Phase change refrigeration material-oriented performance prediction method and related device

Abstract

The invention relates to the field of data phase change material informatics, and discloses a phase change refrigeration material performance prediction method and a related device. A shape memory alloy database about phase change refrigeration material components, process conditions, temperature-induced martensite phase change enthalpy change data and adiabatic temperature change is constructed; establishing a component process-adiabatic temperature change generation model by utilizing a database and a machine learning model; obtaining a phase change and enthalpy change data set by using a component process-adiabatic temperature change generation model; dSC curve features in the phase change and enthalpy change data set are extracted, and low-dimensional phase change features are obtained; based on the low-dimensional phase change characteristics, combining with the process parameters, and utilizing an artificial neural network model to construct an elastic-thermal performance prediction model; and under the condition of given DSC data, obtaining corresponding adiabatic temperature change under the condition of given DSC data by using the elastic-thermal performance prediction model. According to the method, the generation model and the prediction model can be combined under the small sample condition, and the elastic-thermal performance of the shape memory alloy is predicted by taking the temperature-induced martensite phase transformation characteristic as a bridge.

Description

Technical Field

[0001] The present invention relates to the field of data phase change material informatics, and specifically to a method and related device for predicting the performance of phase change refrigeration materials. Background Art

[0002] Elastic caloric refrigeration is the most promising emerging refrigeration technology to replace traditional gas-liquid compression refrigeration. The size of the elasto-caloric effect of the elasto-caloric material directly determines the upper limit of the elasto-caloric refrigeration effect. However, the equipment used to test the elasto-caloric effect using the direct method is complex to set up and the experimental cycle is long, so there are few reports on elasto-caloric effect. The adiabatic temperature change data calculated by the indirect method is often far from the actual value, which limits the development of elasto-caloric materials. Therefore, it is necessary to explore new methods for the development of high elasto-caloric materials.

[0003] In recent years, there have been many studies combining material informatics methods such as machine learning to study the shape memory properties, superelasticity and engineering applications of shape memory alloys. However, due to the small scale of elasto-thermal data, machine learning methods are difficult to apply to elasto-thermal predictions.

[0004] For shape memory alloys, the research on temperature-induced reversible martensitic transformation is much earlier than the research on elastocaloric effect. The essence of both is based on the entropy change between the parent phase and the martensitic phase of shape memory alloys, but the causes are different. The former is temperature-induced martensitic transformation, and the latter is stress-induced martensitic transformation. There is a certain relationship between the two, but it is difficult to describe them using a simple formula. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a method and related devices for predicting the performance of phase change refrigeration materials. The present invention can combine the generation model with the prediction model under small sample conditions, using the temperature-induced martensitic phase transformation characteristics as a bridge to reasonably predict the elastic-thermal properties of shape memory alloys, which has guiding significance for accelerating the research and development of new materials.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows: A method for predicting the performance of phase change refrigeration materials includes the following steps: Construct a database on the composition, process conditions, temperature-induced martensitic phase transformation enthalpy change data and adiabatic temperature change shape memory alloys of phase change refrigeration materials; Using the shape memory alloy database and machine learning model, a composition process-adiabatic temperature change generation model is established; Obtaining a phase change enthalpy change data set using the composition process-adiabatic temperature change generation model; Extract the DSC curve features from the phase change enthalpy change data set to obtain low-dimensional phase change features; Based on the low-dimensional phase change characteristics and combined with process parameters, an elastic-thermal performance prediction model is constructed using an artificial neural network model; Under the condition of given DSC data, the elastic-caloric property prediction model is used to obtain the corresponding adiabatic temperature change under the condition of given DSC data, thereby realizing the prediction of the elastic-caloric property of phase change refrigeration materials.

[0007] Preferably, the composition includes elements contained in the phase change refrigeration material; The process conditions include part or all of whether hot rolling, cold rolling deformation, whether solid solution treatment, whether quenching, aging treatment temperature and aging treatment time.

[0008] Preferably, the shape memory alloy database and the machine learning model are used to establish a composition process-adiabatic temperature change generation model, including: Based on the shape memory alloy database, a variational conditional autoencoder generation model is used to establish a mapping mechanism from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data, and a composition-process-adiabatic temperature change generation model is obtained to achieve latent space generation from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data. The variational conditional autoencoder generation model is used to establish a mapping mechanism from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data. When the composition process-adiabatic temperature change generation model is obtained, the composition, cold rolling deformation, aging treatment temperature and aging treatment time are encoded using actual values ​​as input variational conditional autoencoders; whether hot rolling, solid solution, or quenching is represented by one-hot encoding; By sampling in latent space, the sampled data is passed through a decoder of a composition-process-adiabatic temperature change generation model to generate temperature-induced martensitic phase transformation data with specific composition and process conditions, and used as the phase transformation enthalpy change data set.

[0009] Preferably, extracting DSC curve features from the phase change enthalpy change data set to obtain low-dimensional phase change features includes: Based on the phase change enthalpy change dataset, the synchronous squeezing wavelet transform combined with the variational autoencoder is used to automatically extract the DSC curve features and obtain the low-dimensional phase change features.

[0010] Preferably, based on the phase change enthalpy change data set, the synchronous squeezing wavelet transform combined with the variational autoencoder is used to automatically extract the DSC curve features, including: The DSC curve is processed by synchronous squeezing wavelet transform to generate a time-frequency diagram, which is then input into a variational autoencoder, and the hidden layer variables in the variational autoencoder are taken as low-dimensional phase change features.

[0011] Preferably, based on the low-dimensional phase change characteristics and in combination with process parameters, an elastic-thermal performance prediction model is constructed using an artificial neural network model, including: The low-dimensional phase change characteristics and process parameters are used as the input of the artificial neural network model, and the corresponding adiabatic temperature change is used as the output of the artificial neural network model. The artificial neural network model is trained, and the mapping relationship between DSC data and process conditions and adiabatic temperature change is constructed. The trained artificial neural network model is used as the elastic-thermal performance prediction model.

[0012] Preferably, the artificial neural network model adopts a fully connected neural network, including an input layer, a hidden layer and an output layer; wherein a Dropout layer is added after the hidden layer, and the Dropout layer is selected as a trick for training the fully connected neural network; it includes three hidden layers, and the number of nodes in the three hidden layers is 16, 32 and 16 respectively; the output layer is an adiabatic temperature change value; In each training batch, half of the hidden layer nodes are set to 0 to ignore half of the feature detectors.

[0013] The present invention also provides a system for predicting the performance of phase-change refrigeration materials, comprising: Database construction module: used to construct the composition, process conditions, temperature-induced martensitic phase transformation enthalpy change data and adiabatic temperature change shape memory alloy database of phase change refrigeration materials; The first model building module is used to establish a composition process-adiabatic temperature change generation model using the shape memory alloy database and the machine learning model; A data set acquisition module: used to acquire a phase change enthalpy change data set using the component process-adiabatic temperature change generation model; Feature extraction module: used to extract DSC curve features from phase change enthalpy change data sets to obtain low-dimensional phase change features; The second model building module is used to build an elastic-thermal performance prediction model based on the low-dimensional phase change characteristics and combined with process parameters using an artificial neural network model; Prediction module: used to obtain the corresponding adiabatic temperature change under the condition of given DSC data by using the elasto-caloric performance prediction model, so as to realize the prediction of the elasto-caloric performance of phase change refrigeration materials.

[0014] The present invention also provides an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the phase-change refrigeration material performance prediction method of the present invention as described above.

[0015] The present invention further provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the phase-change refrigeration material performance prediction method of the present invention as described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The performance prediction method for phase change refrigeration materials of the present invention mainly includes data generation and performance prediction. It can realize the rapid and accurate prediction of the elastic and thermal properties of phase change refrigeration materials under small sample conditions, and has important guiding significance for guiding the development of new materials. In order to generate temperature-induced martensitic phase transformation enthalpy change data under specific composition and process parameters, the present invention uses the composition and process parameters as generation model constraints, establishes a composition and process-adiabatic temperature change generation model of temperature-induced martensitic phase transformation enthalpy change under specific composition and process parameters, and realizes the latent space generation from composition and process to temperature-induced martensitic phase transformation data. The present invention proposes to use synchronous extrusion wavelet transform combined with variational autoencoder to realize phase change feature extraction of high-dimensional DSC data, realizes dimensionality reduction of high-dimensional data, and greatly reduces the possibility of overfitting of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a block diagram of an implementation method for predicting the performance of phase change refrigeration materials in an embodiment of the present invention.

[0018] Figure 2 This is a diagram of DSC data feature extraction using synchronous squeezing wavelet transform combined with variational autoencoder in the method of the present invention.

[0019] Figure 3 (a) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on an extreme gradient boosting decision tree; Figure 3 (b) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on a random forest model; Figure 3 (c) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on a gradient boosting regression model; Figure 3 (d) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on an AdaBoost model; Figure 3 (e) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on a CatBoost model; and Figure 3 (f) is a diagram showing the fitting results of the features extracted in an embodiment of the present invention on a fully connected artificial neural network.

[0020] FIG4( a ) is a three-dimensional predicted surface diagram of Hf2A400T2 alloy in an embodiment of the present invention; FIG4( b ) is a projection of the three-dimensional predicted surface in FIG4( a ) on the temperature axis; and FIG4( c ) is a projection of the three-dimensional predicted surface in FIG4( a ) on the strain axis. DETAILED DESCRIPTION

[0021] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and a specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0022] Reference Figure 1 The method for predicting the performance of phase change refrigeration materials of the present invention includes the following steps: Construct a shape memory alloy database of composition process-phase change enthalpy change data-adiabatic temperature change. Based on the above database, develop a machine learning model with good performance to establish a mapping mechanism from composition and process data to temperature-induced martensitic phase change data (DSC curve), establish a composition process-adiabatic temperature change generation model of low-dimensional-high-dimensional mapping, and obtain a large amount of phase change enthalpy change data (DSC curve); Based on the above-mentioned large amount of readily available and inexpensive DSC data, the elastic and thermal properties of shape memory alloys are predicted, the characteristics of DSC curves are extracted, the relationship between temperature-induced martensitic transformation and stress-induced martensitic transformation is established, and a latent heat-adiabatic temperature change prediction model is developed; The elasto-caloric properties of the phase change refrigeration material are predicted based on the phase change enthalpy change-adiabatic temperature change prediction model.

[0023] In the present invention, in order to generate temperature-induced martensitic phase transformation enthalpy change data under specific composition process parameters, the composition process parameters are used as generation model constraints, and a temperature-induced martensitic phase transformation enthalpy change data generation model under specific composition process parameters is established to achieve latent space generation from composition process to temperature-induced martensitic phase transformation data. In addition, a large amount of temperature-induced martensitic phase transformation enthalpy change data is sampled and generated in the latent space based on the generation model. The present invention mainly uses conditional variational autoencoder CVAE to generate phase transformation enthalpy change data.

[0024] Conditional Variational Autoencoder (CVAE) is a generative model based on Variational Autoencoder (VAE), which can generate data with specified features according to given conditions. Compared with the traditional VAE model, CVAE adds conditional inputs in both the encoder and decoder, which makes the generated samples more controllable and can be used for many unsupervised learning tasks.

[0025] The main idea of ​​CVAE is to embed conditional information into the prior distribution of a set of latent variables (z), so that the model can be affected by the conditional information when generating data. Specifically, assuming that the goal is to generate data y from given conditions x and latent variables z, the optimization objective of CVAE can be expressed as:

[0026] in, represents the probability that the decoder generates y, represents the distribution of the encoder mapping x to the latent variable z. Since this integral is difficult to solve, the present invention uses the reparameterization technique to approximate the gradient. Specifically, the present invention reparameterizes the latent variable z by sample noise and expresses it as:

[0027] Where μ and σ are the mean and variance of the encoder network output, respectively. is noise sampled from a standard normal distribution N(0,1). This reparameterization technique allows the gradient to be smoothly propagated back to the network and to change freely during training under constraints on the mean and variance.

[0028] In short, CVAE is a generative model based on conditional information, which can generate data with specified characteristics by embedding conditional information into latent variables. Although its objective function is relatively complex, it can be efficiently trained and generate high-quality samples through reparameterization techniques and other optimization methods.

[0029] In the present invention, composition and process parameters are used as constraints for the CVAE generation model, temperature-induced martensitic phase transformation data are used as generation model data, and the generation model is trained to output spectrum data under specific composition and process parameter conditions.

[0030] Based on the large amount of DSC data generated by the above process and combined with the process condition data, an artificial neural network model is developed to establish the mapping relationship between DSC data and adiabatic temperature change, thereby realizing the rapid prediction of adiabatic temperature change in the service space.

[0031] Example This embodiment takes TiNi-based shape memory alloy as the research object, and establishes a composition, process conditions and their corresponding temperature-induced martensitic transformation data (DSC data) library.

[0032] like Figure 1 As shown, the method for predicting the performance of phase change refrigeration materials in this embodiment includes the following steps: Step 1: Construct a database of DSC curves and adiabatic temperature change of shape memory alloys. The shape memory alloy database used to construct the composition process-DSC curve generation model mainly comes from the accumulation of experiments of the research group. The database includes 89 types of Ti-Ni-based shape memory alloys with different compositions and processes and the corresponding DSC curves. Among them, the composition elements include 9 elements including Ti, Ni, Cu, Hf, Co, Zr, Fe, Pd, and Ta; the process conditions include whether hot rolling, cold rolling deformation, whether solid solution, whether quenching, aging treatment temperature, and aging treatment time, a total of 6 process condition parameters. Since the hot rolling temperature of the samples in the database is 1000℃, and the solid solution treatment conditions are all solid solution at 1000℃ for one hour, the experimental parameters are invariants for the model, and adding them is not helpful for model training; in addition, hot rolling has little effect on the properties of the alloy, so the parameters such as hot rolling temperature, deformation and solid solution temperature time are not used more specifically.

[0033] When constructing features, the composition, cold rolling deformation, aging treatment temperature and aging treatment time use actual values ​​as the input model code; whether hot rolling, solution treatment, or quenching is represented by a single hot code, that is, "yes" is marked as 1, and "no" is marked as 0. The DSC curve is divided into two stages of heating and cooling, with 112 points in each stage, in order to construct the generative model.

[0034] The shape memory alloy database used to construct the DSC curve-adiabatic temperature change prediction model also comes from the accumulation of experiments of the research group. The database includes DSC curves corresponding to 32 types of Ti-Ni-based shape memory alloys with different compositions and processes, as well as a total of 340 adiabatic temperature change data tested under different temperature and strain conditions corresponding to each alloy.

[0035] The adiabatic temperature change database mainly contains the cold-rolled Ti-Ni-based shape memory alloy data. This embodiment uses the characteristic information extracted from the DSC curve and the elastic and thermal test conditions temperature and strain as input, and the corresponding adiabatic temperature change data as output. All DSC curves are processed into 3915 points for subsequent feature extraction. The test temperature conditions of the accumulated adiabatic temperature change data are basically higher than the austenite transformation end temperature Af point, because the superelasticity of the shape memory alloy occurs above the Af temperature; the test strain conditions are concentrated in 1%~8%.

[0036] Step 2. Based on the above database, the encoder of the conditional variational autoencoder (CVAE) generative model is used to learn the distribution of high-dimensional temperature-induced martensitic phase transformation data in the low-dimensional latent space. At the same time, the low-dimensional composition and process data are used as constraints and fused with the latent space distribution. Sampling is performed in the latent space, and the sampled data is passed through the decoder of the generative model to generate temperature-induced martensitic phase transformation data (DSC data) with specific composition and process conditions.

[0037] In this embodiment, the composition and process parameters are used as the constraints of the CVAE generation model, the temperature-induced martensitic phase transformation data is used as the generation model data, and the generation model is trained to output the phase transformation enthalpy change data (DSC curve) under the conditions of specific composition and process parameters.

[0038] Step 3: Based on the large amount of DSC data generated above, the synchronous squeezing wavelet transform combined with the variational autoencoder is used to automatically extract the DSC curve features to obtain low-dimensional phase change features.

[0039] Synchronous squeezing wavelet transform is an effective tool in signal processing. The widely used wavelet transform cannot achieve the optimal resolution when converting the signal into a time-frequency diagram. Therefore, based on the continuous wavelet transform, the complex wavelet coefficients are rearranged in the frequency domain to maintain its reversibility and improve the time-frequency resolution. After rigorous mathematical derivation, the synchronous squeezing wavelet transform method is obtained. The specific method includes the following steps: the first step is to perform a continuous wavelet transform on the signal to obtain the wavelet coefficients; the second step is to obtain the instantaneous frequency from the wavelet coefficients; the third step is to obtain the reorganized spectrum estimation value only on the frequency axis after the rearrangement method, so as to obtain a time-frequency spectrum with more concentrated energy.

[0040] Unlike traditional autoencoders that describe potential energy space by numerical values, variational autoencoders describe potential energy space by probability, showing great application in data generation. Similar to autoencoders, it consists of two parts, an encoder and a decoder. Variational autoencoders use two neural networks to model two probability density distributions. One model is used to transform the original input data to generate a mutated probability distribution of hidden variables, called the inference network; the other model is used to restore the network to generate a probability distribution that approximates the original data, called the generation network. Specifically, the variational autoencoder represents the potential features of a given input as a probability distribution. When decoding from the latent state, a vector is randomly sampled from each latent state distribution to generate a vector as the input of the decoder model. Through the encoding and decoding process described above, a continuous and smooth representation of the latent variables is basically achieved. In order for the decoder model to accurately reconstruct the input, values ​​adjacent to each other in the latent space should correspond to very similar reconstructions. In order to improve the performance of the model, the pre-trained model ResNet50 is used as the encoder and decoder of VAE.

[0041] The specific operation of synchronous squeezing wavelet transform combined variational autoencoder features is as follows: first, the DSC curve is processed by synchronous squeezing wavelet transform to generate a time-frequency diagram, then the time-frequency diagram is input into the variational autoencoder, and the hidden layer variables in the variational autoencoder are taken as features. The feature extraction process is as follows: Figure 2As shown, the number of features is consistent with the wavelet scattering transform method, and finally a 5-dimensional feature vector is obtained.

[0042] Step 4. Based on step 4, an automatic feature extraction method is used to obtain a 5-dimensional feature vector, which is combined with process parameters as the input of the prediction model, and the corresponding adiabatic temperature change is used as the output. The artificial neural network model is trained to construct a mapping relationship between DSC data and process parameters and adiabatic temperature change. Under the condition of given DSC data, the adiabatic temperature change corresponding to any service condition in the service space can be predicted, thereby realizing the rapid prediction of elastic-thermal properties.

[0043] In the step 4, the artificial neural network adopts a fully connected neural network, including an input layer, a hidden layer and an output layer. Among them, due to limited experimental data, in order to avoid overfitting of the model, a Dropout layer is added after the hidden layer. Dropout can be selected as a trick for training deep neural networks. In each training batch, by ignoring half of the feature detectors (letting half of the hidden layer node values ​​be 0), the overfitting phenomenon can be significantly reduced. This method can reduce the interaction between feature detectors (hidden layer nodes), and detector interaction means that some detectors rely on other detectors to play a role. In layman's terms, it is to stop the activation value of a certain neuron with a certain probability p, so that the model can be more generalized because it will not rely too much on certain local features. In the present invention, p takes a value of 0.2. The neural network designed by the present invention includes three hidden layers, with the number of nodes being 16, 32 and 16 respectively, and the output layer is the adiabatic temperature change value.

[0044] Step 5: Predict the optimal elastic and thermal properties of the material based on the trained performance prediction model.

[0045] In order to verify the effectiveness of the method of the present invention, the present invention randomly divides the original data set into a training set and a test set in a ratio of 8:2, uses the training set data to train the model, and uses the test set data to test the model. The prediction results of the six models for the training set and test set data are shown in Figure 3 (a) to Figure 3 (f), where the blue points are the training set data and the red points are the test set data. The horizontal axis in the figure is the actual value obtained by the experiment, and the vertical axis is the predicted value obtained by the model. The closer the data point is to the diagonal line, the higher the accuracy of the model.

[0046] This embodiment compares the fitting effects of the features extracted by the synchronous extrusion wavelet transform combined with the variational autoencoder on six models, namely, extreme gradient boosting decision tree (XGBR), random forest (RF), gradient boosting regression (GBR), AdaBoost, CatBoost, and fully connected artificial neural network (ANN). The intuitive comparison between the model prediction value and the actual value is shown in Figure 3 (a) to Figure 3 (f). As can be seen from the figure, the data set of the present invention has overfitting phenomenon on the XGBR, GBR, and CatBoost models. The random forest model and the AdaBoost model perform well on the training set, but the prediction deviation of the test set is large. Combining the results of the test set and the training set, the best performance is the fully connected artificial neural network model. The fully connected artificial neural network model finally obtained by fitting the features extracted by the synchronous extrusion wavelet transform combined with the variational autoencoder is the best for predicting the elastic-thermal effect of shape memory alloys.

[0047] In order to verify the predictive performance of the model in the unknown service space, the Ti48Ni45Cu5Hf2 alloy (Hf2A400T2) sample, which was cold rolled 40% and then heat treated at 400℃ for 2h, was taken as an example. The model can generate a three-dimensional adiabatic temperature change surface with the test condition temperature and strain as the X-axis and Y-axis, as shown in Figure 4 (a). Through the projection of the surface on the temperature axis and the strain axis (i.e., Figure 4 (b) and Figure 4 (c)), it can be seen that under the same strain test conditions, the adiabatic temperature change first increases and then decreases with the increase of temperature; under the same temperature test conditions, the adiabatic temperature change gradually increases with the increase of strain, which is consistent with the trend obtained in the experiment. The use of this latent heat-adiabatic temperature change model can help researchers predict the elasto-caloric properties of the shape memory alloy system based on the cheap and easily available DSC data before complex elasto-caloric tests, obtain the predicted value of the adiabatic temperature change under any experimental conditions, and judge the elasto-caloric cooling potential of the alloy, which has certain guiding significance for the development of elasto-caloric materials.

[0048] On the other hand, the present invention also provides a system for implementing the above-mentioned method for predicting the performance of phase change refrigeration materials, comprising: Database construction module: used to construct the composition, process conditions, temperature-induced martensitic phase transformation enthalpy change data and adiabatic temperature change shape memory alloy database of phase change refrigeration materials; The first model building module is used to establish a composition process-adiabatic temperature change generation model using the shape memory alloy database and the machine learning model; A data set acquisition module: used to acquire a phase change enthalpy change data set using the component process-adiabatic temperature change generation model; Feature extraction module: used to extract DSC curve features from phase change enthalpy change data sets to obtain low-dimensional phase change features; The second model building module is used to build an elastic-thermal performance prediction model based on the low-dimensional phase change characteristics and combined with process parameters using an artificial neural network model; Prediction module: used to obtain the corresponding adiabatic temperature change under the condition of given DSC data by using the elasto-caloric performance prediction model, so as to realize the prediction of the elasto-caloric performance of phase change refrigeration materials.

[0049] The present invention can also provide a device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and when the processor executes part or all of the computer executable program, the phase change refrigeration material performance prediction method described in the present invention can be implemented.

[0050] On the other hand, the present invention provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method for predicting the performance of phase-change refrigeration materials according to the present invention can be implemented.

[0051] The computer device may be a laptop, tablet computer, desktop computer, mobile phone or workstation.

[0052] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or an off-the-shelf field programmable gate array (FPGA).

[0053] The memory described in the present invention may be an internal storage unit of a laptop computer, tablet computer, desktop computer, mobile phone or workstation, such as a memory or a hard disk; or an external storage unit may be used, such as a mobile hard disk or a flash memory card.

[0054] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0055] The present invention does not elaborate on some of the well-known technologies that belong to those skilled in the art.

[0056] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. Any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention as long as they are within the spirit of the present invention.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the performance of phase change refrigeration materials, characterized in that: The process includes the following: Construct a database on the composition, process conditions, temperature-induced martensitic phase transformation enthalpy change data and adiabatic temperature change shape memory alloys of phase change refrigeration materials; Using the shape memory alloy database and machine learning model, a composition process-adiabatic temperature change generation model is established; Obtaining a phase change enthalpy change data set using the composition process-adiabatic temperature change generation model; Extract the DSC curve features from the phase change enthalpy change data set to obtain low-dimensional phase change features; Based on the low-dimensional phase change characteristics and combined with process parameters, an elastic-thermal performance prediction model is constructed using an artificial neural network model; Under the condition of given DSC data, the elastic-caloric property prediction model is used to obtain the corresponding adiabatic temperature change under the condition of given DSC data, thereby realizing the prediction of the elastic-caloric property of phase change refrigeration materials.

2. A method for predicting performance of phase change refrigeration materials according to claim 1, characterized in that: The composition includes elements contained in the phase change refrigeration material; The process conditions include part or all of whether hot rolling, cold rolling deformation, whether solid solution treatment, whether quenching, aging treatment temperature and aging treatment time.

3. A method for predicting performance of phase change refrigeration materials according to claim 2, characterized in that: Using the shape memory alloy database and machine learning model, a composition process-adiabatic temperature change generation model is established, including: Based on the shape memory alloy database, a variational conditional autoencoder generation model is used to establish a mapping mechanism from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data, and a composition-process-adiabatic temperature change generation model is obtained to achieve latent space generation from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data. The variational conditional autoencoder generation model is used to establish a mapping mechanism from composition and process conditions to temperature-induced martensitic phase transformation enthalpy change data. When the composition process-adiabatic temperature change generation model is obtained, the composition, cold rolling deformation, aging treatment temperature and aging treatment time are encoded using actual values ​​as input variational conditional autoencoders; whether hot rolling, solid solution, or quenching is represented by one-hot encoding; By sampling in latent space, the sampled data is passed through a decoder of a composition-process-adiabatic temperature change generation model to generate temperature-induced martensitic phase transformation data with specific composition and process conditions, and used as the phase transformation enthalpy change data set.

4. The method for predicting performance of phase change refrigeration materials according to claim 1, characterized in that: Extract the DSC curve features from the phase change enthalpy change data set to obtain low-dimensional phase change features, including: Based on the phase change enthalpy change dataset, the synchronous squeezing wavelet transform combined with the variational autoencoder is used to automatically extract the DSC curve features and obtain the low-dimensional phase change features.

5. A method for predicting performance of phase change refrigeration materials according to claim 4, characterized in that: Based on the phase change enthalpy change data set, the synchronous squeezing wavelet transform combined with the variational autoencoder is used to automatically extract the DSC curve features, including: The DSC curve is processed by synchronous squeezing wavelet transform to generate a time-frequency diagram, which is then input into a variational autoencoder, and the hidden layer variables in the variational autoencoder are taken as low-dimensional phase change features.

6. The method for predicting performance of phase change refrigeration materials according to claim 1, characterized in that: Based on the low-dimensional phase change characteristics and combined with process parameters, an elastic-thermal performance prediction model is constructed using an artificial neural network model, including: The low-dimensional phase change characteristics and process parameters are used as the input of the artificial neural network model, and the corresponding adiabatic temperature change is used as the output of the artificial neural network model. The artificial neural network model is trained, and the mapping relationship between DSC data and process conditions and adiabatic temperature change is constructed. The trained artificial neural network model is used as the elastic-thermal performance prediction model.

7. A method for predicting performance of phase change refrigeration materials according to claim 6, characterized in that: The artificial neural network model adopts a fully connected neural network, including an input layer, a hidden layer and an output layer; wherein a Dropout layer is added after the hidden layer, and the Dropout layer is selected as a trick for training the fully connected neural network; it includes three hidden layers, and the number of nodes in the three hidden layers is 16, 32 and 16 respectively; the output layer is an adiabatic temperature change value; In each training batch, half of the hidden layer nodes are set to 0 to ignore half of the feature detectors.

8. A system for predicting the performance of phase-change refrigeration materials, characterized in that: include: Database construction module: used to construct the composition, process conditions, temperature-induced martensitic phase transformation enthalpy change data and adiabatic temperature change shape memory alloy database of phase change refrigeration materials; The first model building module is used to establish a composition process-adiabatic temperature change generation model using the shape memory alloy database and the machine learning model; A data set acquisition module: used to acquire a phase change enthalpy change data set using the component process-adiabatic temperature change generation model; Feature extraction module: used to extract DSC curve features from phase change enthalpy change data sets to obtain low-dimensional phase change features; The second model building module is used to build an elastic-thermal performance prediction model based on the low-dimensional phase change characteristics and process parameters using an artificial neural network model; Prediction module: used to obtain the corresponding adiabatic temperature change under the condition of given DSC data by using the elasto-caloric performance prediction model, so as to realize the prediction of the elasto-caloric performance of phase change refrigeration materials.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for predicting performance of phase-change refrigeration materials according to any one of claims 1 to 7.

10. A storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the method for predicting the performance of phase-change refrigeration materials according to any one of claims 1 to 7 is implemented.

Images