Interpretable Forward and Inverse Design of Particle Spectral Emissivity Using Common Machine-Learning Models

Elzouka, Mahmoud; Yang, Charles; Albert, Adrian; Prasher, Ravi; Lubner, Sean

doi:10.1016/j.xcrp.2020.100259

Cited by 22 publications

(19 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This trend has also been observed in previous ML studies across the physical sciences. [ 23,35 ] While many of the R 2 coefficients are lower than ideal, a large part of this underperformance may be due to its convolution with experimental error. To understand the effects of experimental uncertainty on the predictive power, we considered an additional χ 2 metric, defined as

{\chi ^2} = \frac{1}{N}\sum \frac{{{{({e_{\rm{i}}} - {p_{\rm{i}}})}^2}}}{{{\sigma ^2}}}

, where N is the number of samples, e i is the experimentally‐measured value, p i is the predicted value, and σ i is the experimental uncertainty.…”

Section: Resultsmentioning

confidence: 99%

“…This trend has also been observed in previous ML studies across the physical sciences. [23,35] While many of the R 2 coefficients are lower than ideal, a large part of this underperformance may be due to its convolution with experimental error. To understand the effects of experimental uncertainty on the predictive power, we considered an additional χ 2 metric, defined as 1 ( )…”

Section: Machine Learningmentioning

confidence: 99%

See 1 more Smart Citation

Using Machine Learning to Predict and Understand Complex Self‐Assembly Behaviors of a Multicomponent Nanocomposite

et al. 2022

View full text Add to dashboard Cite

Currently, devices like these are designed and refined via many cycles of trial and error. The nanocomposite design process is particularly arduous because effective medium approximations do not accurately predict the properties of materials built from nanoscopic components. [5,6] In the case of polymer-nanoparticle blends, the spatial arrangement of components determines the overall properties through nanoparticle (NP) coupling, confinement, and organic-inorganic interface effects. [7][8][9] The final NP distribution, achieved via self-assembly, is a complex function of multiple composition and processing variables. [10][11][12] Due to the abundance of non-equilibrium states, knowledge of the kinetic pathway is particularly important for processing polymer-based blends. [13][14][15][16] Reverse engineering a specific nanostructure is challenging because the parameter space of formulations and processing conditions is very large. For example, in blends of block copolymers (BCPs) and NPs, a common class of nanoscopicallystructured composites, the final arrangement is a function of the polymer chain length, the block ratio, the particle size, the relative proportions of the components, the chemical compatibility between them, and the processing conditions. [15,16] Within the last decade, it has been established that adding organic small molecules to a BCP-NP blend facilitates the incorporation of large or anisotropic particles, accelerates assembly, and produces new nanostructures. [17][18][19] These are exciting milestones in the development of functional nanocomposites but, with the addition of small molecules, the parameter space of nanocomposite compositions has grown even larger.The challenges of working with large parameter spaces are not unique to self-assembly. Humans are excellent at finding patterns in low-dimensional data but struggle to understand trends in high-dimensional systems. Machine learning (ML) methods offer ways to predict outcomes and visualize trends in high-dimensional spaces. As these methods do not rely on hard-coded relationships between parameters, they are suited to complex systems without a solid theoretical foundation. Parameter spaces considered large by experimental standards, such as the 7D space described above, are tiny compared to the capabilities of modern ML models. [20] ML methods have recently Blends of nanoparticles, polymers, and small molecules can self-assemble into optical, magnetic, and electronic devices with structure-dependent properties. However, the relationship between a multicomponent nanocomposite's formulation and its assembled structure is complex and cannot be predicted by theory. The blends can be strongly influenced by processing conditions, which can introduce non-equilibrium states. Currently, nanocomposite devices are designed through cycles of experimental trial and error. Machine learning (ML) methods are a compelling alternative because they can use existing datasets to map high-dimensional spaces. These methods do not rely on known relationships b...

show abstract

{\chi ^2} = \frac{1}{N}\sum \frac{{{{({e_{\rm{i}}} - {p_{\rm{i}}})}^2}}}{{{\sigma ^2}}}

, where N is the number of samples, e i is the experimentally‐measured value, p i is the predicted value, and σ i is the experimental uncertainty.…”

Section: Resultsmentioning

confidence: 99%

“…This trend has also been observed in previous ML studies across the physical sciences. [23,35] While many of the R 2 coefficients are lower than ideal, a large part of this underperformance may be due to its convolution with experimental error. To understand the effects of experimental uncertainty on the predictive power, we considered an additional χ 2 metric, defined as 1 ( )…”

Section: Machine Learningmentioning

confidence: 99%

Using Machine Learning to Predict and Understand Complex Self‐Assembly Behaviors of a Multicomponent Nanocomposite

et al. 2022

View full text Add to dashboard Cite

show abstract

“…DL and Machine Learning (ML) have been used, in a broad setting, to solve complex problems ranging from machine vision for self-driving vehicles 32 to automatic speech recognition 33 and spacecraft system optimization 34 – 37 . In the field of optics, DL has been used recently to predict and model plasmonic behavior 31 , 38 – 42 , grating structures 43 , 44 , ceramic metasurfaces 45 , 46 , chiral materials 47 , 48 , particles and nanosturctures 49 – 51 , and to do inverse design 31 , 41 , 50 – 54 . Deep-Learning has also been used extensively in the field of heat transfer for applications such as predicting thermal conductivity 55 , 56 and thermal boundary resistance 57 , studying transport phenomena 58 , optimizing integrated circuits 59 , modelling boiling heat transfer 60 , predicting thermal-optical properties 44 , 61 , 62 , and addressing thermal radiation problems 63 – 66 .…”

Section: Introductionmentioning

confidence: 99%

“…In this paper we propose a methodology based on a DNN to predict the optical properties of micropyramids across a wide design space of geometries, wavelengths, and, most importantly, materials. As opposed to many other studies that provide a deep learning approach to a structure with a single material 38 , 52 , a geometry with fixed materials 44 , 47 , 51 , or a material input defined by one-hot encoding with a random forest 50 , our DNN is designed to predict the optical properties of a vast array of materials and is not constrained by material input. While there are many available machine learning methods 42 , 50 , 61 , 70 – 74 , we choose to utilize the deep neural network approach due to the method’s input flexibility, scalability, and the ability to extrapolate outputs from unseen inputs.…”

Section: Introductionmentioning

confidence: 99%

Deep learning based analysis of microstructured materials for thermal radiation control

Sullivan

Mirhashemi

Lee

2022

Sci Rep

View full text Add to dashboard Cite

Microstructured materials that can selectively control the optical properties are crucial for the development of thermal management systems in aerospace and space applications. However, due to the vast design space available for microstructures with varying material, wavelength, and temperature conditions relevant to thermal radiation, the microstructure design optimization becomes a very time-intensive process and with results for specific and limited conditions. Here, we develop a deep neural network to emulate the outputs of finite-difference time-domain simulations (FDTD). The network we show is the foundation of a machine learning based approach to microstructure design optimization for thermal radiation control. Our neural network differentiates materials using discrete inputs derived from the materials’ complex refractive index, enabling the model to build relationships between the microtexture’s geometry, wavelength, and material. Thus, material selection does not constrain our network and it is capable of accurately extrapolating optical properties for microstructures of materials not included in the training process. Our surrogate deep neural network can synthetically simulate over 1,000,000 distinct combinations of geometry, wavelength, temperature, and material in less than a minute, representing a speed increase of over 8 orders of magnitude compared to typical FDTD simulations. This speed enables us to perform sweeping thermal-optical optimizations rapidly to design advanced passive cooling or heating systems. The deep learning-based approach enables complex thermal and optical studies that would be impossible with conventional simulations and our network design can be used to effectively replace optical simulations for other microstructures.

show abstract

“…Fortunately, the unprecedented development of machine learning has made it a powerful tool for solving complex computing and inverse design problems. Several studies have reported machine learning methods to facilitate the design of structural colors and inverse design of nanostructures and materials to achieve the desired optical response [ 27 , 28 , 29 , 30 , 31 , 32 , 33 ]. So et al [ 28 ] achieved the simultaneous inverse design of materials and structural parameters of core–shell nanoparticles by using neural networks.…”

Section: Introductionmentioning

confidence: 99%

Evaluation and Design of Colored Silicon Nanoparticle Systems Using a Bidirectional Deep Neural Network

Zhou

Wang

et al. 2022

Nanomaterials

View full text Add to dashboard Cite

Silicon nanoparticles (SiNPs) with lowest-order Mie resonance produce non-iridescent and non-fading vivid structural colors in the visible range. However, the strong wavelength dependence of the radiation pattern and dielectric function makes it very difficult to design nanoparticle systems with the desired colors. Most existing studies focus on monodisperse nanoparticle systems, which are unsuitable for practical applications. This study combined the Lorentz–Mie theory, Monte Carlo, and deep neural networks to evaluate and design colored SiNP systems. The effects of the host medium and particle size distribution on the optical and color properties of the SiNP systems were investigated. A bidirectional deep neural network achieved accurate prediction and inverse design of structural colors. The results demonstrated that the particle size distribution flattened the Mie resonance peak and influenced the reflectance and brightness of the SiNP system. The SiNPs generated vivid colors in all three of the host media. Meanwhile, our proposed neural network model achieved a near-perfect prediction of colors with high accuracy of the designed geometric parameters. This work accurately and efficiently evaluates and designs the optical and color properties of SiNP systems, thus accelerating the design process and contributing to the practical production design of color inks, decoration, and printing.

show abstract

Interpretable Forward and Inverse Design of Particle Spectral Emissivity Using Common Machine-Learning Models

Cited by 22 publications

References 52 publications

Using Machine Learning to Predict and Understand Complex Self‐Assembly Behaviors of a Multicomponent Nanocomposite

Using Machine Learning to Predict and Understand Complex Self‐Assembly Behaviors of a Multicomponent Nanocomposite

Deep learning based analysis of microstructured materials for thermal radiation control

Evaluation and Design of Colored Silicon Nanoparticle Systems Using a Bidirectional Deep Neural Network

Contact Info

Product

Resources

About