robot from Main classification, to improve industrial robot
Reliability.
In a first aspect, the present invention provides a kind of wrong autonomous restoration methods for industrial robot, including：
Obtain the processing information of wrong data signal and all wrong data signals corresponding to robot；
The wrong data signal is pre-processed, obtains pretreated wrong data signal；
The pretreated wrong data signal is handled using ELM algorithm quantitative analyses mathematical modeling, to the mistake
Data-signal carries out classification processing, and determines the processing information of the wrong data signal after corresponding classification processing；
Wherein, using ELM algorithm quantitative analysis mathematical modelings classification results rate of accuracy reached to default value.
Alternatively, it is described that the wrong data signal is pre-processed, obtain pretreated wrong data signal
Step, including：
Processing is compressed to wrong data signal using principle component analysis PCA, obtains the data matrix that dimension reduces；
Correspondingly, the pretreated error number is handled using the ELM algorithm quantitative analyses mathematical modeling pre-established
It is believed that number the step of, including：
Classification processing is carried out to the data matrix that dimension reduces using ELM algorithm quantitative analysis mathematical modelings.
Alternatively, processing is compressed to wrong data signal using principle component analysis PCA, obtains the data that dimension reduces
The step of matrix, including：
The wrong data signal is converted into m × n matrix, the matrix of the m × n is standardized, obtained
Matrix after standardization；M, n is the natural number more than 1；
Singular value decomposition is carried out to the matrix after standardization, obtains load vector and score vector；
Load vector and the pivot number of score vector are determined by contribution rate of accumulative total method；
Space load matrix is asked for according to pivot number, and obtains the data matrix of dimension reduction.
Alternatively, it is described that the pretreated mistake is handled using the ELM algorithm quantitative analyses mathematical modeling pre-established
By mistake before the step of data-signal, methods described also includes：
Activation primitive and hidden node number are set, establish ELM algorithm quantitative analysis mathematical modelings；
By ELM algorithm quantitative analysis the application of mathematical model to industry spot, using the wrong data in industry spot to building
Vertical ELM algorithm quantitative analysis mathematical modelings are modified, until the classification of revised ELM algorithms quantitative analysis mathematical modeling
As a result rate of accuracy reached is to the default value.
Alternatively, for containing N number of hidden node, N number of different sample (xi,ti), and the ELM that excitation function is g (x) is calculated
Quantitative analysis mathematical modeling is as shown in formula one：
In formula one, xi=[xi1,xi2…xin]T∈Rn,ti=[ti1,ti2…tin]T∈Rm,wiTo connect i-th of hidden layer
The input weights of node and output node, βiFor i-th of hidden node of connection and the output weights of output node；biIt is hidden for i-th
The deviation of node layer；wi·xiRepresent wiAnd xiInner product；Excitation function g (x) is sigmoid, sine or RBF.
Alternatively, the processing information of wrong data signal and all wrong data signals corresponding to robot is obtained, including：
The species that robot is applicable all robots in industry spot is obtained, and may corresponding to all robot species
The treatment measures of the wrong data signal of appearance；And
The input/output signal of each robot used in industry spot is obtained, the input/output signal includes machine
The wrong data signal of device people；
Correspondingly, the wrong data signal is pre-processed, the step of obtaining pretreated wrong data signal,
Including：
The input/output signal is pre-processed, obtains pretreated wrong data signal.
Second aspect, the present invention provide a kind of wrong autonomous recovery device for robot, including：
Acquiring unit, the processing for obtaining wrong data signal corresponding to robot and all wrong data signals are believed
Breath；
Pretreatment unit, for being pre-processed to the wrong data signal, obtain pretreated error number it is believed that
Number；
Classification processing unit, for handling the pretreatment using the ELM algorithm quantitative analyses mathematical modeling pre-established
Wrong data signal afterwards, classification processing is carried out to the wrong data signal, and determine the processing of corresponding wrong data signal
Information, wherein, using ELM algorithm quantitative analysis mathematical modelings classification results rate of accuracy reached to default value.
Alternatively, the pretreatment unit, is specifically used for, and wrong data signal is pressed using principle component analysis PCA
Contracting is handled, and obtains the data matrix that dimension reduces；
Correspondingly, the classification processing unit, specifically for using the ELM algorithm quantitative analysis mathematical modelings pre-established
The data matrix reduced to dimension carries out classification processing.
The third aspect, the present invention provide a kind of wrong autonomous recovery device for robot, including：Store instruction is deposited
Reservoir, and the processor instructed in memory is loaded and performs, the processor is specifically used for
Obtain the processing information of wrong data signal and all wrong data signals corresponding to robot；
The wrong data signal is pre-processed, obtains pretreated wrong data signal；
The pretreated wrong data signal is handled using the ELM algorithm quantitative analyses mathematical modeling pre-established,
Classification processing is carried out to the wrong data signal, and determines the processing information of corresponding wrong data signal, wherein, using ELM
The rate of accuracy reached of the classification results of algorithm quantitative analysis mathematical modeling is to default value.
It is the device have the advantages that as follows：
The initial data that industry spot collects is pre-processed first in the embodiment of the present invention, to remove live noise
And mode input variable number is reduced, such processing mode subsequently to use ELM algorithm quantitative analysis mathematical modelings speed more
Fast and model precision of prediction is higher.
Recover and processing in addition, the inventive method can quickly and accurately realize that the mistake of industrial robot is autonomous, with
The methods of preceding expert system, is compared, and analytical cycle is short, engineer is simple to operate debugging stage early stage, is built using computer
Mould simultaneously calculates, improves measuring accuracy, improves operating efficiency.
In actual applications, classification processing is carried out using the above method, the input of instrument and the input of manpower can be reduced,
Working strength is small, has saved the input cost of production, while reduce human error.
Brief description of the drawings
Fig. 1 is the schematic diagram of robot working environment；
Fig. 2 is the flow signal for the autonomous restoration methods of the mistake for industrial robot that one embodiment of the invention provides
Figure；
Fig. 3 A are that the flow for the autonomous restoration methods of the mistake for industrial robot that another embodiment of the present invention provides is shown
It is intended to；
Fig. 3 B are the schematic network structure of the ELM algorithm quantitative analysis mathematical modelings used in the embodiment of the present invention；
Fig. 4 is the association schematic diagram for showing training sample set test set；
Fig. 5 is that improved PCA-ELM node in hidden layer influences on training sample set number in the embodiment of the present invention
Matlab analogous diagrams；
Fig. 6 is that the Matlab that improved PCA-ELM node in hidden layer influences on the training time in the embodiment of the present invention is imitated
True figure；
Fig. 7 is the structural representation for the autonomous recovery device of the mistake for industrial robot that one embodiment of the invention provides
Figure.
Embodiment
In order to preferably explain the present invention, in order to understand, below in conjunction with the accompanying drawings, by embodiment, to this hair
It is bright to be described in detail.
All of technologies and scientific terms used here by the article and the those skilled in the art for belonging to the present invention are usual
The implication of understanding is identical.Term used in the description of the invention herein is intended merely to describe specific embodiment
Purpose, it is not intended that in the limitation present invention.Term as used herein " and/or " include one or more related Listed Items
Arbitrary and all combination.
Embodiment for a better understanding of the present invention, the application scenarios of robot are illustrated below.
Body of a motor car industry is made up of a variety of assembled processes, and the part involved by these assembled processes is very smart
, the requirement to precision is very high.That is the robot itself of toilet application can reach very high
Precision, but still the risk of path failure is had, this will cause the damage of primitive part.In order to ensure heat bonding and it is non-hot sticky
Close technique to reach quality standards, bond tool needs to have translational speed quickly and accurately walked, this kind of complicated at present
Motion process is mainly completed by 6 shaft industrial robots for largely carrying instrument.
Pickup and to put part be two kinds of tasks very common in vehicle body industry, if it is desired to completing both with a robot
Business is, it is necessary to define 5 work stations and edit four conventional mobile processes.
(1) work station defines：Work station is that a robot is captured, and places or part is operated specific
Position, 5 work stations are illustrated in the present embodiment, as shown in Figure 1：
Home 1 (origin station 1)：One initial bit being perfectly safe, robot is before crawl instruction is obtained and completes to put
Putting can be always maintained in this position after splicing.Meanwhile on this position, the gripping tool clip that robot carries is in normal
Open state.
Home 2 (origin station 2)：One initial bit being perfectly safe, does not interfere with start bit 1, and robot completes to capture
After instruction and obtain to be always maintained in this position before placing splicing instruction.Meanwhile in the gripping tool of this position robot
Clip is in closure state, strips.
Pick Station (crawl station)：Robot wants the initial placement position of gripper components, and it is usually located at one
The buffer area of individual station.
Drop Station (placement station)：Robot grabs the target location placed needed for part, and it is usually located at
Ad-hoc location or a buffer area on vehicle body.
Scrap Station (scrap station)：If robot made a mistake in course of normal operation cause parts damages or
It is that robot acquiescence places the mean in this position before returning to original position when state is not clear.On this position, robot
Open the clip on gripping tool.
(2) station state defines：Station state is dynamic, and it changes with the movement of robot, defines this
One state be in order to describe it is each station under special time it is whether occupied.In the present invention, following three kinds of states are defined：
Station Ok：Station state is normal and can use, and robot can enter commitment positions；
Station NOK：Station state is abnormal, and robot completes mission failure；
Station Busy：Work station hurries, and represents to be taken by other users.
(3) robot task defines
In industry spot, robot completes specific some work by advance editor, when robot completely completes
When all tasks return to Home1, robot is claimed to complete a circulation, the work of robot is exactly to repeat each to follow
Ring.In the present invention, each circulation contains four tasks, and the robot trajectory of each task is planned in advance：
Crawl task (Pick routine)：Robot is moved to Pick Station grip from Home1 stations；
Capture return task (Pick back routine)：Robot is moved to Home2 from Pick Station strips
Stand；
Placement task：Robot is moved to drop stations from the station strips of Home Station 2；
Place return task：Robot is moved to Home1 stations from drop Station；
Not in track：There is exception in robot, and the position detected is not on the track editted in advance.
(4) tool state defines
Pick-and-place robot be by the sucker on entrained instrument or clip come pick-and-place part, what tool state defined
It is the folding situation of these suckers and clip.The present invention includes four kinds of tool states：
TO：Sucker and clip are in open mode；
Instrument closes：Sucker and clip are in closure state；
Tool abnormality：There is exception in instrument, can not normal folding.
Fig. 2 shows that the flow for the autonomous restoration methods of the mistake for industrial robot that one embodiment of the invention provides is shown
It is intended to, as shown in Fig. 2 the method for the present embodiment comprises the following steps：
201st, the processing information of wrong data signal and all wrong data signals corresponding to robot is obtained.
For example, the processing information of all wrong data signals in the step includes：Robot species, each species
Treatment measures corresponding to robot per a kind of wrong data signal.
In a particular application, the species that robot is applicable all robots in scene, and all machines can be obtained in advance
Processing information of the treatment measures for the wrong data signal being likely to occur corresponding to people's species as all wrong data signals.
In actual applications, the input/output signal of each robot in certain period of time can be obtained in site of deployment,
The input/output signal includes the wrong data signal of robot.
In another optional implementation, the wrong data signal of situ machine people can be obtained.The wrong data
Signal can be likely to occur wrong wrong data signal for what is screened from input/output signal.
Input/output signal includes normal data signal and wrong data signal, wherein, normal data signal may include
The normal signal of TO, the instrument stated close normal signal etc..
Working condition shown in corresponding above-mentioned Fig. 1, the wrong data signal of the present embodiment may include：It is above-mentioned not in track institute
Belong to various wrong data signals corresponding to robot, various wrong data signals corresponding to the above-mentioned affiliated robot of tool abnormality
Deng.
202nd, the wrong data signal is pre-processed, obtains pretreated wrong data signal.
For example, in the present embodiment, wrong data signal can be pre-processed using PCA.If what is obtained is certain
The input/output signal of each robot in period, the input/output signal in certain period of time can now be located in advance
Reason, obtains pretreated wrong data signal.
203rd, the pretreated error number is handled using the ELM algorithm quantitative analyses mathematical modeling that pre-establishes it is believed that
Number, classification processing is carried out to the wrong data signal, and determine the processing of the wrong data signal after corresponding classification processing
Information.
That is, after carrying out classification processing to wrong data signal in the present embodiment, category robot pair can be used
Under each data category answered the treatment measures of wrong data signal to the wrong data signal of data category belonging to respective at
Reason, with solving error data, ensures the use of robot.
As an example it is assumed that the artificial A classes robot of current machine, and wrong data signal is 01 class, then it is wrong from institute
The treatment measures of the wrong data signal of 01 class of A classes robot are selected in the processing information of data-signal, now, using selection
The treatment measures of wrong data signal handle the sorted wrong data signal for belonging to the category, and then faster realize
Solving error.
For example, above-mentioned step 202 may include following sub-steps：
2021st, processing is compressed to wrong data signal using PCA (principle component analysis), obtains the data that dimension reduces
Matrix.
The step 2021 may include following sub-steps：
Such as：A1, the matrix that the wrong data signal is converted into m × n, place is standardized to the matrix of the m × n
Reason, the matrix after being standardized, as shown in Figure 2.
Standardization in the step can be regarded as：The average of data in each dimension is calculated first (using all
Data calculate), the average is subtracted in each dimension afterwards.
A2, singular value decomposition is carried out to the matrix after standardization, obtain load vector and score vector.
A3, load vector and the pivot number of score vector are determined by contribution rate of accumulative total method.
A4, space load matrix asked for according to pivot number, and obtain the data matrix of dimension reduction.
For example, the dimension for the data matrix that usual dimension reduces is more than 4 dimensions in the present embodiment.
Correspondingly, above-mentioned step 203 can be specially：Using the ELM algorithm quantitative analysis mathematical modelings pair pre-established
The data matrix that dimension reduces carries out classification processing.
It should be noted that PCA methods are one of important means of Data Dimensionality Reduction, method is fairly simple, is exactly by sample
Data seek the covariance matrix of a dimension, then solve the characteristic value of this covariance matrix and corresponding characteristic vector, will
These characteristic vectors arrange from big to small according to corresponding characteristic value, form new matrix, are referred to as eigenvectors matrix, also may be used
To be referred to as projection matrix, then sample data is changed with changing projection matrix.K dimension datas, realize the drop to data before taking
Dimension.
Assuming that sample data ties up (forming a r dimensional vector) by r, n sample is shared.R*n matrix As are formed, matrix is each
Row are a samples, and row is each different characteristic dimension.Solve covariance matrix S=AAT, S is r*r square formation, and is
Symmetrical matrix, solves S characteristic vector and corresponding characteristic value, characteristic vector is arranged in order according to characteristic value size, by this
A little characteristic vectors (column vector) form a matrix, are referred to as transition matrix or projection matrix.If before only taking K row feature to
The matrix P of composition is measured, then former state notebook data is transformed into new coordinate space A '=P with projection matrixTA, it is possible to achieve data
Dimensionality reduction.
Contribution rate of accumulative total method：The elemental characteristic value sum and the ratio of all elements characteristic value sum extracted, ratio are got over
Illustrate that its influence is bigger greatly, principal component criterion herein is 80%-95%.
The initial data that industry spot collects is pre-processed first in the present embodiment, to remove live noise and subtract
Few mode input variable number, such processing mode subsequently to use ELM algorithm quantitative analysis mathematical modelings speed faster and
The precision of prediction of model is higher, can quickly and accurately realize that the mistake of industrial robot is autonomous and recover and processing, and former
The methods of expert system, is compared, and improves measuring accuracy, improves operating efficiency.
In a kind of optional implementation, before step 201, ELM algorithm quantitative analysis mathematical modulos can be pre-established
Type (rate of accuracy reached of the classification results of the model arrives default value).It is following by ELM algorithm quantitative analyses mathematical modeling referred to as
ELM models.
That is, it is necessary to correct the classification knot for the ELM models established using random network parameter before step 201
Fruit, can the wrong data according to caused by actual industrial scene be trained amendment, while commissioning staff can adjust the net of ELM models
Network parameter, so that the rate of accuracy reached of the classification results of the ELM models finally used is to default value.
That is, the ELM models that the above-mentioned training sample set pair using industry spot is established using random network parameter are repaiied
Just.
Because each industry spot is different, the data of training sample set can not possibly gather in advance, and for this, ELM models are
Corrected as training sample concentrates the increase of data.For example, when robot obtains first wrong data, ELM models are just
Set up for this data, when obtaining second, using the first two wrong data, ELM models are corrected, with such
Push away, therefore model is being updated, after n-th data are obtained, ELM models reach to the classification results of wrong data
Entirely accurate (i.e. to all wrong data of industrial environment residing for robot can Accurate classification and make respective handling), this
When, makeover process terminates.
That is, the rate of accuracy reached for establishing classification results is established online to the process of 100% ELM models, and
Need commissioning staff to participate in amendment, do not establish and correct offline.
During specific implementation, in the sampled data obtained in actual production process, except weeding out some fault numbers
According to also tending to some noise datas for being brought due to industrial environment and measuring method be present, in order to remove these noises to building
The influence of mould, primary data can be handled using PCA in the present embodiment, the data after processing are subjected to ELM
It the foundation of model, not remove only influence of the noise to established model so, also reduce the dimension of initial data, simplify mould
Type structure, improve precision.PCA-ELM modeling main process be：Initial data is handled with principle component analysis first, its
The secondary data by after processing carry out ELM modelings, so as to complete the combination of principle component analysis and ELM algorithms.
In the present embodiment, because the study makeover process of ELM models once completes, fast without iteration, pace of learning, therefore
PCA-ELM modeling methods equally possess this advantage：It is fast without iteration, modeling speed；(i.e. ELM is calculated the soft-sensing model established
Quantitative analysis mathematical modeling) Generalization Capability height.Here PCA-ELM modeling methods refer to and first data are carried out at PCA
Reason, then carry out the process of ELM model learning correcting process.
The above method can quickly and accurately realize the mistake of industrial robot from Main classification and processing, and former special
Family system the methods of compare, analytical cycle it is short and early stage debugging the stage it is simple to operate, using microcomputer modelling and calculate,
Improve measuring accuracy, improve operating efficiency.
In a kind of possible implementation, before Fig. 1 step 201, the above method also includes following steps：
S1, activation primitive and hidden node number are set, establish ELM algorithm quantitative analysis mathematical modelings, as shown in Figure 3 B
Network configuration figure.
In addition, as the analogous diagram used of the model is shown respectively in Fig. 5 and Fig. 6.
S2, the model of foundation applied in industry spot, now the classification results according to model, using in industry spot
The wrong data of appearance is modified to the ELM models of foundation, until the rate of accuracy reached of the classification results of revised ELM models
To 100%.
Hidden node number, or adjustment activation primitive are can adjust in the makeover process of ELM models.
Correct the wrong data composition training sample set for the ELM models having built up.
Wherein, the training sample concentrates training sample not repeat.As shown in figure 4, Fig. 4 is illustrated that training sample is concentrated
Element number and test set accuracy corresponding relation, wherein, test set be above-mentioned collecting robot people species, machine human factor error
Process signal, and the test set of the treatment measures of all error handle signals of robot.
For example, in the present embodiment for containing N number of hidden node, N number of different sample (xi,ti), and excitation function
For g (x) ELM algorithm quantitative analyses mathematical modeling (following abbreviation ELM models) as shown in formula one：
In formula one, xi=[xi1,xi2…xin]T∈Rn,ti=[ti1,ti2…tin]T∈Rm,wiTo connect i-th of hidden layer
The input weights of node and output node, βiFor i-th of hidden node of connection and the output weights of output node；biIt is hidden for i-th
The deviation of node layer；wi·xiRepresent wiAnd xiInner product；Excitation function g (x) is sigmoid, sin or RBF.
Above-mentioned formula one can be expressed in matrix as:
H β=T formula two
In formula two：
Wherein, H is the hidden layer output matrix of ELM models.
It should be noted that in the modeling process of ELM models, network parameter (the as above He of formula one of ELM models is established
Each parameter in formula two) often need not all it be adjusted by iteration, precondition is to meet that excitation function g (x) is unlimited
Can be micro-.In this course, connection weight w is inputtediWith hidden node offset threshold biIt can randomly select and in whole training process
In do not change, and hidden layer output weights βiIt is to solve to obtain by least square method to be, what least square method solved
System of linear equations is as shown in formula three：
By the correlation theory of least square method, the least square solution of this linear equation of formula three is：
In formula four：H+For the hidden layer output matrix H of ELM models generalized inverse.
Because hidden layer input weight and threshold value randomly select in ELM networks, do not change, without iteration, therefore ELM algorithms are built
The mould time is short, efficiency high；The algorithm of establishing of ELM models is a kind of nonlinear regression modeling method, and Generalization Capability is good.
The main process of PCA-ELM modeling methods is：Initial data is handled with principle component analysis first, secondly will
Data after processing carry out the foundation amendment of ELM models, so as to complete the combination of principle component analysis and ELM algorithms.With reference to
Fig. 3 A introduce specific modeling procedure：
(1) field data (the i.e. input of situ machine people of the principle component analysis to being collected in various industrial process is utilized
Output signal) pre-processed, play a part of dimensionality reduction, denoising to initial data；
(a) field data that industrial process collects is standardized, laid the groundwork for PCA processing below.
Change processing method centered on the mode that accepted standardization is handled in this step.
(b) according to the data matrix X after standardization, its covariance matrix R of input variable matrix is calculated；
(c) covariance matrix of the input variable matrix calculated according to step (b), obtain characteristic value and with characteristic value phase
The characteristic vector answered, and principal component number is determined by contribution rate of accumulative total method and cross-validation method, pay attention to contribution rate of accumulative total method
Standard should be 80%~95%；
(d) according to the principal component number determined in step (c), principal component load matrix P is calculated, and then try to achieve score square
Battle array T=XP, score matrix is that the learning sample that ELM network models are used is established in below step (2).
Above-mentioned steps (1) are to obtain the process of training sample set.
(2) ELM network models are established, as shown in Figure 3 B.
A) apply the data after PCA to establish ELM networks, weights and threshold value random assignment are inputted for hidden layer.
B) hidden layer output matrix H is calculated, and calculates hidden layer output weightsThe output for obtaining ELM networks is estimated
Evaluation, complete PCA-ELM modelings.
The data that step (1) has collected to industry spot have carried out pivot analysis, and have obtained through at pivot analysis
Obtain reconstructing data sample after reason, ELM algorithm modelings are carried out to these reconstructed sample data, i.e., using these reconstructed samples as instruction
Practice data to be trained it with ELM algorithms.
The above method is pre-processed using pca method to the initial data that industry spot collects, existing to remove
Field noise simultaneously reduces mode input variable number, and such processing mode causes modeling speed faster and the precision of prediction of model is got over
It is high.
The modeling of ELM algorithms can be summarized as following three steps：
For the set of data samples with being collected from industrial process, activation primitive is that hidden node number is then to establish
ELM models are ELM algorithm quantitative analysis mathematical modelings：
B1 the hidden node parameter in ELM modeling process) is generated at random：Hidden layer inputs weights and hidden layer threshold value, wherein；
B2 hidden layer output matrix H (the as above H in formula two)) is calculated；
B3) optimal outer power (the i.e. β of above-mentioned formula one of calculating networki)。
That is, in the sampled data obtained in actual production process, except weeding out some fault data, also tend to
In the presence of the noise data that some are brought due to industrial environment and measuring method, in order to remove influence of these noises to modeling,
This section is handled primary data using PCA, and the principal component containing most primary data information (pdi)s is come into generation
It is modeled for initial data, that is, the model established between principal component scores matrix and output matrix, does not remove only so and make an uproar
Influence of the sound to established model, also reduce the dimension, simplified model structure, raising precision of initial data.
Further, since ELM modeling method learning processes are once completed, are fast without iteration, pace of learning, therefore PCA-ELM
Modeling method equally possesses this advantage：It is fast without iteration, modeling speed；The soft-sensing model Generalization Capability established is high.
In addition, the present invention also provides a kind of wrong autonomous recovery device for industrial robot, as shown in fig. 7, the dress
Put including：
Acquiring unit 71 is used for the processing letter for obtaining wrong data signal corresponding to robot and all wrong data signals
Breath；
Pretreatment unit 72 be used for the wrong data signal is pre-processed, obtain pretreated error number it is believed that
Number；
Classification processing unit 73 is used to handle the pretreatment using the ELM algorithm quantitative analyses mathematical modeling pre-established
Wrong data signal afterwards, classification processing is carried out to the wrong data signal, and determine the processing of corresponding wrong data signal
Information.
In a kind of optional implementation, the pretreatment unit 72 is specifically used for using PCA to wrong data signal
Processing is compressed, obtains the data matrix that dimension reduces；
Correspondingly, the classification processing unit 73 is specifically used for using the ELM algorithm quantitative analysis mathematical modulos pre-established
Type carries out classification processing to the data matrix that dimension reduces.
Scene is realized another, the autonomous recovery device of the mistake for industrial robot of the invention may also include：Deposit
The memory of instruction is stored up, and loads and perform the processor instructed in memory, the processor is specifically used for
Obtain the processing information of wrong data signal and all wrong data signals corresponding to robot；
The wrong data signal is pre-processed, obtains pretreated wrong data signal；
The pretreated wrong data signal is handled using the ELM algorithm quantitative analyses mathematical modeling pre-established,
Classification processing is carried out to the wrong data signal, and determines the processing information of corresponding wrong data signal.
It should be noted that the embodiment of the present invention is using PCA-ELA modeling and simulatings and analyzes sample data, of the invention real
Under the industry spot real background for applying example institute foundation, 270 kinds of different type of errors are shared.Training sample set be this 270
Selected in group data, optimal set is reached for train classification models, test set is then for examining resulting mould
The accuracy that type is classified for data, test set include 270 groups of all data, and this data also implied that in training set is same
When be also contained in test set.
In order to simulate the scene in robot debugging stage in real industry spot, data are trained by addition one by one
Collection, because robot can only identify a mistake when making a mistake every time, the sample in new addition training set is to send out
It is selected at random and is not repeated with the sample in known training set in the sample of raw classification error, this is due to that repetition training is same
Kind mistake is insignificant.Only when model reaches 100% to the classification accuracy of test set, training can just terminate, and table 1 is
The model simulated with MATLAB softwares establishes process, and (ELM is single hidden layer in this experiment, the number of hidden nodes 1000, kernel function
For " sin ") once model is established, for working environment of the same race, it is not necessary to repetition training.
Table 1：
As can be seen that can be only with seldom training sample very short using PCA-ELM modeling methods from experimental result
Classification is set to reach 100% in time accurate.
Because the necessary entirely accurate of the action of the robot in real world is reliable, so it is not that accuracy, which is less than 100%,
It is received, therefore influence of the hidden layer node number for model is demonstrated again, it is as shown in table 2 below, have and be a little noted that
Only when node in hidden layer is more than or equal to 50, it is accurate that the system could reach 100% in 270 training samples.
Table 2：
From upper Tables 1 and 2 and Fig. 5, Fig. 6 it could be assumed that：In this experiment, average workout times first drastically subtract
Small, minimum 0.47s, node in hidden layer now is 300, and after this, average workout times increase with node in hidden layer
Increase greatly, but it is relatively slower；Training concentration training number increases and reduced with node in hidden layer needed for average, just starts to subtract
Small speed, slow down after node in hidden layer reaches 200 or so.
Finally it should be noted that：Above-described embodiments are merely to illustrate the technical scheme, rather than to it
Limitation；Although the present invention is described in detail with reference to the foregoing embodiments, it will be understood by those within the art that：
It can still modify to the technical scheme described in previous embodiment, or which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or substitutions, the essence of appropriate technical solution is departed from various embodiments of the present invention technical side
The scope of case.
Claims (9)
- A kind of 1. wrong autonomous restoration methods for industrial robot, it is characterised in that including：Obtain the processing information of wrong data signal and all wrong data signals corresponding to robot；The wrong data signal is pre-processed, obtains pretreated wrong data signal；The pretreated wrong data signal is handled using ELM algorithm quantitative analyses mathematical modeling, to the wrong data Signal carries out classification processing, and determines the processing information of the wrong data signal after corresponding classification processing；Wherein, using ELM algorithm quantitative analysis mathematical modelings classification results rate of accuracy reached to default value.
- 2. according to the method for claim 1, it is characterised in that it is described that the wrong data signal is pre-processed, obtain The step of obtaining pretreated wrong data signal, including：Processing is compressed to wrong data signal using principle component analysis PCA, obtains the data matrix that dimension reduces；Correspondingly, the pretreated error number is handled using the ELM algorithm quantitative analyses mathematical modeling that pre-establishes it is believed that Number the step of, including：Classification processing is carried out to the data matrix that dimension reduces using ELM algorithm quantitative analysis mathematical modelings.
- 3. according to the method for claim 2, it is characterised in that wrong data signal is carried out using principle component analysis PCA Compression handle, obtain dimension reduce data matrix the step of, including：The wrong data signal is converted into m × n matrix, the matrix of the m × n is standardized, obtains standard Matrix after change；M, n is the natural number more than 1；Singular value decomposition is carried out to the matrix after standardization, obtains load vector and score vector；Load vector and the pivot number of score vector are determined by contribution rate of accumulative total method；Space load matrix is asked for according to pivot number, and obtains the data matrix of dimension reduction.
- 4. according to the method for claim 3, it is characterised in that described using the ELM algorithm quantitative analysis numbers pre-established Before the step of learning pretreated wrong data signal described in model treatment, methods described also includes：Activation primitive and hidden node number are set, establish ELM algorithm quantitative analysis mathematical modelings；By ELM algorithm quantitative analysis the application of mathematical model to industry spot, using the wrong data in industry spot to foundation ELM algorithm quantitative analysis mathematical modelings are modified, until the classification results of revised ELM algorithms quantitative analysis mathematical modeling Rate of accuracy reached to the default value.
- 5. according to the method for claim 4, it is characterised in thatFor containing N number of hidden node, N number of different sample (xi,ti), and excitation function is g (x) ELM algorithm quantitative analysis numbers Model is learned as shown in formula one：In formula one, xi=[xi1,xi2…xin]T∈Rn,ti=[ti1,ti2…tin]T∈Rm,wiFor i-th hidden node of connection and The input weights of output node, βiFor i-th of hidden node of connection and the output weights of output node；biFor i-th of hidden node Deviation；wi·xiRepresent wiAnd xiInner product；Excitation function g (x) is sigmoid, sine or RBF.
- 6. method according to any one of claims 1 to 5, it is characterised in that obtain wrong data signal corresponding to robot And the processing information of all wrong data signals, including：Obtain robot and be applicable the species of all robots in industry spot, and be likely to occur corresponding to all robot species Wrong data signal treatment measures；AndThe input/output signal of each robot used in industry spot is obtained, the input/output signal includes robot Wrong data signal；Correspondingly, the wrong data signal is pre-processed, the step of obtaining pretreated wrong data signal, bag Include：The input/output signal is pre-processed, obtains pretreated wrong data signal.
- A kind of 7. wrong autonomous recovery device for robot, it is characterised in that including：Acquiring unit, for obtaining the processing information of wrong data signal corresponding to robot and all wrong data signals；Pretreatment unit, for being pre-processed to the wrong data signal, obtain pretreated wrong data signal；Classification processing unit, for described pretreated using the ELM algorithm quantitative analyses mathematical modeling processing pre-established Wrong data signal, classification processing is carried out to the wrong data signal, and determines the processing information of corresponding wrong data signal, Wherein, using ELM algorithm quantitative analysis mathematical modelings classification results rate of accuracy reached to default value.
- 8. device according to claim 7, it is characterised in that the pretreatment unit, be specifically used for, using pivot analysis Method PCA is compressed processing to wrong data signal, obtains the data matrix that dimension reduces；Correspondingly, the classification processing unit, specifically for using the ELM algorithm quantitative analysis mathematical modelings pre-established to dimension The data matrix that number reduces carries out classification processing.
- A kind of 9. wrong autonomous recovery device for robot, it is characterised in that including：The memory of store instruction, and add The processor instructed in memory is carried and performs, the processor is specifically used forObtain the processing information of wrong data signal and all wrong data signals corresponding to robot；The wrong data signal is pre-processed, obtains pretreated wrong data signal；The pretreated wrong data signal is handled using the ELM algorithm quantitative analyses mathematical modeling pre-established, to institute State wrong data signal and carry out classification processing, and determine the processing information of corresponding wrong data signal, wherein, using ELM algorithms The rate of accuracy reached of the classification results of quantitative analysis mathematical modeling is to default value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710922428.5A CN107696034A (en) | 2017-09-30 | 2017-09-30 | A kind of wrong autonomous restoration methods for industrial robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710922428.5A CN107696034A (en) | 2017-09-30 | 2017-09-30 | A kind of wrong autonomous restoration methods for industrial robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107696034A true CN107696034A (en) | 2018-02-16 |
Family
ID=61184443
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710922428.5A Pending CN107696034A (en) | 2017-09-30 | 2017-09-30 | A kind of wrong autonomous restoration methods for industrial robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107696034A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112109080A (en) * | 2019-06-19 | 2020-12-22 | 发那科株式会社 | Adjustment auxiliary device |
| CN113393211A (en) * | 2021-06-22 | 2021-09-14 | 柳州市太启机电工程有限公司 | Method and system for intelligently improving automatic production efficiency |
| CN114340855A (en) * | 2019-09-05 | 2022-04-12 | 三菱电机株式会社 | Robot action planning system, robot work verification system, and robot action planning method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5715388A (en) * | 1992-09-25 | 1998-02-03 | Kohki Co., Ltd. | Computer system for controlling an industrial robot subsystem which can monitor and detect abnormalities therein |
| CN103604591A (en) * | 2013-11-14 | 2014-02-26 | 沈阳工业大学 | Fault detection method of wheeled mobile robot |
| CN103745093A (en) * | 2013-12-25 | 2014-04-23 | 中国矿业大学 | Principal component analysis-extreme learning machine (PCA-ELM) based coal mine water inrush prediction method |
| CN104635718A (en) * | 2013-11-12 | 2015-05-20 | 沈阳新松机器人自动化股份有限公司 | Robot fault repairing system and method |
| CN105171748A (en) * | 2015-10-21 | 2015-12-23 | 鞍山松意机器人制造有限公司 | Remote state monitoring method and system for robots and robot production line equipment |
| WO2016185589A1 (en) * | 2015-05-20 | 2016-11-24 | 日産自動車株式会社 | Failure diagnostic device and failure diagnostic method |
| CN107169205A (en) * | 2017-05-17 | 2017-09-15 | 东北大学 | A kind of classification model construction method of iron ore |
-
2017
- 2017-09-30 CN CN201710922428.5A patent/CN107696034A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5715388A (en) * | 1992-09-25 | 1998-02-03 | Kohki Co., Ltd. | Computer system for controlling an industrial robot subsystem which can monitor and detect abnormalities therein |
| CN104635718A (en) * | 2013-11-12 | 2015-05-20 | 沈阳新松机器人自动化股份有限公司 | Robot fault repairing system and method |
| CN103604591A (en) * | 2013-11-14 | 2014-02-26 | 沈阳工业大学 | Fault detection method of wheeled mobile robot |
| CN103745093A (en) * | 2013-12-25 | 2014-04-23 | 中国矿业大学 | Principal component analysis-extreme learning machine (PCA-ELM) based coal mine water inrush prediction method |
| WO2016185589A1 (en) * | 2015-05-20 | 2016-11-24 | 日産自動車株式会社 | Failure diagnostic device and failure diagnostic method |
| CN105171748A (en) * | 2015-10-21 | 2015-12-23 | 鞍山松意机器人制造有限公司 | Remote state monitoring method and system for robots and robot production line equipment |
| CN107169205A (en) * | 2017-05-17 | 2017-09-15 | 东北大学 | A kind of classification model construction method of iron ore |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112109080A (en) * | 2019-06-19 | 2020-12-22 | 发那科株式会社 | Adjustment auxiliary device |
| CN114340855A (en) * | 2019-09-05 | 2022-04-12 | 三菱电机株式会社 | Robot action planning system, robot work verification system, and robot action planning method |
| CN114340855B (en) * | 2019-09-05 | 2024-05-07 | 三菱电机株式会社 | Robot action planning system, robot work verification system, and robot action planning method |
| CN113393211A (en) * | 2021-06-22 | 2021-09-14 | 柳州市太启机电工程有限公司 | Method and system for intelligently improving automatic production efficiency |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11338435B2 (en) | Gripping system with machine learning | |
| CN108393888A (en) | control device, robot and robot system | |
| CN108393889A (en) | control device, robot and robot system | |
| CN108393890A (en) | control device, robot and robot system | |
| CN108393891A (en) | control device, robot and robot system | |
| JP6671694B1 (en) | Machine learning device, machine learning system, data processing system, and machine learning method | |
| Ulbrich et al. | The OpenGRASP benchmarking suite: An environment for the comparative analysis of grasping and dexterous manipulation | |
| CN107696034A (en) | A kind of wrong autonomous restoration methods for industrial robot | |
| Ficuciello et al. | Synergy-based policy improvement with path integrals for anthropomorphic hands | |
| CN108710285A (en) | Industrial robot model emulation control method and device | |
| Rahman | Cognitive cyber-physical system (C-CPS) for human-robot collaborative manufacturing | |
| Ehrenmann et al. | Teaching service robots complex tasks: Programming by demonstration for workshop and household environments | |
| CN107598918A (en) | Surface grinding process automatic programming method and device based on milling robot | |
| Yu et al. | Roboassembly: Learning generalizable furniture assembly policy in a novel multi-robot contact-rich simulation environment | |
| Groth et al. | Goal-conditioned end-to-end visuomotor control for versatile skill primitives | |
| Dajles et al. | Teleoperation of a humanoid robot using an optical motion capture system | |
| CN113305845B (en) | Multi-mechanical arm cooperation method | |
| CN114179104A (en) | Picking robot control method and system based on visual identification | |
| Takayanagi et al. | Hierarchical task planning from object goal state for human-assist robot | |
| Kumra et al. | Learning robotic manipulation tasks via task progress based Gaussian reward and loss adjusted exploration | |
| Castro et al. | AdaptPack studio: automatic offline robot programming framework for factory environments | |
| Chongistitvatana et al. | Learning a visual task by genetic programming | |
| CN114594757A (en) | Visual path planning method for cooperative robot | |
| Jin et al. | Shared Control With Efficient Subgoal Identification and Adjustment for Human–Robot Collaborative Tasks | |
| Yuan | Interactive assembly planning in virtual environments |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180216 |
|
| RJ01 | Rejection of invention patent application after publication |