Optimal Layout Method of Strain Gauges for Load Identification
Through the load-oriented recognition method based on the agent model, the position and direction of the strain gauge are optimized, and the problems of load estimation error and noise influence of large or complex mechanical structures are solved, and the accuracy of analysis of complex surface strains and load recognition is improved.
Patent Information
- Application Number
- CN202411230426.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-04
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-09-04
AI Technical Summary
The prior art is difficult to provide optimization methods for the global optimal position and direction of the strain gauge for large or complex mechanical structures, resulting in large measurement errors and noise influences on load estimation, and it is impossible to effectively analyze the isotropic strain conditions of complex surfaces.
Using the load-oriented recognition method based on the agent model, by establishing a geometric finite element model, using the strain state conversion principle to obtain the strain value in any position and direction of the structural surface, construct the continuous mapping relationship between the strain gauge layout and the strain-load coefficient, and optimize the The position and direction of the strain gauge are the objective function of the minimum load recognition error.
Realize the global optimal layout of large or complex mechanical structure strain gauges, improve the accuracy and accuracy of load recognition, can customize the strain gauge installation range and avoid interference, and avoid local optimal solutions caused by initial point selection .
Smart Images

Figure CN118734660B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of optimal layout of strain gauges in load identification of mechanical structures, and relates to an optimal layout method of strain gauges oriented to load identification based on a proxy model. Background Art
[0002] Load identification of major mechanical equipment is a key step to ensure the safe and reliable operation of the equipment. In fact, the most reliable method to obtain the external load of a mechanical structure is actual measurement. However, in actual operation, the force and motion state of the working device of a large mechanical structure are very complex, and it is difficult to obtain the external load acting on the structure by direct measurement. It is necessary to use indirect measurement combined with finite element simulation technology to obtain it. In recent years, load inversion technology based on measured structural response has attracted attention. Since the structural strain response can be measured relatively easily, the external load can be inverted through the measured strain history. However, the position and direction of the strain gauge will affect the quality of the load estimation. When the strain gauge is installed in a position and direction with relatively low sensitivity to the applied load, the load estimation may be greatly affected by measurement errors and noise. Therefore, the layout optimization of the strain gauge is particularly important.
[0003] In response to the above problems, scholars at home and abroad have proposed a variety of strain gauge layout optimization methods for load identification. For example, application number 202311158047.6 takes load identification accuracy as the optimization target, and the finite element node positions in finite element analysis as the candidate set of sensor layout positions, and obtains the optimal sensor layout solution through the particle swarm optimization algorithm; for example, application number 201910670866.6 takes the joint fitness function that takes into account the robust performance of load identification and the sensor distribution index as the optimization target, takes the number and position of sensors as the design variables, and uses the particle swarm optimization algorithm to obtain the most economical and effective sensor layout method.
[0004] Although the above technical solutions can obtain the optimization solution of the strain gauge position, they do not consider the optimization of its direction, and they are all optimization results obtained based on discrete finite element data, so it is impossible to obtain the global optimal result; the accuracy of the results is affected by the size of the mesh in the finite element. If the analysis object is a large mechanical structure, it may cause a large calculation cost to achieve a certain accuracy; in addition, the above technologies are all for the analysis of simple mechanical mechanisms. If it is a mechanical structure with complex surfaces, it is difficult to analyze the isotropic strain of its surface. Therefore, there is an urgent need for an optimization method that can provide the global optimal position and direction of strain gauges for large or complex mechanical structures. Summary of the invention
[0005] In view of the problems existing in the prior art, the present invention provides a strain gauge layout optimization method for load identification based on a proxy model. The method can analyze large or complex mechanical structures and obtain the global optimal layout in the candidate area for strain gauge installation on the structure, which includes the optimal installation position and direction; further, the load on the mechanical structure can be accurately identified based on the optimal strain gauge test data. The present invention fully considers that there are many complex three-dimensional structural surfaces in actual engineering. If only the data on the nodes and the degrees of freedom of the nodes in the finite element are used, the strain conditions at each position on the structural surface in each direction cannot be reflected. Therefore, the strain state conversion principle is used to obtain the strain value of the structural surface at any position in any direction; further, the proxy model is used to construct a continuous mapping relationship between variables, thereby expanding the discrete optimization range into a continuous optimization range; in addition, the constraints of the optimization problem are constructed according to the actual installation conditions, and the load identification error is minimized as the objective function, and finally the global optimal layout is obtained.
[0006] To achieve the above object, the present invention adopts the following technical solutions:
[0007] The optimal layout method of strain gauges for load identification, the optimal layout method of strain gauges, first, establishes a geometric finite element model of the mechanical structure and obtains the units, nodes and strain response data of the structure under the independent action of multiple loads, the strain response data includes the shear strain in the three directions of x, y, and z in the global coordinate system and the shear strain in the three directions of yz, xz, and xy; secondly, creates a local coordinate system of the model surface through unit and node information, and obtains the shear strain in the remaining directions through the strain response data combined with the plane strain state conversion principle; then, when multiple loads are applied to the model independently, the corresponding strain-load coefficient is obtained according to the shear strain in each direction and the corresponding load size, and the proxy model is used to establish a continuous mapping relationship between the strain gauge layout (position and direction) and the strain-load coefficient; finally, selects a suitable installation range as a position constraint, and at the same time constrains the minimum distance between the strain gauges to avoid mutual interference, and optimizes the strain gauge layout with the minimum load identification error as the objective function. The specific steps are as follows:
[0008] The first step is to establish a geometric finite element model based on the physical entity of the mechanical structure, and use finite element analysis to obtain the structural units, nodes, and strain response data under various loads. The details are as follows:
[0009] Step 1.1: According to the geometric characteristics of the physical entity of the mechanical structure, a geometric finite element model is established using 3D modeling software.
[0010] Step 1.2: For the geometric finite element model obtained in step 1.1, remove its internal structural details by appropriately simplifying its screw holes and other tiny features using methods such as geometric merging or deletion to obtain a simplified geometric model.
[0011] Step 1.3: Mesh the simplified geometric finite element model obtained in step 1.2, using shell elements to mesh the model surface, and save the surface mesh elements, node numbers, and node coordinates as a local file.
[0012] Step 1.4: Apply different loads to the simplified geometric finite element model obtained in step 1.2, extract the strain response data on each grid unit and save it as a local file.
[0013] Step 1.5: Set the appropriate strain gauge installation range on the model surface, excluding areas where stress concentration is likely to occur, such as the model edge, the interface between surfaces, and the area near the load position.
[0014] Step 1.6: Extract the unit and node data and the corresponding strain response data within the installation range as analysis data for subsequent steps.
[0015] The second step is to create a local coordinate system at the center of each unit based on the unit and node information within the strain gauge installation range. The details are as follows:
[0016] Step 2.1: Calculate the normal vector of each element by using the coordinate information of each node on each element in the global coordinate system Oxyz .
[0017] Step 2.2: Find the vector orthogonal to the normal vector , and find the vector and Vectors that are orthogonal . (Special note: For a plane, its normal vector is unique, but the number of mutually orthogonal coordinate axes that can be constructed on the plane is infinite. The vector constructed by the present invention is and is one of the ones that meets the conditions and serves as the coordinate reference for subsequent calculations).
[0018] Step 2.3: Get , , After these three vectors are divided by their respective moduli, the direction cosines of the three vectors are calculated as follows: , , , respectively as the x', y', and z' axes of the unit local coordinate system O'x'y'z', and their set is R as shown in formula (1).
[0019] (1);
[0020] in, l ⃗ =[ l 1 l 2 l 3 ] , m ⃗ =[ m 1 m 2 m 3 ] , n ⃗ =[ n 1 n 2 n 3 ] , , , are the components of the corresponding direction cosines.
[0021] The third step is to create a spatial strain state conversion method based on the plane strain state conversion principle and the local coordinate system to obtain the strain value of each unit center in each direction. The details are as follows:
[0022] Step 3.1: Based on the strain response data of each unit in the global coordinate system Oxyz, specifically the shear strain in the x, y, and z directions And the shear strains in the yz, xz, and xy directions , and transform it through formula (2) and formula (3) to obtain the shear strain in the x', y', and z' directions in the unit local coordinate system O'x'y'z' And the shear strains in the y'z', x'z', and x'y' directions , as shown in formulas (2) and (3):
[0023] (2); (3);
[0024] in, , , Direction cosines , , The amount.
[0025] The three shear strains in the global coordinate system and three shear strains The set composed of , like the set of stress components, also obeys the transformation law of the second-order tensor. It can be expressed as formula (4):
[0026] (4);
[0027] The three shear strains in the local coordinate system of the unit And the three shear strains The set composed of According to formulas (1) to (4), we can obtain and The conversion relationship between them is shown in formula (5):
[0028] (5);
[0029] Among them, R is the matrix composed of the three direction cosines corresponding to the unit local coordinate system.
[0030] Step 3.2: According to the plane strain state conversion principle, obtain the strain value in the direction of the angle α between the x' axis of the local coordinate system of the unit and the plane O'x'y' where each unit is located. .
[0031] The plane strain state conversion principle is:
[0032] (6);
[0033] Where α is the strain value required The angle between the x' axis and the local coordinate system of the element.
[0034] The fourth step is to construct the strain-load coefficient matrix of the model according to the load identification principle. The details are as follows:
[0035] Step 4.1: Process and analyze the strain response data obtained in step 1.4 under independent application of different loads. Since most large mechanical structures belong to the category of linear structural systems during operation, there is a linear relationship between load, strain and displacement. Therefore, in the simplified geometric finite element model of step 1.2, the strain in a certain direction at any position is the superposition of the strains in the corresponding directions caused by multiple external loads at the same time. An important assumption here is that the stress acting on the model will never exceed the elastic limit, and the displacement is small enough so that the superposition principle holds.
[0036] The relationship between strain and load can be expressed as: (7);
[0037] Where p is the product of the number of units and the number of directions studied; m is the number of external loads; is the strain matrix, each column represents the shear strain of each unit in each direction under a certain load application; is the strain-load coefficient matrix, each column represents the strain-load coefficient applied to the model independently under a certain load; is the load matrix, and each column represents the load applied on the structure.
[0038] Under a certain load application condition, the load number is set to f, and the strain calculation formula on each unit is:
[0039] (8);
[0040] in, is the shear strain of a unit in a certain direction on the structure; is the strain-load coefficient matrix, which remains constant once the element and the direction of study are determined; is the magnitude of each load applied to the structure.
[0041] From formula (8), we can get: The calculation method is shown in formula (9):
[0042] (9);
[0043] in, for The corresponding strain-load factor.
[0044] When only the first load When the function is applied, formula (8) is:
[0045] (10);
[0046] In formula (10), Indicates that only The shear strain of a unit on the surface of the model in a certain direction under the first load. Since the actual number of applied loads is m, the corresponding shear strain value can be obtained when each load is applied independently. By assembling these data, formula (10) can be expanded to formula (11):
[0047] (11);
[0048] Among them, in the strain matrix In the load matrix, each column represents the corresponding shear strain vector when the corresponding column load is applied independently; In , each column is the load vector when a certain load is applied independently. The magnitude of each independently applied load.
[0049] From formula (11), we can get the strain-load coefficient matrix: The calculation method for each column in is shown in formula (12):
[0050] (12);
[0051] in, .
[0052] Step 4.2: Obtain the corresponding shear strain by applying different loads independently, calculate the corresponding strain-load coefficient by formula (12), and then assemble the strain-load coefficients under these different loads independently to obtain the strain-load coefficient matrix .
[0053] The fifth step is to form a combination of the coordinate value representing the position and the direction vector representing the angle, and then establish a proxy model between the combination and the strain-load coefficient under the independent action of each load. The above coordinate values, direction vectors and strain-load coefficients are all discrete data obtained based on the simplified geometric finite element model, and the proxy model can make the discrete data continuous and construct a functional relationship between the two. The details are as follows:
[0054] Step 5.1: Determine the input variable x, and use the coordinates of the center point of the unit in the selected range on the structure to represent the position, denoted as , p is the product of the number of units and the number of directions studied; at the same time, the third step proposes a method for calculating the strain value of any direction on the tangent plane of each unit. Since the angle α is relative to the O'x'y' coordinate system on the tangent plane at each position (that is, the angle with the x' axis direction is α), it cannot be used as a unified measurement method. Therefore, the present invention uses the unit vector under the global coordinate system Oxyz coordinate system to represent the different angle directions on the tangent plane at each position, which is recorded as According to formula (13), the conversion between direction vector and angle can be realized:
[0055] (13);
[0056] Therefore, the position coordinates and direction vector form the input , which eventually forms a variable set .
[0057] Step 5.2: For different independent load conditions, establish Strain-load factor corresponding to the working condition The proxy model is described as follows:
[0058] Surrogate Model (SM) is a data-driven model that focuses on data and relies on mathematical expressions to respond quickly and accurately without understanding the physical rules between input variables and output responses. Therefore, it can be applied to optimization design and analysis, especially when the mechanism between input and output is unclear and the computational cost is high.
[0059] Many classic SMs have been proposed, such as polynomial regression surface radial basis function (RBF), polynomial regression surface (PRS), Kriging (KRG), support vector regression (SVR). According to different requirements or goals, they can all obtain accuracy or ideal results. Considering the calculation time and accuracy, the present invention uses RBF.
[0060] The RBF is a commonly used multivariate scattered point interpolation prediction algorithm. This method transforms complex high-dimensional problems into simple linear problems through a kernel function. The kernel function is radially symmetric about the center of the training point. The general expression is shown in Formula (14):
[0061] (14);
[0062] In the formula, is the i-th variable With the jth variable The Euclidean distance between , is the basis function that represents the correlation between training points, n is the number of training points, and also the number of functions. represents the i-th weight factor.
[0063] As for the basis function, since the multiquadric basis function (MQ) has a better ability to express the relationship between input and output, the present invention adopts the MQ basis function, as shown in formula (15):
[0064] (15);
[0065] In the basis function is the expansion function of RBF, reflecting the shape of the basis function. This value can be obtained through a unified constant expansion algorithm, as shown in formula (16):
[0066] (16);
[0067] In the formula, is the maximum distance between training points, and N is the number of training points.
[0068] Using R-square (R 2) is used to describe the accuracy of the proxy model, and the calculation formula is as follows:
[0069] (17);
[0070] Among them, R 2 The closer the value of is to 1, the higher the accuracy of the proxy model; is the number of test points; and They are The true value at and the proxy model estimate; is the mean of the surrogate model estimates at all test points.
[0071] Therefore, the proxy model constructed by the present invention is as follows:
[0072] (18);
[0073] in, is the strain-load coefficient under a certain load independent action, Built for a certain load independent and Function expression of the proxy model between them.
[0074] The sixth step is to construct an optimization problem to obtain the optimal strain gauge layout solution. The details are as follows:
[0075] Step 6.1: Minimizing the load identification error is used as the objective function. According to the basic principles of statistics, the relationship between the load identification error, the strain gauge test error, and the strain-load coefficient corresponding to the strain gauge position and orientation is shown in formula (19):
[0076] (19);
[0077] In the formula, is the test data of the strain gauge, is the strain-load factor corresponding to the position and orientation of the strain gauge, is the variance-covariance matrix of the strain, is the variance-covariance matrix of the identified loadings.
[0078] Since strain data is inevitably affected by factors such as strain gauge measurement accuracy and environmental noise, in actual measurement, if the error of each strain data is independent and the standard deviation is , then the covariance matrix of the restored load can be expressed as formula (20):
[0079] (20);
[0080] From the above formula (20), we can see that the accuracy of the identified load is determined by Therefore, in order to minimize the load identification error, The largest, among which is the strain-load coefficient corresponding to the position and direction of the strain gauge. If the optimal result (i.e. the optimal strain gauge layout) is , where n is the number of strain gauges to be installed, the formula must be satisfied:
[0081] (twenty one);
[0082] in, is the shear strain under a certain load, m is the external load number, is the strain-load coefficient corresponding to the position and orientation of the optimal strain gauge layout under a certain independent load.
[0083] Step 6.2: Set constraints. First, each strain gauge must be located on the surface of the structure. If the surface can be expressed as , then it is necessary to satisfy ( is the position coordinate of the strain gauge); secondly, to avoid interference between strain gauges during the actual installation process, the minimum distance between strain gauges must be greater than the length of the strain gauge itself, and the minimum distance is recorded as D; thirdly, the unit vector representing the installation direction of the strain gauge Must be consistent with the normal vector at the installation location Vertical, otherwise the strain gauge cannot be installed on the surface of the structure; finally, set the area where the strain gauge is allowed to be installed .
[0084] Step 6.3: Construct an optimization problem and obtain the optimal solution. Based on the above description, the optimization problem can be expressed as:
[0085] (twenty two);
[0086] Among them, k and s are the numbers of the strain gauges, which are not equal to each other; is the position coordinate of the strain gauge; is the unit vector of the strain gauge installation direction; is the normal vector at the installation position of the strain gauge; The area where strain gauges are allowed to be installed.
[0087] Step 6.4: Set the optimal number of strain gauges n.
[0088] Step 6.5: Use the Sequential Quadratic Programming (SQP) algorithm to solve the problem shown in formula (22). SQP is one of the most effective methods for solving constrained nonlinear optimization problems, with advantages such as good convergence, high computational efficiency, and strong boundary search capabilities. Since the global optimal value searched by the algorithm may depend on the initial point, Latin hypercube sampling (LHS) is used to generate 100 uniform initial points for optimization, and then the optimal solution is selected as the final optimization result. ,in, Indicates the installation position and direction of each strain gauge.
[0089] Step 7: Based on the final optimization result obtained by the above method, the strain data corresponding to the final optimization result is further used to identify the load. The details are as follows:
[0090] Step 7.1: Final optimization result based on step 6.5 , and obtain the strain-load coefficient matrix corresponding to the final optimization result .
[0091] Step 7.2: Apply a load of any magnitude and direction to the original loading position of the model and record the load as a vector , where each component represents a load component under a certain coordinate axis.
[0092] Step 7.3: Obtain the test data matrix corresponding to the optimal strain gauge under the load applied in step 7.2 .
[0093] Step 7.4: Identify the load using the left pseudo-inverse method. The specific process is:
[0094] When the strain-load coefficient matrix is determined, the measured strain data of the structure is used as input to invert the external load of the structure. However, since it is difficult to construct a global strain test, an approximate solution of F is generally obtained by calculating some measuring points. The approximate solution of the load can be expressed as:
[0095] (twenty three);
[0096] in, is the strain matrix of the local measuring point; is the strain-load coefficient matrix corresponding to the local measuring point. When the matrix is full rank, It can be expressed as:
[0097] (twenty four);
[0098] Therefore, the approximate solution of the load is obtained as:
[0099] (25);
[0100] Step 7.5: Verify the rationality of the final optimization results. Calculate the identified loads With actual load The error between them is shown in formula (26), and the identified load component is calculated The actual load component The error percentage between them is shown in formula (27).
[0101] (26);
[0102] (27)
[0103] When the error percentage of each load component 、error When both are less than 10%, it can be shown that the test data corresponding to the strain gauge optimization result can accurately identify the load and verify the rationality of the optimization result.
[0104] The beneficial effects of the present invention are:
[0105] (1) The present invention utilizes the plane strain transformation principle combined with the local coordinate system to create a spatial strain state transformation method, which can calculate the strain at any position on the surface of the structure in any direction, thus solving the problem of difficulty in analyzing the surface strain of complex mechanical structures.
[0106] (2) The present invention constructs a continuous mapping relationship between variables through a proxy model method based on discrete data, thereby expanding the scope of optimization to all possible layouts.
[0107] (3) The present invention can customize the installation range of the strain gauge and the minimum distance between the strain gauges, thereby preventing the strain gauges corresponding to the optimization results from falling in positions that are not suitable for installation and preventing the strain gauges from interfering with each other.
[0108] (4) The present invention adopts Latin hypercube sampling method to generate multiple groups of initial points for optimization respectively, and obtains the final optimal solution from the multiple groups of optimization results, thereby avoiding the optimization result falling into a local optimal solution due to the selection of the initial points. BRIEF DESCRIPTION OF THE DRAWINGS
[0109] Figure 1 It is an overall flow chart of the strain gauge optimal layout method for load identification based on the proxy model provided by the present invention.
[0110] Figure 2It is a schematic diagram of the strain state of the cement pump truck case provided by the present invention when loaded and the surface selected for study; in the figure, G represents the gravity exerted on the structure, and F represents the load exerted on the end of the first boom by the remaining booms of the pump truck; the top surfaces 1 and 2, the side surface 1 and the bottom surface 1 are the structural surfaces selected for study.
[0111] Figure 3 It is a schematic diagram of the installation range of strain gauges on each studied structural surface on the first arm of the cement pump truck case provided by the present invention.
[0112] Figure 4 It is a flow chart of the spatial strain state conversion method provided by the present invention.
[0113] Figure 5 It is the objective function value corresponding to the optimization process of the four surfaces on the pump truck respectively performed by the method of the present invention from 100 initial points and the objective function value corresponding to the optimization results of the four surfaces obtained by the classic method D-Optimal; wherein (a) shows the corresponding result on the top surface 1; (b) shows the corresponding result on the top surface 2; (c) shows the corresponding result on the side surface 1; (d) shows the corresponding result on the bottom surface 1.
[0114] Figure 6 It is the optimal strain gauge layout scheme on the four surfaces of the pump truck obtained by the method of the present invention and the D-Optimal method; wherein (a) shows the corresponding result on the top surface 1; (b) shows the corresponding result on the top surface 2; (c) shows the corresponding result on the side surface 1; and (d) shows the corresponding result on the bottom surface 1.
[0115] Figure 7 It is the maximum load identification error corresponding to the optimal strain gauge layout scheme obtained by the present invention and the 100 strain gauge layout scheme obtained by LHS under the same strain data interference; wherein (a) shows the corresponding result on top surface 1; (b) shows the corresponding result on top surface 2; (c) shows the corresponding result on side surface 1; (d) shows the corresponding result on bottom surface 1. DETAILED DESCRIPTION
[0116] The present invention is described below in conjunction with specific embodiments.
[0117] The overall flow chart of the optimal layout method of strain gauges for load identification based on the agent model is as follows: Figure 1As shown, the present invention uses the first boom section of a cement pump truck as an engineering case to achieve the optimal strain gauge layout on the first boom section and identify the magnitude of the load it receives during operation. The specific implementation method is further described below in conjunction with the accompanying drawings and technical solutions. In this engineering case, there are two loads applied. One is the gravity G received by the first boom section itself, and this value is determined by the structure of the first boom section itself and the mass of cement in the arm; the other is the load F exerted on the end of the first boom section by the remaining boom sections of the pump truck, and both load directions are vertically downward. Among them, the point of gravity is located at the center of gravity of the first boom section. When there is no cement in the arm, the magnitude is 51587.2N; the position where the remaining boom sections apply load to the first boom section is at the end shaft hole. When there is no cement in the arm, the magnitude is 80000N. Combine the two loads into a load vector , when analyzing two loads acting independently on the structure, it can be expressed as F 1 ⃗ =[51587.2 N ,0] , F 2 ⃗ =[0,80000N] The strain state of the first boom of the pump truck under load is as follows: Figure 2 shown.
[0118] In the first step, a geometric model is established based on the physical entity of the first boom section of the pump truck, and finite element analysis is used to obtain the structural units, nodes, and strain response data under various loads.
[0119] Step 1.1: Use 3D modeling software to build a geometric model based on the geometric features of the physical entity of the structure.
[0120] Step 1.2: For the geometric finite element model obtained in step 1.1, remove its internal structural details by appropriately simplifying its screw holes and other tiny features using methods such as geometric merging or deletion to obtain a simplified geometric model.
[0121] Step 1.3: Mesh the simplified geometric finite element model obtained in step 1.2, where the structure surface is divided using shell elements, and the mesh elements, node numbers, and node coordinates of the structure surface are saved as a local file.
[0122] Step 1.4: Apply two loads independently to the simplified geometric finite element model obtained in step 1.2, extract the strain response value of each unit in each degree of freedom and save it as a local file.
[0123] Step 1.5: Set the appropriate strain gauge installation range on the surface of the first boom section of the pump truck, excluding the model edge, the interface between surfaces, and the area near the load position where stress concentration is likely to occur. Figure 3As shown, in the present invention, four areas on the top surface, the side surface and the bottom surface of the first boom are used as the installation range.
[0124] Step 1.6: Extract the unit, node data and corresponding strain data within the installation range as analysis data for subsequent steps.
[0125] In the second step, a local coordinate system is created with the center position of each unit within the installation range.
[0126] Step 2.1: Calculate the normal vector of each element by using the coordinate information of each node on each element in the global coordinate system Oxyz .
[0127] Step 2.2: Find the vector orthogonal to the normal vector , and find the vector and Vectors that are orthogonal . (Special note: For a plane, its normal vector is unique, but the number of mutually orthogonal coordinate axes that can be constructed on the plane is infinite. The vector constructed by the present invention is and is one of the ones that meets the conditions and serves as the coordinate reference for subsequent calculations).
[0128] Step 2.3: Get , , After these three vectors are divided by their respective moduli, the direction cosines of the three vectors are calculated as follows: , , , respectively, as the x', y', and z' axes of the unit local coordinate system O'x'y'z', and their set is R, as shown in formula (1).
[0129] The third step is to create a spatial strain state conversion method based on the plane strain state conversion principle and the local coordinate system to obtain the strain value of each unit center in each direction. The process is as follows Figure 4 shown.
[0130] Step 3.1: Based on the strain response data of each unit in the global coordinate system Oxyz, specifically the shear strain in the x, y, and z directions And the shear strains in the yz, xz, and xy directions , and transform it through formula (2) and formula (3) to obtain the shear strain in the x', y', and z' directions in the unit local coordinate system O'x'y'z' And the shear strains in the y'z', x'z', and x'y' directions , as shown in formulas (2) and (3).
[0131] The three shear strains in the global coordinate system and three shear strains The set composed of , like the set of stress components, also obeys the transformation law of the second-order tensor, so It can be expressed as formula (4).
[0132] The three shear strains in the local coordinate system of the unit And the three shear strains The set composed of According to formulas (1) to (4), we can obtain and The conversion relationship between them is shown in formula (5).
[0133] Step 3.2: According to the plane strain state conversion principle, obtain the strain value in the direction of the angle α between the x' axis of the local coordinate system of the unit and the plane O'x'y' where each unit is located. , as shown in formula (6).
[0134] The fourth step is to construct the strain-load coefficient matrix of the model according to the load identification principle. The details are as follows:
[0135] Step 4.1: Process and analyze the strain response values under the independent action of the two loads obtained in step 1.4. Since most large mechanical structures belong to the category of linear structural systems during operation, there is a linear relationship between load, strain and displacement. Therefore, on the surface of the first boom section of the pump truck, the strain in a certain direction at any position is the superposition of the strains in the corresponding directions caused by multiple external loads at the same time. An important assumption here is that the stress acting on the structure will never exceed the elastic limit, and the displacement is small enough so that the superposition principle is valid.
[0136] According to formula (7) to formula (12), the strain-load coefficient under the independent action of two loads can be obtained.
[0137] Step 4.2: Obtain the corresponding shear strain by independently applying two different loads, calculate the corresponding strain-load coefficient by formula (12), and then assemble the strain-load coefficients under the independent action of the two different loads to obtain the strain-load coefficient matrix.
[0138] The fifth step is to form a combination of the coordinate value representing the position and the direction vector representing the angle, and then establish a proxy model between the combination and the strain-load coefficient under the independent action of two loads. The above coordinate values, direction vectors and strain-load coefficients are all discrete data obtained based on the simplified geometric finite element model, and the proxy model can make the discrete data continuous and construct a functional relationship between the two. The details are as follows:
[0139] Step 5.1: Determine the input variable x, and use the coordinates of the center point of the unit in the selected range on the structure to represent the position, denoted as , p is the product of the number of units and the number of directions studied; at the same time, the third step proposes a method for calculating the strain value of any direction on the tangent plane of each unit. Since the angle α is relative to the O'x'y' coordinate system on the tangent plane at each position (that is, the angle with the x' axis direction is α), it cannot be used as a unified measurement method. Therefore, the present invention uses the unit vector under the global coordinate system Oxyz coordinate system to represent the different angle directions on the tangent plane at each position, which is recorded as According to formula (13), the conversion between direction vector and angle can be realized. Therefore, the position coordinates and direction vector constitute the input: , which eventually forms a variable set .
[0140] Step 5.2: For two independent load cases, establish Strain-load factor corresponding to the working condition The proxy model is described as follows:
[0141] Surrogate Model (SM) is a data-driven model that focuses on data and relies on mathematical expressions to respond quickly and accurately without understanding the physical rules between input variables and output responses. Therefore, it can be applied to optimization design and analysis, especially when the mechanism between input and output is unclear and the computational cost is high.
[0142] Many classic SMs have been proposed, such as polynomial regression surface radial basis function (RBF), polynomial regression surface (PRS), Kriging (KRG), support vector regression (SVR). Depending on different requirements or goals, they can all achieve accuracy or ideal results. Considering the calculation time and accuracy, RBF is used here.
[0143] RBF is a commonly used multivariate scattered point interpolation prediction algorithm. This method transforms complex high-dimensional problems into simple linear problems through the kernel function. The kernel function is radially symmetric about the center of the training point. The general expression is shown in Formula (14).
[0144] As for the basis function, since the multi-quadric basis function (Multiquadric, MQ) has a better ability to express the relationship between input and output, the present invention adopts the MQ basis function, as shown in formula (15).
[0145] Using R-square (R 2 ) is used to describe the accuracy of the proxy model, as shown in formula (17); the constructed proxy model is shown in formula (18).
[0146] In this case, the four surfaces are , The proxy models are constructed under two independent loads, and their accuracy is shown in Table 1:
[0147] Table 1. Accuracy of the proxy models constructed for four surfaces under two independent loading cases.
[0148]
[0149] Among them, the accuracy of all proxy models exceeded 0.99, proving that they can accurately show the relationship between variables.
[0150] The sixth step is to construct an optimization problem and obtain the optimal strain gauge layout solution.
[0151] Step 6.1: Minimizing the load identification error is used as the objective function. According to the basic principles of statistics, the relationship between the load identification error, the strain gauge test error, and the strain-load coefficient corresponding to the strain gauge position and orientation is shown in formula (19).
[0152] Since strain data is inevitably affected by factors such as strain gauge measurement accuracy and environmental noise, in actual measurement, if the error of each strain data is independent and the standard deviation is , then the covariance matrix of the recovered load can be expressed as formula (20), from which it can be seen that the accuracy of the identified load is given by Therefore, in order to minimize the load identification error, The largest, among which is the strain-load coefficient corresponding to the position and direction of the strain gauge. If the optimal result (i.e. the optimal strain gauge layout) is , where n is the number of strain gauges to be installed, formula (21) must be satisfied.
[0153] Step 6.2: Set constraints. First, each strain gauge must be located on the surface of the structure. The four surfaces studied can be expressed as equations , then it is necessary to satisfy ; Secondly, to avoid interference between strain gauges during the actual installation process, the minimum distance between strain gauges must be greater than the length of the strain gauge itself. The minimum distance is D. In this case, D=10mm; Furthermore, the unit vector representing the installation direction of the strain gauge Must be consistent with the normal vector at the installation location Vertical, otherwise the strain gauge cannot be installed on the structural surface. Finally, the optimization result must be within the installation range determined in step 1.4. .
[0154] Step 6.3: Construct the optimization problem and obtain the optimal solution. According to the above description, the optimization problem can be expressed as formula (22).
[0155] Step 6.4: Set the optimal number n of strain gauges on each surface. Generally, the number of strain gauges should be greater than or equal to the number of external loads to be identified. In this case, the number of loads is 2, so the present invention sets n=2.
[0156] Step 6.5: Use the Sequential Quadratic Programming (SQP) algorithm to solve this problem. SQP is one of the most effective methods for solving constrained nonlinear optimization problems, with advantages such as good convergence, high computational efficiency, and strong boundary search capabilities. Since the global optimal value searched by the algorithm may depend on the initial point, Latin hypercube sampling (LHS) is used to generate 100 uniform initial points for optimization, and then the optimal solution is selected as the final optimization result. .like Figure 5 As shown in the figure, the strain gauge layout scheme corresponding to the minimum value of the objective function is selected as the optimal result.
[0157] The optimal layout of strain gauges on the four surfaces of the first arm is as follows: Figure 6 The specific values are shown in Table 2. The layout plan includes the position coordinates and direction vectors of the strain gauge installation.
[0158] Table 2, Optimal layout of strain gauges on four surfaces.
[0159]
[0160] In the seventh step, the optimization results obtained by the above method are used to further identify the load using the strain data corresponding to the results.
[0161] Step 7.1: Obtain the strain-load coefficient matrix corresponding to the optimization results .
[0162] Step 7.2: Apply a new load to the original load position of the first boom section .
[0163] Step 7.3: Obtain the test data matrix corresponding to the optimal strain gauge under the load applied in step 7.2 .
[0164] Step 7.4: Identify the load using the left pseudo-inverse method. The specific principle is:
[0165] When the strain-load coefficient matrix is determined, the measured strain data of the structure is used as input to invert the external load of the structure. However, since it is difficult to construct a global strain test, an approximate solution of F is usually obtained by calculating some measuring points.
[0166] According to formula (23) to formula (25), the approximate solution of load F can be calculated: .
[0167] Step 7.5: Verify the rationality of the optimization results. Calculate the total error between the identified load and the actual load, as shown in formula (26), and calculate the identified load component The actual load component The error percentage between , as shown in formula (27). When the error percentage of each load component is less than 10%, it can be said that the test data corresponding to the strain gauge optimization result can accurately identify the load, verifying the rationality of the optimization result.
[0168] The loads and their errors identified by the strain test data corresponding to the optimal layout of strain gauges on the four surfaces of the first boom of the pump truck are shown in Table 3. It can be seen that the identified load error is close to 0, which verifies the rationality of the optimization result and can accurately identify the load.
[0169] Table 3, loads and their errors identified by strain test data corresponding to the optimal layout of strain gauges on four surfaces.
[0170]
[0171] Step 8: Verify the superiority of the optimization results.
[0172] Step 8.1: Generate 100 sets of strain gauge layout schemes using the LHS method. Combine the above method to obtain the optimal layout scheme, with a total of 101 sets of strain gauge layout schemes.
[0173] Step 8.2: Use the Monte Carlo method to apply 10% disturbance to the test data corresponding to the 101 sets of strain gauge layout schemes obtained in step 8.1; then, use the disturbed data to estimate the load; and compare the influence of the data disturbance on the estimated loads under these 101 sets of schemes. Figure 7As shown, the maximum error values of load identification corresponding to the optimal layout scheme of strain gauges obtained by the present invention are all smaller than the maximum error values of load identification corresponding to 100 sets of layout schemes of strain gauges obtained by LHS, which proves the superiority of this method and its strong robustness in the face of interference.
[0174] Step 8.3: Compare the method proposed in the present invention with the classic optimal layout method for load identification, namely the D-Optimal method based on discrete data. Figure 5 As shown, the optimal objective function values obtained by the D-Optimal method are higher than the optimal objective function values obtained by the present invention, which means that the results obtained by the present invention are better and can identify the load more accurately.
[0175] The above-described embodiments merely express the implementation methods of the present invention, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.
Claims
1. The optimal layout method of strain gauges for load identification is characterized by: The optimal layout method of strain gauges first establishes a geometric finite element model of the mechanical structure, and obtains the unit and node information of the mechanical structure and the strain response data under the independent action of multiple loads. The strain response data includes the shear strain in the three directions of x, y, and z in the global coordinate system and the shear strain in the three directions of yz, xz, and xy; secondly, a local coordinate system of the model surface is created through the unit and node information, and the shear strain in the remaining directions is obtained by combining the strain response data with the plane strain state conversion principle; then, when multiple loads are applied to the model independently, the corresponding strain-load coefficient is obtained according to the shear strain in each direction and the corresponding load size, and the continuous mapping relationship between the strain gauge layout and the strain-load coefficient is established by using the proxy model; finally, a suitable installation range is selected as the position constraint, and the minimum distance between the strain gauges is constrained at the same time, and the strain gauge layout is optimized with the minimum load identification error as the objective function; The specific steps include: The first step is to establish a geometric finite element model based on the physical entity of the mechanical structure, and use finite element analysis to obtain the structural units, nodes, and strain response data under various loads; The second step is to create a local coordinate system at the center of each unit based on the unit and node information within the strain gauge installation range; the details are as follows: Step 2.1: Calculate the normal vector of each element by using the coordinate information of each node on each element in the global coordinate system Oxyz Step 2.2: Find the vector orthogonal to the normal vector And find the vector and Vectors that are orthogonal Step 2.3: Based on The direction cosines of the three vectors are obtained as The x', y', and z' axes are respectively used as the local coordinate system O'x'y'z' of the unit, and the set of them is R; The third step is to create a spatial strain state conversion method based on the plane strain state conversion principle and the local coordinate system to obtain the strain value of each unit center in each direction; the details are as follows: Step 3.1: Based on the strain response data of each unit in the global coordinate system Oxyz, specifically the shear strain ε in the x, y, and z directions x , ε y , ε z And the shear strain γ in the yz, xz, and xy directions yz , γ xz , γ xy , transform it to obtain the shear strain ε in the x', y', and z' directions in the unit local coordinate system O'x'y'z' x′ , ε y′ , ε z′ and the shear strains γ in the y'z', x'z', and x'y' directions y′z′ , γ x′z′ , γ x′y′ ; Three shear strains ε in the global coordinate system x , ε y , ε z and three shear strains γ yz , γ xz , γ xy The set composed of ij It is expressed as formula (4): The three shear strains ε in the local coordinate system of the unit x′ , ε y′ , ε z′ and the three shear strains γ y′z′ , γ x′z′ , γ x′y′ The set composed of i′j′ It is expressed as shown in formula (5): e i′j′ =Rε ij R T (5); Among them, R is the matrix composed of the three direction cosines corresponding to the unit local coordinate system; Step 3.2: Obtain the strain value ε in the direction of the angle α between the x' axis of the local coordinate system of the unit and the plane O'x'y' where each unit is located. α , as shown in formula (6): Where α is the strain value ε α The angle between the x' axis and the local coordinate system of the unit; The fourth step is to construct the strain-load coefficient matrix of the model according to the load identification principle; the details are as follows: Step 4.1: Process and analyze the strain response data obtained in the first step. The relationship between strain and load is expressed as: e p×m =A p×m F m×m (7); Where p is the product of the number of units and the number of directions studied; m is the number of external loads; ε p×m is the strain matrix, each column represents the shear strain of each unit in each direction under a certain load application; A p×m is the strain-load coefficient matrix, each column represents the strain-load coefficient applied independently to the model under a certain load; F m×m is the load matrix, each column represents the load applied on the structure; Under a certain load application condition, the load number is set to f, and the strain calculation formula on each unit is: Among them, ε i,f (i=1,2,…,p) is the shear strain of a unit in a certain direction on the structure; A p×m is the strain-load coefficient matrix. Once the unit and the direction of study are determined, its value remains constant. i,f (i=1,2,…,m) is the magnitude of each load applied to the structure; From formula (8), we can get: i,f The calculation method of (i=1,2,…,p) is shown in formula (9): ε i,f =A i,1 F 1,f +A i,2 F 2,f +…+A i,m F m,f (9); Among them, A i,k (k=1,2,…,m) is ε i,f The corresponding strain-load factor; When only the first load F1 acts, formula (8) is: ε in formula (10) i,1 (i=1,2,…,p) represents the shear strain of a unit on the model surface in a certain direction under the first load of F1 only; Since the actual number of applied loads is m, when each load is applied independently, the corresponding shear strain value can be obtained. These data are assembled and formula (10) is expanded to formula (11): Among them, in the strain matrix ε p×m In the load matrix F, each column represents the corresponding shear strain vector when the corresponding column load is applied independently; m×m In the example, each column is a load vector when a certain load is applied independently. i (i=1,2,…,m) is the magnitude of each independently applied load; The strain-load coefficient matrix A is obtained from formula (11): p×m The calculation method for each column in is shown in formula (12): Where, j = 1, 2, ..., m; Step 4.2: Obtain the corresponding shear strain by applying different loads independently, calculate the corresponding strain-load coefficient by formula (12), and then assemble the strain-load coefficients under these different loads independently to obtain the strain-load coefficient matrix A p×m ; The fifth step is to form a combination of the coordinate value representing the position and the direction vector representing the angle, and then establish a proxy model between the combination and the strain-load coefficient under the independent action of each load; the details are as follows: Step 5.1: Determine the input variable x, and use the coordinates of the unit center point within the selected range on the structure to represent the position, denoted by (c r1 ,c r2 ,c r3 ), (r=1,2,…,p), p is the product of the number of units and the number of directions studied; the unit vector in the global coordinate system Oxyz coordinate system is used to represent the different angles and directions on the tangent plane at each position, denoted as The conversion between direction vector and angle is realized according to formula (13): Then the position coordinates and direction vector form the input x r =(c r1 ,c r2 ,c r3 ,d r1 ,d r2 ,d r3 ), and finally form the variable set X p×6 ; Step 5.2: For different independent load conditions, establish X p×6 Strain-load factor A corresponding to the working condition ·,j , (j=1,2,…,m), as shown in formula (18): A .,j =f j (X p×6 ),(j=1,2,…,m) (18); Among them, A .,j is the strain-load factor under a certain load independent action, f j X constructed under the independent action of a certain load p×6 With A ·,j Functional expression of the proxy model between them; The sixth step is to construct an optimization problem to obtain the optimal strain gauge layout solution; the details are as follows: Step 6.1: Minimizing the load identification error is taken as the objective function; Step 6.2: Set constraints; First, each strain gauge is located on the surface of the structure. If the surface is recorded as the equation F(x, y, z) = 0, then it is necessary to satisfy F(c k1 ,c k2 ,c k3 )=0, where (c k1 ,c k2 ,c k3 ) is the position coordinate of the strain gauge; secondly, the minimum distance between the strain gauges must be greater than the length of the strain gauge itself, and the minimum distance is recorded as D; thirdly, the unit vector representing the installation direction of the strain gauge The normal vector at the installation position vertical; finally, set the area ∑ where strain gauges are allowed to be installed; Step 6.3: Construct an optimization problem and obtain the optimal solution; The optimization problem is expressed as: (22); where k and s are the numbers of the strain gauges and are not equal to each other; (c k1 ,c k2 ,c k3 ) is the position coordinate of the strain gauge; is the unit vector of the strain gauge installation direction; is the normal vector at the strain gauge installation location; ∑ is the area where the strain gauge is allowed to be installed; Step 6.4: Set the optimal number of strain gauges n; Step 6.5: Use the sequential quadratic programming (SQP) algorithm to solve the optimization problem shown in formula (22) and obtain the final optimization result; Step 7: Use the strain data corresponding to the final optimization result to identify the load; the details are as follows: Step 7.1: Based on the final optimization result obtained in step 6.5, obtain the strain-load coefficient matrix corresponding to the final optimization result Step 7.2: Apply a load of any magnitude and direction to the original loading position of the model, and record the load as a vector F = [F1, F2, …, F m ], where each component represents a load component under a certain coordinate axis; Step 7.3: Obtain the test data matrix corresponding to the optimal strain gauge under the load applied in step 7.2 Step 7.4: Identify the loads according to the left pseudo-inverse method; Step 7.5: Verify the rationality of the final optimization results.
2. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: The first step is as follows: Step 1.1: According to the geometric characteristics of the physical entity of the mechanical structure, a geometric finite element model is established using 3D modeling software; Step 1.2: Simplify the geometric finite element model obtained in step 1.1 to obtain a simplified geometric model; Step 1.3: Mesh the simplified geometry finite element model, using shell elements to mesh the model surface, and save the surface mesh elements, node numbers, and node coordinates as a local file; Step 1.4: Apply different loads to the simplified geometric finite element model obtained in step 1.2, extract the strain response data on each grid unit and save it as a local file; Step 1.5: Set the appropriate strain gauge installation range on the model surface, excluding areas prone to stress concentration such as model edges, interfaces between surfaces, and near the load position; Step 1.6: Extract the unit and node information and the corresponding strain response data within the installation range.
3. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: In step 2.3, R is as shown in formula (1); in, l i 、m i 、n i are the components of the corresponding direction cosines, where i = 1, 2, 3.
4. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: In step 3.1, the strain response data is transformed by the following formula to obtain the shear strain ε in the x', y', and z' directions in the unit local coordinate system O'x'y'z' x′ , ε y′ , ε z′ and the shear strains γ in the y'z', x'z', and x'y' directions y′z′ , γ x′z′ , γ x′y′ ; c y′z′ =2(e x l2l3+e y m2m3+e z n2n3)+γ yz (m2n3+m3n2)+γ xz (l2n3+l3n2)+γxy(l2m3+l3m2); c x′z′ =2(e x l1l3+e y m1m3+e z n1n3)+γ yz (m1n3+m3n1)+γ xz (l1n3+l3n1)+γ xy (l1m3+l3m1); c x′y′ =2(e x l1l2+e y m1m2+e z n1n2)+γ yz (m1n2+m2n1)+γ xz (l1n2+l2n1)+γ xy (l1m2+l2m1); Among them, l i 、m i 、n i (i=1,2,3) are direction cosines respectively The amount.
5. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: In step 6.1, the relationship between the load identification error, the strain gauge test error, and the strain-load coefficient corresponding to the position and orientation of the strain gauge is shown in formula (19): In the formula, is the test data of the strain gauge, is the strain-load factor corresponding to the position and orientation of the strain gauge, is the variance-covariance matrix of the strain, is the variance-covariance matrix of the identified loadings; Define the error of each strain data as independent and the standard deviation as σ, then the covariance matrix of the recovery load is expressed as formula (20): From formula (20), we can see that in order to minimize the load identification error, The largest, among which is the strain-load coefficient corresponding to the position and direction of the strain gauge. If the optimal result is {x k =(c k1 ,c k2 ,c k3 ,d k1 ,d k2 ,d k3 )}(k=1,2,...,n), where n is the number of strain gauges to be installed, the formula must be satisfied: Among them, ε j is the shear strain under a certain load, m is the external load number, is the strain-load coefficient corresponding to the position and direction of the optimal strain gauge layout under a certain independent load.
6. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: The specific process of step 7.4 is as follows: When the strain-load coefficient matrix is determined, the measured strain data of the structure is used as input, the external load of the structure is inverted, and the approximate solution of F is obtained by calculating some measuring points; the approximate solution of the load is expressed as: in, is the strain matrix of the local measuring point; is the strain-load coefficient matrix corresponding to the local measuring point. When the matrix is full rank, It can be expressed as: Therefore, the approximate solution of the load is obtained as:
7. The strain gauge optimal layout method for load identification according to claim 1 is characterized in that: In step 7.5, the verification process is as follows: Calculate the load identified in step 7.4 The error between the actual load F and the load F is shown in formula (26); Calculation of identified load components and the actual load component F j The error percentage between (j=1,2,…,m) is shown in formula (27); When the error value is less than 10%, it means that the test data corresponding to the strain gauge optimization result can accurately identify the load.
Citation Information
Patent Citations
Sensor layout method for hypersonic flight vehicle control surface structure distribution load identification
CN110532607A
Multi-type sensor layout optimization method for load performance evaluation under driving of particle swarm optimization algorithm
CN117350096A
Contact analysis method of numerical load tooth surface based on hyperboloid shell element model
CN109145484A
Strain gauge position determination method based on determinant maximization algorithm
CN113836764A
Method for indirectly measuring cable tension based on bollard surface strain
CN117852368A
Cited By
System and method for identifying optimal locations for strain gage placement on a structure
US12656194B2
System and method for identifying optimal locations for strain gage placement on a structure
US20240060835A1
