A spring structure design method and system for a gravity sensitive unit

Abstract

The present invention relates to a spring structure design method and system for a gravity sensitive unit, wherein the system is based on a method, and the method creatively uses a combination of an optimization model and an adaptive elite learning particle swarm algorithm to optimize the design, and after multiple simulation calculations, the sensitivity and linear fit of the structure are used as evaluation indicators to determine the optimal parameter design of a free geometric anti-spring gravity sensitive unit, thereby improving the performance of the structure, solving the technical problem of insufficient performance of the gravity sensitive unit obtained by the existing design method, and optimizing it in terms of geometric layout, thereby improving its measurement accuracy.

Description

Technical Field

[0001] The present invention relates to the field of sensor science and technology, and in particular to a spring structure design method and system for a gravity sensitive unit. Background Art

[0002] The free geometry anti-spring gravity sensitive unit is a physical quantity conversion unit in the micro-electro-mechanical system (MEMS) gravimeter. It has the advantages of light weight, small size, low cost, and low power consumption. It is one of the core components of the MEMS gravimeter, and its performance will directly affect the measurement accuracy of gravity acceleration. The performance of the gravity sensitive unit is closely related to its geometric layout. The gravity sensitive unit is usually a simple geometric layout, so the design freedom is not large, which further constrains the performance improvement. The free geometry layout can fundamentally break through the constraints of simple geometry and obtain an ultra-sensitive sensitive unit.

[0003] A free geometry anti-spring gravity sensitive unit can be formed by adding a free geometry anti-spring gravity sensitive unit to the design. Due to the characteristics of the anti-spring, there are many factors that affect the sensitivity of the free geometry anti-spring gravity sensitive unit, the most important of which is the configuration of the spring.

[0004] However, the role of gravity sensitive units in geophysics and precision measurement auxiliary systems requires optimization of their geometric layout to improve their measurement accuracy. However, the gravity sensitive units obtained by existing design methods often have technical problems of insufficient performance, resulting in unsatisfactory measurement accuracy. Summary of the invention

[0005] The technical problem to be solved by the present invention is how to improve the performance of a gravity sensitive unit by designing a spring structure. In order to overcome the above defects of the prior art, the present invention provides a spring structure design method and system for a gravity sensitive unit, including a spring structure design method for a gravity sensitive unit and a spring structure design system for a gravity sensitive unit.

[0006] The first technical solution of the present invention is to provide a spring structure design method for a gravity sensitive unit, and the technical solution is as follows:

[0007] A method for designing a spring structure of a gravity sensitive unit, wherein the gravity sensitive unit comprises a sensitive mass block and four springs connected to the sensitive mass block via two connecting rods;

[0008] It is characterized by comprising the following steps:

[0009] Step 1, setting the shapes of the four springs to the shapes of the free curves;

[0010] Step 2: using the coordinates of at least five control points of the free curve and the structural parameters of the spring as decision parameters, wherein the structural parameters include the width and structural compression of the spring;

[0011] Step 3: determining the value space of the decision parameter based on the configuration requirements of the spring, and uniformly selecting multiple parameter vectors in the value space;

[0012] Step 4: construct corresponding simulation parameters based on each of the parameter vectors to perform structural simulation on the gravity sensitive unit through a multi-physics field simulation module to obtain a simulation model corresponding to each of the parameter vectors;

[0013] Step 5: Perform model calculation on the simulation model corresponding to each of the parameter vectors by using a finite element analysis method to obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the absolute value of the displacement difference;

[0014] Step 6: construct a search space based on the value space, take the linear combination of the linear regression coefficient and the absolute value of the displacement difference as the fitness, take the fitness as the decision function, and obtain the optimization model;

[0015] Step 7: Based on the corresponding relationship obtained in step 5, an adaptive elite learning particle swarm algorithm is used to solve the optimization model to obtain an optimal solution;

[0016] Step 8: Use the shape of the free curve determined by the optimal solution as the shape of the spring to complete the design.

[0017] This method first constructs a finite element simulation model of a gravity sensitive unit, and parameterizes the spring structure of the sensitive unit, and then designs the objective function according to the absolute value of the displacement difference and the linear regression coefficient, and constructs a search space through the value space. At the same time, the present invention creatively uses the combination of the optimization model and the adaptive elite learning particle swarm algorithm to optimize the design. After multiple simulation calculations, the sensitivity of the structure (i.e., the absolute value of the displacement difference) and the linear fit are used as evaluation indicators to determine the optimal parameter design of the free geometric anti-spring gravity sensitive unit, thereby improving the performance of the structure, solving the technical problem of insufficient performance of the gravity sensitive unit obtained by the existing design method, and optimizing it in geometric layout, thereby improving its measurement accuracy. In addition, since the adaptive elite learning particle swarm algorithm has a simple principle, is easy to implement, and requires fewer parameters to be adjusted, compared with other computational intelligence algorithms, it has a faster convergence speed and can obtain a set of better solutions in a shorter time, greatly improving the computational efficiency.

[0018] In a possible implementation manner, the free curve is a B-spline curve or a Bezier curve;

[0019] In the B-spline curve or Bezier curve, changing one point does not affect all curve segments, thus greatly saving simulation time.

[0020] In a possible implementation manner, the step five includes the following steps:

[0021] Step 5.1, meshing the simulation models corresponding to the parameter vectors according to the structural characteristics of the gravity sensitive units to obtain the corresponding meshed models;

[0022] Step 5.2, constructing the equilibrium equations for the grid models corresponding to the parameter vectors respectively, and obtaining the corresponding equilibrium equations with the sensitive axial displacement of the sensitive mass block as the unknown quantity;

[0023] Step 5.3, with an initial step size of 10 Gal, increase the gravity acceleration g from 0 Gal to 970 Gal, and then assign a value to the gravity acceleration g according to the rule of increasing the gravity acceleration g from 974 Gal to 984 Gal with a step size of 0.1 Gal, so that the solver solves the equilibrium equations corresponding to each of the parameter vectors respectively, and obtains the linear regression relationship between the sensitive axial displacement of the sensitive mass block corresponding to each of the parameter vectors and the gravity acceleration g;

[0024] Step 5.4, respectively calculating the linear regression coefficients of the linear regression relationships corresponding to the parameter vectors obtained in step 5.3, and obtaining the linear regression coefficients corresponding to the parameter vectors;

[0025] Step 5.5, obtaining the corresponding relationship between the decision parameter and the linear regression coefficient based on all the parameter vectors and the linear regression coefficient corresponding to each of the parameter vectors by a numerical approximation method;

[0026] Step 5.6, obtaining the displacement difference absolute value corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector obtained in step 5.3;

[0027] The absolute value of the displacement difference is the absolute value of the difference between the displacement of the sensitive mass block along the sensitive axis when the gravitational acceleration g value is 974Gal and the displacement of the sensitive mass block along the sensitive axis when the gravitational acceleration g value is 984Gal;

[0028] Step 5.7, obtaining the corresponding relationship between the decision parameter and the absolute value of the displacement difference based on all the parameter vectors and the absolute value of the displacement difference corresponding to each of the parameter vectors by a numerical approximation method;

[0029] Furthermore, through finite element analysis, a calculation method for the absolute value of the displacement difference in the sensitive axis direction of the sensitive mass block under the action of gravitational acceleration and the linear regression determination coefficient of the displacement and gravitational acceleration within the working range is constructed, and simulation calculations are performed to obtain the results. In this way, the objective function used to optimize the gravity-sensitive unit structure of the spring is constructed, forming a reasonable optimization model, which provides a basis for subsequent processing and achieves predictability of accuracy.

[0030] In a possible implementation manner, the step seven includes the following steps:

[0031] Step 7.1, use a population size of N The particle swarm randomly takes the search space N vectors as the initial positions of the particles, assign initial velocities to the particles respectively, and set the maximum number of iterations;

[0032] Step 7.2, using the position update mode of the particle swarm optimization algorithm as the position update mode of the particles, and defining four strategies to generate learning objectives in four different ways to guide the particles to perform search updates, the four strategies being C-strategy, O-strategy, D-strategy and A-strategy;

[0033] Step 7.3, dividing the particle group into four subgroups, and executing the C-strategy, the O-strategy, the D-strategy, and the A-strategy respectively in a one-to-one allocation manner, and taking the initial position and initial velocity of each particle as its last position update result and velocity update result respectively;

[0034] Step 7.4: Let the particles in each subgroup obtain their own learning goals according to their respective executed strategies, and then obtain the current speed update result and current position update result of each particle based on the last position update result and speed update result of each particle and the current learning goal of each particle;

[0035] Step 7.5, respectively obtain the decision parameters of this time with the current position update result of each particle, obtain the current linear regression coefficient of each particle using the corresponding relationship between the decision parameters and the linear regression coefficient, and simultaneously obtain the current absolute value of the displacement difference of each particle using the corresponding relationship between the decision parameters and the absolute value of the displacement difference;

[0036] Step 7.6, respectively determine whether the current position update result of each particle belongs to the search space, remove all particles that do not belong to the search space, take all particles that belong to the search space as the current remaining particle group, and perform fitness calculations to obtain the fitness of each particle in the current remaining particle group;

[0037] Step 7.7, sorting all particles of the remaining particle swarm from best to worst according to the current individual historical optimal fitness of all particles of the remaining particle swarm, and obtaining a sorting result;

[0038] Step 7.8: According to the sorting result of step 7.7, select several particles from the remaining particle group as elite particles in descending order, determine the subgroups to which these elite particles belong, and traverse these elite particles in the following manner to compete and eliminate them:

[0039] If an elite particle belongs to subgroup s, a particle is randomly selected from all subgroups except subgroup s, and the individual historical optimal fitness of the elite particle is compared with that of the selected particles. If the elite particle wins, the current historical optimal position of the elite particle is used as the final position update result of the selected particles. Otherwise, the speed update result and position update result of the selected particle are retained, where s=1, 2, 3, 4;

[0040] Step 7.9, determine whether the maximum number of iterations has been reached. If so, the optimal fitness obtained this time and the particle position corresponding to the optimal fitness are used as the optimal solution of the optimization model according to the sorting result; if not, the remaining particle group is used as the particle group updated next time without changing the learning target update strategy executed by each particle therein, and the current speed update result and the current position update result of each particle therein are used as their respective last speed update result and last position update result, and return to execute step 7.4;

[0041] This scheme uses four strategies to generate learning objectives in four different ways to guide particles to search and update, and after obtaining the current velocity update results and position update results of each particle, it conducts competition and elimination among particles in the manner of steps 7.8. This not only ensures the accuracy and reliability of the solution, but also ensures that the algorithm has strong convergence.

[0042] In a possible implementation, in step 7.2:

[0043] The C-strategy is: randomly select two particles, obtain the particle with the best current individual historical optimal fitness among the two particles, perform binomial crossover between the individual historical optimal position of this particle and the individual historical optimal position of the particle currently executing the update, and obtain the learning target of the particle currently executing the update accordingly;

[0044] The O-strategy is: using the adjustment mode of the ring topology structure to adjust the positional relationship between the particles, so that each particle has two surrounding particles, each particle is connected with its two surrounding particles to form a small area of ​​the particle, and the current local historical optimal position of the particle in each small area is found, and the current global historical optimal position is found by comparing the current local historical optimal position of the particle in each small area, and the historical optimal position of the particle at the current global historical optimal position is orthogonally combined with the position of the particle currently performing the update, so as to construct the learning target;

[0045] The D-strategy is: constructing a learning goal with the speed update strategy of the particle swarm optimization algorithm;

[0046] The A-strategy is: using the adjustment mode of the ring topology structure to adjust the positional relationship between the particles, so that each particle has two surrounding particles, each particle is connected with its two surrounding particles to form a small area of ​​the particle, and the learning target is constructed by the weighted average of the individual historical optimal positions of each small area;

[0047] in,

[0048] Each speed update result of particles executing other than the D-strategy is composed of the last speed update result item and the difference item between the current learning target of the particle and the current position of the particle. The particles executing the D-strategy update their speed in the speed update mode of the particle swarm optimization algorithm.

[0049] Among the four proposed speed update strategies, A-strategy and C-strategy are suitable for global search, while D-strategy and O-strategy are suitable for local search. They will allow some particles with better performance to learn from more targets, thereby jumping out of the local optimum or finding a better solution. At the same time, they can also maintain the diversity of the population, because when particles in other subgroups are eliminated, elite particles will take its place. In the next search, there will be two identical particles in the population that will execute completely different strategies, so the solutions searched must be different.

[0050] In a possible implementation, after executing step 7.3, the following steps need to be performed before executing step 7.4:

[0051] The position index of each particle in each subgroup is marked respectively;

[0052] Furthermore, a recording scheme for each particle is provided, which is beneficial to improving the efficiency of updating the design process.

[0053] In a possible implementation, when executing step 7.8, when some particles are selected from each subgroup for comparison with elite particles for the first time, the position indexes of these selected particles relative to their respective subgroups are recorded, so that from the next time onwards, particles are selected from these position indexes for comparison with elite particles each time;

[0054] Compared with the existing technology, the above scheme is conducive to maintaining population diversity and preventing premature convergence.

[0055] Another technical solution of the present invention is to provide a spring structure design system for a gravity sensitive unit, and the technical solution is as follows:

[0056] A spring structure design system for a gravity sensitive unit, based on the spring structure design method for a gravity sensitive unit in the present invention, comprises:

[0057] An optimization building module, used for building and / or storing an optimization model corresponding to the spring design of the gravity sensitive unit;

[0058] A calculation module, used to obtain an optimal solution corresponding to the optimization model, and obtain shape design parameters of the spring based on the optimal solution;

[0059] in,

[0060] The optimization construction module is communicatively connected with the calculation module.

[0061] This system is based on the spring structure design method of the gravity sensitive unit of the present invention, adopts the optimization construction module to realize the optimization modeling, and finally obtains the optimal decision parameters corresponding to the optimization model through the mutual cooperation with the calculation module to complete the shape design of the spring. In addition, it solves the technical problem of insufficient performance of the gravity sensitive unit obtained by the existing design method, realizes the optimization of the geometric layout, and improves the measurement accuracy.

[0062] In a possible implementation, the optimization construction module includes:

[0063] A parameter module is configured to use the coordinates of at least five control points of the free curve and structural parameters of the spring including the width of the spring and the structural compression amount of the spring as decision parameters;

[0064] A value taking module, configured to determine a value taking space of the decision parameter based on the configuration requirements of the spring;

[0065] A simulation module is configured to uniformly select a plurality of parameter vectors in the value space, and perform structural simulation on the gravity sensitive unit to obtain a simulation model corresponding to each of the parameter vectors;

[0066] A corresponding module is configured to perform model calculation on the simulation model corresponding to each of the parameter vectors by a finite element analysis method, so as to obtain a corresponding relationship between the decision parameter and the linear regression coefficient and a corresponding relationship between the decision parameter and the absolute value of the displacement difference;

[0067] A model module is configured to construct a search space, and use a linear combination of the linear regression coefficient and the absolute value of the displacement difference as fitness, and use the fitness as a decision function to obtain an optimization model;

[0068] in,

[0069] The value module is connected to the parameter module for communication, the simulation module is connected to the value module for communication, the corresponding module is connected to the simulation module for communication, and the model module is connected to the corresponding module for communication;

[0070] Then, the optimal model corresponding to the spring design of the gravity-sensitive unit is constructed and / or stored through parameter setting of the parameter module, constraint determination of the value module, structural simulation of the simulation module, corresponding relationship generation of the corresponding module and final model construction of the model module in sequence, which has execution orderliness and model accuracy, thereby improving the efficiency of model construction.

[0071] In a possible implementation, the corresponding module includes:

[0072] The grid unit is configured to grid the simulation model corresponding to each of the parameter vectors according to the structural characteristics of the gravity sensitive unit to obtain the corresponding grid models;

[0073] A balance unit is configured to construct a balance equation for each grid model corresponding to each of the parameter vectors to obtain a corresponding balance equation;

[0074] The solver is set to increase the gravity acceleration g from 0Gal to 970Gal with an initial step size of 10Gal, and then assign a value to the gravity acceleration g according to a rule of increasing the gravity acceleration g from 974Gal to 984Gal with a step size of 0.1Gal, respectively solve the equilibrium equations corresponding to the parameter vectors, and obtain a linear regression relationship between the sensitive axial displacement of the sensitive mass block corresponding to each of the parameter vectors and the gravity acceleration g;

[0075] A regression unit is configured to calculate the linear regression coefficients of each linear regression relationship according to the linear regression relationships corresponding to each of the parameter vectors, obtain the linear regression coefficients corresponding to each of the parameter vectors, and obtain the corresponding relationship between the decision parameter and the linear regression coefficient based on all of the parameter vectors and the linear regression coefficients corresponding to each of the parameter vectors;

[0076] An absolute unit is configured to obtain the absolute value of the displacement difference corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector, and obtain the corresponding relationship between the decision parameter and the absolute value of the displacement difference based on all the parameter vectors and the absolute value of the displacement difference corresponding to each parameter vector;

[0077] in,

[0078] The grid unit is communicatively connected to the simulation module, the balancing unit is communicatively connected to the grid unit, the solver is communicatively connected to the balancing unit, the regression unit is communicatively connected to the solver, the absolute unit is communicatively connected to the solver, and the model module is communicatively connected to the regression unit and the absolute unit;

[0079] Then, a method for calculating the absolute value of the displacement difference in the sensitive axis direction of the sensitive mass block under the action of gravity acceleration and the linear regression determination coefficient of the displacement and gravity acceleration within the working range is constructed through finite element analysis, and simulation calculations are performed to obtain the results, so as to construct the objective function used in optimizing the spring gravity sensitive unit structure, thereby improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 This is a schematic diagram of the structure of the gravity sensitive unit on which the present invention is based;

[0081] Figure 2 is a schematic diagram of a spring of the present invention;

[0082] Figure 3 This is a flow chart of a spring structure design method for a gravity sensitive unit disclosed in Embodiment 1 of the present invention;

[0083] Figure 4 This is a flow chart corresponding to step 5 in the method disclosed in Example 1 of the present invention;

[0084] Figure 5 This is a schematic diagram of the grid division of the gravity sensitive unit according to the first embodiment of the present invention;

[0085] Figure 6 It is a corresponding relationship image between the axial displacement y of the sensitive mass block and the gravitational acceleration g corresponding to a certain parameter vector in the first embodiment of the present invention;

[0086] Figure 7 This is a flowchart corresponding to step 7 of the method disclosed in Example 1 of the present invention;

[0087] Figure 8 The schematic diagram of the area allocation of the ring topology structure listed in the first embodiment of the present invention;

[0088] Fig. 9 A schematic diagram of particle swarm learning target update strategy allocation according to the first embodiment of the present invention;

[0089] Fig.10 This is a schematic diagram of the process corresponding to step 7.8 of the first embodiment of the present invention;

[0090] Fig.11 This is a schematic diagram of the structure of a spring structure design system for a gravity sensitive unit disclosed in the second embodiment of the present invention;

[0091] Fig.12 This is a schematic diagram of the optimized construction module structure disclosed in the second embodiment of the present invention;

[0092] Fig.13 This is a schematic diagram of the computing module structure disclosed in the second embodiment of the present invention.

[0093] Description of reference numerals;

[0094] 101. Connecting rod; 102. Spring; 103. Compression mechanism; 104. Frame; 105. Sensitive mass block. DETAILED DESCRIPTION

[0095] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0096] In the description of the embodiments of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments of the present application can be understood according to specific circumstances.

[0097] In the embodiments of the present application, unless otherwise clearly specified and limited, the communication connection between the first feature and the second feature refers to the transmission of information between the first feature and the second feature, and this information transmission can be either unidirectional or bidirectional, and the communication connection can be achieved by wire electrical connection, radio connection, electrical connection of electromagnetic media (such as semiconductors), communication achieved by channels, etc. In addition, the optimal solution of the optimization model includes the optimal position obtained in the search space and the fitness corresponding to the optimal position, and the fitness is the decision function value.

[0098] The technical solution of the present invention will be clearly and completely described below using two embodiments in conjunction with the accompanying drawings of the present invention.

[0099] Embodiment 1:

[0100] like Figure 1 As shown, the gravity sensitive unit is a highly symmetrical structure with uniform thickness in the z-axis direction (the z-axis direction is perpendicular to the paper and outward), which has the effect of suppressing the crosstalk in the horizontal axis. The gravity sensitive unit comprises a sensitive mass block 105 with a metal grating at the center of the surface, two connecting rods 101 symmetrically arranged at both ends of the sensitive mass block 105, four springs 102, four compression mechanisms 103 and a frame 104. The sensitive axis direction of the sensitive unit is the y-axis direction. The sensitive mass block 105 with a large mass is connected to four springs 102 through a connecting rod 101. The four springs 102 have the same shape. One end of two of the springs 102 is connected to one of the connecting rods 101 at the same time, and one end of the other two springs 102 is connected to another connecting rod 101 at the same time. The other ends of the four springs 102 are respectively connected to one of the compression mechanisms 103. The two springs 102 connected to the same connecting rod 101 are symmetrically distributed. The compression mechanism 103 can actively compress the spring 102 in the x-axis direction for a certain distance, thereby reducing the stiffness of the system (i.e., the gravity sensitive unit) in the y-axis direction.

[0101] The schematic diagram of spring 102 is as follows Figure 2 As shown, compared with the bending anti-spring structure in the existing MEMS gravimeter sensitive unit, the spring 102 is more flexible and can obtain a structure with a novel shape. The final shape of the spring 102 will be determined by inverse design.

[0102] The free curve may be a B-spline curve or a Bezier curve. As an implementation method, a B-spline curve is used in this embodiment.

[0103] Based on this, this embodiment provides a spring structure design method for a gravity sensitive unit, see Figure 3 As shown, the method comprises the following steps:

[0104] Step 1: Set the four springs 102 to have the same shape, which is the shape of a free curve, that is, the shape of a B-spline curve. The situation of a Bezier curve is similar.

[0105] Step 2: using the coordinates of at least five control points of the B-spline curve and structural parameters of the spring 102 including the width of the spring 102 and the structural compression amount of the spring 102 as decision parameters.

[0106] Step 3: Determine the value space of the decision parameter based on the configuration requirements of the spring 102, and evenly select multiple parameter vectors in the value space;

[0107] As a specific example, based on the engineering implementation requirements of the shape and structure of the spring 102, the following area is obtained as the value space:

[0108] ,

[0109] ,

[0110] ,

[0111] ,

[0112] ,

[0113] ,

[0114] in, , and The coordinates of three of the five control points selected for the B-spline curve, and is the ordinate of the other two control points, w is the width of the spring 102, B is the structural compression amount of the spring 102.

[0115] Step 4: construct simulation parameters based on each parameter vector respectively, perform structural simulation on the gravity sensitive unit through a multi-physics field simulation module, and obtain a simulation model corresponding to each parameter vector;

[0116] Since the gravity-sensitive unit has the same thickness everywhere and is axisymmetric, the model can be simplified to two dimensions, and only half of it needs to be analyzed. The structural deformation to be analyzed is independent of the time scale and frequency, so multi-physics field simulation software (such as COMSOL software) can be used for simulation, which contains a steady-state solver.

[0117] Specifically, the parameters set during modeling are the corresponding parameters of the parameter vector, which may include: 7 coordinates of the B-spline curve control points, of which 5 control points are listed above, and the remaining two are selected at will, in addition to the width w of the anti-spring and the compression amount of the anti-spring structure B The B-spline curve may be a third-order quadratic B-spline curve; structural parameters of the sensitive mass block 105, the connecting rod 101, the compression mechanism 103 and the frame 104 (Note: in order to more accurately obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the corresponding relationship between the decision parameter and the absolute value of the displacement difference, the structural parameters of the sensitive mass block 105, the connecting rod 101, the compression mechanism 103 and the frame 104 should be set to fixed values ​​based on the idea of ​​the control variable method).

[0118] Step 5: Perform model calculations on the simulation models corresponding to each parameter vector by using the finite element analysis method to obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the corresponding relationship between the decision parameter and the absolute value of the displacement difference, where:

[0119] The linear regression coefficient is the linear regression coefficient of the sensitive axial displacement of the sensitive mass block 105 and the gravitational acceleration g. The change of the gravitational acceleration g includes ;

[0120] The absolute value of the displacement difference is the absolute value of the difference between the displacement of the sensitive mass block 105 along the sensitive axis when the gravitational acceleration g value is 974Gal and the displacement of the sensitive mass block 105 along the sensitive axis when the gravitational acceleration g value is 984Gal.

[0121] See also Figure 4 As shown, in this embodiment, step five includes the following steps:

[0122] Step 5.1: According to the structural characteristics of the gravity-sensitive unit, the simulation model corresponding to each parameter vector is meshed to obtain the corresponding meshed model.

[0123] When meshing the simulation model, the spring 102 can be densely meshed in combination with the structural characteristics of the gravity sensitive unit, and smoothly transition to the sparser mesh at the connection between the connecting rod 101 and the sensitive mass block 105, such as Figure 5 shown.

[0124] Step 5.2: construct the equilibrium equations for the grid models corresponding to the parameter vectors respectively, and obtain the corresponding equilibrium equations with the sensitive axial displacement of the sensitive mass block 105 as the unknown quantity.

[0125] Step 5.3, with an initial step size of 10Gal, increase the gravity acceleration g from 0Gal to 970Gal, then increase the gravity acceleration g from 970Gal to 974Gal, and then increase the gravity acceleration g from 974Gal to 984Gal with a step size of 0.1Gal. The gravity acceleration g is assigned a value in this way, and the solver is used to solve the equilibrium equations corresponding to each parameter vector, and the linear regression relationship between the sensitive axial displacement of the sensitive mass block 105 corresponding to each parameter vector and the gravity acceleration g is obtained.

[0126] The equilibrium equations of each simulation are established when the spring 102 is in a deformed state. When establishing the equilibrium equation, the gravity acceleration g is gradually loaded, with an initial step of 10Gal, starting from 0Gal and increasing to 970Gal, and then increasing the gravity acceleration from 974Gal to 984Gal with a step of 0.1Gal. Then, the corresponding relationship between the axial displacement y of the sensitive mass block 105 and the gravity acceleration g under the gravity acceleration of 974-984Gal is obtained. Figure 6 The corresponding relationship between the axial displacement y of the sensitive mass block 105 and the gravitational acceleration g corresponding to a certain parameter vector is shown. Accordingly, when the decision parameter is a different value, the corresponding relationship will show different forms, and a corresponding relationship has a linear regression coefficient and an absolute value of the displacement difference corresponding to it.

[0127] Step 5.4: Calculate the linear regression coefficients of each linear regression relationship according to the linear regression relationships corresponding to each parameter vector obtained in step 5.3, and obtain the linear regression coefficients corresponding to each parameter vector. For the calculation method of the linear regression coefficient, please refer to the relevant books of mathematical statistics (such as Wei Laisheng's version of mathematical statistics, Science Press).

[0128] Step 5.5: Based on all parameter vectors and the linear regression coefficients corresponding to each parameter vector, obtain the corresponding relationship between the decision parameters and the linear regression coefficients. The specific corresponding relationship can be obtained by numerical approximation or related calculation methods. This is an existing mathematical calculation method and will not be elaborated here.

[0129] Step 5.6: Obtain the absolute value of the displacement difference corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector obtained in step 5.3.

[0130] Step 5.7: Based on all parameter vectors and the absolute value of the displacement difference corresponding to each parameter vector, obtain the corresponding relationship between the decision parameter and the absolute value of the displacement difference.

[0131] Step 6. Based on the linear demand and working stress demand, the range space of the linear regression coefficient and the maximum working stress of the gravity sensitive unit are determined respectively. The direct product space of the range space and the value space is used as the search space. The linear combination of the linear regression coefficient and the absolute value of the displacement difference is used as the fitness. The fitness is used as the decision function to obtain the optimization model.

[0132] In the specific parameter configuration, the linear requirement corresponds to the range requirement of the linear regression coefficient, which is usually required to be no less than 0.9; while the working stress requirement does not exceed 300. Then determine the range space of the linear regression coefficient and the maximum working stress of the gravity sensitive unit. The direct product space of the range space and the value space is used as the search space, and the linear combination of the linear regression coefficient and the absolute value of the displacement difference is used as the fitness, and the fitness is used as the decision function. The obtained optimization model is expressed as follows:

[0133] Max ,

[0134] st ,

[0135] ,

[0136] ,

[0137] ,

[0138] ,

[0139] ,

[0140] ,

[0141] ,

[0142] Among them, F is the fitness, a is the scale factor, is stress, is the linear regression coefficient.

[0143] Step 7: Based on the corresponding relationship between the decision parameters and the linear regression coefficients obtained in step 5 and the corresponding relationship between the decision parameters and the absolute value of the displacement difference, the adaptive elite learning particle swarm algorithm is used to solve the optimization model and obtain the optimal solution.

[0144] See also Figure 7 As shown, in this embodiment, step seven includes the following steps:

[0145] Step 7.1, use a population size of N The particle swarm is used to search the space NA vector is used as the initial position of each particle, and each particle is assigned an initial velocity, and the maximum number of iterations is set.

[0146] Step 7.2: Use the position update method of the particle swarm optimization algorithm as the position update method of each particle, that is, use the following update formula to update the position of each particle:

[0147] ,

[0148] In the formula, t Represents the number of updates, For particles i No. j Dimensional location, For particles i The speed of j dimensional velocity component;

[0149] Then four strategies are defined to generate learning objectives in four different ways to guide particles to search and update. These four strategies are:

[0150] The C-strategy is to randomly select two particles, obtain the particle with the best current individual historical optimal fitness among the two particles, perform binomial crossover between the individual historical optimal position of this particle and the individual historical optimal position of the particle currently executing the update, and obtain the learning target of the particle currently executing the update.

[0151] The O-strategy is: use the adjustment mode of the ring topology structure to adjust the positional relationship between the particles so that each particle has two surrounding particles. Each particle is connected with its two surrounding particles to form a small area of ​​the particle. The current local historical optimal position of the particle in each small area is found. The current global historical optimal position is found by comparing the current local historical optimal position of the particles in each small area. The historical optimal position of the particle at the current global historical optimal position is orthogonally combined with the position of the particle currently executing the update to construct the learning goal.

[0152] D -strategy is: update the strategy to build the learning goal at the speed of particle swarm optimization algorithm;

[0153] A -Strategy: Use the adjustment mode of the ring topology structure to adjust the positional relationship between particles so that each particle has two surrounding particles. Each particle is connected with its two surrounding particles to form a small area of ​​the particle. The learning goal is constructed by the weighted average of the individual historical optimal positions of each small area.

[0154] Each speed update result of particles executing other than the D-strategy is composed of the last speed update result item and the difference item between the current learning target of the particle and the current position of the particle. Particles executing the D-strategy update their speed in the speed update method of the particle swarm optimization algorithm.

[0155] The C-strategy can be called a cross strategy, the O-strategy can be called an orthogonal strategy, the D-strategy can be called a dynamic strategy, and the A-strategy can be called a weighted average strategy. Global search strategies include A-strategy and C-strategy; local search strategies include D-strategy and O-strategy.

[0156] The principle of area allocation in the ring topology in O-strategy and A-strategy is as follows Figure 8 As shown in the figure, i-1, i, etc. represent particle position indexes or indicators. Figure 8 It can be seen that each particle has two surrounding particles, such as particle i The surrounding particles are particles i-1 and particles i+1 , each particle together with its two surrounding particles constitutes a small area of ​​the particle, Figure 8 The elliptical area centered on a certain particle listed in represents the small area corresponding to the particle.

[0157] Step 7.3: Based on the four strategies formulated in step 7.2, the particle swarm is divided into four sub-groups, and the C-strategy, O-strategy, D-strategy, and A-strategy are respectively executed in a one-to-one allocation manner, and the initial position and initial velocity of each particle are respectively used as the last position update result and velocity update result.

[0158] Fig. 9 The following is a schematic diagram of the distribution of learning target update strategies for each sub-group particle swarm. The following will combine it with specific particles to explain in detail the learning target construction and speed update of each strategy. Fig. 9 As shown, the particle swarm is divided into four subgroups, which are respectively recorded as subgroup 1, subgroup 2, subgroup 3 and subgroup 4, and respectively implement C-strategy, O-strategy, D-strategy and A-strategy.

[0159] Subgroup 1 uses the C-strategy to construct learning goals. i The learning goal of the particle whose learning goal is to be updated No. j The values ​​for the dimensions are:

[0160] ,

[0161] ,

[0162] Among them, “||” represents a binomial crossover operation. is the firsti The individual historical optimal position of the particle whose learning target is to be updated j The value of the dimension, , is the individual historical optimal position of two randomly selected particles in subgroup I. j The value of the dimension, better(,) The speed update result of each particle in subgroup 1 is composed of the last speed update result and the difference between the particle's current learning target and the particle's current position, that is, i The speed update formula of the particle whose speed is to be updated is:

[0163] ,

[0164] In the formula, Represents the last velocity update result of the particle. Represents the velocity weight of the particles, which can be generated adaptively. Represents the difference between the learning target of the corresponding particle and the current position of the particle. is the target factor, Represents a random number between 0 and 1.

[0165] Subgroup 2 uses the O-strategy to construct learning goals. i The learning goal of the particle whose learning goal is to be updated for:

[0166] ,

[0167] in, represents an orthogonal experiment, For the i The individual historical optimal position of the particle whose learning target is to be updated, is the historical optimal position of the particle in the current historical optimal position of subgroup 2. The velocity update result of each particle in subgroup 2 is composed of the last velocity update result item and the difference item between the current learning target of the particle and the current position of the particle, that is, i The speed update formula of the particle whose speed is to be updated is:

[0168] ,

[0169] In the formula, Represents the last velocity update result of the particle. represents the velocity weight of particles in subgroup 2, Represents the difference between the particle's current learning target and the particle's current position. represents the learning objective of the corresponding particle. jThe value of the dimension.

[0170] When subgroup 3 uses the D-strategy to construct learning objectives, its principle is the same as that of the particle swarm optimization algorithm. This strategy dynamically adjusts the inertia weight and acceleration coefficient based on the distribution of particles in subgroup 3, and introduces the mutation method in evolutionary computing to make particles jump out of the local optimum. i The velocity update formula of the particle to be updated is:

[0171] ,

[0172] In the formula, Representative i The historical optimal position of the particle to be updated, Representative particles i The learning objectives of this strategy are j All particles in subgroup 3 are judged by calculating the Euclidean distance between particles, and then inertia weights are assigned according to different states. and acceleration factor , Different values ​​are used, and particles in a convergent state are allowed to mutate to test whether they have fallen into a local optimum. For details, please refer to pages 162-164 of "Introduction to Artificial Intelligence" edited by Wang Wanliang.

[0173] Subgroup 4 uses the A-strategy to construct learning goals. i The learning goal of the particle whose learning goal is to be updated No. j The values ​​for the dimensions are:

[0174] ,

[0175] ,

[0176] in, l is the number of particles in subgroup 4, is a random number between 0 and 1, Representative particles k The current individual historical optimal position j The speed update result of each particle in subgroup 4 is composed of the last speed update result and the difference between the current learning target of the particle and the current position of the particle. That is, the speed update formula of the i-th particle to be updated in subgroup 4 is:

[0177] ,

[0178] In the formula, Represents the last velocity update result of the corresponding particle, is the velocity weight of particles in subgroup 4, which can be set adaptively. Represents the difference between the particle's current learning target and the particle's current position.

[0179] As a specific implementation, in this embodiment, after executing step 7.3, before executing step 7.4, the following steps need to be performed: the position index of each particle in each subgroup is calibrated respectively in the subgroup to which it belongs. This is also to comply with the learning objectives, speed and position updates of different particles, so as to facilitate operation.

[0180] Step 7.4: Based on the allocation results of step 7.3, the particles in each subgroup obtain their own learning goals according to the strategies they execute, and obtain the current speed update results and current position update results of each particle based on the last position update results of each particle, the speed update results of each particle, the current learning goals of each particle and the speed update method of the strategies executed by each particle.

[0181] Step 7.5, respectively obtain the decision parameters of this time with the current position update result of each particle, use the corresponding relationship between the decision parameters and the linear regression coefficient to obtain the current linear regression coefficient of each particle, and use the corresponding relationship between the decision parameters and the absolute value of the displacement difference to obtain the current absolute value of the displacement difference of each particle.

[0182] Since the current position update result of each particle is obtained based on the previous update result, including the coordinates of the 7 control points, the width of the spring 102, the structural compression of the spring 102, the linear regression coefficient and the working stress, and the decision parameters include the coordinates of the 7 control points, the width of the spring 102, and the structural compression of the spring 102, the decision parameters of each particle this time are included in the current position update result, so that the decision parameters of this time can be obtained.

[0183] Step 7.6: Determine whether the position update result of each particle belongs to the search space, remove all particles that do not belong to the search space, take all particles that belong to the search space as the remaining particle group, and perform fitness calculations to obtain the fitness of each particle in the remaining particle group.

[0184] Step 7.7: Sort all particles in the remaining particle swarm from best to worst according to their current individual historical optimal fitness to obtain a sorting result.

[0185] Step 7.8: According to the sorting result of step 7.7, select several elite particles from the remaining particle group from the best to the worst, determine the subgroups to which these elite particles belong, and conduct competition and elimination among particles in the following manner:

[0186] If it belongs to subgroup s, a particle is randomly selected from all subgroups except subgroup s, and the individual historical optimal fitness of the elite particle and these selected particles is compared. If the elite particle wins, the current historical optimal position of the elite particle is used as the final position update result of these selected particles. Otherwise, the selected particle is retained, where s=1, 2, 3, 4. The schematic diagram corresponding to this process is as follows Fig.10 shown.

[0187] See also Fig.10 As shown, when executing step 7.8, when some particles are selected from each subgroup for comparison with the elite particles, the position indexes of these selected particles relative to their respective subgroups are recorded, so that starting from the next time, particles are selected from these position indexes for comparison with the elite particles each time.

[0188] In addition, the method of selecting several elite particles from the remaining particle group from the best to the worst can also be: the sizes of the four divided subgroups are , , and , then the corresponding elite particle numbers are , , and ,in:

[0189] ,

[0190] In the formula, t is the number of iterations, is the maximum number of iterations, and Represent the maximum adjustment parameter and the minimum adjustment parameter respectively. The value is 0.93854. The value is 0.19692.

[0191] Step 7.9, determine whether the maximum number of iterations has been reached. If so, the optimal fitness obtained this time and the particle position corresponding to the optimal fitness are used as the optimal solution of the optimization model according to the sorting result; if not, the remaining particle group is used as the particle group updated next time without changing the learning target update strategy executed by each particle therein, and the current speed update result and the current position update result of each particle therein are used as their respective last speed update result and last position update result, and return to execute the said step 7.4.

[0192] Step 8: The shape of the free curve determined by the optimal solution is used as the shape of the spring 102 to complete the design.

[0193] In essence, the four strategies of this embodiment are integrated through the adaptive elite learning strategy (hereinafter referred to as AEL). The core idea of ​​AEL is to eliminate some particles and let elite particles, that is, some particles with better performance in the population, replace these particles to learn more goals.

[0194] The design method of this embodiment combines finite element analysis, optimization and particle swarm algorithm. First, a finite element simulation model of a gravity sensitive unit is constructed, and the spring structure of the gravity sensitive unit is parameterized. Then, the objective function and constraint conditions are designed according to the absolute value of the displacement difference in the sensitive axis direction of the sensitive mass block under the action of gravity acceleration within a certain range of variation and the linear regression coefficient of the displacement and gravity acceleration within the range of variation. At the same time, this design method creatively uses the combination of the optimization model and the adaptive elite learning particle swarm algorithm to optimize the design. After multiple simulation calculations, the sensitivity of the structure (i.e., the absolute value of the displacement difference) and the linear fit are used as evaluation indicators to determine the optimal parameter design of the free geometric anti-spring gravity sensitive unit, thereby improving the performance of the structure, solving the technical problem of insufficient performance of the gravity sensitive unit obtained by the existing design method, and optimizing it in terms of geometric layout, thereby improving its measurement accuracy.

[0195] Embodiment 2:

[0196] Based on the spring structure design method of a gravity sensitive unit proposed in the first embodiment, this embodiment further proposes a spring structure design system for a gravity sensitive unit, see Fig.11 , Fig.12 and Fig.13 As shown, the design system includes an optimization building module and a calculation module, wherein the optimization building module is used to build and / or store the optimization model corresponding to the design of the spring 102 of the gravity sensitive unit; the calculation module is used to obtain the optimal solution corresponding to the optimization model, and obtain the shape design parameters of the spring 102 based on the optimal solution. The optimization building module is in communication connection with the calculation module.

[0197] See also Fig.12 As shown, in this embodiment, the optimization construction module includes a parameter module, a value module, a simulation module, a corresponding module and a model module, wherein the value module is communicatively connected with the parameter module, the simulation module is communicatively connected with the value module, the corresponding module is communicatively connected with the simulation module, and the model module is communicatively connected with the corresponding module. The functions of each module are as follows:

[0198] The parameter module is configured to use the coordinates of at least five control points of the free curve and the structural parameters of the spring 102 including the width of the spring 102 and the structural compression amount of the spring 102 as decision parameters;

[0199] The value module is configured to determine the value space of the decision parameter based on the configuration requirements of the spring 102;

[0200] The simulation module is configured to uniformly select multiple parameter vectors in the value space, and perform structural simulation on the gravity sensitive unit to obtain a simulation model corresponding to each parameter vector;

[0201] The corresponding module is configured to perform model calculations on the simulation models corresponding to each parameter vector respectively through a finite element analysis method, so as to obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the corresponding relationship between the decision parameter and the absolute value of the displacement difference;

[0202] The model module is set to determine the range space of the linear regression coefficient and the maximum working stress of the gravity-sensitive unit based on the linear demand and the working stress demand respectively, and the direct product space of the range space and the value space is used as the search space. The linear combination of the linear regression coefficient and the absolute value of the displacement difference is used as the fitness, and the fitness is used as the decision function to obtain the optimization model.

[0203] Please continue to see Fig.12 As shown, in this embodiment, the corresponding modules include a grid unit, a balancing unit, a solver, a regression unit and an absolute unit, wherein the grid unit is connected to the simulation module for communication, the balancing unit is connected to the grid unit for communication, the solver is connected to the balancing unit for communication, the regression unit is connected to the solver for communication, the absolute unit is connected to the solver for communication, and the model module is connected to the regression unit and the absolute unit for communication. The functions of each unit are as follows:

[0204] The grid unit is set to grid the simulation model corresponding to each parameter vector according to the structural characteristics of the gravity sensitive unit, and obtain the corresponding grid model;

[0205] The balance unit is set to construct the balance equation for each grid model corresponding to each parameter vector, and obtain the corresponding balance equation;

[0206] The solver is set to increase the gravity acceleration g from 0Gal to 970Gal, then from 970Gal to 974Gal, and then from 974Gal to 984Gal in a step of 0.1Gal to assign values ​​to the gravity acceleration g, solve the equilibrium equations corresponding to each parameter vector, and obtain the linear regression relationship between the sensitive axial displacement of the sensitive mass block 105 and the gravity acceleration g corresponding to each parameter vector;

[0207] The regression unit is configured to calculate the linear regression coefficients of each linear regression relationship according to the linear regression relationship corresponding to each parameter vector, obtain the linear regression coefficient corresponding to each parameter vector, and obtain the corresponding relationship between the decision parameter and the linear regression coefficient based on all the parameter vectors and the linear regression coefficient corresponding to each parameter vector;

[0208] The absolute unit is configured to obtain the absolute value of the displacement difference corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector, and obtain the corresponding relationship between the decision parameter and the absolute value of the displacement difference based on all parameter vectors and the absolute value of the displacement difference corresponding to each parameter vector;

[0209] See also Fig.13 As shown, in this embodiment, the calculation module includes a setting unit, a storage unit, an allocation unit, an update unit, a sorting unit, a sorting unit, an elimination unit and a judgment unit, wherein the setting unit is connected to the model module in communication, the storage unit is connected to the setting unit in communication, the allocation unit is connected to the storage unit in communication, the update unit is connected to the allocation unit in communication, the sorting unit is connected to the update unit in communication, the sorting unit is connected to the sorting unit in communication, the elimination unit is connected to the sorting unit in communication, and the judgment unit is connected to the elimination unit in communication. The functions of each unit are as follows:

[0210] The setting unit is configured to set the population size of the particle group, set the initial positions of the particles respectively, assign initial velocities to the particles respectively, and set the maximum number of iterations;

[0211] The storage unit is configured to store the position update mode of each particle in the particle swarm, the four strategies and the speed update mode when executing each strategy;

[0212] The allocation unit is configured to divide the particle group into four subgroups, and respectively execute the four strategies stored in the storage unit in a one-to-one allocation manner, and use the initial position and initial velocity of each particle as the last position update result and velocity update result;

[0213] The updating unit is set to be based on the allocation result of the allocation unit, and controls the particles in each subgroup to obtain their respective learning goals according to the strategies they execute, and obtains the current speed update result and the current position update result of each particle based on the last position update result of each particle, the speed update result of each particle, the current learning goal of each particle and the speed update method of the strategy executed by each particle, and then obtains the current decision parameters based on the current position update result of each particle, and obtains the current linear regression coefficient of each particle by using the corresponding relationship between the decision parameter and the linear regression coefficient, and obtains the current displacement difference absolute value of each particle by using the corresponding relationship between the decision parameter and the absolute value of the displacement difference;

[0214] The sorting unit is configured to respectively determine whether the position update result of each particle belongs to the search space, remove all particles that do not belong to the search space, take all particles that belong to the search space as the remaining particle group, and perform fitness calculations to obtain the fitness of each particle in the remaining particle group;

[0215] The sorting unit is set to sort all the particles of the remaining particle group from best to worst according to the current individual historical optimal fitness of all the particles of the remaining particle group, and obtain the sorting result;

[0216] The elimination unit is set to select several elite particles from the remaining particle group from the best to the worst according to the sorting result, and determine the subgroups to which these elite particles belong respectively. If they come from subgroup s, a particle is randomly selected from all subgroups except subgroup s, and the individual historical optimal fitness of the elite particle is compared with that of these selected particles. If the elite particle wins, the current historical optimal position of the elite particle is used as the final position update result of these selected particles. Otherwise, the selected particles are retained, where s=1, 2, 3, 4;

[0217] The judgment unit is configured to judge whether the maximum number of iterations has been reached. If so, the optimal fitness obtained this time and the particle position corresponding to the optimal fitness are used as the optimal solution of the optimization model according to the sorting result; if not, the remaining particle group is used as the particle group updated next time without changing the learning target update strategy executed by each particle therein, and the current speed update result and the current position update result of each particle therein are used as their respective last speed update result and last position update result, and transmitted to the update unit.

[0218] The system disclosed in this embodiment is based on the spring structure design method of the gravity sensitive unit of the first embodiment, and adopts the optimization construction module to realize the adaptive optimization modeling of the gravity sensitive unit. Through the cooperation with the calculation module, the optimal decision parameters corresponding to the optimization model are finally obtained to complete the shape design of the spring. In this way, the technical problem of insufficient performance of the gravity sensitive unit obtained by the existing design method is solved, and the optimization of the geometric layout is realized, and the measurement accuracy is improved.

[0219] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0220] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for designing a spring structure of a gravity sensitive unit, the gravity sensitive unit comprising a sensitive mass block (105) and four springs (102) connected to the sensitive mass block (105) via two connecting rods (101); Features: The steps include: Step 1, setting the shapes of the four springs (102) to the shapes of free curves; Step 2: using the coordinates of at least five control points of the free curve and the structural parameters of the spring (102) as decision parameters, wherein the structural parameters include the width and structural compression of the spring (102); Step three, determining the value space of the decision parameter based on the configuration requirements of the spring (102), and uniformly selecting a plurality of parameter vectors in the value space; Step 4: construct corresponding simulation parameters based on each of the parameter vectors to perform structural simulation on the gravity sensitive unit through a multi-physics field simulation module to obtain a simulation model corresponding to each of the parameter vectors; Step 5: Perform model calculation on the simulation model corresponding to each of the parameter vectors by using a finite element analysis method to obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the absolute value of the displacement difference; The linear regression coefficient is the linear regression coefficient of the sensitive axial displacement of the sensitive mass block (105) and the gravitational acceleration g. The change of the gravitational acceleration g includes ; The absolute value of the displacement difference is the absolute value of the difference between the displacement of the sensitive mass block (105) along the sensitive axis when the gravitational acceleration g value is 974Gal and the displacement of the sensitive mass block (105) along the sensitive axis when the gravitational acceleration g value is 984Gal; Step 6: construct a search space based on the value space, take the linear combination of the linear regression coefficient and the absolute value of the displacement difference as the fitness, take the fitness as the decision function, and obtain the optimization model; Step 7: Based on the corresponding relationship obtained in step 5, an adaptive elite learning particle swarm algorithm is used to solve the optimization model to obtain an optimal solution; Step eight, using the shape of the free curve determined by the optimal solution as the shape of the spring (102) to complete the design; The step five comprises the following steps: Step 5.1, meshing the simulation models corresponding to the parameter vectors according to the structural characteristics of the gravity sensitive units to obtain the corresponding meshed models; Step 5.2, constructing a balance equation for each grid model corresponding to each of the parameter vectors, and obtaining a corresponding balance equation with the sensitive axial displacement of the sensitive mass block (105) as an unknown quantity; Step 5.3, increasing the gravity acceleration g from 0Gal to 970Gal with an initial step length of 10Gal, and then increasing the gravity acceleration g from 974Gal to 984Gal with a step length of 0.1Gal to assign a value to the gravity acceleration g, so that the solver solves the equilibrium equation corresponding to each of the parameter vectors respectively, and obtains a linear regression relationship between the sensitive axial displacement of the sensitive mass block (105) corresponding to each of the parameter vectors and the gravity acceleration g; Step 5.4, respectively calculating the linear regression coefficients of the linear regression relationships corresponding to the parameter vectors obtained in step 5.3, and obtaining the linear regression coefficients corresponding to the parameter vectors; Step 5.5, obtaining the corresponding relationship between the decision parameter and the linear regression coefficient based on all the parameter vectors and the linear regression coefficient corresponding to each of the parameter vectors by a numerical approximation method; Step 5.6, obtaining the displacement difference absolute value corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector obtained in step 5.3; The absolute value of the displacement difference is the absolute value of the difference between the displacement of the sensitive mass block (105) along the sensitive axis direction when the gravitational acceleration g value is 974Gal and the displacement of the sensitive mass block (105) along the sensitive axis direction when the gravitational acceleration g value is 984Gal; Step 5.7: Obtain the corresponding relationship between the decision parameter and the absolute value of the displacement difference based on all the parameter vectors and the absolute value of the displacement difference corresponding to each parameter vector through a numerical approximation method.

2. The method for designing a spring structure of a gravity sensitive unit according to claim 1, characterized in that: The free curve is a B-spline curve or a Bezier curve.

3. The method for designing a spring structure of a gravity sensitive unit according to claim 2, characterized in that: The step seven comprises the following steps: Step 7.1: Use a population size of N The particle swarm randomly takes the search space N vectors as the initial positions of the particles, assign initial velocities to the particles respectively, and set the maximum number of iterations; Step 7.2, using the position update mode of the particle swarm optimization algorithm as the position update mode of the particles, and defining four strategies to generate learning objectives in four different ways to guide the particles to perform search updates, the four strategies being C-strategy, O-strategy, D-strategy and A-strategy; Step 7.3, dividing the particle group into four subgroups, and executing the C-strategy, the O-strategy, the D-strategy, and the A-strategy respectively in a one-to-one allocation manner, and taking the initial position and initial velocity of each particle as the last position update result and velocity update result respectively; Step 7.4, the particles in each subgroup obtain their own learning goals according to their respective executed strategies, and then obtain the current speed update result and current position update result of each particle based on the last position update result and speed update result of each particle and the current learning goal of each particle; Step 7.5, respectively obtain the decision parameters of this time with the current position update result of each particle, obtain the current linear regression coefficient of each particle using the corresponding relationship between the decision parameters and the linear regression coefficient, and simultaneously obtain the current absolute value of the displacement difference of each particle using the corresponding relationship between the decision parameters and the absolute value of the displacement difference; Step 7.6, respectively determine whether the current position update result of each particle belongs to the search space, remove all particles that do not belong to the search space, take all particles that belong to the search space as the current remaining particle group, and perform fitness calculations to obtain the fitness of each particle in the current remaining particle group; Step 7.7, sorting all particles of the remaining particle swarm from best to worst according to the current individual historical optimal fitness of all particles of the remaining particle swarm, and obtaining a sorting result; Step 7.8: According to the sorting result of step 7.7, select several elite particles from the remaining particle group in descending order, determine the subgroups to which these elite particles belong, and conduct competition and elimination among particles in the following manner: If it belongs to subgroup s, a particle is randomly selected from all subgroups except subgroup s, and the individual historical optimal fitness of the elite particle is compared with that of these selected particles. If the elite particle wins, the current historical optimal position of the elite particle is used as the final position update result of these selected particles. Otherwise, the selected particle is retained, where s=1, 2, 3, 4; Step 7.9, determine whether the maximum number of iterations has been reached. If so, the optimal fitness obtained this time and the particle position corresponding to the optimal fitness are used as the optimal solution of the optimization model according to the sorting result; if not, the remaining particle group this time is used as the particle group updated next time without changing the learning target update strategy executed by each particle therein, and the current speed update result and the current position update result of each particle therein are used as their respective last speed update result and last position update result, and return to execute step 7.

4.

4. The method for designing a spring structure of a gravity sensitive unit according to claim 3, characterized in that: In step 7.2: The C-strategy is: randomly select two particles, obtain the particle with the best current individual historical optimal fitness among the two particles, perform binomial crossover between the individual historical optimal position of this particle and the individual historical optimal position of the particle currently executing the update, and obtain the learning target of the particle currently executing the update accordingly; The O-strategy is: using the adjustment mode of the ring topology structure to adjust the positional relationship between the particles, so that each particle has two surrounding particles, each particle is connected with its two surrounding particles to form a small area of ​​the particle, and the current local historical optimal position of the particle in each small area is found, and the current global historical optimal position is found by comparing the current local historical optimal position of the particle in each small area, and the historical optimal position of the particle at the current global historical optimal position is orthogonally combined with the position of the particle currently performing the update, so as to construct the learning target; The D-strategy is: constructing a learning goal with the speed update strategy of the particle swarm optimization algorithm; The A-strategy is: using the adjustment mode of the ring topology structure to adjust the positional relationship between the particles, so that each particle has two surrounding particles, each particle is connected with its two surrounding particles to form a small area of ​​the particle, and the learning target is constructed by the weighted average of the individual historical optimal positions of each small area; in, Each speed update result of particles executing other than the D-strategy is composed of the last speed update result item and the difference item between the current learning target of the particle and the current position of the particle. The particles executing the D-strategy update their speed in the speed update method of the particle swarm optimization algorithm.

5. The method for designing a spring structure of a gravity sensitive unit according to claim 4, characterized in that: After executing step 7.3, the following steps need to be performed before executing step 7.4: The position index of each particle in each subgroup is calibrated respectively.

6. The method for designing a spring structure of a gravity sensitive unit according to claim 5, characterized in that: When executing step 7.8, when some particles are selected from each subgroup for comparison with the elite particles for the first time, the position indexes of these selected particles relative to their respective subgroups are recorded, so that starting from the next time, particles are selected from these position indexes for comparison with the elite particles each time.

7. A spring structure design system for a gravity sensitive unit, characterized in that: The spring structure design method of the gravity sensitive unit according to any one of claims 1 to 6 comprises: An optimization construction module, used for constructing and / or storing an optimization model corresponding to the design of the spring (102) of the gravity sensitive unit; A calculation module, used to obtain an optimal solution corresponding to the optimization model, and to obtain shape design parameters of the spring (102) based on the optimal solution; in, The optimization construction module is communicatively connected with the calculation module.

8. The spring structure design system of the gravity sensitive unit according to claim 7, characterized in that: The optimization construction module includes: The parameter module uses the coordinates and structural parameters of at least five control points of the free curve as decision parameters; A value taking module, which determines a value taking space of the decision parameter based on a configuration requirement of the spring (102); A simulation module, which uniformly selects a plurality of parameter vectors in the value space, and based on the structure of the gravity sensitive unit, uses each of the parameter vectors as a simulation parameter to perform structural simulation on the gravity sensitive unit, and obtains a simulation model corresponding to each of the parameter vectors; The corresponding module performs model calculation on the simulation model corresponding to each of the parameter vectors by a finite element analysis method to obtain the corresponding relationship between the decision parameter and the linear regression coefficient and the corresponding relationship between the decision parameter and the absolute value of the displacement difference; A model module, which determines the range space of the linear regression coefficient and the maximum working stress of the gravity sensitive unit respectively based on the linear demand and the working stress demand, uses the direct product space of the range space and the value space as the search space, uses the linear combination of the linear regression coefficient and the absolute value of the displacement difference as the fitness, uses the fitness as the decision function, and obtains the optimization model; in, The value acquisition module is communicatively connected to the parameter module, the simulation module is communicatively connected to the value acquisition module, the corresponding module is communicatively connected to the simulation module, and the model module is communicatively connected to the corresponding module.

9. The spring structure design system of the gravity sensitive unit according to claim 8, characterized in that: The corresponding modules include: A grid unit, which grids the simulation models corresponding to the parameter vectors according to the structural characteristics of the gravity sensitive unit to obtain the corresponding grid models; A balancing unit constructs a balancing equation for each grid model corresponding to each of the parameter vectors to obtain a corresponding balancing equation; A solver is provided, which increases the gravity acceleration g from 0Gal to 970Gal with an initial step size of 10Gal, and then increases the gravity acceleration g from 974Gal to 984Gal with a step size of 0.1Gal to assign a value to the gravity acceleration g, respectively solves the equilibrium equation corresponding to each of the parameter vectors, and obtains a linear regression relationship between the sensitive axial displacement of the sensitive mass block (105) corresponding to each of the parameter vectors and the gravity acceleration g; A regression unit, calculating the linear regression coefficients of each linear regression relationship according to the linear regression relationships corresponding to each of the parameter vectors, obtaining the linear regression coefficients corresponding to each of the parameter vectors, and obtaining the corresponding relationship between the decision parameter and the linear regression coefficient based on all the parameter vectors and the linear regression coefficients corresponding to each of the parameter vectors; an absolute unit, for obtaining the absolute value of the displacement difference corresponding to each parameter vector based on the linear regression relationship corresponding to each parameter vector, and obtaining the corresponding relationship between the decision parameter and the absolute value of the displacement difference based on all the parameter vectors and the absolute value of the displacement difference corresponding to each parameter vector; in, The grid unit is communicatively connected to the simulation module, the balancing unit is communicatively connected to the grid unit, the solver is communicatively connected to the balancing unit, the regression unit is communicatively connected to the solver, the absolute unit is communicatively connected to the solver, and the model module is communicatively connected to the regression unit and the absolute unit.

Images