A reliability method for complex structural systems based on a multi-level distributed collaborative agent model

Abstract

The present invention discloses a complex structural system reliability method based on a multi-level distributed collaborative agent model, which is applied to the field of aircraft engine reliability technology. The method includes the following steps: first, decompose the multi-level complex structural system into single-level and single-level distributed sub-problems; second, coordinate them to establish collaborative sub-problems according to the correlation between the various levels and levels of the complex structural system; third, use the finite element simulation method to determine the multi-dimensional input variables and multiple output responses of each sub-problem; then, use the integrated linkage sampling technology to synchronously extract the multi-dimensional input variables and multiple output responses, and establish a multi-level agent model suitable for each sub-problem; finally, use the established multi-level agent model to complete the reliability analysis of the complex structural system. Compared with the prior art, the present invention can effectively deal with the key high nonlinearity and strong coupling problems in the reliability analysis of complex structural systems, which brings convenience to improving calculation efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft engine reliability, and specifically relates to a reliability method for a complex structural system with multiple assembly components and multiple failure modes, and in particular to a reliability method for a complex structural system based on a multi-level distributed collaborative agent model. Background Art

[0002] Complex rotating parts such as aviation gas turbine engines are often complex structural systems composed of multiple components, each of which has multiple failure modes. Their design problems involve complex design analysis problems at multiple levels. For example, the bladed rotor system is composed of a disk, blades, main shaft, etc., which involves the design and analysis of multiple components; under the action of multiple field loads such as aerodynamics, machinery and temperature, the failure mode of the same component is manifested as low-cycle fatigue, high-cycle fatigue, creep and other failure modes. For such complex structural systems, collaborative design at all levels is required to achieve good economic results. The deterministic collaborative design of complex structural systems is already cumbersome. If reliability design is carried out on this basis, the computational efficiency and computational accuracy will become the bottleneck restricting its engineering application. If reliability design is carried out independently at each level and level of the complex structural system, the computational efficiency can be improved, but the entire system is split, and the result is bound to be not a high-precision reliability design; if the reliability design of the complex structural system is directly carried out as a whole, not only will the existing computing platform be unable to withstand such a large amount of calculation, but it will also not conform to the process of complex system design.

[0003] The application of surrogate models is an effective way to improve the computational efficiency of reliability design of complex structural systems. However, the reliability design of complex structural systems involves the coupling relationship between random variables, the high nonlinearity between random variables and analysis targets, and the correlation between various levels. If a "big" surrogate model is established for the complex structural system, although the computational efficiency is improved compared with the finite element or real structural model, the established surrogate model is difficult to accurately reflect the complex mapping relationship between the analysis target and the random variables, and cannot effectively deal with strong coupling, high nonlinearity and other problems, so that the established overall surrogate model is also facing severe tests of accuracy and efficiency. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the present invention provides a complex structure system reliability method based on a multi-level distributed collaborative agent model to improve the calculation accuracy and efficiency of the complex structure system reliability.

[0005] The technical solution adopted by the present invention is: a complex structure system reliability method based on a multi-level distributed collaborative agent model, comprising:

[0006] Step S1, decomposing the reliability problem of a complex structural system involving multiple analysis levels, each of which contains multiple analysis levels, into distributed sub-problems of a single level and a single level; the analysis level refers to: reliability design of multiple components and multiple failure modes involved in a complex structural system assembled from multiple components, each of which has multiple failure modes; the analysis level refers to: reliability analysis of each component and each mode contained in each analysis level of the complex structural system;

[0007] Step S2, according to the correlation between various levels and grades of the complex structural system, coordinate them to establish a collaborative sub-problem; wherein the correlation between various levels and grades of the complex structural system refers to the deformation coordination relationship between components and the correlation between multiple failure modes;

[0008] Step S3, considering multiple physical field variables and multiple randomness of life models, using finite element simulation method to perform deterministic analysis of fatigue life of complex structural systems, and determining multi-dimensional input variables and multiple output responses of each sub-problem;

[0009] Step S4, synchronously sampling the multi-dimensional input variables and the multiple output responses using an integrated linkage sampling technique;

[0010] Step S5, selecting a multi-objective proxy model or a hybrid proxy model suitable for each sub-problem according to the characteristics of each level of sub-problems, such as explicit or implicit, high or low nonlinearity, strong or weak coupling, etc.; the multi-objective proxy model includes a scalar target proxy model, a vector target proxy model, etc.; the hybrid proxy model includes a polynomial model, a support vector regression model, a Kriging model, and an artificial neural network model;

[0011] Step S6, using a two-step error control technique to search for optimal parameters of the established agent model, and establishing a multi-level collaborative agent model;

[0012] Step S7, based on the multi-level collaborative agent model, a numerical simulation method is used to complete the reliability analysis of the complex structure.

[0013] The complex structure system reliability method based on the multi-level distributed collaborative agent model provided by the present invention also has the following characteristics:

[0014] The step S1 comprises the following sub-steps:

[0015] Step S11, decomposing the "macro" problem of reliability analysis of complex structural systems into "micro" distributed sub-problems at multiple analysis levels according to the mapping relationship of each layer;

[0016] Step S12, decomposing into multiple “micro” distributed sub-problems at multiple analysis levels according to the characteristics of the dependent variables within the layer;

[0017] The complex structure system reliability method based on the multi-level distributed collaborative agent model provided by the present invention also has the following characteristics:

[0018] Wherein said step S3 comprises the following sub-steps:

[0019] Step S31, select the environmental load, material properties, and life model parameters of the complex structure system as random variables. The environmental load includes: rotation speed, temperature; the material properties include: elastic modulus, material density, thermal conductivity, thermal expansion coefficient; the life model parameters include fatigue strength index, fatigue ductility index, fatigue strength coefficient, and fatigue ductility coefficient;

[0020] Step S32, establishing a finite element model of a complex structural system;

[0021] Step S33, based on the radial temperature loading criterion and the temperature continuity distribution law, the input variables in step S31 are imported into the finite element model, a steady-state thermal analysis is performed on the complex structure system, and the temperature distribution of the complex structure system is obtained;

[0022] Step S34, then transmitting the obtained temperature load to the structural force field, completing the thermal-solid coupling deterministic analysis of the complex structural system, obtaining the maximum stress and strain distribution of the complex structural system, and determining the multidimensional input variables and multiple output responses of each sub-problem through the maximum stress and strain amplitude;

[0023] The complex structure system reliability method based on the multi-level distributed collaborative agent model provided by the present invention also has the following characteristics:

[0024] The integrated linkage sampling technology in step S4 is: using the Latin hypercube sampling method to synchronously sample the multi-dimensional input variables and multiple output responses of the complex structure system for multiple times.

[0025] The complex structure system reliability method based on the multi-level distributed collaborative agent model provided by the present invention also has the following characteristics:

[0026] Among them, in step S6, the specific steps of the two-step error control technology are: first, use the heuristic algorithm to search for the initial undetermined parameters, and then use the gradient descent algorithm to train the final undetermined parameters. The heuristic algorithm includes simulated annealing algorithm, particle swarm optimization algorithm, genetic algorithm, ant colony algorithm, etc.; the gradient descent algorithm includes: stochastic gradient descent algorithm, Newton method, quasi-Newton method, conjugate gradient method, etc.;

[0027] The complex structure system reliability method based on the multi-level distributed collaborative agent model provided by the present invention also has the following characteristics:

[0028] Wherein, the step S7 includes the following sub-steps:

[0029] Step S71, using the established multi-level collaborative agent model to replace the real complex structural system for reliability analysis;

[0030] Step S72: According to the allowable values ​​of the targets at each level, the limit state equation of the complex structural system is established. Assuming that the allowable output response of the structure is [Y], multiple uncertain variables such as material properties, environmental loads, and life prediction models are used as input random variables x, and the number of flight cycles of the blade disk structure is used as the output response y(x). Based on the proposed proxy model function, a multi-level proxy model is established, and the limit state function of the structure can be characterized as:

[0031] G(x)=[Y]-y(x) (3)

[0032] Step S73, based on the integrated linkage sampling technology, large-scale sampling is performed on the multi-dimensional input variables and multiple output responses to obtain one-to-one corresponding sample data, and the sample data is brought into the established multi-level proxy model to calculate the output response of each level target;

[0033] Step S74, based on the Monte Carlo simulation concept, a reliability model of the complex structural system is established:

[0034]

[0035] In the formula, I r [G(x l )] is the security domain indicator function, G(x l )≥0 indicates that the structure is safe, G(x l )<0 indicates structural failure; N r is the number of sample points in the security domain; N is the total number of samples; x l is the lth sample point l=1,2,…,N; R represents reliability; st represents constraint conditions. Step S75, based on the data in step S73 and the reliability model in step S74, calculate the fatigue life under 99.87% reliability.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] Through this multi-level splitting and linkage combination, it is possible to effectively deal with the high nonlinearity and strong coupling problems of complex structural systems, which brings convenience to improving calculation efficiency and accuracy. The specific reasons are as follows: compared with the "big" model of the reliability analysis problem of complex structural systems, the coupling relationship of the sub-problems after decomposition by level and level is weaker and the degree of nonlinearity is also lower, which is conducive to reducing modeling complexity and improving approximation efficiency; multi-level and multi-level distributed agent models and collaborative agent models are established respectively, which effectively realizes the multi-level explicitness of complex structural systems and provides convenience for improving calculation accuracy; through the synchronous sampling of all parameters and the optimization solution of model hyperparameters, the validity of samples and the accuracy of modeling are effectively guaranteed; different types of agent models are established for different sub-problems, which can maximize the potential and advantages of multiple types of agent models while avoiding the limitations of single-type agent models. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of the complex structure system reliability method based on the multi-level distributed collaborative agent model of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with specific embodiments, but is not limited to the specific embodiments.

[0040] The following takes the creep fatigue reliability analysis of the turbine blade disk structure of an aircraft engine as an example to further illustrate the reliability method of a complex structural system based on a multi-level distributed collaborative agent model of the present invention. The specific steps are as follows:

[0041] like Figure 1 As shown, the complex structure system reliability method based on the multi-level distributed collaborative agent model of the present invention includes the following steps:

[0042] Step 1, decomposing the reliability problem of the turbine blade disk structure involving multiple analysis levels, each analysis level containing multiple analysis levels, into distributed sub-problems of a single level and a single level; the analysis level refers to: the reliability design of multiple components and multiple failure modes involved in the turbine blade disk structure assembled from multiple components and each component having multiple failure modes, i.e., the reliability design of multiple components and multiple failure modes of the turbine blade disk structure; the analysis level refers to: the analysis level contained in each analysis level of the turbine blade disk structure, i.e., the reliability analysis of each component and each mode, i.e., the reliability analysis of each component and each failure mode of the turbine blade disk structure;

[0043] This embodiment includes the following sub-steps:

[0044] Step 11, decomposing the "macro" problem of turbine blade disk structure reliability analysis into "micro" distributed sub-problems at multiple analysis levels according to the mapping relationship of each layer;

[0045] Step 12, decompose into “micro” distributed sub-problems at multiple analysis levels according to the characteristics of the dependent variables within the layer;

[0046] In this embodiment, the complex structure system is a turbine blade structure assembled from multiple components, each component having multiple failure modes. The turbine blade structure is composed of two components, the wheel disc and the blade. Under the action of complex loads such as fluid load, thermal load, solid load, etc., each component of the turbine blade may have failure modes such as low cycle fatigue, high cycle fatigue, creep, etc.;

[0047] Step 2: Coordinate and establish collaborative sub-problems based on the correlations between various levels and grades of the complex structural system; the correlations between various levels and grades of the complex structural system refer to the deformation coordination relationship between components and the correlation between multiple failure modes;

[0048] Step 3, considering the multi-physics field variables and the multiple randomness of the life model, the finite element simulation method is used to perform a deterministic analysis of the fatigue life of the turbine blade disk structure to determine the multi-dimensional input variables and multiple output responses of each sub-problem;

[0049] The step 3 includes the following sub-steps:

[0050] Step 31, select the turbine blade disk structure environmental load, material properties, and life model parameters as random variables. The environmental load includes: rotation speed, temperature; the material properties include: elastic modulus, material density, thermal conductivity, thermal expansion coefficient; the life model parameters include fatigue strength index, fatigue ductility index, fatigue strength coefficient and fatigue ductility coefficient;

[0051] Step 32, establishing a finite element model of a complex structural system;

[0052] Step 33, based on the radial temperature loading criterion and the temperature continuity distribution law, the random variables in step S31 are imported into the finite element model as input variables, a steady-state thermal analysis is performed on the complex structure system, and the temperature distribution of the complex structure system is obtained;

[0053] Step 34, then transmit the obtained temperature load to the structural force field, complete the thermal-solid coupling deterministic analysis of the turbine blade disk structure, obtain the maximum stress and strain distribution of the turbine blade disk structure, and determine the multi-dimensional input variables and multiple output responses of each sub-problem through the maximum stress and strain amplitude;

[0054] In this embodiment, it is found through deterministic analysis that the blade root portion with the greatest stress is prone to low-cycle fatigue failure, and the tenon tooth portion with higher stress and temperature is prone to creep fatigue failure.

[0055] Step 4, using an integrated linkage sampling technique to synchronously sample the multidimensional input variables and multiple output responses; the integrated linkage sampling technique is: using a Latin hypercube sampling method to synchronously sample the multidimensional input variables and multiple output responses of the turbine blade disk structure multiple times.

[0056] Step 5: Select a multi-objective proxy model or hybrid proxy model suitable for each sub-problem according to the characteristics of each level of sub-problems, such as explicit or implicit, high or low nonlinearity, strong or weak coupling, etc.;

[0057] In this embodiment, the multi-target proxy model includes a scalar target proxy model, a vector target proxy model, etc.; the hybrid proxy model includes a polynomial model, a support vector regression model, a Kriging model, and an artificial neural network model;

[0058] Step 6, using the two-step error control technology to search for the optimal parameters of the established agent model, and establish a multi-level collaborative agent model; in step 6, the specific steps of the two-step error control technology are: first, using the heuristic algorithm to search for the initial undetermined parameters, and then using the gradient descent algorithm to train the final undetermined parameters. The heuristic algorithms include simulated annealing algorithm, particle swarm optimization algorithm, genetic algorithm, ant colony algorithm, etc.; the gradient descent algorithms include: stochastic gradient descent algorithm, Newton method, quasi-Newton method, conjugate gradient method, etc.;

[0059] In this embodiment, the differential evolutionary algorithm (DE) is first used to search for the initial parameters to be determined, and then the gradient descent algorithm (GD) is used to train the final parameters to be determined. Its advantage is that the DE algorithm uses cooperation and competition between vector populations to find the global optimal solution. Its differential-based mutation operation and competitive survival strategy can effectively reduce the search complexity and avoid premature convergence. After the DE algorithm is used to search for the optimal initial network parameters, the network needs to be further trained to ensure the prediction accuracy of the model;

[0060] Step 7: Based on the multi-level collaborative agent model in step 6, a numerical simulation method is used to complete the reliability analysis of the turbine blade disk structure.

[0061] The step 7 includes the following sub-steps:

[0062] Step 71, using the established multi-level collaborative agent model to replace the actual turbine blade disk structure for reliability analysis;

[0063] Step 72, based on the allowable values ​​of the targets at each level, the limit state equation of the turbine blade disk structure is established. Assuming that the allowable output response of the structure is [Y], multiple uncertain variables such as material properties, environmental loads, and life prediction models are used as input random variables x, and the number of flight cycles of the blade disk structure is used as the output response y(x). Based on the proposed proxy model function, a multi-level proxy model is established, and the limit state function of the structure can be characterized as follows:

[0064] G(x)=[Y]-y(x) (5)

[0065] Step 73, based on the integrated linkage sampling technology, a large-scale sampling is performed on the multi-dimensional input variables and the multiple output responses to obtain one-to-one corresponding sample data, and the sample data is brought into the established multi-level proxy model to calculate the output response of each level target;

[0066] Step 74, based on the Monte Carlo simulation concept, establish a reliability model of the complex structural system:

[0067]

[0068] In the formula, I r [G(x l )] is the security domain indicator function, G(x l )≥0 indicates that the structure is safe, G(x l )<0 indicates structural failure; N r is the number of sample points in the security domain; N is the total number of samples; x l is the lth sample point l=1,2,…,N; R represents reliability; st represents constraint condition.

[0069] Step 75 , based on the sample data in step 73 and the reliability model in step 74 , calculate the fatigue life of the turbine blade disk at a reliability of 99.87%.

[0070] The above embodiments are provided only for the purpose of describing the present invention, rather than for limiting the scope of the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A complex structural system reliability method based on a multi-level distributed collaborative agent model, characterized in that: The following steps are involved: Step S1, decomposing the reliability problem of a complex structural system involving multiple analysis levels, each of which contains multiple analysis levels, into distributed sub-problems of a single level and a single level; the analysis level refers to: a multi-component, multi-failure mode reliability design involved in a complex structural system assembled from multiple components, each of which has multiple failure modes; the analysis level refers to: reliability analysis of each component and each mode contained in each analysis level of the complex structural system; Step S2, according to the correlation between various levels and grades of the complex structural system, coordinate them to establish a collaborative sub-problem; wherein the correlation between various levels and grades of the complex structural system refers to the deformation coordination relationship between components and the correlation between multiple failure modes; Step S3, considering multiple physical field variables and multiple randomness of life models, using finite element simulation method to perform deterministic analysis of fatigue life of complex structural systems, and determining multi-dimensional input variables and multiple output responses of each sub-problem; Step S4, synchronously sampling the multi-dimensional input variables and the multiple output responses using an integrated linkage sampling technique; Step S5, selecting a multi-objective proxy model or a hybrid proxy model suitable for each sub-problem according to the characteristics of explicit or implicit, high or low nonlinearity, strong or weak coupling of each level of sub-problem; the multi-objective proxy model is a scalar target proxy model or a vector target proxy model; the hybrid proxy model is a polynomial model, a support vector regression model, a Kriging model or an artificial neural network model; Step S6, using a two-step error control technique to search for optimal parameters of the established agent model, and establishing a multi-level collaborative agent model; Step S7, based on the multi-level collaborative agent model, a numerical simulation method is used to complete the reliability analysis of the complex structural system.

2. The complex structure system reliability method based on the multi-level distributed collaborative agent model according to claim 1 is characterized by: The step S1 includes the following sub-steps: Step S11, decomposing the macro problem of reliability analysis of complex structural systems into micro-distributed sub-problems at multiple analysis levels according to the mapping relationships of each layer; Step S12, decomposing into micro-distributed sub-problems at multiple analysis levels according to the characteristics of the dependent variables within the layer.

3. The complex structure system reliability method based on the multi-level distributed collaborative agent model according to claim 1 is characterized by: The step S3 includes the following sub-steps: Step S31, selecting environmental loads, material properties, and life model parameters of the complex structural system as random variables; the environmental loads include: rotation speed and temperature; the material properties include: elastic modulus, material density, thermal conductivity, and thermal expansion coefficient; the life model parameters include fatigue strength index, fatigue ductility index, fatigue strength coefficient, and fatigue ductility coefficient; Step S32, establishing a finite element model of a complex structural system; Step S33, based on the radial temperature loading criterion and the temperature continuity distribution law, the random variables in step S31 are imported into the finite element model as input variables, a steady-state thermal analysis is performed on the complex structure system, and the temperature distribution of the complex structure system is obtained; In step S34, the obtained temperature load is then transmitted to the structural force field to complete the thermal-solid coupling deterministic analysis of the complex structural system, obtain the maximum stress and strain distribution of the complex structural system, and determine the multidimensional input variables and multiple output responses of each sub-problem through the maximum stress and strain amplitude.

4. The complex structure system reliability method based on the multi-level distributed collaborative agent model according to claim 1 is characterized by: The integrated linkage sampling technology in step S4 is: using the Latin hypercube sampling method to synchronously sample the multi-dimensional input variables and multiple output responses of the complex structure system multiple times.

5. The complex structure system reliability method based on the multi-level distributed collaborative agent model according to claim 1 is characterized by: In step S6, the specific steps of the two-step error control technology are: first, using a heuristic algorithm to search for initial unknown parameters, and then using a gradient descent algorithm to train final unknown parameters; wherein the heuristic algorithm is a simulated annealing algorithm, a particle swarm optimization algorithm, a genetic algorithm or an ant colony algorithm; and the gradient descent algorithm is a stochastic gradient descent algorithm, a Newton method, a quasi-Newton method or a conjugate gradient method.

6. The complex structure system reliability method based on the multi-level distributed collaborative agent model according to claim 1 is characterized by: The step S7 includes the following sub-steps: Step S71, using the established multi-level collaborative agent model to replace the real complex structural system for reliability analysis; Step S72, based on the allowable values ​​of the targets at each level, the limit state equation of the complex structural system is established. Assuming that the allowable output response of the structure is [Y], the multiple uncertain variables of material properties, environmental loads, and life prediction models are used as the input random variable x, and the number of flight cycles of the blade disk structure is used as the output response y(x). Based on the proposed proxy model function, a multi-level proxy model is established, and the limit state function of the structure can be characterized as: G(x)=[Y]-y(x) (1) Step S73, based on the integrated linkage sampling technology, large-scale sampling is performed on the multi-dimensional input variables and multiple output responses to obtain one-to-one corresponding sample data, and the sample data is brought into the multi-level proxy model established in step S72 to calculate the output response of each level target; Step S74, based on the Monte Carlo simulation concept, a reliability model of the complex structural system is established: In the formula, I r [G(x l )] is the security domain indicator function, G(x l )≥0 indicates that the structure is safe, G(x l )<0 indicates structural failure; N r is the number of sample points in the security domain; N is the total number of samples; x l is the lth sample point l=1,2,…,N; R represents reliability; st represents the constraint condition; Step S75, based on the data in step S73 and the reliability model in step S74, the fatigue life at a reliability of 99.87% is calculated.

Images