A creep curve prediction method

Abstract

The invention discloses a creep curve prediction method, comprising the following steps: step 1, performing creep experiments under different experimental conditions, extracting creep curves obtained from multiple tests, extracting required condition parameters and creep strain to construct a data set; step 2, fitting the data set in step 1 with the Larson-Miller parameter equation; step 3, taking the condition parameters of the data set constructed in step 1 as input and the creep strain as output, selecting network training to obtain a creep curve prediction model; step 4, combining the Larson-Miller equation obtained by fitting in step 2 with the creep curve prediction model obtained in step 3, to realize creep curve prediction under the conditions covered by the data set. The method can realize high-precision prediction of creep curves, and provide a more accurate reference for understanding the creep characteristics of materials, material design and life assessment throughout the whole process.

Description

Technical Field

[0001] The invention relates to the field of material science and engineering technology, in particular to a creep curve prediction method. Background Art

[0002] During the service process, the blades of gas turbines are often affected by high temperature and centrifugal loads, which are prone to creep deformation and failure, seriously affecting the safety of the unit. Since the 1950s, the creep curves and life prediction research of various structural materials have been widely carried out. The θ projection method, which appeared in 1982, established the relationship between creep strain and creep time by phenomenologically describing the creep curve. It has good life prediction ability and is widely used in various types of industrial alloys. However, in the actual creep process, the true creep stress increases with the increase of deformation, and it is difficult to use the θ projection method to accurately predict the constant stress creep curve. Moreover, this method is relatively scattered in the relationship between temperature and stress, and it is difficult to apply to curves with large creep strain. Chinese patent CN108256179A discloses a method of combining actual measurement with extrapolation to predict creep curves using a power exponential model. Chinese patent CN105910921A discloses a method based on a newly created creep model to achieve high-precision creep curve prediction.

[0003] In recent years, data-driven machine learning methods have attracted widespread attention in the prediction of material service life. Chinese patent CN115035963A discloses a method based on creep condition creep time creep rupture life data set, using BPNN neural network and Larson Miller parameter equation to achieve data-driven accurate prediction of creep curve of high-temperature alloy. However, this method lacks the reflection of key information of creep process, and the prediction results are not accurate enough with the measured values ​​(relative error is 20%). Summary of the invention

[0004] The purpose of the present invention is to provide a creep curve prediction method for the above-mentioned problems, which can achieve high-precision prediction of creep curves and provide a more accurate reference for understanding the creep characteristics of materials, material design and life assessment throughout the process.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A creep curve prediction method comprises the following steps:

[0007] Step 1: Conduct creep experiments under different experimental conditions, extract creep curves obtained from multiple tests, extract required condition parameters and creep strain to construct a data set;

[0008] Step 2, fitting the Larson-Miller parameter equation to the data set in step 1;

[0009] Step 3: Taking the conditional parameters of the data set constructed in step 1 as input and the creep strain as output, network training is selected to obtain a creep curve prediction model;

[0010] 4. Combine the Larson-Miller equation obtained by fitting in step 2 and the creep curve prediction model obtained in step 3 to realize the creep curve prediction under the data set coverage condition.

[0011] Specifically, the Larson-Miller parameter equation is an empirical formula used to predict the creep life of materials at constant temperature. It simplifies the prediction of creep life by combining temperature and stress into one parameter. Neural networks are a powerful machine learning tool that can capture and simulate complex nonlinear relationships in data. For creep prediction problems, neural networks can learn the creep behavior of materials under different temperature and stress conditions through training data, and then predict the creep life under unseen conditions.

[0012] When the Larson-Miller parameter equation is combined with a neural network, the empirical knowledge of the physical model is actually embedded into the data-driven neural network model. This combination provides a reliable starting point for the neural network, which helps the network learn faster and improves the accuracy of prediction. Through the Larson-Miller parameter equation, more synthetic data can be generated based on limited experimental data. These synthetic data can be used as training samples for the neural network. The synthetic data can help the neural network better understand the creep behavior of materials under different conditions, thereby enhancing the learning ability of the network. In addition, although the Larson-Miller parameter equation can describe some aspects of the creep life of materials, it may not fully capture all nonlinear relationships. Neural networks are good at learning nonlinear relationships, thereby improving the accuracy of predictions. They can supplement and improve the parts that the Larson-Miller parameter equation fails to cover, and automatically adjust internal parameters during training to minimize prediction errors. This process can be regarded as an adaptive optimization of the material constant C in the Larson-Miller parameter equation. In addition, the structure of the neural network model, such as the number of layers, the number of neurons, the activation function, etc., can be adjusted according to specific problems to obtain the best prediction performance.

[0013] Optionally, the required condition parameters in step 1 include creep temperature, creep stress, creep time, creep rupture life, and transition time between the second and third stages of creep. Among them, temperature is one of the key factors affecting the creep behavior of the material. As the temperature increases, the atomic diffusion rate increases, making the material more susceptible to creep. The stress level is directly related to the dislocation movement and lattice deformation inside the material, and is the driving force for creep. In the process of constructing the data set, creep time refers to a series of time from loading to rupture. In the data set, the input creep time and creep strain are a series of time-strain data recorded during the experiment. Creep rupture life is the time required for a material to rupture at constant temperature and stress, and it is a key indicator for evaluating the long-term service performance of the material. Creep is usually divided into three stages: the primary stage (deceleration stage), the second stage (steady-state stage), and the third stage (acceleration stage). In the second stage (steady-state stage) of creep, the creep rate of the material is relatively stable, so the long-term behavior of the material can be predicted by monitoring the creep rate. When the material enters the third stage (acceleration stage), the creep rate increases rapidly, which usually indicates that the material is about to rupture. The transition time of the second and third stages can be regarded as an early warning signal, marking the critical transition of the creep performance of the material and the imminent entry into the rapid failure stage. Therefore, the transition time of the second and third stages can be used as an important indicator of the remaining life of the material. This scheme can verify and improve the creep prediction model and improve the accuracy of the prediction by accumulating data on the transition time of the second and third stages of material creep under different conditions.

[0014] Optionally, select 10 creep curves obtained from the tests in step 1. These 10 creep curves should be able to represent the diversity of the entire data set. They should cover different creep temperatures, creep stresses, creep times, creep rupture lives, creep second and third stage transition times, and different creep strains. The data collection and processing methods should be consistent to ensure that the condition parameters and creep strains are obtained under the same measurement standards.

[0015] Alternatively, the creep rupture life can be obtained by fitting in step 2 and creep second and third stage transition time Equation for creep rupture life and creep second and third stage transition time These are two extremely important time points in the creep process. is the time when the material fails completely, and It is the moment when the material enters the accelerated creep stage from steady-state creep. These two time points are directly related to the failure behavior of the material, so accurately predicting them is the key to understanding the entire creep process and can more accurately capture the creep behavior of the material under different stresses.

[0016] Alternatively, the equation is: , ,in is the creep rupture life, is the transition time between the second and third stages of creep, T is the creep temperature, and σ is the creep stress. It is a common variant used in industrial applications. Represents stress The polynomial, C LM Refers to the creep constant of a material, which is often related to the material itself.

[0017] Alternatively, in step 3, RNN neural network training is selected to obtain a creep curve prediction model. RNN is specially designed to process sequence data. Creep curve is time series data, in which the data at each time point is related to the previous time point. RNN is able to consider previous information at each time step through its loop structure, and creep is a process that evolves over time, so RNN neural network improves the performance of the model in predicting creep curves. In addition, the Larson-Miller parameter equation is an empirical formula for predicting the creep life of materials at specific temperatures and times. This equation involves time as one of the variables, so processing time-related data is its key requirement. RNN can remember previous states and use this information for prediction, which is more suitable for processing time series data. The combination of the two plays a greater role in creep prediction tasks.

[0018] Alternatively, in step 3, an RNN network with 10 to 40 neurons in each layer is selected for training to obtain a creep curve prediction model. Too large or too small a number of neurons will lead to a sharp drop in the accuracy of curve fitting. Based on the reported literature and the existing research foundation of this scheme, the number of neurons is usually controlled in the range of 5-50. The number of neurons is different for different creep data and selected neural network types. Compared with traditional neural networks (ANNs), RNNs can capture dynamic features in time series data, so more neurons are usually required to effectively learn these complex time dependencies. However, compared with long short-term memory networks (LSTMs), RNNs cannot effectively handle long-term dependency problems due to their relatively simple structure. Therefore, in terms of the selection of the number of neurons, their complexity and capacity should usually be between ANN and LSTM. This scheme systematically screens the number of neurons and finally determines the optimization range of 10 to 40 neurons. For different data sets, appropriate adjustments within this range can significantly improve the prediction accuracy of the model.

[0019] Optionally, step 4 includes the following steps:

[0020] Step 4.1: Use the Larson-Miller equation obtained in step 2 to obtain the creep rupture life under the conditions to be determined. and creep second and third stage transition time Specifically, we first construct a data set using the data from the training set, from which we can obtain: creep temperature, creep stress, creep rupture life, creep second and third stage transition time. We use this data to fit two LM curves. By fitting curves from limited data, we can theoretically obtain the life and t23 value under any temperature stress combination from these two curves. Through the two LM curves, we first obtain the creep rupture life of the test set under the creep conditions to be predicted. And the transition time of the second and third stages of creep .

[0021] Step 4.2: Substitute the conditions to be determined in step 4.1 and the creep rupture life and creep second and third stage transition time Input the creep curve prediction model obtained in step 2 to obtain the creep strain under the desired condition. The predicted creep curve can be obtained based on the creep time and creep strain. Construct a time series from the beginning to the rupture to predict creep temperature, creep stress, and creep rupture life , creep second and third stage transition time , creep time (time series) is used as the input of the RNN model, and the creep strain value corresponding to the creep time is obtained, which is the output result of the RNN model. The predicted creep curve can be obtained based on the creep time and creep strain.

[0022] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0023] The creep curve prediction method provided by the present invention combines the Larson-Miller equation and the network training creep curve prediction model, which can make full use of the advantages of physical knowledge and data-driven models, improve the prediction ability, generalization ability and interpretability of the model, and reduce the experimental cost and development cycle. Combined with the RNN neural network, the time series characteristics in the data can be captured. The combination of the two can simultaneously utilize the power of physical knowledge and data-driven to improve the accuracy and reliability of the prediction. While maintaining the flexibility and strong learning ability of the RNN model, physical constraints are introduced to improve the practicality and effectiveness of the model in actual engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0025] Figure 1 This is the structure diagram of the RNN neural network used.

[0026] Figure 2 This is a graph of the neural network training loss values ​​in the present invention.

[0027] Figure 3 It is a LM curve diagram of the creep rupture life and the transition time of the second and third stages of creep according to an embodiment of the present invention.

[0028] Figure 4 This is a creep curve error comparison diagram obtained in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0030] All features disclosed in this specification, or steps in all methods or processes disclosed, except mutually exclusive features and / or steps, can be combined in any manner.

[0031] Any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0032] Example

[0033] A method for predicting creep curves of MarM247 oriented nickel-based high-temperature alloy used as a blade material for a gas turbine specifically comprises the following steps:

[0034] 1. Construct a data set: Conduct creep experiments under different experimental conditions, and obtain 10 corresponding creep curves through creep data, obtain creep temperature, creep stress, creep time, creep rupture life, creep second and third stage transition time, creep strain and other data, and establish a corresponding data set.

[0035] 2. Perform the following operations on the data sets in step 1: , The fitting of the two formulas gives the temperature-stress-creep rupture life and the second and third stage transition time of temperature-stress-creep By using these two equations, the rupture time and the second and third stage transition times under any creep condition can be determined.

[0036] 3. Take the creep temperature, creep stress, creep time, creep rupture life, creep second and third stage transition time in the data set of step 1 as input, and the creep strain as output, and select RNN network training to obtain the creep curve prediction model. The RNN network structure is as follows: Figure 1 , the network training loss value is as follows Figure 2 .

[0037] 4. First, use the data of the training set to construct a data set, from which we can obtain: creep temperature, creep stress, creep rupture life, creep second and third stage transition time, and use this data to fit two LM curves, such as Figure 3 As shown in the figure, the curves are fitted from the limited data. Theoretically, the life and t23 values ​​under any temperature stress combination can be obtained from these two curves. Through the two LM curves, the creep rupture life of the test set under the creep condition to be predicted is first obtained. And the transition time of the second and third stages of creep , then, the creep rupture life Construct a time series from the beginning to the rupture to predict creep temperature, creep stress, and creep rupture life , creep second and third stage transition time , creep time (time series) as the input of the RNN model, and the creep strain value corresponding to the creep time is obtained, which is the output result of the RNN model. According to the creep time and creep strain, the predicted creep curve can be obtained, which can realize the creep curve prediction of any temperature stress state under the data set coverage conditions, and then obtain any interruption time under the conditions. The relative error between the predicted curve and the experimental curve is ±10%, such as Figure 4 shown.

[0038] The present invention is not limited to the above-mentioned specific embodiments, but extends to any new features or any new combination disclosed in this specification, as well as any new method or process steps or any new combination disclosed.

Claims

1. A creep curve prediction method, characterized in that: Step 1, creep experiments are carried out under different creep temperatures and creep stress experimental conditions, 10 creep curves obtained from the tests are extracted, and the required condition parameters and creep strain are extracted to construct a data set. The required condition parameters include creep temperature, creep stress, creep time, creep rupture life, and transition time of the second and third stages of creep; Step 2: Fit the Larson-Miller parameter equation to the data set in step 1 to obtain the creep rupture life and creep second and third stage transition time The equation of Step 3: Take the conditional parameters of the data set constructed in step 1 as input and the creep strain as output, and select the RNN neural network training to obtain the creep curve prediction model; Step 4: Combine the Larson-Miller parameter equation obtained by fitting in step 2 with the creep curve prediction model obtained in step 3 to achieve creep curve prediction under the condition of data set coverage.

2. The creep curve prediction method according to claim 1, characterized in that: The equation is: , ,in is the creep rupture life, is the transition time between the second and third stages of creep, T is the creep temperature, σ is the creep stress, C LM Refers to the material creep constant.

3. The creep curve prediction method according to claim 1, characterized in that: In step 3, a RNN network with 10 to 40 neurons in each layer is selected for training to obtain a creep curve prediction model.

4. The creep curve prediction method according to claim 1, characterized in that: Step 4 includes the following steps: Step 4.1: Obtain the creep rupture life under the desired conditions using the Larson-Miller parameter equation obtained in step 2 and creep second and third stage transition time ; Step 4.2: Substitute the conditions to be determined in step 4.1 and the creep rupture life and creep second and third stage transition time The creep curve prediction model obtained in step 3 is input to obtain the creep strain under the desired condition, and the predicted creep curve is obtained according to the creep time and the creep strain.

Images