CN104283393B - Method for optimizing structure parameter of single-winding magnetic suspension switch reluctance machine - Google Patents

Abstract

The invention discloses a method for optimizing the structural parameters of a single-winding magnetic levitation switched reluctance motor, which belongs to the technical field of magnetic levitation switched reluctance motors. Select the structural parameters to be optimized and determine the sample data set; use the extreme learning machine algorithm to train the sample data set to construct the motor model; take the parameters to be optimized as the optimization object, and optimize the parameters of the motor model with the suspension force and torque as the optimization objectives. The invention uses the extreme learning machine of the single hidden layer feedforward neural network to perform model identification without iteration, and can train the motor model quickly and with high precision, and realizes the coordinated optimization of suspension force and torque by using a multi-objective optimization algorithm parametric design.

Description

A Method for Optimizing Structural Parameters of Single Winding Magnetic Levitation Switched Reluctance Motor

technical field

The invention discloses a method for optimizing structural parameters of a single-winding magnetic suspension switched reluctance motor, belonging to the technical field of magnetic suspension switched reluctance motors.

Background technique

Since the end of the 20th century, the magnetic levitation switched reluctance motor has received extensive attention from researchers. During this period, the magnetic levitation switched reluctance motor with double winding structure is the most researched. The radial force winding and the torque winding are stacked on the same stator pole, so that the radial force winding does not occupy an independent axial space. Because this kind of motor has the advantages of no lubrication, no wear, high power and ultra-high speed operation, it is very suitable for aerospace, high-speed machine tools, flywheel energy storage and other fields. However, the strong coupling between the main winding and the suspension winding in the magnetic levitation switched reluctance motor with double winding structure makes the mathematical modeling and control algorithm of the motor more complicated; the additional suspension winding increases the difficulty of the motor structure design; the suspension The addition of windings results in additional power amplifiers and associated electrical subsystems, increasing the complexity of the control circuit design. Aiming at the above-mentioned shortcomings of double-winding magnetic levitation motors, NASA, Dresden Technical University in Germany and Kyungsung University in South Korea have successively carried out research on single-winding magnetic levitation switched reluctance motors. The structure of the single-winding magnetic levitation switched reluctance motor is basically the same as that of the switched reluctance motor. The main difference is that the coils on multiple stator poles in the switched reluctance motor are excited in series, while any stator pole in the single-winding magnetic levitation switched reluctance motor The coils on the motor are independently excited to achieve the purpose of levitating and rotating the motor rotor.

For the structural optimization design of the magnetic levitation switched reluctance motor, the methods that have been proposed include: using the design method combining theoretical analysis and finite element simulation to optimize the structure of the magnetic levitation switched reluctance motor, but because the parameter optimization requires a large number of call calculations model to obtain its output, so the calculation efficiency is low; another method is: use the learning algorithm training of support vector machine to establish a magnetic levitation switched reluctance motor model, and then use the genetic algorithm to optimize the performance of a certain motor for the motor model. Optimizing, this method solves the problem of low computational efficiency of the former, but in the absence of an intelligent algorithm to optimize the parameters of the support vector machine, the accuracy of the motor model established by the support vector machine training is average, and it is difficult to meet the actual needs. In addition, as a single-objective optimization algorithm, the genetic algorithm can only achieve a single maximization of the suspension force or torque, and the separate optimization of the two will inevitably weaken each other.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method for optimizing the structural parameters of a single-winding magnetic levitation switched reluctance motor in view of the shortcomings of the above-mentioned background technology. The data training of the model is carried out with the extreme learning machine of the single hidden layer feedforward neural network, and the multi-objective optimization is realized by using the non-dominated sorting genetic algorithm.

The present invention adopts following technical scheme for realizing above-mentioned purpose of the invention:

A method for optimizing structural parameters of a single-winding magnetic levitation switched reluctance motor, comprising the following steps:

Step 1. Calculate the initial structural parameters of the single-winding magnetic levitation switched reluctance motor. The initial structural parameters include: the initial value of the outer diameter of the rotor D _a0 , the initial value of the inner diameter of the rotor D _i0 , the initial value of the actual length of the iron core l _a0 , and the initial value of the outer diameter of the stator D _s0 , initial value of air gap length g ₀ , initial value of stator pole arc β _s0 , initial value of rotor pole arc β _r0 , initial value of stator yoke thickness h _cs0 , initial value of rotor yoke thickness h _cr0 ;

Step 2, perform finite element simulation on the motor model, select the structural parameters to be optimized, and determine the sample data set;

Step 3, use the extreme learning machine algorithm to train the sample data set to build the motor model,

The extreme learning machine algorithm: based on the principle that the number of hidden layer nodes is less than the number of training samples, the number of hidden layer nodes is determined by a fixed-step search, Sigmoid or RBF is selected as the activation function, and the parameter to be optimized is the extreme learning machine The input data, and the suspension force and torque corresponding to the value of the parameters to be optimized are used as the output data of the extreme learning machine to start training the sample data set;

Step 4, taking the parameters to be optimized as the optimization object, and optimizing the parameters of the motor model with the levitation force and torque as the optimization objectives.

Further, step 2 performs finite element simulation according to the following method:

Step 2-1, establish the finite element model of the single-winding maglev switched reluctance motor with the initial structural parameters calculated in step 1, and obtain the torque current component i _m by simulating the finite element model: establish the initial structural parameters in Ansoft finite element software The single-winding magnetic levitation switched reluctance motor model in , under the condition that the three phases are turned on in turn and the rated voltage U _N is applied, the simulation obtains the effective value of the current of the one-phase winding only during the conduction period, which is left for the subsequent establishment of the single-winding magnetic levitation The finite element model of the switched reluctance motor is used for excitation, and the effective value of the current of the one-phase winding only during the conduction period is the torque current component i _m ;

Step 2-2, establish the switched reluctance motor model with the initial structural parameters in the Rmxprt module of Ansoft, record the rotor mass m _r automatically calculated by the software, and use the rotor mass m _r to compare the single-winding magnetic levitation switched reluctance in Matlab/Simulink The mass of the rotor in the suspension system of the motor is set, and then the radial suspension force value range F for successful rotor suspension is obtained by the simulation of the suspension system. In Ansoft, the model of the single-winding magnetic levitation switched reluctance motor is established with the initial structural parameters. The numerical value interval F of the levitation force is simulated to obtain approximate values of the levitation current components i _sα and i _sβ ;

Step 2-3, under the condition that the actual length of the iron core l _a and the outer diameter of the stator D _s are constant, apply the excitation current to the winding in the finite element model, the excitation current includes the torque current component i _m and the levitation current component i _sα , i _sβ , changing the values of rotor outer diameter D _a , rotor inner diameter D _i , air gap length g, stator pole arc β _s , rotor pole arc β _r , stator yoke thickness h _cs , and rotor yoke thickness h _cr Obtain the rule that levitation force and torque vary with other structural parameters, and select structural parameters that have different effects on levitation force and torque as parameters to be optimized x ₁ , x ₂ ,… _xi ,…,x _n , i≤n,n =1,…,7, x _i is any structural parameter among rotor outer diameter, rotor inner diameter, air gap length, stator pole arc, rotor pole arc, stator yoke thickness, and rotor yoke thickness. Levitation force and torque vary with the structure The structural parameters whose parameters are monotonically increasing or monotonically decreasing do not need to be optimized, and the structural parameters whose suspension force monotonically increases (or monotonically decreases) and torque monotonically decreases (or monotonically increases) should be selected as parameters to be optimized;

In step 2-4, the values of different parameters to be optimized and their corresponding levitation force F _s and torque T are combined into a sample data set (x ₁ ,x ₂ ,…,x _n ,F _s ,T), and the training data The set is used as the sample data set of the extreme learning machine.

Further, step 4 uses the non-dominated sorting genetic algorithm to perform multi-objective optimization on the motor model.

As a further optimization scheme for the optimization method of the structural parameters of the single-winding magnetic levitation switched reluctance motor, the structural parameters of the single-winding magnetic levitation switched reluctance motor are calculated according to the traditional calculation method of the structural parameters of the switched reluctance motor. In order to further illustrate the structural parameters of the traditional switched reluctance motor Calculation method, structure parameters are defined as follows:

Determine the rated power, rated speed, and efficiency according to the design and application of the single-winding magnetic levitation switched reluctance motor, and obtain the magnetic load, electric load, winding current coefficient, square wave current coefficient, coefficient 1, coefficient 2, and coefficient based on the empirical value range of each variable 3. Coefficient 4, Coefficient 5, specific value of air gap length, step 1 according to formula (1):

Determine the initial value of the structural parameters, k _i is the winding current coefficient, k _m is the square wave current coefficient, P _N is the rated power, n _N is the rated speed, η is the efficiency, B _δ is the magnetic load, A is the electric load, b _ps is the stator pole width, b _pr is the rotor pole width, λ ₁ , λ ₂ , λ ₃ , λ ₄ , λ ₅ are coefficients, λ ₁ =0.5～3.0, λ ₂ =0.5～0.55, λ ₃ =0.4～ 0.5, λ ₄ =1.2～1.4, λ ₅ =1.2～1.4, km _≈0.8 , _ki ≈0.5, B _δ ＝0.3～0.6, A＝15000～50000,

In order to meet the self-starting capability requirements of single-winding maglev switched reluctance motors, the values of rotor pole arc β _r and stator pole arc β _s must satisfy formula (2):

N _r is the number of rotor teeth, m is the number of current phases,

So far, the specific structural parameters of the single-winding magnetic levitation switched reluctance motor are obtained, and the structural parameters are recorded as: the initial value of the rotor outer diameter D _a0 , the initial value of the rotor inner diameter D _i0 , the initial value of the actual length of the iron core l _a0 , the outer diameter of the stator diameter initial value D _s0 , air gap length initial value g ₀ , stator pole arc initial value β _s0 , rotor pole arc initial value β _r0 , stator yoke thickness initial value h _cs0 , rotor yoke thickness initial value h _cr0 , where D _si0 =D _a0 +2g ₀ , each structural parameter constitutes the initial structural parameter.

The present invention adopts the above-mentioned technical scheme, and has the following beneficial effects:

(1) The extreme learning machine of the single hidden layer feedforward neural network is used for data training of the model without iteration;

(2) The parameter design of synergistic optimization of suspension force and torque is realized.

Description of drawings

Fig. 1 shows the specific process of the structural optimization design of the single-winding magnetic levitation switched reluctance motor proposed by the present invention.

Fig. 2(a), Fig. 2(b), Fig. 2(c), Fig. 2(d), Fig. 2(e), Fig. 2(f) are the coincident positions of stator and rotor pole center lines, and the rotor has no eccentricity The suspension force and the relationship curve of each parameter.

Figure 3(a), Figure 3(b), Figure 3(c), Figure 3(d), Figure 3(e), Figure 3(f) are the relationship curves between the average torque and various parameters.

Figure 4(a), Figure 4(b), and Figure 4(c) are the cross-sectional view of the motor before optimization, the cross-sectional view of the optimized motor 1, and the cross-sectional view of the optimized motor 2, respectively.

Figure 5 shows the comparison of the motor output suspension force before and after optimization.

Figure 6 shows the comparison of motor output torque before and after optimization.

detailed description

The technical solution of the specific process shown in FIG. 1 will be described in detail below with reference to FIGS. 2 to 6 .

1. When the predetermined performance parameters of the motor are: rated power P _N = 1.1kW, rated speed n _N = 2000r/min, rated voltage U _N = 220V, rated efficiency η = 0.8, according to SRM (Switched Reluctance Motor, switched reluctance motor ) The structural parameters of the traditional calculation method can be obtained 12/8 structure SWBSRM (Single Winding Bearingless Switched Reluctance Motor, single winding magnetic levitation switched reluctance motor) The initial value of the structural parameters is: rotor outer diameter D _s = 137mm, rotor outer diameter D _a = 70mm , rotor inner diameter D _i =31.5mm, stator yoke thickness h _cs =10mm, rotor yoke thickness h _cr =10mm, air gap length g=0.3mm, stator pole arc β _s =15°, rotor pole arc β _r =15° , The actual length of iron core l _a =70mm, each pole stator winding N=85 turns.

2. Determine the torque current component im = _4.7A by the above method, because the principle of the two levitation force components in the vertical coordinate system is the same, so here only set the levitation current component i _sx = 1.88A in the direction of the x coordinate axis.

3. Set the motor excitation in the finite element simulation with i _m = 4.7A, i _sx = 1.88A, and i _sy = 0A, and analyze the change law of the levitation force and torque obtained with the increase of the parameters when the structural parameters change, specifically:

(1) Levitation force increases monotonously with the increase of rotor inner diameter D _i , rotor outer diameter _Da , stator yoke thickness h _cs , and rotor yoke thickness h _cr ; levitation force decreases monotonously with the increase of air gap length g; in addition, the stator As the pole arc β _s increases, the levitation force first increases and then tends to be stable, and the change law of the levitation force with the rotor pole arc β _r is consistent with β _s . Figure 2(a), Figure 2(b), Figure 2(c), Figure 2(d), Figure 2(e) and Figure 2(f) show the relationship between force and various parameters;

(2) The torque increases monotonously with the increase of the rotor outer diameter D _a and the stator yoke thickness h _cs ; the torque decreases with the increase of the rotor inner diameter D _i , the rotor yoke thickness h _cr , and the air gap length g; the torque decreases with the stator pole The arc β _s and the rotor pole arc β _r increase, and the trend is to increase first and then decrease. Among them, the relationship curves between the average torque and various parameters are shown in Figure 3(a), Figure 3(b), Figure 3(c), Shown in Figure 3(d), Figure 3(e), and Figure 3(f);

Finally, the rotor inner diameter D _i , rotor yoke thickness h _cr , stator pole arc β _s , and rotor pole arc β _r are selected as optimization objects.

4. Finite element simulation, changing the inner diameter D _i , rotor yoke thickness h _cr , stator pole arc β _s , rotor pole arc β _r , the output is (F _s , T), and the sample data (D _i ,h _cr , β _s , β _r , F _s , T).

5. Use ELM (Extrem Learning Machine, extreme learning machine) to train the above sample data to obtain a motor model, the input of which is (D _i , h _cr , β _s , β _r ), and the output is (F _s , T).

6. Use NSGA (Non dominated Sorting Genetic Algorithm, Non-Dominated Sorting Genetic Algorithm) to optimize and obtain the optimized single-winding magnetic levitation switched reluctance motor, because the optimal result obtained by optimization is a set (the "optimal solution obtained by multi-objective optimization) " is "generalized optimal", because the optimization objectives are likely to contradict each other, so the "optimal solution" obtained is often a set), and two specific examples are:

(1) Optimized motor 1: rotor outer diameter D _s =137mm, rotor outer diameter D _a =70mm, rotor inner diameter D _i =31.5mm, stator yoke thickness h _cs =10mm, rotor yoke thickness h _cr =12mm, air gap length g=0.3mm, stator pole arc β _s =23.42°, rotor pole arc β _r =20.53°, actual iron core length l _a =70mm, stator winding N=85 turns per pole;

(2) Optimize motor 2: rotor outer diameter D _s =137mm, rotor outer diameter D _a =70mm, rotor inner diameter D _i =40mm, stator yoke thickness h _cs =10mm, rotor yoke thickness h _cr =11.329mm, air gap length g=0.3mm, stator pole arc β _s =23.287°, rotor pole arc β _r =20.14°, core actual length l _a =70mm, stator winding N=85 turns per pole.

7. Figure 4(a), Figure 4(b), and Figure 4(c) show the cross-sectional views of the motor before and after optimization, and Figure 5 and Figure 6 show the curves of the levitation force and torque before and after optimization with the rotor angle. By comparison, we can see that

8. Shown are the change curves of levitation force and torque with the rotor angle before and after optimization. The comparison shows that the curve of optimized motor 1 and the curve of optimized motor 2 basically coincide. Within the range, that is, at all rotor position angles (-22.5°～22.5°), the suspension force increases by about 200N, which greatly enhances the radial suspension stability of the motor. In other words, under the same suspension force, the suspension force can be reduced The current input required for the motor suspension process is small, thus reducing the power consumption of the motor's radial suspension operation; at the same time, in the range of rotor position angle (-22.5°～-13.5°) and (13.5～22.5°), the motor 1 , 2 The output torque of the original motor is increased by about 2N m, which shows that the optimized motor can adapt to a larger load torque, and the torque ripple has been greatly improved compared with the original motor, which makes the optimized motor speed It is more stable, and to a certain extent reduces the limitation of large torque ripple on the application of motors. Therefore, through the implementation of this optimization method, the radial levitation force and output torque of the motor are greatly enhanced, the radial levitation performance of the motor is improved, and the operating power consumption and torque ripple are reduced.

The above-mentioned embodiment only optimizes the parameters of the 12/8 single-winding magnetic levitation switched reluctance motor, and the parameters of the other single-winding magnetic levitation switched reluctance motors can be optimally designed by using the technical solution of the present invention.

In summary, the present invention has the following beneficial effects:

(1) The data training of the model is carried out with the extreme learning machine of the single hidden layer feedforward neural network, without iteration, and the motor model can be trained quickly and with high precision;

(2) The multi-objective algorithm is used to realize the optimal parameter design of suspension force and rotation rejection at the same time.

Claims (4)

1. a kind of optimization method of simplex winding magnetic suspension switched reluctance motor structural parameters is it is characterised in that comprise the steps：

Step 1, calculates the initial structure parameter of simplex winding magnetic suspension switched reluctance motor, and initial structure parameter includes：Outside rotor Footpath initial value D_a0, rotor internal diameter initial value D_i0, iron core physical length initial value l_a0, stator outer diameter initial value D_s0, at the beginning of gas length Initial value g₀, stator polar arc initial value β_s0, rotor pole arc initial value β_r0, stator yoke thickness initial value h_cs0, rotor yoke thickness initial value h_cr0；

Step 2, carries out finite element simulation to motor model, chooses structural parameters to be optimized, determines sample data set；

Step 3, builds motor model using extreme learning machine Algorithm for Training sample data set,

Described extreme learning machine algorithm：Number of training is less than as principle with node in hidden layer, in the way of fixed step size search Determine node in hidden layer, select Sigmoid or RBF as excitation function, defeated with parameter to be optimized as extreme learning machine Enter data, the output data with suspending power corresponding with parameter values to be optimized and torque as extreme learning machine starts to train sample Notebook data collection；

Step 4, with parameter to be optimized as optimization object, with suspending power, that torque enters line parameter for optimization aim to motor model is excellent Change.

2. the optimization method of simplex winding magnetic suspension switched reluctance motor structural parameters according to claim 1, its feature exists In step 2 carries out finite element simulation as follows：

Step 2-1, sets up the finite element of simplex winding magnetic suspension switched reluctance motor with the calculated initial structure parameter of step 1 Model, obtains torque current component i to FEM (finite element) model emulation_m；

Step 2-2, obtains suspending power numerical intervals F with the rotor quality emulation in FEM (finite element) model, then by suspending power numerical value area Between F emulation obtain levitating current component i_sα、i_sβ；

Step 2-3, in the case that iron core physical length, stator outer diameter are constant, applies excitation to the winding in FEM (finite element) model Electric current, exciting current includes torque current component i_mWith levitating current component i_sα、i_sβ, simulation analysis obtain suspending power, torque with The rule of remaining structural parameters change, chooses the structural parameters different on suspending power, torque impact as parameter x to be optimized₁, x₂,…x_i,…,x_n, i<N, n=1 ..., 7, x_iFor rotor diameter, rotor internal diameter, gas length, stator polar arc, rotor pole arc, Any one structural parameters in stator yoke thickness, rotor yoke thickness；

Step 2-4, by different parameter values to be optimized and its corresponding suspending power F_s, torque T assemble sample data set (x₁, x₂,…,x_n,F_s,T).

3. the optimization method of simplex winding magnetic suspension switched reluctance motor structural parameters according to claim 1 and 2, its feature It is, step 4 carries out multi-objective optimization using non-dominated sorted genetic algorithm to motor model.

4. the optimization method of simplex winding magnetic suspension switched reluctance motor structural parameters according to claim 3, its feature exists In step 1 is according to formula：

$\left\{\begin{array}{c}{D}_a=\sqrt[3]{\frac{6.1}{1.05{\mathrm{\&lambda;}}_1{B}_{\mathrm{\&delta;}}A}\frac{k_i}{k_m}\frac{P_N(1+\mathrm{\&eta;})/2\mathrm{\&eta;}}{n_N}}\\ l_a={\mathrm{\&lambda;}}_1{D}_a\\ D_s=D_a/{\mathrm{\&lambda;}}_2\\ D_i={\mathrm{\&lambda;}}_3{D}_a\\ D_{si}=D_a+2g\\ b_{ps}=D_{si}\sin \frac{{\mathrm{\&beta;}}_s}{2}\\ b_{pr}=D_a\sin \frac{{\mathrm{\&beta;}}_r}{2}\\ h_{cr}={\mathrm{\&lambda;}}_4\frac{b_{pr}}{2}\\ h_{cs}={\mathrm{\&lambda;}}_5\frac{b_{ps}}{2}\end{array}\right.$

Determine structural parameters initial value, k_iFor winding current coefficient, k_mFor square wave current coefficient, P_NFor rated power, n_NFor specified Rotating speed, η is efficiency, B_δFor magnetic loading, A is electric load, b_psFor stator pole-core width, b_prFor rotor pole-core width, λ₁、λ₂、λ₃、 λ₄、λ₅For coefficient,

For meeting the requirement of simplex winding magnetic suspension switched reluctance motor self-startup ability, rotor pole arc β_r, stator polar arc β_sValue must Must meet:

N_rFor rotor tooth pole number, m is the electric current number of phases.