CN104932847B - A kind of spatial network 3D printing algorithm - Google Patents
A kind of spatial network 3D printing algorithm Download PDFInfo
- Publication number
- CN104932847B CN104932847B CN201510310560.1A CN201510310560A CN104932847B CN 104932847 B CN104932847 B CN 104932847B CN 201510310560 A CN201510310560 A CN 201510310560A CN 104932847 B CN104932847 B CN 104932847B
- Authority
- CN
- China
- Prior art keywords
- neuron
- input
- space
- network
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Abstract
The invention discloses a kind of spatial network 3D printing algorithm, extends self-organizing feature silver snake algorithm first, finds out the similar degree between input data automatically, similar input is configured nearby on network, forms the network that reaction is selectively given to input data.Learning algorithm construction SOM spatial network 3D printing algorithms based on self-organizing feature map；The present invention can make the multithreading tasks such as calculating of the printer to itself, communication, control, man-machine interaction carry out comprehensive coordinate and Optimal Decision-making, the optimum allocation of self-ability is carried out with reference to itself current controlling feature and task feature, thread conflict and task contradiction are at utmost avoided, and with the ability for finding faults itself and suggesting to solution of being out of order.
Description
Technical field
The invention belongs to 3D printing technique field, is related to a kind of spatial network 3D printing algorithm.
Background technology
Existing 3D printing control method is opened loop control, or common PID control method, these methods do not possess to more
The calculating of the task and contradiction of kind conflict carries out the ability of intelligent optimization, that is, 3D printer can not be felt to oneself state
Know so that the most important functions Coordinated Play of itself and being automatically disengaged when multiple threads clash.When calculating, lead to
An organic whole can not be formed during the content weave ins such as letter, control, man-machine interaction, more can not be out of 3D printing control
Support to form intelligent network between multiple printers on core, realize the distributed operation between 3D printer, and people can only be relied on
For manually adjust with separation structure design and separation structure individually print, when there is thread conflict or task contradiction in 3D printer without
Method provides fault flag and solution.
The content of the invention
It is an object of the invention to provide a kind of spatial network 3D printing algorithm, solves current 3D printing control method
It can not support to form intelligent network between multiple printers from the kernel that 3D printing controls, realize the distribution between 3D printer
The problem of formula operates.
The technical solution adopted in the present invention is as follows:
A kind of spatial network 3D printing algorithm, first extend self-organizing feature silver snake algorithm, find out automatically input data it
Between similar degree, similar input is configured nearby on network, forms the network that reaction is selectively given to input data.
Learning algorithm construction SOM spatial network 3D printing algorithms based on self-organizing feature map：
STEP1：Fourier's thermic vibrating screen is imported by the use of random number as scale parameter, is set between input layer and mapping layer
Weights initial value, using input layer as variable X, by the use of mapping layer be used as variable Y, carry out gradient space evolution, so as to be formed
Input sheaf space and mapping sheaf space.It is less that gradient space is assigned to m input neuron to output neuron connection weight
Weights.Choose the set S of " the adjacent neuron " in j space of output neuronj.Wherein Sj(0) nerve at moment t=0 is represented
The set of first j spaces " adjacent neuron ", Sj(t) set of moment t space " adjacent neuron " is represented.Area of space Sj(t) with
The growth of time and constantly reduce；
STEP2：More dimensional input vector X=(x1,x2,x3,…,xm)TInput sheaf space is given as data；
STEP3：Calculate the weight vector of mapping sheaf space and the distance (Euclidean distance) in input vector space.In mapping layer
Space, calculate the weight vector space of each neuron and the Euclidean distance in input vector space.J-th neuron of mapping layer and
The distance of input vector is as shown in Equation 1
Formula 1
In formula, wijI neurons for input layer and the weights between the j neurons of mapping layer.By calculating, one is obtained
Neuron with minimum range, is designated as j*, that is, determine some unit k so that for arbitrary j, there is dk=min (dj),
And provide it and abut neuronal ensemble；
STEP4：The study of weights 2 corrects output neuron j* and its " adjacent neuron " weights as the following formula；
Δ w_ij=w_ij (t+1)-w_ij (t)-η (t) (x_i (t)-w_ij (t)) (formula 2)
In formula 2, η is one and is more than 0 constant for being less than 1, and change over time gradually decreases to 0；
Or(formula 3)
STEP5：Calculate output Ok
Ok=f (min | | x-wj||)
In formula, f (*) is generally 0~1 function or other nonlinear functions；
STEP6：Judge whether to reach requirement set in advance.Algorithm terminates if requirement is reached；
Otherwise, return to step (2), next round study is carried out.
After above-mentioned algorithm, beneficial effects of the present invention are as follows：
Patent of the present invention discloses one kind according to the deficiency of prior art and product can be widely used in various shaping works
SOM spatial network 3D printing algorithms in the 3D printer of skill, it is a kind of printer oneself state and parameter to be perceived
With the intelligent control algorithm of coordination, the Method of printings such as SLA, SLS, FDM, LOM, 3DP are provided and supported comprehensively, printer can be made
The multithreading tasks such as calculating, communication, control, man-machine interaction to itself carry out comprehensive coordinate and Optimal Decision-making, work as with reference to itself
Preceding controlling feature and task feature carry out the optimum allocation of self-ability, at utmost avoid thread conflict and task contradiction,
And with the ability for finding faults itself and suggesting to solution of being out of order.
Brief description of the drawings
Shown in Fig. 1 is the 3D printing network topology structure figure in the present invention；
Shown in Fig. 2 is the 3D adjoining neuron direct range figures in the present invention；
Shown in Fig. 3 is 3DSOM competitions triumph neuron distribution map in the present invention.
Embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
The structure of SOM spatial network 3D printing algorithms
Self-organizing feature silver snake algorithm is extended first, the similar degree between input data is found out automatically, by similar input
Configured nearby on network, form the network that reaction is selectively given to input data.Based on self-organizing feature map
Practise algorithm construction SOM spatial network 3D printing algorithms：
Step 1：
Fourier's thermic vibrating screen (known) is imported by the use of random number (known) as scale parameter, sets input layer and mapping
The initial value of weights between layer, using input layer as variable X, by the use of mapping layer as variable Y, progress gradient space evolution, from
And form input sheaf space and mapping sheaf space.To m input neuron gradient space is assigned to output neuron connection weight
Less weights.Choose the set S of " the adjacent neuron " in j space of output neuronj, wherein Sj(0) moment t=0 is represented
Neuron j spaces " adjacent neuron " set, Sj(t) set of moment t space " adjacent neuron " is represented.Area of space
Sj(t) growth over time and constantly reduce.
Step 2：
The initial value of the weights between input layer and mapping layer is set, using input layer as variable X, more dimensional input vectors
X=(x1,x2,x3,…,xm)TInput sheaf space is given as data；
Step 3：
Calculate mapping sheaf space weight vector and input vector space distance (Euclidean distance) (be here say two to
Measure the distance in mapping sheaf space).In mapping sheaf space, weight vector space and the input of each mapping sheaf space neuron are calculated
The Euclidean distance of vector space.J-th of neuron of mapping layer and the distance of input vector are as shown in Equation 1
Formula 1
In formula, wijI neurons for input layer and the weights between the j neurons of mapping layer.By calculating, one is obtained
Neuron with minimum range, is designated as j*, that is, determine some unit k so that for arbitrary j, there is dk=min (dj)；
dk：：The k neurons of mapping layer and the distance of input vector.j：One has the neuron of minimum range, and gives
Go out it and abut neuronal ensemble；
Step 4：
The study of neuron weights, 2 correct output neuron j* and its " adjacent neuron " weights as the following formula.
Δ w_ij=w_ij (t+1)-w_ij (t)-η (t) (x_i (t)-w_ij (t)) (formula 2)
In formula 2, η is one and is more than 0 constant for being less than 1, and change over time gradually decreases to 0
Or(formula 3)
Calculate output Ok
Ok=f (min | | x-wj||)
In formula, f (*) is generally 0~1 function or other nonlinear functions.
Step 5：
Judge whether to reach requirement (display can be configured on a printer) set in advance itself.If reach requirement
Then algorithm terminates；Otherwise, return to step (2), next round study is carried out.
The example of the SOM spatial network 3D printing algorithms of the present invention and expansion application are as follows：
Lower example gives an out data set containing 8 3D printing system fault samples, has 8 spies in each fault sample
Sign, it is above to refer to respectively：Maximum pressure (P1), secondary maximum pressure (P2), wave-shape amplitude (P3), rising edge width (P4)、
Width (the P of waveform widths (P5), maximum repercussions6), the area (P of waveform7), rise spray power (P8), carry out intelligence using SOM networks
Can control and corresponding fault diagnosis.Actual control sample is as listed in table 1 (data have normalized).
The actual control sample of the 3D printing system of table 1.
The step of carrying out control in real time and tracing trouble using SOM neutral net 3D printings control method is as follows：
1. selection standard controls sample；
2. pair each standard control sample learns, after study terminates, the neuron with maximum output is marked with
The mark of the control action and failure；
3. sample to be detected is input in SOM neutral nets；
4. if output neuron is controlled in the position of output layer with certain standard and the position of fault sample is identical, illustrate to be checked
Sample is there occurs corresponding failure and can use corresponding control method.The sample to be tested used in this example is:
| Test | |
| T1 | 0.9512 |
| T2 | 1.0000 |
| T3 | 0.9458 |
| T4 | -0.4215 |
| T5 | 0.4218 |
| T6 | 0.9511 |
| T7 | 0.9645 |
| T8 | 0.8941 |
The training step of 3DSOM networks is 500；
Cluster result：
The data result of survey to be measured is shown as 25, namely testing data has been incorporated into T1 this failure cause.
3D printing network topology structure is as shown in figure 1, adjacent neuron is directly as shown in Figure 2 apart from situation.SOM's is competing
Sharp neuron distribution situation of competing for first place is as shown in Figure 3.
As shown in Figure 1, competition layer neuron has 6*6=36 neuron.Blueness represents neuron in Fig. 3, and red represents
Directly connected between neuron, the color in each rhombus represents the distance of distance between neuron, from yellow to black, face
The distance between the deeper explanation neuron of color is more remote.From this it can be seen that the SOM spatial network 3D printings that this patent is announced
Algorithm can effectively detect the task feature that own components network or 3D printing network are formed in 3D printer, operation spy
Sign, and the distribution of control thread and the decision-making of task can be detected and correspond to, so that 3D printer can be maximum
The ability of itself is played, these tasks and thread can be then arbitrary calculating, communication, control and man-machine interaction instruction and count
According to whole 3D printing process can be run through.
It is also an advantage of the present invention that：
1. one of features of the present invention is with pe array, the time for receiving 3D printer inputs, and shape
" discriminant function " of these paired signals, so that the control of 3D printer possesses to the sensitiveness of time and intelligent.
2. with selection mechanism is compared, it is provided with 3D printing control method of the invention and " differentiates letter for what is compared
Number ", it can make the calculating of task, contradiction of the 3D printer to a variety of conflicts in kernel carry out intelligent selection processing, and
One processing unit with larger function input value of selection, so as to which the moment makes 3D printer have optimal state.
3. with local interconnection function, have in 3D printing control method of the invention and be used to encourage selected place simultaneously
Unit and its closest processing unit are managed, it can make each part of 3D printer from kernel, cover calculating, be logical
The contents such as letter, control, man-machine interaction form an organic whole, and printer is realized intelligent intercommunication, can equally made multiple
Intelligent network is formed between printer, realizes interconnecting between printer.
4. Self Adaptive Control and decision process, 3D printing control method of the invention is interior to be had for correcting energized place
The parameter of unit is managed, corresponds to the input value of specific input " discriminant function " to increase it, calculating, decision-making and communication can be made certainly
Move and realize Automatic Optimal in complex environment and conflict, so as to complete the adaptive process of control and decision-making.
Described above is only the better embodiment to the present invention, not makees any formal limit to the present invention
System, any simple modification that every technical spirit according to the present invention is made to embodiment of above, equivalent variations and modification,
Belong in the range of technical solution of the present invention.
Claims (1)
1. a kind of spatial network 3D printing algorithm, it is characterised in that extend self-organizing feature silver snake algorithm first, find out automatically defeated
Enter the similar degree between data, similar input is configured nearby on network, form and input data is selectively given instead
The network answered, the learning algorithm construction SOM spatial network 3D printing algorithms based on self-organizing feature map；
STEP1：Fourier's thermic vibrating screen is imported by the use of random number as scale parameter, sets the power between input layer and mapping layer
The initial value of value, using input layer as variable X, by the use of mapping layer as variable Y, gradient space evolution is carried out, so as to form input
Sheaf space and mapping sheaf space, the less weights of gradient space are assigned to output neuron connection weight to m input neuron,
Choose the set S of " the adjacent neuron " in j space of output neuronj, wherein Sj(0) represent that the neuron j at moment t=0 is empty
Between " adjacent neuron " set, Sj(t) set of moment t space " adjacent neuron ", area of space S are representedj(t) with when
Between growth and constantly reduce；
STEP2：More dimensional input vector X=(x1,x2,x3,…,xm)TInput sheaf space is given as data；
STEP3：Calculate the weight vector of mapping sheaf space and the distance (Euclidean distance) in input vector space：It is empty in mapping layer
Between, calculate the weight vector space of each neuron and the Euclidean distance in input vector space, j-th of neuron of mapping layer and defeated
The distance of incoming vector is as shown in Equation 1：
In formula, wijI neurons for input layer and the weights between the j neurons of mapping layer, by calculating, obtaining one has
The neuron of minimum range, is designated as j*, that is, determine some unit k so that for arbitrary j, there is dk=min (dj), and give
Go out it and abut neuronal ensemble；
STEP4：The study of weights：2 amendment output neuron j as the following formula*And its " adjacent neuron " weights,
Δωij=ωij(t+1)-ωij(t)-η(t)(xi(t)-ωij(t)) (formula 2)
In formula 2, η is one and is more than 0 constant for being less than 1, and change over time gradually decreases to 0；
STEP5：Calculate output 0k
0k=f (min | | x-wj||)
In formula, f (*) is generally 0~1 function or other nonlinear functions；
STEP6：Judge whether to reach requirement set in advance, algorithm terminates if requirement is reached, and otherwise, returns to STEP2, enters
Row next round learns.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510310560.1A CN104932847B (en) | 2015-06-08 | 2015-06-08 | A kind of spatial network 3D printing algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510310560.1A CN104932847B (en) | 2015-06-08 | 2015-06-08 | A kind of spatial network 3D printing algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104932847A CN104932847A (en) | 2015-09-23 |
| CN104932847B true CN104932847B (en) | 2018-01-19 |
Family
ID=54120026
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510310560.1A Active CN104932847B (en) | 2015-06-08 | 2015-06-08 | A kind of spatial network 3D printing algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104932847B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107561936B (en) * | 2017-08-28 | 2018-06-19 | 三维泰柯（厦门）电子科技有限公司 | The mixing class rapid shaping framework method of internal control review type association control |
| US11487271B2 (en) * | 2019-04-23 | 2022-11-01 | Dassault Systemes Simulia Corp. | Machine learning with fast feature generation for selective laser melting print parameter optimization |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0385873B1 (en) * | 1989-03-01 | 1997-05-28 | Fujitsu Limited | A learning system in a neuron computer |
| CN102402712A (en) * | 2011-08-31 | 2012-04-04 | 山东大学 | Robot reinforced learning initialization method based on neural network |
-
2015
- 2015-06-08 CN CN201510310560.1A patent/CN104932847B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0385873B1 (en) * | 1989-03-01 | 1997-05-28 | Fujitsu Limited | A learning system in a neuron computer |
| CN102402712A (en) * | 2011-08-31 | 2012-04-04 | 山东大学 | Robot reinforced learning initialization method based on neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104932847A (en) | 2015-09-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103620624B (en) | For the method and apparatus causing the local competition inquiry learning rule of sparse connectivity | |
| Pasqualetti et al. | Controllability metrics, limitations and algorithms for complex networks | |
| Jin et al. | Robust adaptive hierarchical insensitive tracking control of a class of leader-follower agents | |
| Klimm et al. | Resolving structural variability in network models and the brain | |
| CN107977706A (en) | Modularized distribution type artificial neural network | |
| Zhou et al. | Cluster synchronization on multiple nonlinearly coupled dynamical subnetworks of complex networks with nonidentical nodes | |
| CN104932847B (en) | A kind of spatial network 3D printing algorithm | |
| Van Segbroeck et al. | Learning to coordinate in complex networks | |
| CN110286694A (en) | A kind of unmanned plane formation cooperative control method of more leaders | |
| CN109165724A (en) | A kind of gradient neural network based decline the number of iterations prediction technique and device | |
| CN106096911A (en) | A kind of based role have people/unmanned collaborative command and control system and method thereof | |
| KR102424729B1 (en) | Water quality analysis system using labeling of water quality data and learning of artificial neural networks | |
| Cheng et al. | Selecting pinning nodes to control complex networked systems | |
| CN106067077A (en) | A kind of load forecasting method based on neutral net and device | |
| Zhang et al. | Generalized outer synchronization between non-dissipatively coupled complex networks with different-dimensional nodes | |
| CN102708407A (en) | Self-adaptive hybrid multi-objective evolutionary method on basis of population decomposition | |
| Trendafilov et al. | The role of the environment in collective perception: A generic complexity measure | |
| CN103886340A (en) | Sample set training method for online learning | |
| CN106156891A (en) | A kind of Scheduling method based on Pareto multiple target ant colony optimization algorithm | |
| CN106204381A (en) | Study cloud paas platform modeling method based on hadoop | |
| CN105468862B (en) | A method of building Aided Design Expert System in silk relics field | |
| Allaoui et al. | Optimization of unsupervised classification by evolutionary strategies | |
| Bahavarnia et al. | Row-column sparse linear quadratic controller design via bi-linear rank penalty technique and non-fragility notion | |
| CN109636240A (en) | A kind of decoupling model that participates in evaluation and electing suitable for water quality assessment | |
| da Fontoura Costa et al. | Topographical maps as complex networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |