CN104932847A - Spatial network 3D printing algorithm - Google Patents

Abstract

The invention discloses a spatial network 3D printing algorithm, comprising the steps as follows: firstly extending self organizing feature silver snake algorithm; automatically finding similarity among input data, performing nearby configuration to similar input on the network to form a network for selectively giving feedback to the input data, and configuring SOM spatial network 3D printing algorithm based on a learning algorithm of self organizing feature mapping. The spatial network 3D printing algorithm of the invention could be used for performing overall coordination and optimized decision to multithreading tasks such as self calculation, communication, control and human-computer interaction, and performing optimum allocation of self ability by combining with current control feature and task feature so as to avoid threading conflict and task conflict to the largest extent. The spatial network 3D printing algorithm of the invention has the ability of finding self faults and giving a fault solving suggestion.

Description

A kind of spatial network 3D print algorithms

Technical field

The invention belongs to 3D printing technique field, relate to a kind of spatial network 3D print algorithms.

Background technology

Existing 3D print control program is opened loop control, or common PID control method, these methods do not possess the ability of the task of multiple conflict and the calculating of contradiction being carried out to intelligent optimization, and namely 3D printer cannot carry out perception to oneself state thus make the most important functions Coordinated Play of self and automatically disengage when multiple thread clashes.An organic whole cannot be formed when content weave ins such as calculating, communication, control, man-machine interactions, more cannot support to form intelligent network between multiple printer from the kernel of 3D Print Control, realize the distributed operation between 3D printer, and can only rely on artificial manual adjustment and separation structure design and separation structure print separately, fault flag and solution cannot be provided when thread conflict or task contradiction appear in 3D printer.

Summary of the invention

The object of the present invention is to provide a kind of spatial network 3D print algorithms, solve current 3D print control program to support to form intelligent network between multiple printer from the kernel of 3D Print Control, realize the problem of the distributed operation between 3D printer.

The technical solution adopted in the present invention is as follows:

A kind of spatial network 3D print algorithms, first expands self-organizing feature silver snake algorithm, automatically finds out the similar degree between input data, similar input configured nearby on network, form input data selectively to the network with reaction.Learning algorithm structure SOM spatial network 3D print algorithms based on self-organizing feature map:

STEP1: import Fourier's thermic vibrating screen as scale parameter by random number, the initial value of the weights between setting input layer and mapping layer, using input layer as variable X, with mapping layer as variable Y, carry out gradient space evolution, thus form input layer space and mapping layer space.Connecting weights to m input neuron to output neuron gives gradient space less weights.Choose the S set of " the adjacent neuron " in an output neuron j space _j.Wherein S _j(0) set in the neuron j space " adjacent neuron " of moment t=0 is represented, S _jt () represents the set of moment t space " adjacent neuron ".Area of space S _jt () constantly reduces along with the growth of time;

STEP2: multidimensional input vector X=(x ₁, x ₂, x ₃..., x _m) ^tas data to input layer space;

STEP3: calculate the weight vector in mapping layer space and the distance (Euclidean distance) in input vector space.In mapping layer space, calculate the Euclidean distance in each neuronic weight vector space and input vector space.The jth neuron of mapping layer and input vector distance as shown in Equation 1

$d_j=\left|\right|X-W_j\left|\right|=\sqrt{\underset{i=1}{\overset{m}{\Σ}}{(x_i\left(t\right)-{\mathrm{\ω}}_{ij}(t\left)\right)}^2}$ Formula 1

In formula, w _ijfor the weights between the i neuron of input layer and the j neuron of mapping layer.By calculating, obtaining the neuron that has minor increment, being designated as j ^*, namely determine certain unit k, make, for arbitrary j, have d _k=min (dj), and provide its adjacent neuronal ensemble;

STEP4: the study of weights. revise output neuron j* and " adjacent neuron " weights thereof by following formula 2;

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 greater than 0 constant being less than 1, along with time variations drops to 0 gradually;

$\mathrm{\η}\left(t\right)=\frac{1}{t}$ Or $\mathrm{\η}\left(t\right)=0.2(1-\frac{t}{10000})$ (formula 3)

STEP5: calculate and export O _k

O _k＝f(min||x-w _j||)

In formula, f (*) is generally 0 ~ 1 function or other nonlinear functions;

STEP6: judge whether to reach the requirement preset.If reach requirement, algorithm terminates;

Otherwise, return step (2), carry out next round study.

After adopting above-mentioned algorithm, beneficial effect of the present invention is as follows:

Patent of the present invention discloses a kind of SOM spatial network 3D print algorithms that can be widely used in the 3D printer of various moulding process according to the deficiency of prior art and product, it is a kind of intelligent control algorithm printer oneself state and parameter being carried out to perception and coordination, to SLA, SLS, FDM, LOM, the Method of printings such as 3DP provide comprehensive support, printer can be made the calculating of self, communication, control, the multithreading tasks such as man-machine interaction carry out comprehensive coordinate and Optimal Decision-making, the controlling feature current in conjunction with self and task feature carry out the optimum allocation of self-ability, at utmost avoid thread conflict and task contradiction, and there is the ability finding faults itself and advise to solution of being out of order.

Accompanying drawing explanation

Shown in Fig. 1 is that 3D in the present invention prints network topology structure figure;

Shown in Fig. 2 is that 3D in the present invention adjoins neuron direct range figure;

Shown in Fig. 3 is 3DSOM competition triumph neuron distribution plan in the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

The structure of SOM spatial network 3D print algorithms

First expand self-organizing feature silver snake algorithm, automatically find out the similar degree between input data, similar input is configured nearby on network, form input data selectively to the network with reaction.Learning algorithm structure SOM spatial network 3D print algorithms based on self-organizing feature map:

Step 1:

Fourier's thermic vibrating screen (known) is imported as scale parameter by random number (known), the initial value of the weights between setting input layer and mapping layer, using input layer as variable X, with mapping layer as variable Y, carry out gradient space evolution, thus form input layer space and mapping layer space.Connecting weights to m input neuron to output neuron gives gradient space less weights.Choose the S set of " the adjacent neuron " in an output neuron j space _j, wherein S _j(0) set in the neuron j space " adjacent neuron " of moment t=0 is represented, S _jt () represents the set of moment t space " adjacent neuron ".Area of space S _jt () constantly reduces along with the growth of time.

Step 2:

The initial value of the weights between setting input layer and mapping layer, using input layer as variable X, multidimensional input vector X=(x ₁, x ₂, x ₃..., x _m) ^tas data to input layer space;

Step 3:

Calculate the weight vector in mapping layer space and the distance (Euclidean distance) (being say the distance of two vectors in mapping layer space) in input vector space here.In mapping layer space, calculate the Euclidean distance in neuronic weight vector space, each mapping layer space and input vector space.The jth neuron of mapping layer and input vector distance as shown in Equation 1

$d_j=\left|\right|X-W_j\left|\right|=\sqrt{\underset{i=1}{\overset{m}{\Σ}}{(x_i\left(t\right)-{\mathrm{\ω}}_{ij}(t\left)\right)}^2}$ Formula 1

In formula, w _ijfor the weights between the i neuron of input layer and the j neuron of mapping layer.By calculating, obtaining the neuron that has minor increment, being designated as j ^*, namely determine certain unit k, make, for arbitrary j, have d _k=min (dj);

D _k:: the k neuron of mapping layer and input vector distance.J: one neuron with minor increment, and provide its adjacent neuronal ensemble;

Step 4:

The study of neuron weights, revises output neuron j* and " adjacent neuron " weights thereof by following formula 2.

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 greater than 0 constant being less than 1, along with time variations drops to 0 gradually

$\mathrm{\η}\left(t\right)=\frac{1}{t}$ Or $\mathrm{\η}\left(t\right)=0.2\left(1-\frac{t}{10000}\right)$ (formula 3)

Calculate and export O _k

O _k＝f(min||x-w _j||)

In formula, f (*) is generally 0 ~ 1 function or other nonlinear functions.

Step 5:

Judge whether to reach the requirement (can carry out on a printer arranging display) itself preset.If reach requirement, algorithm terminates; Otherwise, return step (2), carry out next round study.

Example and the expansive approach of SOM spatial network 3D print algorithms of the present invention are as follows:

Lower example gives an out data set containing 8 3D print system fault samples, has 8 features, mentions before being respectively: maximum pressure (P in each fault sample ₁), secondary maximum pressure (P ₂), wave-shape amplitude (P ₃), rising edge width (P ₄), the width (P of waveform widths (P5), maximum repercussions ₆), the area (P of waveform ₇), rise spray power (P ₈), use SOM network to carry out Based Intelligent Control and corresponding fault diagnosis.Working control sample is (data are normalization) as listed in table 1.

Table 1. 3D print system working control sample

The step that application SOM neural network 3D print control program carries out control and tracing trouble is in real time as follows:

1. selection standard controls sample;

2. control sample to each standard to learn, after study terminates, the neuron with maximum output is marked with to the mark of this control action and fault;

3. sample to be detected is input in SOM neural network;

If 4. output neuron controls identical with the position of fault sample with certain standard in the position of output layer, illustrate that sample to be checked there occurs corresponding fault 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 network is 500;

Cluster result:

The data result of survey to be measured is shown as 25, and also namely testing data has been incorporated in this failure cause of T1.

3D prints network topology structure as shown in Figure 1, and adjacent neuron directly apart from situation as shown in Figure 2.The competition triumph neuron distribution situation of SOM as shown in Figure 3.

As shown in Figure 1, competition layer neuron has 6*6=36 neuron.In Fig. 3, blueness represents neuron, and redness represents between neuron and directly connects, and the color in each rhombus represents the distance of the spacing of neuron, and from yellow to black, color illustrates that the distance between neuron is far away more deeply.From this example can see SOM spatial network 3D print algorithms that this patent is announced can effectively to detect in 3D printer form task feature, the operating characteristics that own components network or 3D print network, and can detect and control the distribution of thread and the decision-making of task with corresponding, thus enable 3D printer farthest play the ability of self, these tasks and thread can be then arbitrary calculating, communicate, control and man-machine interaction instruction and data, can run through whole 3D print procedure.

The present invention be advantageous in that:

1. one of feature of the present invention is for having pe array, for receiving the time input of 3D printer, and forms " discriminant function " to these signals, thus the control of 3D printer has been possessed the susceptibility of time and intelligent.

2. there is alternative mechanism, " discriminant function " for comparing is provided with in 3D print control program of the present invention, it can make 3D printer carry out intelligent selection process to the calculating of the task of multiple conflict, contradiction in kernel, and select a processing unit with larger function input value, thus the moment makes 3D printer have optimum state.

3. there is local interconnect function, have for encouraging by the processing unit selected and the most contiguous processing unit thereof simultaneously in 3D print control program of the present invention, it can make each ingredient of 3D printer from kernel, contain the contents such as calculating, communication, control, man-machine interaction forms an organic whole, printer realizes intelligent intercommunication, can make equally to form intelligent network between multiple printer, realize interconnecting between printer.

4. adaptive control and decision process, there is in 3D print control program of the present invention the parameter for revising energized processing unit, to increase the input value that it corresponds to specific input " discriminant function ", calculating, decision-making can be made and communicate automatically in complex environment and conflict, to realize Automatic Optimal, thus complete the adaptive process of control and decision-making.

The above is only to better embodiment of the present invention, not any pro forma restriction is done to the present invention, every any simple modification done above embodiment according to technical spirit of the present invention, equivalent variations and modification, all belong in the scope of technical solution of the present invention.

Claims (1)

1. a spatial network 3D print algorithms, it is characterized in that, first self-organizing feature silver snake algorithm is expanded, automatically the similar degree between input data is found out, similar input is configured nearby on network, form to input data selectively to the network with reaction, based on the learning algorithm structure SOM spatial network 3D print algorithms of self-organizing feature map;

STEP1: import Fourier's thermic vibrating screen as scale parameter by random number, the initial value of the weights between setting input layer and mapping layer, using input layer as variable X, with mapping layer as variable Y, carry out gradient space evolution, thus form input layer space and mapping layer space, to output neuron, weights are connected to m input neuron and gives gradient space less weights, choose the S set of " the adjacent neuron " in an output neuron j space _j, wherein S _j(0) set in the neuron j space " adjacent neuron " of moment t=0 is represented, S _jt () represents the set of moment t space " adjacent neuron ", area of space S _jt () constantly reduces along with the growth of time;

STEP2: multidimensional input vector X=(x ₁, x ₂, x ₃..., x _m) ^tas data to input layer space;

STEP3: calculate the weight vector in mapping layer space and the distance (Euclidean distance) in input vector space, in mapping layer space, calculate the Euclidean distance in each neuronic weight vector space and input vector space, the jth neuron of mapping layer and input vector distance as shown in Equation 1:

$d_j=||X-W_j||=\sqrt{\underset{i=1}{\overset{m}{\Σ}}{\left(x_i\left(t\right)-{\mathrm{\ω}}_{ij}\left(t\right)\right)}^2}$ Formula 1

In formula, w _ijfor the weights between the i neuron of input layer and the j neuron of mapping layer, by calculating, obtaining the neuron that has minor increment, being designated as j ^*, namely determine certain unit k, make, for arbitrary j, have d _k=min (dj), and provide its adjacent neuronal ensemble;

STEP4: the study of weights. revise output neuron j* and " adjacent neuron " weights thereof by following formula 2.

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 greater than 0 constant being less than 1, along with time variations drops to 0 gradually;

$\mathrm{\η}\left(t\right)=\frac{1}{t}$ Or $\mathrm{\η}\left(h\right)=0.2(1-\frac{t}{10000})$ (formula 3)

STEP5: calculate and export O _k

O _k＝f(min||x-w _j||)

In formula, f (*) is generally 0 ~ 1 function or other nonlinear functions;

STEP6: judge whether to reach the requirement preset, if reach requirement, algorithm terminates, otherwise, return step (2), carry out next round study.