CN103617329A - Assembly time evaluating method based on artificial neural network and virtual assembly - Google Patents
Assembly time evaluating method based on artificial neural network and virtual assembly Download PDFInfo
- Publication number
- CN103617329A CN103617329A CN201310671532.3A CN201310671532A CN103617329A CN 103617329 A CN103617329 A CN 103617329A CN 201310671532 A CN201310671532 A CN 201310671532A CN 103617329 A CN103617329 A CN 103617329A
- Authority
- CN
- China
- Prior art keywords
- assembling
- assembly
- matrix
- network
- virtual
- 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
Images
Abstract
The invention discloses an assembly time evaluating method based on an artificial neural network and a virtual assembly. The method comprises the step of establishing a virtual working platform, planning a DELMIA assembly sequence and a DELMIA path and generating an assembly path, the step of generating the assembly sequence through a constraint matrix method and establishing the constraint relation among components, the step of preassembling the product components through a DELMIA assembly module, establishing a priority dismounting matrix, extracting the assembly sequence meeting the condition from the priority dismounting matrix and planning the assembly sequence through the strong data collecting capacity of the matrix, the step of establishing a virtual assembly work environment and the virtual assembly action process, the step of measuring the assembly path and calculating the assembly time, the step of establishing an assembly time evaluating model, the step of establishing the model and standardizing a sample, and the step of evaluating the sample. The assembly time evaluating method solves the problems that the product assembly period is long, cost is high, and the effect is poor.
Description
Technical field
The invention belongs to Computer Applied Technology field, relate in particular to a kind of installation time evaluation method based on artificial neural network and virtual assembling.
Background technology
Along with the development of computer-aided design (CAD) and manufacturing technology becomes sharp weapon that can change assembly industry inefficiency in product design, manufacturing process.Assembling manufacturing industry in the past, the assembling experience that can only accumulate over a long period by assembler completes produces assembling.Traditional optimization means is all only based upon again a stage stagnation that model is imitated, such assembly technology design optimization method, it is not merely long problem experimental period, take the resources of production too much, the uncertainty of simulation process environment changes all can reduce the cogency of this model experiment method.Not enough condition in these technology, all the development along with computing machine has occurred to change fast.Under computer virtual environment, assemble Sequence Planning, cost is low, take few, the extraneous artificial interference factor of resource does not almost have.This will be by the evolutionary path and the future that are assembly sequence optimization.
At present, a lot of Aeronautics and Astronautics enterprise all will assemble and plan that this technology produces utilization in the middle of putting into reality under virtual condition in the world.According to reliable data, show: this technology has produced huge benefit.Lower to a great extent assembly period, reduce rework rate, reduced generally whole production cost.Because it has so big impact to whole production assembling process, cause a lot of Automobile Enterprises, munitions factory all competitively to put under virtual condition and assemble and plan the use of this technology and research and develop.
At home, the research of virtual assembling starting is more late, and because the required hardware price of Virtual Assembling Technology is expensive, most of research institutions are carrying out theory study and research.Each colleges and universities of recent year are more active to the research of Virtual Assembling Technology, the grinding of some schools that Zhejiang University and the Central China University of Science and Technology be representative of wherein take is more active, and interim theoretical result and preliminary application have been obtained, particularly the Tan Rong of Zhejiang University teaches and student, in virtual assembling field, set up a whole set of and had the theory of practical value, the existing fruit that is ground into by domestic scholars is summarized as follows.
In order to solve, from CAD system, forward part model the loss problem of product information virtual assembly system to, the Liu Zhenyu of Zhejiang University, Tan Jianrong have proposed the product attribute in virtual assembling and behavioural information to be divided into gas producing formation, characteristic layer, geometry topological layer and display layer.By the data-mapping in product level information model, realize between the level of product information associatedly, guaranteed when product design information is transformed to virtual assembly system by CAD system can not lose simultaneously.
Be similarly product under solution reality environment and lacked assembly features information and realize the dynamic assembling under virtual environment, strong people in building of Zhejiang University waits the extractive technique that proposes Product Assembly characteristic information under reality environment.This technology is by the extraction that modelling feature is inherited, the method such as rule-based Boundary Match identification and heuristic definition realizes Product Assembly characteristic information under virtual environment.
It seems now, the research of this technology is some defects more or less of old existence still, yet recent years, leader in several industries is the Bedford of Stanford university professor for example, the Swift professor of Hull university, has proposed some and has had a novelty viewpoint for this technology.But Lucas is DFA, these three kinds of classical methods of Boothroyd-Dewhurst DFA and Hitachi AEM DFA are included in this technology, every kind of method has the direction that stresses separately, although by research and the exploration of all angles, also some breakthroughs have been obtained, but the assembling capacity that is difficult to the product under the different assembly environments of reflection is the problem of their facings, if select assembling operating mode ideally, unavoidably can be subject to the impact of the foeign elements such as environment, that conventionally can cause Assembly of the parts operating mode can not judgement property, thereby be difficult to the equipment effect in the various situations of reflection, therefore also mainly as a kind of assay instrument.Although this method is at the initial stage of product design, just can bring into play certain effect, but because analysis contrast and the design for assembly of different assemble mechanisms is the assembly manipulations for initial part, after certain phase process, can bring supervision afterwards " this bad impact, to such an extent as to cause a part of parts to do over again.
Analyze in sum, home products design for assembly system also has the following disadvantages now: 1) can not integrate the many factors in assembling process, generally can not be from the assembling of integral body taking into account system, cannot be by each part, parts and component units are effectively connected it in Product Assembly process with assembly technology, enforcement resource.2), seldom relevant for the innovative approach of evaluation system result, cause now also not having the application system of a set of maturation.3) cannot carry out on-line optimization to evaluation system, can only to it, analyze from result.4) CAD system can't realize perfect integration with DFA at present.Can't each step in assembling process be evaluated and be analyzed at present, thereby cannot to assembling problem, evaluate from integral body.Although current stage people pay close attention to exploration that can assembly evaluation system and grind, and still overcome a lot of critical problems because need have always, also immature such as this assessment technique development, technical merit is comparatively backward, needs the factor of consideration too loaded down with trivial details.
This technology development at home has at present been subject to obstruction, why causing this situation is due to nonstandard commercial system, to domestic, carry out technical information blockade abroad, the at present domestic design software towards assembling lacks very much, also there is to very large gap with exploitation with comparing abroad in the application of software, temporarily do not have sufficient theory to come support assemblies principle of design and design stability, and the personnel that are engaged in that are correlated with are also very rare, therefore, assembling capacity assessment technique uses all in the urgent need to research in theoretical developments and engineering practice.
Tradition assembly technology design need to be set up a large amount of physical models could determine design to carry out the test of mould dress, makes that the Product Assembly cycle is long, cost is high, and effect is also poor.
Summary of the invention
The object of the embodiment of the present invention is to provide a kind of installation time evaluation method based on artificial neural network and virtual assembling, being intended to solve the design of traditional assembly technology need to set up a large amount of physical models and could determine design to carry out the test of mould dress, cannot carry out on-line optimization to evaluation system, can only to it, analyze from result, CAD system can't realize perfect integration with DFA at present, make that the Product Assembly cycle is long, cost is high, and effect poor problem also.
The embodiment of the present invention is achieved in that a kind of installation time evaluation method based on artificial neural network and virtual assembling, should the installation time evaluation method based on artificial neural network and virtual assembling comprise the following steps:
Step 1, sets up virtual work platform, planning DELMIA assembling sequence, and planning DELMIA path, generates assembly path;
Step 2, uses constraint matrix method to generate assembling sequence, sets up the restriction relation between part; Then use DELMIA load module, product parts is carried out to pre-assembled, a vertical preferential dismounting matrix, then therefrom extracts the assembling sequence satisfying condition, and utilizes so the powerful data collection capability of matrix to plan assembling sequence;
Step 3, the assembly path planning of the emulation examination dismounting based on CATIA/DELMIA: set up virtual assembly work environment, create virtual assembling course of action;
Step 4, the measurement of assembly path and the calculating of installation time;
Step 5, the foundation of installation time evaluation model: the foundation of model, the standardization of sample;
Step 6, evaluates sample.
Further, in step 1, use CATIA to carry out modeling to assembly and virtual work platform, concrete grammar is: after product component design finishes, the parts that each design department designs are put into a virtual environment according to coordinate setting unification, after integrating, all parts can obtain the product that an assembling completes, then being handed over to Assembling Production department is planned whole product establishment assembling sequence and carries out virtual assembling according to knowhow in the past by millwright performer, finally formulate a whole set of assembling programme, scheme comprises assembling action and the required built-up member of whole product in the different time periods, then assembler carries out practical set production under the guidance of this scheme, if running in process of production assembling planning problem is committed to problem technology establishment department again and carries out secondary modification.
Further, DELMIA assembling Sequence Planning in step 1, be divided into two part: DPE and DPM, DPE is process planning platform, DPM is emulation platform, while carrying out assembly process planning design in DPE module, for the product resource information table generating under DPE module must import to EBOM in the template of technological design in order.
Further, the generation method of resource information table is: will manufacture resource and import in DPE environment, then will be according to the practical set technique in reality is produced, rely on knowhow under DPE, to create out a detailed assembly technology information table, in the process of just carrying out, the structure relevant to technique added in the middle of technological process simultaneously, when finishing, PPR Hub database is preserved and deposited in to the assembly technology of having planned, DPE is the pretreatment stage of DPM, and DPM has the effect of checking to DPE, belong to Qualify Phase.
Further, the concrete grammar of step 2 is:
First use constraint matrix method to generate assembling sequence, set up the restriction relation between part; Then use DELMIA load module, product parts is carried out to pre-assembled, a vertical preferential dismounting matrix, then therefrom extracts the assembling sequence satisfying condition, and utilizes the powerful data collection capability of matrix to plan assembling sequence;
The method that generates the constraint matrix employing of assembly is: by { x,-x, y,-y, z,-z} direction is as assembly direction, by setting up the interference matrix of each direction, show the relation constrained each other between its parts, the generation of interference matrix is the first step of Sequence Planning, parts meeting geometric constraint relation whether in assembling process, the dismoutability of judgement part on six coordinate directions, the program of interference checking, by each part respectively to { x,-x, y,-y, z, the dismounting of-z} axle, generate along { x,-x, y,-y, z, the free interference matrix of-z} direction, after interference checking, generate { x,-x, y,-y, z, the corresponding free interference matrix of-z}6 direction.
Further, in step 2, based on examination, tearing the integrated concrete steps of method interference matrix open is:
Six free interference matrix FIM ' [N] [N] are carried out to occasional combination and become new matrix F IM ' [6N] [N], by the value of FIM ' [N] [N], the element of FIM ' [6N] [N] is carried out to initialization;
Calculate the row vector mould in FIM ' | FIi'|;
If | FIi'|=0, will | part that FIi'|=0 is expert at carries out new definition identifier p
k, k=i%N(% represents except rear remainder) and count the detachable array C of current round, and make integrated interference matrix element II
kj=FI
kj, j=(0,1 ..., N-1), II
k, N=i/N (/ represent remove then round) result is 0 expression+directions X, 1 expression-directions X, then after transforming by ± 1 expression ± X direction, the rest may be inferred can obtain the preferential disassembly direction of part in integrated interference matrix; Otherwise turn d;
If i > 6N-1, upgrades element p in C
kthe row, column value of corresponding FIM ' [6N] [N] matrix, makes: FI '
m, j=0 (m=k, k+N, k+2N ...; J=0,1 ..., N 1), FI '
j,n=0 (n=k, k+N, k+2N ...; J=0,1 ... N-1); Otherwise turn b;
If j, saving result; Otherwise turn b;
Entered after integrated, obtain new preferential dismounting matrix, in integrated interference matrix, the preferential disassembly direction in the end data of row represents, by ± 1 expression ± x direction, by ± 2 expression ± y directions, by ± 3 expression ± z directions, according to algorithm principle, preferential disassembly direction is to get rid of and to try to achieve with all directions matrix.
Further, the concrete grammar that assembling sequence forms is: the artificial planning of dismounting sequence, based on free interference matrix, in unloading process, do not need the feasibility of disassembly path again to check:
The first step, first carries out standardization to integrated dismounting matrix P;
Second step, the mould of row vector in calculating P (i=1,2 ..., N), if | C
i|=0, show that part i is removable, if part i can not be subject to the obstruction of other parts when preferential disassembly direction is dismantled, by parts P corresponding to part i
iput in the set of detachable parts, otherwise, the 3rd step entered;
The 3rd step, establishes i=i+1, if i > is N, represents that the mould of all row is all complete by computing in the computing of front-wheel, and this takes turns complete as calculated, enters the 4th step, otherwise second step;
The 4th step, use mouse to get residue parts, drag, if occurred without interference situation, the sequence number of part is added in dismounting sequence, after dilatory detection, parts are proposed, the row vector of parts is all set to 0, represent that parts are disassembled, can not affect again the dismounting of subsequent parts;
The 5th step, observes whether all row vectors have all completed analysis, and all parts are all dismantled, if all parts have been dismantled, the parts label generating is inverted, and the assembly sequences generation of assembly completes.
Further, in step 3, the calculating of installation time, by the generation of assembly path above, in DELMIA software, can measure path, in daily Assembling Production, the translational speed of a part is confirmable, path according to measuring gained, calculates installation time by path and assembling speed:
T: the installation time of single part;
S
n: the assembly path of single part;
V: the mobile assembling speed of single part.
Further, in step 4 first the sample of input be through hidden layer neuron transport function after input layer computing, then propagate into again this output layer, if the output of expecting can not be obtained, will proceed to the second backpropagation so; Concrete grammar is:
The first step, input signal one direction is propagated, and the input layer of network has I node, and hidden layer has J node, and output layer has K node, establishes X
p(X
p1, X
p2..., X
pl), O
jrepresent network input (O
p1, O
p2... O
pk), T
p=(t
p1, t
p2... t
pk) represent respectively actual output and the desired output of network, wherein p=(1,2 ..., P) sample number is P, (O
p1, O
p2..., O
pj) be the output of hidden node, represent i=(1,2 ..., I) individual input layer to j=(1,2 ..., the J) weights of individual hidden node, represent j hidden node to k=(1,2 ..., the K) weights of individual output layer node;
The excitation function of network is selected sigmoid function
for P sample:
The output layer of network is output as:
So far BP network has just completed the approximate mapping of I dimension space vector to K dimension space;
Second step, error signal backpropagation: now can use the error function of Square-type, the sample error of p is:
Have P sample, the error of the overall situation will be:
The 3rd step, the variation of output layer weights, adopts cumulative errors BP algorithm to adjust w
jk, global error E is diminished:
So η represents speed and the 0 < η <1 of study, now the error signal of definition is just expressed as:
Wherein:
So:
Thereby will obtain various neurons, at the formula of the weights adjustment of output layer, be:
The 4th step, the variation of hidden layer weights:
Definition error signal is:
Wherein:
So:
Thereby shadow layer weights regulate formula to be:
。
Further, in step 5, for BP neural network, the algorithm using roughly can be divided into basic gradient descent algorithm and emulation function:
Utilization function in BP network:
The initialization function of network: the initialization function of network is all used in the initial value of giving for threshold values vector b before the network weight matrix W to created and network training, parameter value affects training speed and the convergence of algorithm, in NNToolbox, the call format of netinit function: net=init (net);
Parameter net and variable net are respectively network that newff sets up and the network after initialization, wherein this is initialized as default initialization, be about to [1,1] random number between is composed with the weight matrix W of the linear neuron layer in network and threshold values vector b, with non-linear hour neuron layer, can carry out initialization according to widrow Nyuyen method, in the time of need to calling network and set up function newff, just can automatically complete, special initialization can be by init function to applying;
Trial run function: the network having trained is come to evaluation computing and application by emulation function, the form of calling of this function is: after network has been set up, need to have a trial run function to come Neural Networks Solution, form is: a=sim (net, p);
Training function: complete if BP network has been set up, and input, output data are provided.
Installation time evaluation method based on artificial neural network and virtual assembling provided by the invention, has adopted interference matrix method to determine assembling sequence, has used integrated interference matrix method, makes more intuitively more succinct; It is example that certain model engine is take in the present invention, and with the virtual basis that is assembled into, the interference condition producing during according to part movement uses matrix method to generate dismounting constraint matrix, and then under the condition of virtual assembling, assembling sequence and assembly path is planned; Finally utilize the Neural Network Toolbox in MALAB to evaluate proposed installation time, find out and comparatively reasonably assemble sequence and assembly path, having solved the design of traditional assembly technology need to set up a large amount of physical models and could determine design to carry out the test of mould dress, make that the Product Assembly cycle is long, cost is high, and effect poor problem also.The present invention is based on DELMIA emulation platform, assembly path and installation time are planned, thereby make the optimization of assembly technology more directly perceived, more accurate, more low-cost.In the present invention, also use artificial neural network algorithm to evaluate the installation time having obtained, make the assembly technology optimization drawing more there is cogency.
Accompanying drawing explanation
Fig. 1 is the installation time evaluation method process flow diagram based on artificial neural network and virtual assembling that the embodiment of the present invention provides.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with embodiment, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.
Below in conjunction with drawings and the specific embodiments, application principle of the present invention is further described.
As shown in Figure 1, the installation time evaluation method based on artificial neural network and virtual assembling of the embodiment of the present invention comprises the following steps:
S101: set up virtual work platform, planning DELMIA assembling sequence, planning DELMIA path, generates assembly path;
S102: use constraint matrix method to generate assembling sequence, set up the restriction relation between part; Then use DELMIA load module, product parts is carried out to pre-assembled, a vertical preferential dismounting matrix, then therefrom extracts the assembling sequence satisfying condition, and utilizes so the powerful data collection capability of matrix to plan assembling sequence;
S103: the assembly path planning of the emulation examination dismounting based on CATIA/DELMIA: set up virtual assembly work environment, create virtual assembling course of action;
S104: the measurement of assembly path and the calculating of installation time;
S105: the foundation of installation time evaluation model: the foundation of model, the standardization of sample;
S106: sample is evaluated.
Concrete steps of the present invention are:
Step 1, CATIA modeling:
Use the model of CATIA software creation model engine.CATIA and DELMIA can be at enterprising line operates of platform, and the exchange between data is especially without any problem.To the common use of these two softwares, can make subject study below more convenient.
This model engine has 15 one-level assembly parts, is respectively that flywheel, bent axle, cylinder, piston, piston are buckled, gas admittance valve, air outlet valve, gear, crankcase, lighter, cylinder head, fuel tank, air strainer, flywheel lid, exhaust box.Model after engine assembly.Select one-level assembly to simulate assembling.
Step 2, the assembling Sequence Planning based on constraint matrix:
The present invention, by interfering detection method and matrix method to combine in assembling process under virtual environment, shows assembly restriction with matrix.So significantly improve the accuracy rate that assemble sequence geometric feasibility is judged, have the good method that finds can collect assembly relation:
1) generation of free interference matrix:
First first select the direction of (± x, ± y, ± z) axle, choosing of direction must cover all disassembly directions; If cij=1, represents that part i dismantles the obstruction that is subject to part j in the direction; If cij=0, represents that part i dismantles the obstruction that can not be subject to part j in the direction; Six free matrixes are respectively:
2) integrated interference matrix:
These six free interference matrixes are merged into an integrated interference matrix;
First, from these six free interference matrixes, can find out part 1(cylinder) quality volume is maximum and coordinate the maximum part of connecting relation, so be defined as basic part; By part 1(cylinder) element value of corresponding row is all set to 1, to determine that it tears open in process and be finally disassembled in examination; Then, six free interference matrixes are above synthesized.
Obtain integrated interference matrix IIM, in this integrated interference matrix, the preferential disassembly direction that the data representation of last row obtains after reasoning, + directions X represents with 1,-directions X represents with-1, and+Y-direction represents with+2, and-Y-direction represents with-2, + Z direction represents with+3, and-Z direction represents with-3.
3) the manual planning of dismounting sequence:
Manual planning to the dismounting sequence of integrated interference matrix, then, by sequence upset, draws assembling sequence.Via step, can join sequence in the hope of following several assemblings:
Sequence one: (1,2,9,7,8,3,4,6,11,10,12,14,13,5,15)
Sequence two: (1,2,7,8,3,4,9,6,12,14,11,10,13,5,15)
Sequence three: (1,2,9,7,8,3,4,6,12,14,11,10,13,15,5)
Sequence four: (1,2,7,8,3,4,9,6,12,14,11,10,13,15,5)
The above several groups of assemblings that obtain are joined sequence and are carried out assembly path planning below;
Step 3, the assembly path planning of the emulation examination dismounting based on CATIA/DELMIA:
1) create virtual assembly work environment;
2) creating all assembling actions that will create of virtual assembling course of action is all to complete on the virtual work platform of creating in the above, so just in a virtual working environment, completes assembling, and concrete grammar is:
(1) set up Process Library
Setting up Process Library, be equivalent to set up a file that holds action, in menu bar, select " newly " order, then in the dialog box of beating, select Process Library, under the window of Process Library, carry out the establishment to the name of each assembly parts assembling process and assembly relation.The content of the foundation of assembly relation comprises the assembly relation that line is capable and parallel.
(2) set up Process Plan: wound Process Plan, its effect is that Process Library and assembly manipulation are connected, method is: by clicking Process icon on assembly tree, add operational motion;
(3) according to each assembling, move and make emulation: after having assembled, each assembling action is made to emulation animation, enter the use of PDM/Assembly Process Simulate module and assemble action, record assembling action.
Step 4, the measurement of assembly path and the calculating of installation time: the length of this assembly path can measure, then, divided by the average translational speed of 0.5 m/s, draw installation time.According to engineering experience, in order to ensure high-level efficiency and the stability of assembling process, the part of " light, little " can preferentially be dismantled, so classify installation time and assembly parts quality as following table.The quality (kg) of 1 to No. 15 assembly parts is respectively: (20,12.4,0.1,0.1,7.8,8.2,3.7,1.6,1.8,6.2,3.7,6.7,14.6,6.4,3.2)
According to four kinds of above-mentioned assembling sequences, try to achieve following four kinds of installation times and distribute, as (a) in table 1 with (b):
Table 1 a
Table 1 b
Step 5, the foundation of installation time evaluation model:
1) foundation of model
According to the installation time evaluation model having established above, the present invention adopts three layers of BP neural network, and use Neural Network Toolbox GUI solution to evaluate installation time, because current BP is to neural network, also do not have definite theory to determine scheme about research network hidden layer node number, if quantity is little, the Nonlinear Processing poor effect that the impact bringing is network; If quantity is large, can make again network learning procedure veryer long, on result precision without absolute impact, in the present invention through after computing repeatedly with and experimental data compare that to show that error sum of squares is defined as 10 more reasonable.Because last appraisement system is set to an evaluation of estimate, so output layer neuron node number is set to 1.
2) standardization of sample, because installation time and assembly quality are all that to be worth less evaluation better.So it is as follows that installation time is carried out to standardization:
T
ifor standardization desired value; x
ibe i sample value; x
minfor i minimum value can getting; x
maxfor i maximal value can getting.
Assembly parts quality is carried out to standardization, and to carry out standardization as follows:
M is standardization desired value; m
ibe i sample value; m
alwaysfor assembly gross mass, the gross mass of this assembly is 96.5kg.Because indivedual one-level assembly parts quality are too small, so do not affect on the basis of result, be not multiplied by ten convenient calculating below.The present invention analyzed above, and installation time and assembly quality are to influence each other and the relation conditioning each other, and the importance in assembly technology design is also no less important, so two samples are merged to standardization, desired value U
i=T
i* M
i.
Step 6, the evaluation of sample:
Matrix by after standardization, is converted to ordered series of numbers, is respectively:
a= （0 、0.63 、0.54、0.47 、0.45 、0.68 、0.63 、0.62 、0.58、 0.52、0.63 、0.60 、0.58、0.67、0.54）
b= （0 、0.64 、0.63 、0.57、0.48 、0.46 、0.51 、0.48 、、0.68 、0.54、0.44 、0.61 、0.54、0.58、0.61）
c= （0 、0.67 、0.56 、0.47 、0.48 、0.61 、0.59 、0.62 、0.53 、0.62 、0.47 、0.64 、0.46、0.52、0.60）
d= （0 、0.62 、0.63 、0.60 、0.45 、0.45 、0.58 、0.53 、0.52 、0.56 、0.63 、0.60 、0.53、0.60、0.56）
Will be above vectorial as input vector, with the network net_work1 having trained, carry out emulation above.
Installation time evaluation method based on artificial neural network and virtual assembling provided by the invention, adopted interference matrix method to determine assembling sequence, in the present invention, used a kind of integrated interference matrix method, make this method more intuitively more succinct, it is example that certain model engine is take in the present invention, with the virtual basis that is assembled into, the interference condition producing during according to part movement uses matrix method to generate dismounting constraint matrix, and then under the condition of virtual assembling, assembling sequence and assembly path are planned, finally utilize the Neural Network Toolbox in MALAB to evaluate installation time proposed by the invention, find out and comparatively reasonably assemble sequence and assembly path, thereby having solved the design of traditional assembly technology need to set up a large amount of physical models and could determine design to carry out the test of mould dress, make the Product Assembly cycle long, cost is high, and effect poor problem also, the solution that Virtual Assembling Technology is the problems referred to above provides new solution, the present invention is based on DELMIA emulation platform, assembly path and installation time are planned, thereby make the optimization of assembly technology more directly perceived, more accurate, more low-cost, in the present invention, also use artificial neural network algorithm to evaluate the installation time having obtained, make the assembly technology optimization drawing more there is cogency.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any modifications of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.
Claims (10)
1. the installation time evaluation method based on artificial neural network and virtual assembling, is characterized in that, should the installation time evaluation method based on artificial neural network and virtual assembling comprise the following steps:
Step 1, sets up virtual work platform, planning DELMIA assembling sequence, and planning DELMIA path, generates assembly path;
Step 2, uses constraint matrix method to generate assembling sequence, sets up the restriction relation between part; Then use DELMIA load module, product parts is carried out to pre-assembled, a vertical preferential dismounting matrix, then therefrom extracts the assembling sequence satisfying condition, and utilizes so the powerful data collection capability of matrix to plan assembling sequence;
Step 3, the assembly path planning of the emulation examination dismounting based on CATIA/DELMIA: set up virtual assembly work environment, create virtual assembling course of action;
Step 4, the measurement of assembly path and the calculating of installation time;
Step 5, the foundation of installation time evaluation model: the foundation of model, the standardization of sample;
Step 6, evaluates sample.
2. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, in step 1, use CATIA to carry out modeling to assembly and virtual work platform, concrete grammar is: after product component design finishes, the parts that each design department designs are put into a virtual environment according to coordinate setting unification, after integrating, all parts can obtain the product that an assembling completes, then being handed over to Assembling Production department is planned whole product establishment assembling sequence and carries out virtual assembling according to knowhow in the past by millwright performer, finally formulate a whole set of assembling programme, programme comprises assembling action and the required built-up member of whole product in the different time periods, then assembler carries out practical set production under the guidance of this scheme, if running in process of production assembling planning problem is committed to problem technology establishment department again and carries out secondary modification.
3. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, DELMIA assembling Sequence Planning in step 1, be divided into two part: DPE and DPM, DPE is process planning platform, DPM is emulation platform, while carrying out assembly process planning design in DPE module, for the product resource information table generating under DPE module must import to EBOM in the template of technological design in order.
4. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 3, it is characterized in that, the generation method of resource information table is: will manufacture resource and import in DPE environment, then will be according to the practical set technique in reality is produced, rely on knowhow under DPE, to create out a detailed assembly technology information table, in the process of just carrying out, the structure relevant to technique added in the middle of technological process simultaneously, when finishing, PPR Hub database is preserved and deposited in to the assembly technology of having planned, DPE is the pretreatment stage of DPM, and DPM has the effect of checking to DPE, belong to Qualify Phase.
5. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, is characterized in that, the concrete grammar of step 2 is:
First use constraint matrix method to generate assembling sequence, set up the restriction relation between part; Then use DELMIA load module, product parts is carried out to pre-assembled, a vertical preferential dismounting matrix, then therefrom extracts the assembling sequence satisfying condition, and utilizes the powerful data collection capability of matrix to plan assembling sequence;
The method that generates the constraint matrix employing of assembly is: by { x,-x, y,-y, z,-z} direction is as assembly direction, by setting up the interference matrix of each direction, show the relation constrained each other between its parts, the generation of interference matrix is the first step of Sequence Planning, parts meeting geometric constraint relation whether in assembling process, the dismoutability of judgement part on six coordinate directions, the program of interference checking, by each part respectively to { x,-x, y,-y, z, the dismounting of-z} axle, generate along { x,-x, y,-y, z, the free interference matrix of-z} direction, after interference checking, generate { x,-x, y,-y, z, the corresponding free interference matrix of-z}6 direction.
6. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, is characterized in that, tears the integrated concrete steps of method interference matrix open to be in step 2 based on examination:
Six free interference matrix FIM ' [N] [N] are carried out to occasional combination and become new matrix F IM ' [6N] [N], by the value of FIM ' [N] [N], the element of FIM ' [6N] [N] is carried out to initialization;
Calculate the row vector mould in FIM ' | FIi'|;
If | FIi'|=0, will | part that FIi'|=0 is expert at carries out new definition identifier p
k, k=i%N(% represents except rear remainder) and count the detachable array C of current round, and make integrated interference matrix element II
kj=FI
kj, j=(0,1 ..., N-1), II
k, N=i/N (/ represent remove then round) result is 0 expression+directions X, 1 expression-directions X, then after transforming by ± 1 expression ± X direction, the rest may be inferred can obtain the preferential disassembly direction of part in integrated interference matrix; Otherwise turn d;
If i > 6N-1, upgrades element p in C
kthe row, column value of corresponding FIM ' [6N] [N] matrix, makes: FI '
m, j=0 (m=k, k+N, k+2N ...; J=0,1 ..., N 1), FI '
j,n=0 (n=k, k+N, k+2N ...; J=0,1 ... N-1); Otherwise turn b;
If j, saving result; Otherwise turn b;
Entered after integrated, obtain new preferential dismounting matrix, in integrated interference matrix, the preferential disassembly direction in the end data of row represents, by ± 1 expression ± x direction, by ± 2 expression ± y directions, by ± 3 expression ± z directions, according to algorithm principle, preferential disassembly direction is to get rid of and to try to achieve with all directions matrix.
7. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, the concrete grammar that assembling sequence forms is: the artificial planning of dismounting sequence, based on free interference matrix, in unloading process, do not need the feasibility of disassembly path again to check:
The first step, first carries out standardization to integrated dismounting matrix P;
Second step, the mould of row vector in calculating P (i=1,2 ..., N), if | C
i|=0, show that part i is removable, if part i can not be subject to the obstruction of other parts when preferential disassembly direction is dismantled, by parts P corresponding to part i
iput in the set of detachable parts, otherwise, the 3rd step entered;
The 3rd step, establishes i=i+1, if i > is N, represents that the mould of all row is all complete by computing in the computing of front-wheel, and this takes turns complete as calculated, enters the 4th step, otherwise second step;
The 4th step, use mouse to get residue parts, drag, if occurred without interference situation, the sequence number of part is added in dismounting sequence, after dilatory detection, parts are proposed, the row vector of parts is all set to 0, represent that parts are disassembled, can not affect again the dismounting of subsequent parts;
The 5th step, observes whether all row vectors have all completed analysis, and all parts are all dismantled, if all parts have been dismantled, the parts label generating is inverted, and the assembly sequences generation of assembly completes.
8. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, in step 3, the calculating of installation time by the generation of assembly path above, can be measured path in DELMIA software, in daily Assembling Production, the translational speed of a part is confirmable, and the path according to measuring gained, calculates installation time by path and assembling speed:
T: the installation time of single part;
S
n: the assembly path of single part;
V: the mobile assembling speed of single part.
9. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, in step 4 first input sample be through hidden layer neuron transport function after input layer computing, then propagate into again this output layer, if the output of expectation can not be obtained, will proceed to the second backpropagation so; Concrete grammar is:
The first step, input signal one direction is propagated, and the input layer of network has I node, and hidden layer has J node, and output layer has K node, establishes X
p(X
p1, X
p2..., X
pl), O
jrepresent network input (O
p1, O
p2... O
pk), T
p=(t
p1, t
p2... t
pk) represent respectively actual output and the desired output of network, wherein p=(1,2 ..., P) sample number is P, (O
p1, O
p2..., O
pj) be the output of hidden node, represent i=(1,2 ..., I) individual input layer to j=(1,2 ..., the J) weights of individual hidden node, represent j hidden node to k=(1,2 ..., the K) weights of individual output layer node;
The excitation function of network is selected sigmoid function
for P sample:
The output layer of network is output as:
So far BP network has just completed the approximate mapping of I dimension space vector to K dimension space;
Second step, error signal backpropagation: now can use the error function of Square-type, the sample error of p is:
Have P sample, the error of the overall situation will be:
The 3rd step, the variation of output layer weights, adopts cumulative errors BP algorithm to adjust w
jk, global error E is diminished:
So η represents speed and the 0 < η <1 of study, now the error signal of definition is just expressed as:
Wherein:
So:
Thereby will obtain various neurons, at the formula of the weights adjustment of output layer, be:
The 4th step, the variation of hidden layer weights:
Definition error signal is:
Wherein:
So:
Thereby shadow layer weights regulate formula to be:
10. the installation time evaluation method based on artificial neural network and virtual assembling as claimed in claim 1, it is characterized in that, in step 5, for BP neural network, the algorithm using roughly can be divided into basic gradient descent algorithm and emulation function:
Utilization function in BP network:
The initialization function of network: the initialization function of network is all used in the initial value of giving for threshold values vector b before the network weight matrix W to created and network training, parameter value affects training speed and the convergence of algorithm, in NNToolbox, the call format of netinit function: net=init (net);
Parameter net and variable net are respectively network that newff sets up and the network after initialization, wherein this is initialized as default initialization, be about to [1,1] random number between is composed with the weight matrix W of the linear neuron layer in network and threshold values vector b, with non-linear hour neuron layer, can carry out initialization according to widrow Nyuyen method, in the time of need to calling network and set up function newff, just can automatically complete, special initialization can be by init function to applying;
Trial run function: the network having trained is come to evaluation computing and application by emulation function, the form of calling of this function is: after network has been set up, need to have a trial run function to come Neural Networks Solution, form is: a=sim (net, p);
Training function: complete if BP network has been set up, and input, output data are provided.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310671532.3A CN103617329A (en) | 2013-12-12 | 2013-12-12 | Assembly time evaluating method based on artificial neural network and virtual assembly |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310671532.3A CN103617329A (en) | 2013-12-12 | 2013-12-12 | Assembly time evaluating method based on artificial neural network and virtual assembly |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103617329A true CN103617329A (en) | 2014-03-05 |
Family
ID=50168032
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310671532.3A Pending CN103617329A (en) | 2013-12-12 | 2013-12-12 | Assembly time evaluating method based on artificial neural network and virtual assembly |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103617329A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104122796A (en) * | 2014-07-17 | 2014-10-29 | 中国科学院自动化研究所 | Intelligent assembly sequence planning method |
| CN104462705A (en) * | 2014-12-19 | 2015-03-25 | 西安工业大学 | Fixture assembly sequence planning method based on particle swarm optimization algorithm |
| CN104504471A (en) * | 2014-12-22 | 2015-04-08 | 华北电力大学 | Assembly sequence planning method and assembly sequence planning system |
| CN105205278A (en) * | 2015-10-09 | 2015-12-30 | 河海大学常州校区 | Method for automatically judging detaching directions of part in mechanical product assembling model |
| CN105947232A (en) * | 2016-06-06 | 2016-09-21 | 电子科技大学 | Aircraft fuselage component assemblability evaluation method taking assembling resource influences into consideration |
| CN106249705A (en) * | 2015-06-10 | 2016-12-21 | 通用汽车环球科技运作有限责任公司 | Assembly system configures |
| CN107256009A (en) * | 2017-06-30 | 2017-10-17 | 武汉理工大学 | A kind of Digital product model Intelligent assembly system based on deep learning |
| CN108416471A (en) * | 2018-02-26 | 2018-08-17 | 成都科木信息技术有限公司 | Intelligence computation method for supply chain |
| CN109508484A (en) * | 2018-10-26 | 2019-03-22 | 北京理工大学 | A kind of assembly, method for dismounting and device |
| CN110097909A (en) * | 2018-01-30 | 2019-08-06 | 慧与发展有限责任合伙企业 | The virtualization of memristor dot product engine |
| CN112182705A (en) * | 2020-09-28 | 2021-01-05 | 成都希盟泰克科技发展有限公司 | BIM technology-based simulation model data analysis method |
| CN113391598A (en) * | 2021-06-28 | 2021-09-14 | 哈尔滨工业大学 | Virtual assembly simulation method and system |
| US11321605B2 (en) * | 2017-11-13 | 2022-05-03 | Dassault Systemes Solidworks Corporation | Automatic assembly mate creation for frequently-used components |
| CN115859768A (en) * | 2022-10-12 | 2023-03-28 | 武汉理工大学 | Dynamic assembly job workshop workpiece completion time prediction method and device |
| CN116663204A (en) * | 2023-07-31 | 2023-08-29 | 南京航空航天大学 | Offline programming method, system and equipment for robot milling |
| TWI829807B (en) * | 2018-11-30 | 2024-01-21 | 日商東京威力科創股份有限公司 | Hypothetical measurement equipment, hypothetical measurement methods and hypothetical measurement procedures for manufacturing processes |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739478A (en) * | 2008-11-17 | 2010-06-16 | 北京卫星环境工程研究所 | Spacecraft assembly simulation technique-based virtual assembly system and virtual assembly method |
| TW201322149A (en) * | 2011-11-22 | 2013-06-01 | Univ Shu Te | Method of establishing system equivalent model combined with Volterra system and its computer program product |
-
2013
- 2013-12-12 CN CN201310671532.3A patent/CN103617329A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739478A (en) * | 2008-11-17 | 2010-06-16 | 北京卫星环境工程研究所 | Spacecraft assembly simulation technique-based virtual assembly system and virtual assembly method |
| TW201322149A (en) * | 2011-11-22 | 2013-06-01 | Univ Shu Te | Method of establishing system equivalent model combined with Volterra system and its computer program product |
Non-Patent Citations (3)
| Title |
|---|
| HUANG Y M等: "Disassembly matrix for disassembly processes of products", 《INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH》 * |
| 冯禹等: "基于人工神经网络技术的产品可装配性评价", 《计算机集成制造系统CIMS》 * |
| 杨骥: "飞机构件虚拟装配技术研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104122796A (en) * | 2014-07-17 | 2014-10-29 | 中国科学院自动化研究所 | Intelligent assembly sequence planning method |
| CN104462705A (en) * | 2014-12-19 | 2015-03-25 | 西安工业大学 | Fixture assembly sequence planning method based on particle swarm optimization algorithm |
| CN104504471A (en) * | 2014-12-22 | 2015-04-08 | 华北电力大学 | Assembly sequence planning method and assembly sequence planning system |
| CN104504471B (en) * | 2014-12-22 | 2018-05-25 | 华北电力大学 | Assembly sequence-planning method and apparatus |
| CN106249705A (en) * | 2015-06-10 | 2016-12-21 | 通用汽车环球科技运作有限责任公司 | Assembly system configures |
| CN106249705B (en) * | 2015-06-10 | 2019-12-06 | 通用汽车环球科技运作有限责任公司 | Assembly system configuration |
| CN105205278A (en) * | 2015-10-09 | 2015-12-30 | 河海大学常州校区 | Method for automatically judging detaching directions of part in mechanical product assembling model |
| CN105205278B (en) * | 2015-10-09 | 2018-07-27 | 河海大学常州校区 | The detachable direction automatic judging method of part in engineering goods assembling model |
| CN105947232A (en) * | 2016-06-06 | 2016-09-21 | 电子科技大学 | Aircraft fuselage component assemblability evaluation method taking assembling resource influences into consideration |
| CN105947232B (en) * | 2016-06-06 | 2018-03-20 | 电子科技大学 | A kind of airframe components assembling ability evaluating for considering Assembling resource and influenceing |
| CN107256009B (en) * | 2017-06-30 | 2019-08-02 | 武汉理工大学 | A kind of Digital product model Intelligent assembly system based on deep learning |
| CN107256009A (en) * | 2017-06-30 | 2017-10-17 | 武汉理工大学 | A kind of Digital product model Intelligent assembly system based on deep learning |
| US11321605B2 (en) * | 2017-11-13 | 2022-05-03 | Dassault Systemes Solidworks Corporation | Automatic assembly mate creation for frequently-used components |
| CN110097909A (en) * | 2018-01-30 | 2019-08-06 | 慧与发展有限责任合伙企业 | The virtualization of memristor dot product engine |
| CN108416471A (en) * | 2018-02-26 | 2018-08-17 | 成都科木信息技术有限公司 | Intelligence computation method for supply chain |
| CN109508484A (en) * | 2018-10-26 | 2019-03-22 | 北京理工大学 | A kind of assembly, method for dismounting and device |
| CN109508484B (en) * | 2018-10-26 | 2023-08-25 | 北京理工大学 | Assembling and disassembling method and device |
| TWI829807B (en) * | 2018-11-30 | 2024-01-21 | 日商東京威力科創股份有限公司 | Hypothetical measurement equipment, hypothetical measurement methods and hypothetical measurement procedures for manufacturing processes |
| CN112182705A (en) * | 2020-09-28 | 2021-01-05 | 成都希盟泰克科技发展有限公司 | BIM technology-based simulation model data analysis method |
| CN113391598A (en) * | 2021-06-28 | 2021-09-14 | 哈尔滨工业大学 | Virtual assembly simulation method and system |
| CN115859768A (en) * | 2022-10-12 | 2023-03-28 | 武汉理工大学 | Dynamic assembly job workshop workpiece completion time prediction method and device |
| CN115859768B (en) * | 2022-10-12 | 2023-05-02 | 武汉理工大学 | Method and device for predicting work piece finishing time of dynamic assembly workshop |
| CN116663204A (en) * | 2023-07-31 | 2023-08-29 | 南京航空航天大学 | Offline programming method, system and equipment for robot milling |
| CN116663204B (en) * | 2023-07-31 | 2023-10-17 | 南京航空航天大学 | Offline programming method, system and equipment for robot milling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103617329A (en) | Assembly time evaluating method based on artificial neural network and virtual assembly | |
| Tao et al. | Digital twin modeling | |
| Wang et al. | Mechanical assembly planning using ant colony optimization | |
| Akpan et al. | The confirmed realities and myths about the benefits and costs of 3D visualization and virtual reality in discrete event modeling and simulation: A descriptive meta-analysis of evidence from research and practice | |
| Long et al. | Modeling and distributed simulation of supply chain with a multi-agent platform | |
| Di Filippo et al. | Generative Design for project optimization (S). | |
| CN109086930A (en) | A kind of user power utilization behavior analysis method based on electric power K line chart and depth network | |
| CN109409561A (en) | The construction method of Multiple Time Scales time series collaborative forecasting model | |
| CN105843991B (en) | Stage HLA simulated members combined system and method | |
| CN108121530A (en) | A kind of conceptual design analysis method of multidisciplinary complex product | |
| Gao et al. | A data structure for studying 3D modeling design behavior based on event logs | |
| CN103617563B (en) | A kind of water supply network theoretical based on geo-statistic spatial analysis is without monitoring node pressure determination statement | |
| CN107491508A (en) | A kind of data base querying time forecasting methods based on Recognition with Recurrent Neural Network | |
| Zhu | Implementation of support-vector machine algorithm to develop a model for electronic commerce energy regulatory system | |
| CN106096145A (en) | A kind of complication system mathematics library based on state space and analysis environments | |
| CN113298911B (en) | Graphical concept modeling method based on lambda rule | |
| CN111597176A (en) | Teaching simulation training method and system for delaying supply chain generation | |
| Calkins et al. | Knowledge-based engineering (KBE) design methodology at the undergraduate and graduate levels | |
| Li et al. | Identifying the development trends and technological competition situations for digital twin: A bibliometric overview and patent landscape analysis | |
| CN104133808B (en) | User behavior uniformity degree measuring method based on complicated correspondence system | |
| Clyde et al. | Turnable formalism in object-oriented systems analysis: meeting the needs of both theoreticians and practitioners | |
| CN116167740A (en) | Digital twin model construction method for intelligent maintenance of pumped storage power station | |
| US7778807B2 (en) | Simulation model generator of causality phenomena and events | |
| Jin et al. | Execution and evaluation of enterprise models in IEM/MO2GO based on Petri net | |
| Waghmode et al. | A collective study of PCA and neural network based on COCOMO for software cost estimation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140305 |