System layer reliability modeling method based on data-physics combined driving

Abstract

The invention provides a data-physics combined drive system layer reliability modeling method. Damage parameter data are obtained through an external load working condition of a mechanical system and a finite element model; selecting a reference variable, and respectively establishing a system external load-reference variable proxy model and a reference variable-part damage parameter proxy model by applying a machine learning model based on the damage parameter data set; acquiring external load data information, and acquiring probability distribution of reference variables through an external load-reference variable machine learning model; reflecting a failure correlation relationship among parts in the system through a reference variable-part damage parameter machine learning model; based on the probability distribution of the reference variables and the failure correlation between the parts, a system layer reliability modeling method is applied to establish a data-physics combined drive system layer reliability model so as to solve the system reliability under the complex load working condition, and the accuracy and rationality of system reliability evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical system reliability assessment, and in particular to a data-physics jointly driven system-level reliability modeling method. Background Art

[0002] The system is composed of parts. However, the system reliability is not completely determined by the reliability of the parts. The relationship between system reliability and part reliability is also related to the failure correlation between parts. In the reliability problem, a system in which each part fails independently is called an independent failure system. In such a system, the failure of each part (or any component of a certain level of the system, such as subsystems, parts, etc., which can be collectively referred to as "units") is a random event that is independent of each other. The traditional system reliability model is established under the assumption that each part fails independently. The typical system reliability analysis method is to calculate the reliability of the parts with the help of an interference model, or to determine the reliability of the parts through test data estimation, and then establish a system reliability model based on the logical structure of the system (series, parallel, voting system, etc.) under the assumption that "the failures of each part in the system are independent of each other". This system reliability model directly constructed from the reliability of parts cannot reflect the important role of failure correlation. The traditional thinking method is to calculate the system reliability based on the reliability of parts, and the assumption of independent failure of parts is indispensable, otherwise it is difficult to establish a system reliability model. However, almost all mechanical systems are not independent failure systems. Only when the loads borne by the parts in the system are unrelated (no logical connection between the loads, independent of each other), or the load is a deterministic quantity (no randomness), the failure of the parts in the system is independent of each other. In general, the failure events of the parts in the system are statistically related random events. The degree of correlation of the failure of the parts in the system depends on the relative size of the uncertainty of the load (workload) and the uncertainty of the strength (the ability of the part to resist failure). Except in the case where the uncertainty of the load is very small (especially relative to the uncertainty of the part strength), the traditional system reliability model will have obvious errors.

[0003] In recent years, the ideas and methods of system engineering have been further developed and applied in many fields. In system reliability analysis, "system-level" or "system-layer" reliability analysis and modeling methods have also been proposed. Applying such ideas and methods can not only easily establish an accurate model of system reliability, but also avoid making assumptions that do not conform to the actual situation during the modeling process, such as "the failure of components is independent of each other", which indeed shows the connotation of "the whole is simpler than the sum of its parts" in modeling.

[0004] Taking the stress-strength interference model as an example, when each part in the system is under a certain stress sUnder the action of , the conditional failure probability of the parts is completely determined by their strength distribution. In this case, the failure of each part is independent of each other. The load-damage parameter mapping relationship is a key link in the system-level reliability model, which can effectively reflect the failure correlation between the parts in the system. However, there is usually a complex nonlinear relationship between the load of the mechanical system and the damage parameters of the parts, so the load-damage parameter mapping relationship is often difficult to obtain. In addition, for systems with multiple external loads, how to solve multi-layer integrals is a key issue in solving the system reliability model. At present, there is no more effective multi-layer integral solution method or tool, which limits the promotion and application of such system reliability models in engineering. On the other hand, most of the existing system reliability models are suitable for the design stage and do not have the function of real-time update during the service of the system, which makes it difficult to meet the needs of high-precision reliability evaluation of the system under complex load conditions. Summary of the invention

[0005] Based on this, it is necessary to propose a data-physics jointly driven system-level reliability modeling method to address the technical problem that the existing technology has not taken into account that the existing technology is difficult to meet the needs of high-precision reliability assessment of systems under complex load conditions.

[0006] In a first aspect, a data-physics joint driven system layer reliability modeling method is provided, including: Acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system; Selecting a baseline variable, and based on the damage parameter data, establishing a system external load-benchmark variable proxy model and a baseline variable-part damage parameter proxy model through machine learning technology; Acquire external load data information according to sensor data, and input the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable; Inputting the reference variable into the reference variable-part damage parameter proxy model to obtain the failure correlation relationship between the parts in the mechanical system; Based on the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint-driven system-level reliability model, and the reliability of the mechanical system is determined according to the system-level reliability model. Optionally, the step of acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system comprises: Acquiring attribute information of the mechanical system, and establishing a finite element model of the mechanical system according to the attribute information; External load condition data of the mechanical system determined by orthogonal test design method; Finite element analysis is performed on the mechanical system based on the finite element model and the load condition, and combined with the damage parameter model, damage parameter data of each part in the mechanical system corresponding to the external load condition data is obtained.

[0007] Optionally, the step of selecting a baseline variable and establishing a system external load-baseline variable proxy model and a baseline variable-part damage parameter proxy model by machine learning technology based on the damage parameter data includes: Select any part in the mechanical system as a reference part, use the damage parameter corresponding to the reference part in the damage parameter data as a reference variable, establish and train a machine learning model based on the damage parameter data using a machine learning technique, with an external load variable as an input variable and a reference variable as an output variable, to obtain an external load-reference variable proxy model of the system; A machine learning model with a baseline variable as an input variable and the damage parameters of each part in the mechanical system as an output variable is established and trained based on the damage parameter data through machine learning technology to obtain the baseline variable-part damage parameter proxy model. Optionally, the step of acquiring external load data information according to sensor data, and inputting the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable includes: determining a probability distribution of an external load variable based on the external load data information; Monte Carlo sampling method is applied to obtain external load combinations according to the probability distribution of external load variables; Acquire a benchmark variable corresponding to the external load combination through the system external load-benchmark variable proxy model; A statistical analysis method is applied to obtain a probability distribution of the benchmark variables corresponding to the external load combination. Optionally, the step of applying a system-level reliability modeling method based on the probability distribution and the failure correlation to establish a data-physics joint-driven system-level reliability model, and determining the reliability of the mechanical system according to the system-level reliability model includes: Based on the load-strength interference theory, according to the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint driven system-level reliability model.

[0008] Optionally, the mathematical representation of the system layer reliability model is:

[0009] In the formula, R s is the system reliability;v b is the baseline variable; g ( v b ) is the probability density function of the benchmark variable; s i ( v b ) indicates parts i The damage parameter is the reference variable v b Function of S i For the i The strength of a part indicates its ability to resist damage and has a game relationship with the damage parameter, not limited to the strength of the material or structure; f i ( S i ) is the i The strength of each part S i The probability density function of n is the number of parts in the system.

[0010] The present application obtains damage parameter data of the mechanical system based on the external load condition data of the mechanical system and the finite element model of the mechanical system; selects a baseline variable, and based on the damage parameter data, establishes a system external load-benchmark variable proxy model and a baseline variable-part damage parameter proxy model through machine learning technology; obtains external load data information according to sensor data, and inputs the external load data information into the system external load-benchmark variable proxy model to obtain the probability distribution of the baseline variable; inputs the baseline variable into the baseline variable-part damage parameter proxy model to obtain the failure correlation between the parts in the mechanical system; based on the probability distribution and the failure correlation, applies a system-level reliability modeling method to establish a data-physics joint-driven system-level reliability model, and determines the reliability of the mechanical system according to the system-level reliability model. By using a reference variable to replace multiple load random variables, multiple outer integrals in the traditional system layer reliability model can be converted into a single outer integral, solving the problem that the traditional system layer reliability model is difficult to solve with multi-layer integrals; establishing a complex mapping relationship between the reference variable and the damage parameters of each part solves the problem that the complex nonlinear relationship between the mechanical system load and the damage parameters of the parts is difficult to obtain, reflecting the failure correlation between the parts in the system; the reliability model can be updated in real time during the service stage to obtain a more accurate real-time reliability of the mechanical system. It has better engineering practicality and higher application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] in: Figure 1 A flowchart of a data-physical joint driven system layer reliability modeling method is provided for an embodiment of the present invention; Figure 2 A schematic diagram of a finite element model in a system-level reliability modeling method driven by a data-physics joint drive is provided for an embodiment of the present invention; Figure 3 A schematic diagram of a dangerous part of a mechanical system in a data-physics joint-driven system-level reliability modeling method is provided for an embodiment of the present invention; Figure 4 A schematic diagram of the topological structure of the system external load-reference variable model provided in an embodiment of the present invention; Figure 5 A prediction effect diagram of the system external load-benchmark variable model provided in an embodiment of the present invention; Figure 6 A schematic diagram of the topological structure of a mapping relationship model between a reference variable and a component damage parameter provided in an embodiment of the present invention; Figure 7 A prediction effect diagram of a benchmark variable-part damage parameter mapping relationship model provided in an embodiment of the present invention; Figure 8 The reliability evaluation result of the steam turbine rotor system provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The present invention is described in detail below through specific embodiments.

[0015] See also Figure 1 As shown, Figure 1 A data-physical joint-driven system layer reliability modeling method provided in an embodiment of the present invention includes: S101, acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system; S102, selecting a reference variable, and establishing a system external load-reference variable proxy model and a reference variable-part damage parameter proxy model through machine learning technology based on the damage parameter data; S103, acquiring external load data information according to sensor data, and inputting the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable; S104, inputting the reference variable into the reference variable-part damage parameter proxy model to obtain the failure correlation relationship between the parts in the mechanical system; S105. Based on the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint-driven system-level reliability model, and the reliability of the mechanical system is determined according to the system-level reliability model. The damage parameter data of the mechanical system is obtained based on the external load condition data of the mechanical system and the finite element model of the mechanical system; a baseline variable is selected, and based on the damage parameter data, a system external load-benchmark variable proxy model and a baseline variable-part damage parameter proxy model are established through machine learning technology; external load data information is obtained according to sensor data, and the external load data information is input into the system external load-benchmark variable proxy model to obtain the probability distribution of the baseline variable; the baseline variable is input into the baseline variable-part damage parameter proxy model to obtain the failure correlation between the parts in the mechanical system; based on the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint driven system-level reliability model, and the reliability of the mechanical system is determined according to the system-level reliability model. By using a reference variable to replace multiple load random variables, multiple outer integrals in the traditional system layer reliability model can be converted into a single outer integral, solving the problem that the traditional system layer reliability model is difficult to solve with multi-layer integrals; establishing a complex mapping relationship between the reference variable and the damage parameters of each part solves the problem that the complex nonlinear relationship between the mechanical system load and the damage parameters of the parts is difficult to obtain, reflecting the failure correlation between the parts in the system; the reliability model can be updated in real time during the service stage to obtain a more accurate real-time reliability of the mechanical system. It has better engineering practicality and higher application value.

[0016] In a possible implementation manner, the step of acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system includes: Acquiring attribute information of the mechanical system, and establishing a finite element model of the mechanical system according to the attribute information; External load condition data of the mechanical system determined by orthogonal test design method; Finite element analysis is performed on the mechanical system based on the finite element model and the load condition, and combined with the damage parameter model, damage parameter data of each part in the mechanical system corresponding to the external load condition data is obtained.

[0017] In a possible implementation, the step of selecting a baseline variable and establishing a system external load-baseline variable proxy model and a baseline variable-part damage parameter proxy model by machine learning technology based on the damage parameter data includes: Select any part in the mechanical system as a reference part, use the damage parameter corresponding to the reference part in the damage parameter data as a reference variable, establish and train a machine learning model based on the damage parameter data using a machine learning technique, with an external load variable as an input variable and a reference variable as an output variable, to obtain an external load-reference variable proxy model of the system; A machine learning model with a baseline variable as an input variable and the damage parameters of each part in the mechanical system as an output variable is established and trained based on the damage parameter data through machine learning technology to obtain the baseline variable-part damage parameter proxy model. In a possible implementation manner, the step of acquiring external load data information according to sensor data, and inputting the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable includes: determining a probability distribution of an external load variable based on the external load data information; Monte Carlo sampling method is applied to obtain external load combinations according to the probability distribution of external load variables; Acquire a benchmark variable corresponding to the external load combination through the system external load-benchmark variable proxy model; A statistical analysis method is applied to obtain a probability distribution of the benchmark variables corresponding to the external load combination. In a possible implementation, the step of applying a system-level reliability modeling method based on the probability distribution and the failure correlation to establish a data-physics joint-driven system-level reliability model, and determining the reliability of the mechanical system according to the system-level reliability model includes: Based on the load-strength interference theory, according to the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint driven system-level reliability model.

[0018] In a possible implementation manner, the mathematical representation of the system layer reliability model is:

[0019] In the formula, R s is the system reliability; v b is the baseline variable; g ( v b ) is the probability density function of the benchmark variable; s i ( v b ) indicates parts i The damage parameter is the reference variable v b Function of S i For the i The strength of a part indicates its ability to resist damage and has a game relationship with the damage parameter, not limited to the strength of the material or structure; f i ( S i ) is the i The strength of each part S i The probability density function of n is the number of parts in the system.

[0020] Exemplarily, the implementation steps of the data-physics joint driven system layer reliability modeling method include: Finite element modeling is performed on the turbine rotor system (mechanical system) to be evaluated. The established finite element model is as follows: Figure 2 shown.

[0021] According to the actual load conditions, the stress conditions of the system are simulated, and the dangerous parts of the system are determined according to the stress state, such as Figure 3 As shown, the results show that the stress at the stress relief groove and blade root groove of the turbine rotor system is relatively large, which is a dangerous part of the system.

[0022] According to the variation range of different types of external loads, several representative load conditions are designed by using the orthogonal test design method. In this embodiment, for the steam turbine rotor, the parameters that can be monitored include steam temperature, steam pressure, speed, load and other parameters. In the finite element simulation, steam temperature, speed and load are mainly used to apply the load. Therefore, orthogonal test sampling is performed on the three load parameters of steam temperature, speed and load, and they are combined into several groups of representative load conditions.

[0023] Finite element analysis of the steam turbine rotor is carried out under several designed representative load conditions to obtain the damage parameters of the dangerous parts under various load conditions. Combined with the damage calculation model, the cumulative damage of each dangerous part under various load conditions is obtained to obtain the proxy model data set.

[0024] The accumulated damage of the stress relief groove was selected as the benchmark variable.

[0025] A machine learning model is established with the external load variable as the input variable and the cumulative damage of the stress release groove as the output variable, which is called the system external load-reference variable proxy model (proxy model 1). The machine learning model established in this embodiment is a three-layer artificial neural network model, such as Figure 4 As shown, the input layer of the artificial neural network model includes 7 nodes, the hidden layer includes 3 nodes, and the output layer includes 1 node. In this embodiment, the machine learning model includes but is not limited to the artificial neural network model and other high-performance machine learning models, which are not limited here. The prediction accuracy of the artificial neural network model is verified by the cross-validation method. The data in the data set is divided into 10 groups, and 9 groups of data are extracted to train the artificial neural network model respectively, and the remaining 1 group of data is used to verify the artificial neural network model. The prediction effect of the artificial neural network model is shown in FIG. Figure 5 The root mean square error (RMSE) and Pearson correlation coefficient (R 2 ) can represent the prediction accuracy of the model. The smaller the RMSE, the better the R 2 The closer it is to 1, the better the model effect. The prediction results show that the artificial neural network model has a good prediction effect.

[0026] According to the actual working condition data collected by the sensor, the statistical analysis method is applied to obtain the probability distribution of various external loads, and the Monte Carlo sampling method is applied to obtain several load parameters, which are combined into several groups of load conditions. The cumulative damage value of the stress release groove corresponding to each group of load conditions is output through the proxy model 1, and the statistical analysis method is applied to obtain the probability distribution of the cumulative damage value of the stress release groove, that is, the probability density function of the benchmark variable is obtained.

[0027] A machine learning model is established with the baseline variable as the input variable and the cumulative damage of each dangerous part as the output variable, which is called the baseline variable-part damage parameter mapping agent model (agent model 2). The machine learning model established in this embodiment is a three-layer artificial neural network model, such as Figure 6As shown, the input layer of the artificial neural network model includes 2 nodes, the hidden layer includes 3 nodes, and the output layer includes 11 nodes. In this embodiment, the machine learning model includes but is not limited to the artificial neural network model and other high-performance machine learning models, which are not limited here. 90% of the data is used to train the artificial neural network model, and the remaining 10% of the data is used to verify the artificial neural network model. The prediction effect of the artificial neural network model is shown in FIG. Figure 7 As shown in Figure 2. The prediction results show that the artificial neural network model has a good prediction effect.

[0028] Based on the obtained proxy model and benchmark variable probability distribution, a data-physics-co-driven system-level reliability model is established:

[0029] In the formula, R s ( m )for m System reliability corresponding to the start-stop process; D rml for m The logarithmic value of the cumulative damage of the stress relief slot corresponding to the start-stop process; h rm ( D rml )for D rml The probability density function of b is the number of dangerous parts; D iml ( D rml )for m Dangerous parts corresponding to the start-stop process i The cumulative damage logarithm of D rml Function of T is the critical value of cumulative damage, which can be obtained through experiments; f 0( T ) is the probability density function of the standard normal distribution; μ Til is the logarithmic mean of the cumulative damage threshold; σ Til is the logarithmic standard deviation of the cumulative damage threshold.

[0030] Solve the system layer reliability model to obtain the reliability-start-stop number relationship of the steam turbine rotor system in this embodiment as follows: Figure 8 shown.

[0031] Specifically, during the design phase of a mechanical system, the system reliability can be calculated based on a given load probability distribution. During the service phase of a mechanical system, the statistical characteristics of the load can be calculated in real time based on sensor monitoring data, and the probability distribution of the baseline variables can be updated to achieve real-time updating of the reliability model and obtain the real-time reliability of the mechanical system.

Claims

1. A data-physical joint driven system-level reliability modeling method, characterized in that: include: Acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system; Selecting a baseline variable, and based on the damage parameter data, establishing a system external load-benchmark variable proxy model and a baseline variable-part damage parameter proxy model through machine learning technology; Acquire external load data information according to sensor data, and input the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable; Inputting the reference variable into the reference variable-part damage parameter proxy model to obtain the failure correlation relationship between the parts in the mechanical system; Based on the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint-driven system-level reliability model, and the reliability of the mechanical system is determined according to the system-level reliability model.

2. The data-physical joint driven system layer reliability modeling method according to claim 1 is characterized in that: The step of acquiring damage parameter data of the mechanical system based on external load condition data of the mechanical system and a finite element model of the mechanical system comprises: Acquiring attribute information of the mechanical system, and establishing a finite element model of the mechanical system according to the attribute information; External load condition data of the mechanical system determined by orthogonal test design method; Finite element analysis is performed on the mechanical system based on the finite element model and the load condition, and combined with the damage parameter model, damage parameter data of each part in the mechanical system corresponding to the external load condition data is obtained.

3. The data-physical joint driven system layer reliability modeling method according to claim 1 is characterized in that: The steps of selecting a baseline variable and establishing a system external load-baseline variable proxy model and a baseline variable-part damage parameter proxy model through machine learning technology based on the damage parameter data include: Select any part in the mechanical system as a reference part, use the damage parameter corresponding to the reference part in the damage parameter data as a reference variable, establish and train a machine learning model based on the damage parameter data using a machine learning technique, with an external load variable as an input variable and a reference variable as an output variable, to obtain an external load-reference variable proxy model of the system; A machine learning model with a baseline variable as an input variable and the damage parameters of each part in the mechanical system as an output variable is established and trained based on the damage parameter data through machine learning technology to obtain the baseline variable-part damage parameter proxy model.

4. The data-physical joint driven system layer reliability modeling method according to claim 1 is characterized in that: The step of acquiring external load data information according to sensor data, and inputting the external load data information into the system external load-reference variable proxy model to obtain the probability distribution of the reference variable includes: determining a probability distribution of an external load variable based on the external load data information; Monte Carlo sampling method is applied to obtain external load combinations according to the probability distribution of external load variables; Acquire a benchmark variable corresponding to the external load combination through the system external load-benchmark variable proxy model; A statistical analysis method is applied to obtain a probability distribution of the benchmark variables corresponding to the external load combination.

5. The data-physical joint driven system layer reliability modeling method according to claim 1 is characterized in that: The step of applying a system-level reliability modeling method based on the probability distribution and the failure correlation to establish a data-physics joint-driven system-level reliability model, and determining the reliability of the mechanical system according to the system-level reliability model includes: Based on the load-strength interference theory, according to the probability distribution and the failure correlation, a system-level reliability modeling method is applied to establish a data-physics joint driven system-level reliability model.

6. The data-physical joint driven system layer reliability modeling method according to claim 1 is characterized in that: The mathematical representation of the system-level reliability model is: In the formula, R s is the system reliability; v b is the baseline variable; g ( v b ) is the probability density function of the benchmark variable; s i ( v b ) indicates parts i The damage parameter is the reference variable v b Function of S i For the i The strength of a part indicates its ability to resist damage and has a game relationship with the damage parameter, not limited to the strength of the material or structure; f i ( S i ) is the i The strength of each part S i The probability density function of n is the number of parts in the system.

Images