Deep learning-based titanium alloy blade rheological property prediction method

Abstract

The invention discloses a deep learning-based titanium alloy blade rheological property prediction method, and belongs to the technical field of titanium alloy hot working stress analysis, and the method specifically comprises the following steps: 1, cutting a plurality of high-temperature tensile samples from a titanium alloy material after diffusion welding; the titanium alloy high-temperature tensile sample is tested, and rheological curves under different deformation conditions are obtained; step 2, constructing a data set by using the tensile condition parameters and the rheological curve data obtained in the step 1, performing normalization processing on the data set, transforming true stress to an interval of [0, 1], and dividing the data set into a training set and a test set; 3, sending the data set divided in the step 2 into a deep learning model, and outputting the deep learning model based on a training set learning result; and 4, for the deep learning model trained in the step 3, carrying out rheological property prediction by using the trained model. According to the method, the high-temperature rheological property of the TC4 titanium alloy is accurately predicted through the deep learning model.

Description

Technical Field

[0001] The present invention relates to the technical field of titanium alloy thermal processing analysis, which belongs to the technical field of titanium alloy thermal processing stress analysis, and in particular to a method for predicting the rheological properties of titanium alloy blades based on deep learning. Background Art

[0002] Titanium alloys are widely used in the aerospace industry due to their high strength, thermal stability, and corrosion resistance. High-strength structural components and parts that require lightweighting are mostly made of titanium alloys.

[0003] Titanium alloys can be divided into α-type, β-type, and α+β-type types based on their organizational structure. α+β-type titanium alloys are particularly outstanding in terms of lightweighting, reducing the weight of the same component by approximately 9% compared to the other two types. TC4 titanium alloy is a typical α+β-type titanium alloy and is the most widely used of all titanium alloys. TC4 titanium alloy has strong deformation resistance and high resilience at room temperature, making it difficult to process at room temperature. However, at high temperatures, its plasticity and formability are significantly improved, so high-temperature forming is often used to obtain the desired shape and titanium alloy components.

[0004] TC4 titanium alloy for aviation undergoes complex thermal-mechanical coupling during the forming process, and its high-temperature rheological response varies greatly under different conditions. Therefore, in actual production, studying the high-temperature rheological response of TC4 titanium alloy plays a guiding role in optimizing process parameters. The high-temperature rheological response of TC4 titanium alloy during the forming process is affected by many factors, such as process parameters such as deformation temperature and deformation rate, and the initial microstructure of the material. An accurate prediction model for the high-temperature rheological properties of TC4 titanium alloy can help optimize the thermal deformation parameters of titanium alloy, and can also simulate and predict the plastic deformation process of titanium alloy.

[0005] With the rapid development of computer technology and data analysis techniques, materials informatics has become a research hotspot that has garnered significant attention. Advanced artificial intelligence methods are gradually entering the traditional field of materials forming, demonstrating significant potential in reducing experimental costs and shortening development cycles. Developing a deep learning-based method for predicting the high-temperature rheological response of titanium alloys under complex deformation conditions is of great significance for accurately predicting the high-temperature rheological response of titanium alloys under complex deformation conditions and optimizing forming process parameters. Summary of the Invention

[0006] In order to accurately predict the rheological response of TC4 titanium alloy during high-temperature forming, the present invention provides a TC4 titanium alloy rheological response prediction method based on deep learning. The main feature of this method is to accurately predict the high-temperature rheological properties of TC4 titanium alloy through a deep learning model.

[0007] The present invention is achieved in that:

[0008] The present invention is achieved through the following technical solutions.

[0009] The present invention provides a method for predicting the rheological properties of titanium alloy blades based on deep learning, which comprises the following steps:

[0010] Step 1: Cutting a number of high-temperature tensile specimens from the titanium alloy material after diffusion welding; using a UTM 5504X universal electronic testing machine to perform high-temperature tensile tests on the titanium alloy high-temperature tensile specimens at different temperatures and strain rates to obtain rheological curves under different deformation conditions;

[0011] Step 2: Using the tensile condition parameters and rheological curve data obtained in step 1, a data set is constructed. The deformation temperature, deformation rate, and true strain are used as inputs, and the true stress is used as output to construct the data set required for the training model. The data set is normalized, the true stress is transformed to the interval [0, 1], and the data set is divided into a training set and a test set. The training set is used to train the model to obtain the optimal parameters, and the test set is used to evaluate the model prediction performance.

[0012] Step 3: The data set divided in step 2 is fed into a deep learning model, and the self-attention mechanism is used to analyze the contextual information of each data point in the model. The model is trained based on the training set. Softmax is used in the model to transform each element into a probability distribution, and the element with the largest probability is used as the output result, and a deep learning model based on the learning results of the training set is output;

[0013] Step 4: For the deep learning model trained in step 3, the coefficient of determination R is used. 2 To quantitatively evaluate the model prediction performance, complete the model establishment, and use the trained model to predict rheological properties.

[0014] Furthermore, the titanium alloy sample is TC4 (Ti-6Al-4V) titanium alloy after diffusion welding. TC4 (Ti-6Al-4V) titanium alloy belongs to α+β type titanium alloy, has high specific strength, thermal stability, and corrosion resistance, and is widely used in aviation structural parts.

[0015] Furthermore, in the high temperature tensile test, the gauge section size of the titanium alloy tensile specimen is 25 mm × 6 mm × 1 mm.

[0016] Furthermore, the different temperatures and strain rates in the high temperature tensile compression simulation test are 700°C, 750°C, 800°C, and 850°C, and the strain rates are 0.001s -1 , 0.01s -1 , 0.1s -1, stretched to fracture, a total of 12 groups of tests; the test steps are to first heat the titanium alloy sample at a heating rate of 10℃ / s, keep it warm for 3 minutes after reaching the specified temperature, and then stretch it at the required strain rate until the sample breaks, and then water-cool the sample to retain the high-temperature structure of the material.

[0017] Normalization, self-attention mechanism, Softmax change and determination coefficient R described in steps 2, 3 and 4 2 The calculation formula is:

[0018]

[0019] Among them, x i * is the normalized value of the i-th sample in the data set; x i is the i-th sample; x max and x min are the maximum and minimum values ​​in the sample data respectively; Q, K and V are three matrices obtained by random linear transformation of the input data; d is the scale factor, which is 64; z i is the i-th element in the output vector; j is the sum of all elements in the vector.

[0020] Furthermore, the deep learning model described in step 3 selects a Transformer deep learning model, which includes a 3-layer encoder and a 1-layer decoder, and a multi-head attention mechanism with 4 heads in each encoder and decoder. The training set is trained to obtain the corresponding model parameters, and the Transformer model trained based on the existing training set is output.

[0021] The beneficial effects of the present invention compared with the prior art are:

[0022] Compared with the traditional constitutive model prediction method, which is difficult to accurately predict the high-temperature rheological response of TC4 titanium alloy under complex conditions, the present invention adopts a deep learning method, which can not only accurately predict the high-temperature rheological response of TC4 titanium alloy under complex deformation conditions, but also shorten the modeling time, solving the problem of accurately predicting the rheological response of TC4 titanium alloy under complex deformation conditions.

[0023] The present invention makes full use of the advantages of deep learning models, takes deformation conditions and true strain as input, and outputs true stress. There is no need to directly construct an intermediate complex mapping relationship model, which reduces modeling time and improves prediction accuracy.

[0024] This paper studies the rheological stress of TC4 titanium alloy during the thermal deformation process after diffusion welding. By establishing a material rheological property prediction model based on deep learning, it can provide assistance for optimizing the hot forming parameters of titanium alloy blades, which is of great help in understanding the evolution law of the thermal deformation behavior of titanium alloys, and has a guiding role in the optimization of process parameters. It can provide guidance and assistance for the actual production of titanium alloy blades.

[0025] The present invention can flexibly select the deformation temperature and strain rate according to actual conditions, and can simulate all working conditions in the production process of titanium alloy blades, and has universal applicability. At the same time, the method of the present invention is not limited to TC4 titanium alloy after diffusion welding, and can also be applied to other titanium alloys to predict the rheological behavior of the titanium alloy during thermal deformation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is the TC4 rheological curve after diffusion welding in an embodiment of a method for predicting rheological properties of titanium alloy blades based on deep learning of the present invention;

[0027] Figure 2 This is a comparison chart of predicted values ​​and experimental values ​​of a data set in an embodiment of a method for predicting rheological properties of titanium alloy blades based on deep learning of the present invention;

[0028] Figure 3 This is a comparison diagram of experimental rheological curves and predicted rheological curves at different deformation temperatures and deformation rates in an embodiment of a method for predicting rheological properties of titanium alloy blades based on deep learning of the present invention;

[0029] Figure 4 This is an error distribution histogram in an embodiment of a method for predicting rheological properties of titanium alloy blades based on deep learning in the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the following examples are given to further illustrate the present invention in detail. It should be noted that the specific implementation described here is only used to explain the present invention and is not intended to limit the present invention.

[0031] This embodiment takes the high-temperature stretching of TC4 titanium alloy after diffusion welding as an example, and the specific implementation steps are as follows:

[0032] Step 1: Use wire cutting to cut a 25mm×6mm×1mm gauge length specimen from the diffusion-bonded TC4 titanium alloy. High-temperature tensile testing was performed on the specimen using a UTM 5504X universal electronic testing machine at temperatures of 700°C, 750°C, 800°C, and 850°C, with a strain rate of 0.001s. -1 , 0.01s -1 , 0.1s -1, stretched to fracture, a total of 12 groups of tests; the test steps are to first heat the titanium alloy sample at a heating rate of 10℃ / s, heat it to the specified temperature and keep it warm for 3 minutes, then stretch it at the required strain rate until the sample breaks, and then water-cool the sample to retain the high-temperature structure of the material. The true stress-strain curves under different deformation conditions are obtained as follows Figure 1 shown.

[0033] Step 2: Use the tensile condition parameters and rheological curve data obtained in step 1 to construct a data set. Use deformation temperature, deformation rate, and true strain as input and true stress as output to construct the data set required for training the model. Normalize the data set and transform the true stress to the [0, 1] interval. This operation can effectively improve the model training speed and improve the model performance. It can be expressed as:

[0034]

[0035] The dataset is divided into a training set and a test set with a ratio of 80%:20%. The training set is used to train the model to obtain the optimal parameters, and the test set is used to evaluate the model prediction performance.

[0036] Step 3: The data set divided in step 2 is fed into the Transformer model, which contains a 3-layer encoder and a 1-layer decoder. Each encoder and decoder contains a 4-head multi-head attention mechanism. The training set is trained to obtain the corresponding model parameters and output. The Transformer model trained based on the existing training set uses the self-attention mechanism to analyze the context information of each data point in the model, which can be expressed as:

[0037]

[0038] The model is trained based on the training set. Softmax is used in the model to transform each element into a probability distribution, which can be expressed as:

[0039]

[0040] The element with the highest probability is used as the output result, and the deep learning model based on the learning results of the training set is output;

[0041] Step 4: For the deep learning model trained in step 3, the coefficient of determination R is used. 2 To quantitatively evaluate the model prediction performance, it can be expressed as:

[0042]

[0043] Complete the model establishment and use the trained model to predict rheological properties.

[0044] like Figure 2 As shown in , in this case, the test set prediction result R2 is , and the training set prediction result R2 is . The model prediction performance is good, and there is no overfitting phenomenon, which shows that the model has good generalization ability. In order to further analyze the deviation distribution of the model's prediction results for all samples, the error distribution histogram of the model is drawn with the number of samples in a certain interval as the vertical axis and the error interval as the horizontal axis, as shown in Figure 3 As shown,

[0045] Figure 4 The figure shows a comparison of the experimental rheological curves and the predicted rheological curves at different deformation temperatures and deformation rates. The predicted curves are basically consistent with the experimental curves, which further illustrates that the model has good prediction accuracy.

[0046] The above process is high-temperature stretching of TC4 titanium alloy after diffusion welding, which is used for a deep learning-based prediction method for the rheological properties of titanium alloy blades. The same prediction method can also be used for different deformation parameters, and can realize the simulation of all working conditions in the production process of titanium alloy blades. It has universal applicability. At the same time, the method of the present invention is not limited to TC4 titanium alloy after diffusion welding, and can also be applied to other titanium alloys to predict the rheological behavior of the titanium alloy during thermal deformation.

[0047] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principles of the present invention. These improvements should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the rheological properties of titanium alloy blades based on deep learning, characterized in that: The method described is: Step 1: Cutting a number of high-temperature tensile specimens from the titanium alloy material after diffusion welding; using a UTM 5504X universal electronic testing machine to perform high-temperature tensile tests on the titanium alloy high-temperature tensile specimens at different temperatures and strain rates to obtain rheological curves under different deformation conditions; Step 2: Using the tensile condition parameters and rheological curve data obtained in step 1, a data set is constructed. The deformation temperature, deformation rate, and true strain are used as inputs, and the true stress is used as output to construct the data set required for the training model. The data set is normalized, the true stress is transformed to the interval [0, 1], and the data set is divided into a training set and a test set. The training set is used to train the model to obtain the optimal parameters, and the test set is used to evaluate the model prediction performance. Step 3: The data set divided in step 2 is fed into a deep learning model, and the self-attention mechanism is used to analyze the contextual information of each data point in the model. The model is trained based on the training set. Softmax is used in the model to transform each element into a probability distribution, and the element with the largest probability is used as the output result, and a deep learning model based on the learning results of the training set is output; Step 4: For the deep learning model trained in step 3, the coefficient of determination R is used. 2 To quantitatively evaluate the model prediction performance, complete the model establishment, and use the trained model to predict rheological properties.

2. The method for predicting the rheological properties of titanium alloy blades based on deep learning according to claim 1, characterized in that: The different temperatures and strain rates in the high temperature tensile compression simulation test are 700℃, 750℃, 800℃, 850℃, and the strain rates are 0.001s -1 , 0.01s -1 , 0.1s -1 , stretched to fracture, a total of 12 groups of tests; the test steps are to first heat the titanium alloy sample at a heating rate of 10℃ / s, keep it warm for 3 minutes after reaching the specified temperature, and then stretch it at the required strain rate until the sample breaks, and then water-cool the sample to retain the high-temperature structure of the material.

3. The method for predicting the rheological properties of titanium alloy blades based on deep learning according to claim 2, characterized in that: The titanium alloy sample is TC4 (Ti-6Al-4V) titanium alloy after diffusion welding. TC4 (Ti-6Al-4V) titanium alloy belongs to α+β type titanium alloy, has high specific strength, good heat resistance, corrosion resistance and formability, and is widely used in aviation structural parts.

4. The method for predicting rheological properties of titanium alloy blades based on deep learning according to claim 1, characterized in that: Normalization, self-attention mechanism, Softmax change and determination coefficient R in steps 2, 3 and 4 2 The calculation formula is: Among them, x i * is the normalized value of the i-th sample in the data set; x i is the i-th sample; x max and x min are the maximum and minimum values ​​in the sample data respectively; Q, K and V are three matrices obtained by random linear transformation of the input data; d is the scale factor, which is 64; z i is the i-th element in the output vector; j is the sum of all elements in the vector; y i is the predicted value of the i-th sample.

5. The method for predicting rheological properties of titanium alloy blades based on deep learning according to claim 1, characterized in that: The steps are a high-temperature tensile test, and the gauge section size of the titanium alloy tensile specimen is 25mm×6mm×1mm.

6. The method for predicting rheological properties of titanium alloy blades based on deep learning according to claim 1, characterized in that: The model uses the Transformer deep learning model, which consists of a 3-layer encoder and a 1-layer decoder. Each encoder and decoder contains a 4-head multi-head attention mechanism. It trains and learns the training set, obtains the corresponding model parameters, and outputs the Transformer model trained based on the existing training set.

Images