CN105956235A - Optimum design method for ultrasonic machining special cutter based on SVR-PSO - Google Patents

Abstract

The invention discloses an optimum design method for an ultrasonic machining special cutter based on SVR (support vector regression)-PSO (particle swarm optimization). Through establishing an approximation model of structural response, and quantitatively establishing mapping relations between cutter structure parameters and the structural response, the method provides a solution for accurately optimizing and designing a cutter structure.

Description

Ultrasonic machining special type knife Optimization Design based on SVR-PSO

Technical field

The present invention relates to a kind of Ultrasonic machining based on SVR (support vector regression)-PSO (particle cluster algorithm) Special type knife Optimization Design.

Background technology

Ultrasonic machining technology is one of special processing technology, for the unmanageable composite wood of conventional machining techniques Material, Ultrasonic machining technology tends to obtain beyond thought effect.Cellular composite material because of its construction features and Component feature, is a kind of typical difficult-to-machine material.The application of Ultrasonic machining technology efficiently solves honeycomb The problems such as present in composite tradition Milling Process, environmental pollution is serious, machined surface quality is poor.Super In sound processing assembly, Ultrasonic machining special type knife is an extremely important part.The architectural feature harmony of cutter Learn performance and directly influence suface processing quality and the working (machining) efficiency of cellular composite material, and cutter itself Service life.Being found by the reading of lot of documents, current scholars are extensive to the research of Ultrasonic machining cutter Apply FInite Element and laboratory method, the former summarizes the tool structure parameter change to structural response qualitatively Law, provides thinking for cutter parameters optimization, and the latter, by processed and applied in kind, determines reasonably knot Structure parameter.But owing to the time cost of FInite Element and the cost of manufacture of laboratory method are limited, it is difficult to obtain cutter Structural parameters and the quantitative relationship of structural response, this often makes final design result be difficult to reach optimum.

Summary of the invention

In order to solve existing Ultrasonic machining special type knife method for designing exists high time cost and height is fabricated to This, and it is difficult to the problem quantifying tool structure parameter with structural response relation, the present invention provides one more Fast, the Optimization Design of accurate Ultrasonic machining special type knife structure.

In order to solve above-mentioned technical problem, the present invention adopts the following technical scheme that based on SVR-PSO ultrasonic Machining of special cutter Optimization Design, comprises the following steps:

Step one: utilize ABAQUS finite element secondary exploitation technology to set up cutter parameters and optimize plug-in unit, utilize this Plug-in unit obtains a number of tool structure parameter variable and the sample data of corresponding structural response variable；

Step 2: based on above-mentioned sample data, uses support vector regression method to obtain cutter structure ginseng The approximate model of mapping relations between number and structural response；

Step 3: improve particle cluster algorithm, promotes particle cluster algorithm convergence rate during optimizing；

Step 4: set up the optimization object function of cutter structure design；

Step 5: cutter structure is optimized by the particle cluster algorithm of application enhancements, obtains Tool Design Excellent parameter.

Described step 2 is to use gaussian radial basis function as the kernel function of Support vector regression model, profit With the svr function of the scikit-learn module under Python platform, the sample data obtained is trained Obtain approximate model.

After described step 3 is improved population realize formula:

$\begin{array}{cc}{x}_i^{k+1}=x_i^k+v& \left|v\right|>0.1\end{array}$

$\begin{array}{cc}{x}_i^{k+1}=x_i^k\&PlusMinus;0.1& \left|v\right|<0.1\end{array}$

$v_i^{k+1}={\mathrm{\&omega;v}}_i^k+c_1{r}_1(p_i^k-x_i^k)+c_2{r}_2(p^{k\&CenterDot;g}-x_i^k)$

In formula,Representing location parameter during i-th particle kth time iteration, v represents speed parameter now,Representing speed parameter during i-th particle kth time iteration, what ω represented is inertia weight, inertia weight number Inertia weight linear decrease strategy, p is used in value^k·gRepresent history globally optimal solution,Represent particle i iteration To the history optimal solution of kth time, r₁And r₂It is the random number of scope [0,1] so that this change of iteration input value every time Measurer has certain randomness, c₁And c₂Represent the empirical learning factor and the social learning factor respectively.

In described step 4, optimization object function is:

Y (l, d, m)=1-[0.3 σ (l, d, m)+0.3 (1-U₁(l,d,m))+0.3U₂(l,d,m)

+0.1(1-f(l,d,m))]

In formula, l is tool length, and m is MAT'L mark, and d is cutter thickness, σ (l, d, m) be maximum stress, U₁(l, d m) are tool nose's amplitude, U₂(l, d, m) be point of a knife lateral displacement, and (l, d m) are natural frequency to f.

The present invention is by setting up the approximate model of structural response, and quantitative tool structure parameter of setting up rings with structure The mapping relations answered, for cutter structure is optimized design provides solution more accurately.

Accompanying drawing explanation

Fig. 1 is the maximum stress datagram of training sample；

Fig. 2 is the tool nose's amplitude datagram of training sample；

Fig. 3 is the point of a knife lateral displacement datagram of training sample；

Fig. 4 is the natural frequency datagram of training sample；

Fig. 5 is that cutter optimization designs flowchart.

Detailed description of the invention

Present invention Ultrasonic machining based on SVR-PSO special type knife Optimization Design, comprises the following steps:

Step one: input variable is tool length, cutter material, cutter thickness, output variable is maximum answering Power, resonant frequency, tool nose's amplitude, point of a knife lateral displacement.Between tool length variable selection 25 55mm scope Every 15 of 2mm, thickness is 1.5mm, 2mm two kinds, and material selects YW2 hard alloy and high-speed steel two Plant material.Cutter is joined by the ultrasonic cut cutter assembly emulation plug-in unit utilizing finite element secondary development to set up Numberization traversal emulation obtains the training samples such as maximum stress, tool nose's amplitude, point of a knife lateral displacement, natural frequency This.The data of different dimensions in sample are normalized so that data normalization is in [0,1] or [-1,1] In the range of.

Step 2: use gaussian radial basis function (RBF) as the kernel function of Support vector regression model, The sample data obtained is trained obtaining by the svr function utilizing the scikit-learn module of Python Approximate model.By approximation mould knowable to the match value to the cutter maximum stress that 2mm thickness material is hard alloy The sample point error of fitting of type is less than 0.1%.

Step 3: for the cutter in the present invention, design accuracy is set to 0.1mm can meet reality Situation.Therefore, the randomly generated value of particle is set as the integral multiple of 0.1 in the algorithm, such as particle cluster algorithm In flight speed parameter Vi take 0.1 integral multiple, its less than 0.1 time, its flight speed is reset to Minima 0.1.After improvement population realize formula:

$\begin{array}{cc}{x}_i^{k+1}=x_i^k+v& \left|v\right|>0.1\\ x_i^{k+1}=x_i^k\&PlusMinus;0.1& \left|v\right|<0.1\end{array}---\left(5\right)$

$v_i^{k+1}={\mathrm{\&omega;v}}_i^k+c_1{r}_1(p_i^k-x_i^k)+c_2{r}_2(p^{k\&CenterDot;g}-x_i^k)---\left(6\right)$

In formula,Representing location parameter during i-th particle kth time iteration, v represents speed parameter now,Representing speed parameter during i-th particle kth time iteration, what ω represented is inertia weight, inertia weight number Inertia weight linear decrease strategy, p is used in value^k·gRepresent history globally optimal solution,Represent particle i iteration To the history optimal solution of kth time, r₁And r₂It is the random number of scope [0,1] so that this change of iteration input value every time Measurer has certain randomness, c₁And c₂Represent the empirical learning factor and the social learning factor respectively.Through improving After particle cluster algorithm relatively standard particle group algorithm for, the iterations standard particle to be less than needed for convergence Group's algorithm, improves particle cluster algorithm convergence rate during optimizing.

Step 4: the output of the sample that finite element simulation obtains is carried out data analysis and determines object function Such as formula (2)

Y (l, d, m)=1-[0.3 σ (l, d, m)+0.3 (1-U₁(l,d,m))+0.3U₂(l,d,m)

+0.1(1-f(l,d,m))] (2)

In formula, l is tool length, and m is MAT'L mark (0 represents hard alloy, and 1 represents high-speed steel), and d is Cutter thickness, (l, d m) are maximum stress, U to σ₁(l, d m) are tool nose's amplitude, U₂(l, d m) are the horizontal position of point of a knife Moving, (l, d m) are natural frequency to f.Output function is all [0,1] normal data.So Y (l, d, value m) is [0,1] Interior dimensionless constant.

Then optimization object function is:

min:Y(l,d,m) (3)

$\begin{array}{cc}s.t.& \begin{array}{c}l\&Element;(25,55)\\ d=1.5,2\\ m=0.1\end{array}\end{array}---\left(4\right)$

Step 5: input variable has three parameters in practical problem of the present invention, and only tool length is continuous Variable, material and thickness is discrete variable, by three structural parameters are mapped to setting one-dimensional continuously To solve Discrete Variables Optimization in the variable space (25,145), this variable space is divided into a length of 30 Four parts with corresponding 4 class tool length excursion (25,55) and variation lengths, i.e. (l, d, m) → x. Use support vector machine and particle cluster algorithm Combinatorial Optimization method for designing (SVR-POS) empty in one-dimensional continuous variable Between object function is carried out optimizing, the result that draws after the optimizing result corresponding with cutter parameters is cutter Structural parameters optimal value.

Claims (4)

1. Ultrasonic machining special type knife Optimization Design based on SVR-PSO, it is characterised in that: include following step Rapid:

Step one: utilize ABAQUS finite element secondary exploitation technology to set up cutter parameters and optimize plug-in unit, utilize this Plug-in unit obtains a number of tool structure parameter variable and the sample data of corresponding structural response variable；

Step 2: based on above-mentioned sample data, uses support vector regression method to obtain tool structure parameter And the approximate model of mapping relations between structural response；

Step 3: improve particle cluster algorithm, promotes particle cluster algorithm convergence rate during optimizing；

Step 4: set up the optimization object function of cutter structure design；

Step 5: cutter structure is optimized by the particle cluster algorithm of application enhancements, obtains the optimum of Tool Design Parameter.

Ultrasonic machining special type knife Optimization Design based on SVR-PSO the most according to claim 1, its It is characterised by: described step 2 is to use gaussian radial basis function as the core letter of Support vector regression model Number, utilizes the svr function of the scikit-learn module of Python to be trained the sample data obtained Obtain approximate model.

Ultrasonic machining special type knife Optimization Design based on SVR-PSO the most according to claim 1, its Be characterised by: after described step 3 is improved population realize formula:

$\begin{array}{cc}{x}_i^{k+1}=x_i^k+v& \left|v\right|>0.1\end{array}$

$\begin{array}{cc}{x}_i^{k+1}=x_i^k\&PlusMinus;0.1& \left|v\right|<0.1\end{array}$

$v_i^{k+1}={\mathrm{\&omega;v}}_i^k+c_1{r}_1(p_i^k-x_i^k)+c_2{r}_2(p^{k\&CenterDot;g}-x_i^k)$

In formula,Representing location parameter during i-th particle kth time iteration, v represents speed parameter now,Generation Speed parameter during table i-th particle kth time iteration, what ω represented is inertia weight, and inertia weight is numerically Use inertia weight linear decrease strategy, p^k·gRepresent history globally optimal solution,Represent particle i iteration to kth Secondary history optimal solution, r₁And r₂It is that the random number of scope [0,1] is so that each this variable of iteration input value has There are certain randomness, c₁And c₂Represent the empirical learning factor and the social learning factor respectively.

Ultrasonic machining special type knife Optimization Design based on SVR-PSO the most according to claim 1, its It is characterised by: in described step 4, object function is:

Y (l, d, m)=1-[0.3 σ (l, d, m)+0.3 (1-U₁(l,d,m))+0.3U₂(l,d,m)

+0.1(1-f(l,d,m))]

In formula, l is tool length, and m is MAT'L mark, and d is cutter thickness, and (l, d m) are maximum stress, U to σ₁(l,d,m) For tool nose's amplitude, U₂(l, d, m) be point of a knife lateral displacement, and (l, d m) are natural frequency to f.