Next Article in Journal
Plasma-Etched Black GaAs Nanoarrays with Gradient Refractive Index Profile for Broadband, Omnidirectional, and Polarization-Independent Antireflection
Previous Article in Journal
Electrospinning and Partial Etching Behaviors of Core–Shell Nanofibers Directly Electrospun on Mesh Substrates for Application in a Cover-Free Compact Air Filter
Previous Article in Special Issue
Flow Behavior of Nanoparticle Agglomerates in a Fluidized Bed Simulated with Porous-Structure-Based Drag Laws
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Nanomaterials
Volume 14
Issue 13
10.3390/nano14131153
nanomaterials-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Preparation of Thermochromic Vanadium Dioxide Films Assisted by Machine Learning

by
Gaoyang Xiong
,
Haining Ji
*,
Yongxing Chen
,
Bin Liu
,
Yi Wang
,
Peng Long
,
Jinfang Zeng
,
Jundong Tao
and
Cong Deng
School of Physics and Optoelectronics, Xiangtan University, Xiangtan 411105, China
*
Author to whom correspondence should be addressed.
Nanomaterials 2024, 14(13), 1153; https://doi.org/10.3390/nano14131153 (registering DOI)
Submission received: 14 June 2024 / Revised: 3 July 2024 / Accepted: 4 July 2024 / Published: 6 July 2024
(This article belongs to the Special Issue Nanomaterials for Chemical Engineering (Volume III))
Browse Figures
Versions Notes

Abstract

:
In recent years, smart windows have attracted widespread attention due to their ability to respond to external stimuli such as light, heat, and electricity, thereby intelligently adjusting the ultraviolet, visible, and near-infrared light in solar radiation. VO2(M) undergoes a reversible phase transition from an insulating phase (monoclinic, M) to a metallic phase (rutile, R) at a critical temperature of 68 °C, resulting in a significant difference in near-infrared transmittance, which is particularly suitable for use in energy-saving smart windows. However, due to the multiple valence states of vanadium ions and the multiphase characteristics of VO2, there are still challenges in preparing pure-phase VO2(M). Machine learning (ML) can learn and generate models capable of predicting unknown data from vast datasets, thereby avoiding the wastage of experimental resources and reducing time costs associated with material preparation optimization. Hence, in this paper, four ML algorithms, namely multi-layer perceptron (MLP), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB), were employed to explore the parameters for the successful preparation of VO2(M) films via magnetron sputtering. A comprehensive performance evaluation was conducted on these four models. The results indicated that XGB was the top-performing model, achieving a prediction accuracy of up to 88.52%. A feature importance analysis using the SHAP method revealed that substrate temperature had an essential impact on the preparation of VO2(M). Furthermore, characteristic parameters such as sputtering power, substrate temperature, and substrate type were optimized to obtain pure-phase VO2(M) films. Finally, it was experimentally verified that VO2(M) films can be successfully prepared using optimized parameters. These findings suggest that ML-assisted material preparation is highly feasible, substantially reducing resource wastage resulting from experimental trial and error, thereby promoting research on material preparation optimization.

1. Introduction

Vanadium dioxide (VO2) is a typical transition metal oxide with over ten crystalline phases (A, B, D, P, M, etc.) [1,2,3,4,5,6]. However, only VO2(M) undergoes a metal-insulator transition at 68 °C, drawing significant attention from researchers [7,8,9]. When the temperature rises to the critical temperature, VO2(M) rapidly transforms from a low-temperature monoclinic structure to a high-temperature tetragonal structure, resulting in a sudden change in the optical and electrical properties of VO2. VO2(M) has a wide range of applications in various fields, including optoelectronic switches [10,11], smart windows [12,13], military camouflage [14,15,16,17], and spacecraft thermal control [18,19]. Thus, the preparation of high-purity VO2(M) films has long been a key concern in VO2 research, as it determines the effectiveness of applications of VO2 film.
Magnetron sputtering, a widely employed physical vapor deposition method, is considered a promising technique for manufacturing VO2(M) films due to its high uniformity and strong adhesion to the substrate [20,21]. However, the preparation of high-purity VO2(M) through magnetron sputtering requires precise control of sputtering parameters, such as sputtering power, substrate temperature, and the ratio of the oxygen to argon flow rate [22]. Therefore, the precise control of these deposition conditions is extremely important for achieving high-purity VO2(M) films. However, the complex management of deposition conditions during VO2 preparation introduces uncertainty in the sedimentary results. Thus, it is vital to explore and optimize the preparation process of the VO2(M) films. In the field of materials science, significant efforts from predecessors have led to the accumulation of abundant experimental and computational data for material preparation, containing numerous experimental parameters and conditions. Consequently, it is urgently needed to manage this vast dataset for robust data handling methods. Machine learning (ML), a typical data-driven pattern research method [23,24], fulfills the requirements for a big data statistical analysis. It is a powerful tool for learning from training existing datasets, enabling the extraction of underlying patterns and facilitating regression or classification on previously unseen data. ML has been widely applied in various fields of materials science, such as discovering new compounds or molecules with desired properties [25], material synthesis [26,27,28], material structure design [29,30], and material structure and performance prediction [31,32,33,34]. Owing to its potent capabilities, its application in aiding the preparation of high-purity VO2(M) materials holds significant prospects.
In this paper, multi-layer perceptron (MLP), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGB) were implemented to train the preparation dataset of VO2 obtained from laboratory experiments and the literature. Each model was evaluated based on indicators such as specificity, accuracy, and recall to optimize the final deposition process. Additionally, a feature importance analysis was conducted on the dataset to explore the influence of different deposition conditions on the results. The results revealed that XGB demonstrated the highest classification performance on this dataset, achieving a prediction accuracy of 88.52% on the test dataset. The XGB model can effectively instruct the preparation of high-purity VO2(M). This paper offers insights into the use of ML methods for aiding the targeted generation of VO2 materials, while the proposed prediction model introduces novel approaches for optimizing the research of material preparation technologies.

2. Results and Discussion

2.1. Optimized Synthesis Frame Design

During the experimental procedures, it is usually desired to achieve an effective preparation of the target material under predetermined deposition conditions. However, under varied experimental conditions, fixed parameters may fail to yield optimal results. Thus, to avoid unnecessary experimental errors and minimize costs, it is crucial to determine a model capable of adjusting experimental conditions and predicting results accurately. In this article, an ML-assisted approach for depositing VO2 films by magnetron sputtering was presented (Figure 1). The conventional magnetron sputtering method entails introducing specific proportions of O2 and argon gas into a high-vacuum environment, where the target material undergoes oxidation and is subsequently deposited onto the substrate surface via magnetron sputtering to produce VO2(M). In this study, ML can aid in optimizing the deposition parameters for VO2(M) films. Before sputtering, predictive models were employed to forecast deposition parameters, thereby determining the likelihood of achieving pure-phase VO2(M) under these conditions and using the predicted probability as a benchmark for adjusting comprehensive parameters. The high probability of sedimentation parameters in the experiments significantly diminished the uncertainty of the results and minimized resource waste caused by trial and error.

2.2. Dataset Establishment

Data preparation is a fundamental aspect of optimizing material preparation. In this article, a dataset was constructed, including 203 experimental data points from laboratory experiments and the relevant literature. The crystal structure of VO2 was identified by comparing the X-ray diffraction (XRD) data with the JCPDS standard card, serving as the foundation for data acquisition [35]. Table 1 presents the experimental parameters utilized in the preparation of VO2(M) via magnetron sputtering, mainly including sputtering power, substrate temperature, ratio of the oxygen to argon flow rate, reaction pressure, target material, and substrate.
Before model training, the dataset was preprocessed, including encoding the string data and assigning numerical values of “1” and “0” to pure-phase VO2(M) and non-VO2(M), respectively. The scatter plot matrix is a visual aid used to analyze the relationships between multiple variables, as illustrated in Figure 2. A Pearson correlation coefficient analysis was employed to assess the degree of interdependence among variables in the dataset. The correlation between the six features obtained is shown in Figure 3. The positive and negative values in the figure represent positive and negative correlations between features, respectively. Based on the analysis results, the correlation coefficients between the selected features approached 0, indicating a high degree of independence among each feature, thereby reducing the data redundancy that arises from highly correlated data and the problem of feature weight allocation during model training.

2.3. Classification Model

Following dataset establishment and preprocessing, it was necessary to train four ML models (MLP, XGB, SVM, and RF) to accurately predict experimental results. Before inputting the dataset, it needed to be segmented, with 75% randomly allocated as the training set and the remaining 25% as the validation set. After building the model framework, it was essential to adjust the hyperparameters of the model [36], search for hyperparameters of each model through a grid search, and then fine-tune based on validation set performance, aiming to determine the optimal hyperparameters and ensure the best predictive performance. Additionally, to prevent overfitting in model predictions, ten-fold cross-validation was conducted for each model. Subsequently, the test set was utilized to comprehensively assess the performance of each model. Eventually, the optimal model was chosen based on the evaluation results and employed to guide the optimization of the magnetron sputtering process. This model can predict material preparation results and minimize the uncertainty of deposition experiments.

2.4. Model Selection and Performance Evaluation

To optimize magnetron sputtering, a thorough performance evaluation was conducted on the ML model, and the optimal model based on a dataset with high confidence was selected. This study utilized four ML algorithms—MLP, XGB, SVM, and RF—for training based on the original dataset. Previous studies have demonstrated the widespread application and effectiveness of these four models in the optimization of material preparation, particularly on small datasets [27,37,38]. To identify the conditions for the optimal model, this article assessed six indicators, the receiver operating characteristic curve (ROC), area under the ROC curve (AUC), accuracy, specificity, recall, and F1 score, to evaluate the performance of the four models. Among these indicators, ROC assessed the model’s predictive ability, with the AUC value representing its efficacy. Accuracy measured the percentage of correctly predicted samples, specificity quantified the percentage of correctly predicted negative samples, recall determined the percentage of correctly predicted positive samples, and the F1 score represented the harmonic average of accuracy and recall. The results of the model evaluation are depicted in Figure 4a. Overall, these results indicated rapid convergence of all ML models without encountering overfitting or underfitting issues, and they exhibited outstanding predictive capabilities.
Among the four trained models, the XGB model exhibited the highest prediction accuracy, reaching 88.52%, which is higher than previous relevant reports [39,40], followed by the RF model at 86.89%. The F1 score was a comprehensive evaluation indicator of accuracy and recall. Comparing the F1 scores of the four models, XGB still achieved the highest score of 0.78. Regarding other evaluation indicators, XGB still maintained the highest recall rate as well as specificity, which were 0.74 and 0.98, respectively. The ROC curve was considered one of the most important evaluation tools for classification models. Usually, the closer the curve is to the upper left corner, the better its classification performance. Figure 4b illustrates the ROC curves of the four models, demonstrating that the XGB model exhibited the curve closest to the upper left corner, with the highest AUC value of 0.90 among them. Based on the above, this study further investigated the optimization of preparation pathways using XGB models. Figure 5a displays the learning curve of the XGB model. As the sample size increased, the model scored steadily on both the training and validation sets, demonstrating the feasibility of using the XGB model without overfitting problems. The confusion matrix is obtained according to the dataset and prediction results, as shown in Figure 5b, and the specificity and sensitivity can be calculated from the matrix, showing that the XGB model has a good performance.
This discovery indicated a significant correlation between the prediction results of the XGB model and the experimental results reported in the published literature. In order to evaluate the predictive performance of the XGB model, the experimental conditions on the preparation of VO2 by magnetron sputtering were validated in the recent literature. The prediction results under various experimental conditions are shown in Table 2. Among these conditions, the first three conditions have been confirmed to successfully prepare pure-phase VO2(M), while the latter two conditions cannot. These findings indicated a significant correlation between the prediction results of the XGB model and the experimental results reported in the published literature.

2.5. Material Synthesis Pathway Optimization

To optimize the feature parameters of the preparation process, the Shapley Additive exPlanning (SHAP) method was used to extract the importance of features from the six feature parameters. SHAP is a tool for interpreting model outputs. By analyzing the impact of each feature parameter on the output, the sum of Shapley values (contributions) of all feature parameters can be obtained, which is the magnitude of importance [41]. Figure 6 demonstrates that substrate temperature (Temp) had the most significant impact on the purity of VO2(M), followed by the ratio of the oxygen to argon flow rate (O2:Ar) and sputtering power (Power), whereas substrate type (Sub) and target type (Tar) had the least impact. To optimize the range of feature parameters, a dataset containing 6,375,000 virtual experimental condition data points was generated. The XGB model is employed to predict the probability of successfully preparing VO2(M) from these data. Subsequently, the data with a probability greater than 90% were filtered out, resulting in 1,527,400 high-success-rate data points. The range of values for each feature parameter is shown in Table 3. By selecting the characteristic parameter values within this range, the preparation purity of VO2(M) can be optimized to a certain extent, accelerating the controllable preparation speed of the materials.

2.6. Model Verification

Furthermore, under the guidance of the XGB model, a set of experimental parameters with high success rates were selected to verify the reliability of the model. The specific experimental conditions are a sputtering power of 150 W, substrate temperature of 550 °C, ratio of the oxygen to argon flow rate of 1.6, reaction pressure of 0.4 Pa, target material of vanadium metal, substrate of SiO2, and deposition time of 2 h. According to the prediction of the XGB model, the success rate under these experimental conditions was 93.61%. Under these conditions, magnetron sputtering experiments were conducted, and the products were analyzed using the Bruker D8 advanced X-ray diffractometer. The obtained XRD patterns are shown in Figure 7a. The main diffraction peaks at 2θ degrees of 27.8° and 42.3° corresponded to the (011) and (210) orientations of the VO2(M) phase, respectively. The high-resolution XPS spectrum of VO2(M) is shown in Figure 7b. The spectrum of V2p can be resolved into two independent peaks, where the V2p3/2 peak was 515.8 eV, corresponding to the valence state of V4+ ions in VO2. The O1s peak at 530.0 eV is usually attributed to the O2− in the sample, which is consistent with the previous results [42,43]. Combined with the XRD results, it can be further determined that the VO2(M) was obtained. Figure 7c represents the SEM image of the obtained VO2(M) films via magnetron sputtering. The film surface was composed of spherical-like grains from ~80 nm to ~260 nm, with typical grain size of ~180 nm. Due to the transition characteristics of VO2 from the low-temperature semiconductor phase to high-temperature metal phase, its resistance changes by nearly two orders of magnitude during heating and cooling between 20 °C and 120 °C (Figure 7d). These experimental findings demonstrate that high-purity VO2(M) materials can be successfully prepared with the assistance of the XGB model, validating the ability of the trained ML model to promote the preparation of film materials.
Table 2. Prediction of recently published literature parameters using XGB model.
Table 2. Prediction of recently published literature parameters using XGB model.
NumberPower
(W)		Temp
(°C)		O2:ArPress
(Pa)		TarSubModel
Predict		Reference
1906500.0131.33VAl2O386.23%Zhang C et al. [44]
21006500.0131.33VSi85.50%Zhang C et al. [45]
3805800.061.00VSi71.47%Ma X et al. [46]
41005000.0251.00V2O3Al2O328.64%Yang Z et al. [47]
52006000.051.00VSi30.41%Xiong Y et al. [48]
Table 3. Optimizing the range of features. The 0, 1, 2, and 3 of Tar represent V, V2O5, VO2, and V2O3, respectively, and the 0, 1, 2, 3, and 4 of Sub represent Al2O3, NaCa glass, Si, SiO2, and stainless steel, respectively.
Table 3. Optimizing the range of features. The 0, 1, 2, and 3 of Tar represent V, V2O5, VO2, and V2O3, respectively, and the 0, 1, 2, 3, and 4 of Sub represent Al2O3, NaCa glass, Si, SiO2, and stainless steel, respectively.
ParametersMinMax
Power (W)110210
Temp (℃)500700
O2:Ar0.12.2
Press (Pa)0.152.9
Tar0 or 1, 2, 3
Sub0 or 1, 2, 3, 4

3. Conclusions

Four ML models were selected to train the dataset, and their performance was comprehensively evaluated in this study. Among them, the XGB model exhibited the highest prediction accuracy, reaching 88.52%. The comprehensive parameter space was explored by utilizing the trained XGB model. Then, the range of each feature value was determined with a high probability of success in the preparation process, thus offering experimental guidelines. A feature importance analysis was conducted using the SHAP method, revealing that substrate temperature had a significant impact on the purity of VO2(M). Subsequently, experimental verification was performed. The experimental findings confirmed the successful preparation of pure-phase VO2(M) films. In conclusion, this study illustrated that machine learning can effectively facilitate the experimental preparation process, thereby minimizing the uncertainty of experimental results and reducing resource consumption associated with trial and error. This study provides methodological guidance for optimizing material preparation, which has enormous potential in advancing materials science.

Author Contributions

Software, Y.C.; Validation, G.X., H.J. and Y.W.; Investigation, G.X., H.J. and P.L.; Writing—original draft, G.X. and H.J.; Writing—review & editing, G.X., H.J., Y.C., B.L., Y.W., P.L., J.Z., J.T. and C.D.; Funding acquisition, H.J. All authors have read and agreed to the published version of the manuscript.

Funding

This work was financially supported by the National Natural Science Foundation of China (Nos. 51902276, 62005234), the Natural Science Foundation of Hunan Province (Nos. 2019JJ50583, 2023JJ30585), the Scientific Research Fund of Hunan Provincial Education Department (No. 21B0111), and the Hunan Provincial Innovation Foundation for Postgraduate (No. QL20220158).

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Li, Z.; Ren, Y.; Mo, L.; Liu, C.; Hsu, K.; Ding, Y.; Zhang, X.; Li, X.; Hu, L.; Ji, D.; et al. Impacts of oxygen vacancies on zinc ion intercalation in VO2. ACS Nano 2020, 14, 5581–5589. [Google Scholar] [CrossRef] [PubMed]
  2. Cui, F.; Zhao, J.; Zhang, D.; Fang, Y.; Hu, F.; Zhu, K. VO2(B) nanobelts and reduced graphene oxides composites as cathode materials for low-cost rechargeable aqueous zinc ion batteries. Chem. Eng. J. 2020, 390, 124118. [Google Scholar] [CrossRef]
  3. Vaseem, M.; Zhen, S.; Yang, S.; Li, W.; Shamim, A. Development of VO2-nanoparticle-based metal-insulator transition electronic ink. Adv. Electron. Mater. 2019, 5, 1800949. [Google Scholar] [CrossRef]
  4. Chen, L.; Yang, Z.; Huang, Y. Monoclinic VO2(D) hollow nanospheres with super-long cycle life for aqueous zinc ion batteries. Nanoscale 2019, 11, 13032–13039. [Google Scholar] [CrossRef] [PubMed]
  5. Wang, Z.; Yu, K.; Feng, Y.; Qi, R.; Ren, J.; Zhu, Z. VO2(P)-V2C (MXene) grid structure as a lithium polysulfide catalytic host for high-performance Li-S battery. ACS Appl. Mater. Interfaces 2019, 11, 44282–44292. [Google Scholar] [CrossRef] [PubMed]
  6. Cao, C.; Gao, Y.; Luo, H. Pure single-crystal rutile vanadium dioxide powders: Synthesis, mechanism and phase-transformation property. J. Phys. Chem. C 2008, 112, 18810–18814. [Google Scholar] [CrossRef]
  7. Ji, H.; Liu, D.; Cheng, H.; Zhang, C.; Yang, L.; Ren, D. Infrared thermochromic properties of monoclinic VO2 nanopowders using a malic acid-assisted hydrothermal method for adaptive camouflage. RSC Adv. 2017, 7, 5189–5194. [Google Scholar] [CrossRef]
  8. Zhang, H.; Xiao, X.; Lu, X.; Chai, G.; Sun, Y.; Zhan, Y.; Xu, G. A cost-effective method to fabricate VO2(M) nanoparticles and films with excellent thermochromic properties. J. Alloys Compd. 2015, 636, 106–112. [Google Scholar] [CrossRef]
  9. Xiao, X.; Huan, C.; Cheng, H.; Lu, Y.; Qi, S.; Zhan, Y.; Liu, H.; Xu, G. A novel method to improve phase transition of VO2 thermochromic films by Cs0.32WO3 nanoparticles. Mater. Lett. 2019, 249, 95–98. [Google Scholar] [CrossRef]
  10. Li, G.; Xie, D.; Zhong, H.; Zhang, Z.; Fu, X.; Zhou, Q.; Li, Q.; Ni, H.; Wang, J.; Guo, E.J.; et al. Photo-induced non-volatile VO2 phase transition for neuromorphic ultraviolet sensors. Nat. Commun. 2022, 13, 1729. [Google Scholar] [CrossRef]
  11. Khan, Z.; Singh, P.; Ansari, S.A.; Manippady, S.R.; Jaiswal, A.; Saxena, M. VO2 nanostructures for batteries and supercapacitors: A review. Small 2021, 17, 2006651. [Google Scholar] [CrossRef] [PubMed]
  12. Zhao, Y.; Ji, H.; Ou, Y.; Wang, Y.; Chen, Y.; Tao, J.; Liu, B.; Lu, M.; Huang, Y.; Wang, J. Novel sunlight-driven Cu7S4/VO2 composite films for smart windows. J. Mater. Chem. C 2024, 12, 2534–2543. [Google Scholar] [CrossRef]
  13. Ji, H.; Zhao, Y.; Lu, M.; Tao, J.; Chen, Y.; Ou, Y.; Wang, Y.; Mao, Y. Novel warm/cool-tone switchable VO2-based smart window composite films with excellent optical performance. Ceram. Int. 2023, 49, 22630–22635. [Google Scholar] [CrossRef]
  14. Ji, H.; Liu, D.; Cheng, H. Infrared optical modulation characteristics of W-doped VO2(M) nanoparticles in the MWIR and LWIR regions. Mater. Sci. Semicond. Process. 2020, 119, 105141. [Google Scholar] [CrossRef]
  15. Wang, Y.; Ji, H.; Chen, Y.; Liu, B.; Huang, J.; Lu, M.; Ou, Y.; Zhao, Y.; Tao, J.; Huang, Y.; et al. Artificially adjustable radiative cooling device with environmental adaptability. Ceram. Int. 2023, 49, 40297–40304. [Google Scholar] [CrossRef]
  16. Liu, D.; Cheng, H.; Xing, X.; Zhang, C.; Zheng, W. Thermochromic properties of W-doped VO2 thin films deposited by aqueous sol-gel method for adaptive infrared stealth application. Infrared Phys. Technol. 2016, 77, 339–343. [Google Scholar] [CrossRef]
  17. Liu, D.; Ji, H.; Peng, R.; Cheng, H.; Zhang, C. Infrared chameleon-like behavior from VO2(M) thin films prepared by transformation of metastable VO2(B) for adaptive camouflage in both thermal atmospheric windows. Sol. Energy Mater. Sol. Cells 2018, 185, 210–217. [Google Scholar] [CrossRef]
  18. Kim, H.; Cheung, K.; Auyeung, R.C.; Wilson, D.E.; Charipar, K.M.; Piqué, A.; Charipar, N.A. VO2-based switchable radiator for spacecraft thermal control. Sci. Rep. 2019, 9, 11329. [Google Scholar] [CrossRef] [PubMed]
  19. Long, L.; Taylor, S.; Wang, L. Enhanced infrared emission by thermally switching the excitation of magnetic polariton with scalable microstructured VO2 metasurfaces. ACS Photonics 2020, 7, 2219–2227. [Google Scholar] [CrossRef]
  20. Li, X.; Zhang, S.; Yang, L.; Li, X.; Chen, J.; Huang, C. A convenient way to reduce the hysteresis width of VO2(M) nanomaterials. New J. Chem. 2017, 41, 15260–15267. [Google Scholar] [CrossRef]
  21. Kelly, P.J.; Arnell, R.D. Magnetron sputtering: A review of recent developments and applications. Vacuum 2000, 56, 159–172. [Google Scholar] [CrossRef]
  22. Ho, H.; Lai, Y.; Chen, K.; Dao, T.D.; Hsueh, C.H.; Nagao, T. High quality thermochromic VO2 films prepared by magnetron sputtering using V2O5 target with in situ annealing. Appl. Surf. Sci. 2019, 495, 143436. [Google Scholar] [CrossRef]
  23. Ahmad, T.; Madonski, R.; Zhang, D.; Huang, C.; Mujeeb, A. Data-driven probabilistic machine learning in sustainable smart energy/smart energy systems: Key developments, challenges, and future research opportunities in the context of smart grid paradigm. Renew. Sustain. Energy Rev. 2022, 160, 112128. [Google Scholar] [CrossRef]
  24. Pugliese, R.; Regondi, S.; Marini, R. Machine learning-based approach: Global trends, research directions, and regulatory standpoints. Data Sci. Manag. 2021, 4, 19–29. [Google Scholar] [CrossRef]
  25. Jordan, M.I.; Mitchell, T.M. Machine learning: Trends, perspectives, and prospects. Science 2015, 349, 255–260. [Google Scholar] [CrossRef] [PubMed]
  26. Chen, Y.; Ji, H.; Lu, M.; Liu, B.; Zhao, Y.; Ou, Y.; Wang, Y.; Tao, J.; Zou, T.; Huang, Y.; et al. Machine learning guided hydrothermal synthesis of thermochromic VO2 nanoparticles. Ceram. Int. 2023, 49, 30794–30800. [Google Scholar] [CrossRef]
  27. Lu, M.; Ji, H.; Zhao, Y.; Chen, Y.; Tao, J.; Ou, Y.; Wang, Y.; Huang, Y.; Wang, J.; Hao, G. Machine learning-assisted synthesis of two-dimensional materials. ACS Appl. Mater. Interfaces 2022, 15, 1871–1878. [Google Scholar] [CrossRef]
  28. Wang, J.; Lu, M.; Chen, Y.; Hao, G.; Liu, B.; Tang, P.; Yu, L.; Wen, L.; Ji, H. Machine Learning-Assisted Large-Area Preparation of MoS2 Materials. Nanomaterials 2023, 13, 2283. [Google Scholar] [CrossRef] [PubMed]
  29. Chen, C.; Gu, G.X. Machine learning for composite materials. MRs Commun. 2019, 9, 556–566. [Google Scholar] [CrossRef]
  30. MDieb, T.; Hou, Z.; Tsuda, K. Structure prediction of boron-doped graphene by machine learning. J. Chem. Phys. 2018, 148, 241716. [Google Scholar] [CrossRef]
  31. Botu, V.; Ramprasad, R. Adaptive machine learning framework to accelerate ab initio molecular dynamics. Int. J. Quantum Chem. 2015, 115, 1074–1083. [Google Scholar] [CrossRef]
  32. De Jong, M.; Chen, W.; Notestine, R.; Persson, K.; Ceder, G.; Jain, A.; Asta, M.; Gamst, A. A statistical learning framework for materials science: Application to elastic moduli of k-nary inorganic polycrystalline compounds. Sci. Rep. 2016, 6, 34256. [Google Scholar] [CrossRef] [PubMed]
  33. Pilania, G.; Wang, C.; Jiang, X.; Rajasekaran, S.; Ramprasad, R. Accelerating materials property predictions using machine learning. Sci. Rep. 2013, 3, 2810. [Google Scholar] [CrossRef] [PubMed]
  34. Sai, N.J.; Rathore, P.; Chauhan, A. Machine learning-based predictions of fatigue life for multi-principal element alloys. Scr. Mater. 2023, 226, 115214. [Google Scholar] [CrossRef]
  35. Wang, X.J.; Li, H.D.; Fei, Y.J.; Wang, X.; Xiong, Y.Y.; Nie, Y.X.; Feng, K.A. XRD and Raman study of vanadium oxide thin films deposited on fused silica substrates by RF magnetron sputtering. Appl. Surf. Sci. 2001, 177, 8–14. [Google Scholar] [CrossRef]
  36. Bischl, B.; Binder, M.; Lang, M.; Pielok, T.; Richter, J.; Coors, S.; Thomas, J.; Ullmann, T.; Becker, M.; Boulesteix, A.L.; et al. Hyperparameter optimization: Foundations, algorithms, best practices and open challenges. arXiv 2021, arXiv:2107.05847. [Google Scholar] [CrossRef]
  37. Tang, B.; Lu, Y.; Zhou, J.; Chouhan, T.; Wang, H.; Golani, P.; Xu, M.; Xu, Q.; Guan, C.; Liu, Z. Machine learning-guided synthesis of advanced inorganic materials. Mater. Today 2020, 41, 72–80. [Google Scholar] [CrossRef]
  38. Zhang, Y.; Ling, C. A strategy to apply machine learning to small datasets in materials science. Npj Comput. Mater. 2018, 4, 25. [Google Scholar] [CrossRef]
  39. Song, Y.; Lindsay, J.; Zhao, Y.; Nasiri, A.; Louis, S.Y.; Ling, J.; Hu, M.; Hu, J. Machine learning based prediction of noncentrosymmetric crystal materials. Comput. Mater. Sci. 2020, 183, 109792. [Google Scholar] [CrossRef]
  40. Cruz, J.A.; Wishart, D.S. Applications of machine learning in cancer prediction and prognosis. Cancer Inform. 2006, 2, 743338146. [Google Scholar] [CrossRef]
  41. Zhou, Y.; Wu, W.; Wang, H.; Zhang, X.; Yang, C.; Liu, H. Identification of soil texture classes under vegetation cover based on Sentinel-2 data with SVM and SHAP techniques. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2022, 15, 3758–3770. [Google Scholar] [CrossRef]
  42. Sirvent, P.; Pérez, G.; Guerrero, A. Efficient VO2(M) synthesis to develop thermochromic cement-based materials for smart building envelopes. Mater. Chem. Phys. 2021, 269, 124765. [Google Scholar] [CrossRef]
  43. Xiao, X.; Zhang, H.; Chai, G.; Sun, Y.; Yang, T.; Cheng, H.; Chen, L.; Miao, L.; Xu, G. A cost-effective process to prepare VO2(M) powder and films with superior thermochromic properties. Mater. Res. Bull. 2014, 51, 6–12. [Google Scholar] [CrossRef]
  44. Zhang, C.; Koughia, C.; Güneş, O.; Luo, J.; Hossain, N.; Li, Y.; Cui, X.; Wen, S.J.; Wong, R.; Yang, Q.; et al. Synthesis, structure and optical properties of high-quality VO2 thin films grown on silicon, quartz and sapphire substrates by high temperature magnetron sputtering: Properties through the transition temperature. J. Alloys Compd. 2020, 848, 156323. [Google Scholar] [CrossRef]
  45. Zhang, C.; Koughia, C.; Li, Y.; Cui, X.; Ye, F.; Shiri, S.; Sanayei, M.; Wen, S.J.; Yang, Q.; Kasap, S. Near-zero IR transmission of VO2 thin films deposited on Si substrate. Appl. Surf. Sci. 2018, 440, 415–420. [Google Scholar] [CrossRef]
  46. Ma, X.; Liu, X.; Li, H.; Zhang, A.; Huang, M. Influence of oxygen flow rate on metal–insulator transition of vanadium oxide thin films grown by RF magnetron sputtering. Appl. Phys. A 2017, 123, 1–6. [Google Scholar] [CrossRef]
  47. Yang, Z.; Yang, Q.; Yang, L.; Dai, B.; Xia, F.; Wang, P.; Guo, S.; Gao, G.; Xu, L.; Zhang, Y.; et al. Effect of thickness on infrared optical property of VO2 film deposited by magnetron sputtering. Sci. China Technol. Sci. 2020, 63, 1591–1598. [Google Scholar] [CrossRef]
  48. Xiong, Y.; Wen, Q.; Chen, Z.; Tian, W.; Wen, T.L.; Jing, Y.L.; Yang, Q.H.; Zhang, H.W. Tuning the phase transitions of VO2 thin films on silicon substrates using ultrathin Al2O3 as buffer layers. J. Phys. D Appl. Phys. 2014, 47, 455304. [Google Scholar] [CrossRef]
Nanomaterials 14 01153 g001
Figure 1. Machine learning optimization framework for VO2(M) magnetron sputtering preparation.
Figure 1. Machine learning optimization framework for VO2(M) magnetron sputtering preparation.
Nanomaterials 14 01153 g001
Nanomaterials 14 01153 g002
Figure 2. Scatter plot matrices of collected data. The 0, 1, 2, and 3 of Tar represent V, V2O5, VO2, and V2O3, respectively, and the 0, 1, 2, 3, and 4 of Sub represent Al2O3, NaCa glass, Si, SiO2, and stainless steel, respectively.
Figure 2. Scatter plot matrices of collected data. The 0, 1, 2, and 3 of Tar represent V, V2O5, VO2, and V2O3, respectively, and the 0, 1, 2, 3, and 4 of Sub represent Al2O3, NaCa glass, Si, SiO2, and stainless steel, respectively.
Nanomaterials 14 01153 g002
Nanomaterials 14 01153 g003
Figure 3. Correlation analysis of dataset infrared heat map.
Figure 3. Correlation analysis of dataset infrared heat map.
Nanomaterials 14 01153 g003
Nanomaterials 14 01153 g004
Figure 4. (a) Bar chart of model performance comparison; (b) ROC curve of each model.
Figure 4. (a) Bar chart of model performance comparison; (b) ROC curve of each model.
Nanomaterials 14 01153 g004
Nanomaterials 14 01153 g005
Figure 5. (a) Learning curve of XGB model; (b) confusion matrix of XGB model.
Figure 5. (a) Learning curve of XGB model; (b) confusion matrix of XGB model.
Nanomaterials 14 01153 g005
Nanomaterials 14 01153 g006
Figure 6. The SHAP value for each feature.
Figure 6. The SHAP value for each feature.
Nanomaterials 14 01153 g006
Nanomaterials 14 01153 g007
Figure 7. (a) XRD pattern; (b) XPS spectrum; (c) SEM image; (d) resistivity vs. temperature of VO2(M) prepared with magnetron sputtering.
Figure 7. (a) XRD pattern; (b) XPS spectrum; (c) SEM image; (d) resistivity vs. temperature of VO2(M) prepared with magnetron sputtering.
Nanomaterials 14 01153 g007
Table 1. Reaction conditions and labeling comparison table.
Table 1. Reaction conditions and labeling comparison table.
MarkReaction Conditions
PowerThe sputtering power of magnetron sputtering
TempThe temperature of the substrate
O2:ArThe ratio of oxygen to argon flow rate
PressMagnetron sputtering reaction pressure
TarSputtering target material
SubSputtering reaction substrate
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Xiong, G.; Ji, H.; Chen, Y.; Liu, B.; Wang, Y.; Long, P.; Zeng, J.; Tao, J.; Deng, C. Preparation of Thermochromic Vanadium Dioxide Films Assisted by Machine Learning. Nanomaterials 2024, 14, 1153. https://doi.org/10.3390/nano14131153

AMA Style

Xiong G, Ji H, Chen Y, Liu B, Wang Y, Long P, Zeng J, Tao J, Deng C. Preparation of Thermochromic Vanadium Dioxide Films Assisted by Machine Learning. Nanomaterials. 2024; 14(13):1153. https://doi.org/10.3390/nano14131153

Chicago/Turabian Style

Xiong, Gaoyang, Haining Ji, Yongxing Chen, Bin Liu, Yi Wang, Peng Long, Jinfang Zeng, Jundong Tao, and Cong Deng. 2024. "Preparation of Thermochromic Vanadium Dioxide Films Assisted by Machine Learning" Nanomaterials 14, no. 13: 1153. https://doi.org/10.3390/nano14131153

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Article metric data becomes available approximately 24 hours after publication online.
Back to TopTop