Machine Learning Surrogate Model Research Articles

Generalised Latent Assimilation in Heterogeneous Reduced Spaces with Machine Learning Surrogate Models

Reduced-order modelling and low-dimensional surrogate models generated using machine learning algorithms have been widely applied in high-dimensional dynamical systems to improve the algorithmic efficiency. In this paper, we develop a system which combines reduced-order surrogate models with a novel data assimilation (DA) technique used to incorporate real-time observations from different physical spaces. We make use of local smooth surrogate functions which link the space of encoded system variables and the one of current observations to perform variational DA with a low computational cost. The new system, named generalised latent assimilation can benefit both the efficiency provided by the reduced-order modelling and the accuracy of data assimilation. A theoretical analysis of the difference between surrogate and original assimilation cost function is also provided in this paper where an upper bound, depending on the size of the local training set, is given. The new approach is tested on a high-dimensional (CFD) application of a two-phase liquid flow with non-linear observation operators that current Latent Assimilation methods can not handle. Numerical results demonstrate that the proposed assimilation approach can significantly improve the reconstruction and prediction accuracy of the deep learning surrogate model which is nearly 1000 times faster than the CFD simulation.

Machine learning surrogate models for strain-dependent vibrational properties and migration rates of point defects

Machine learning surrogate models employing atomic environment descriptors have found wide applicability in materials science. In our previous work, this approach yielded accurate and transferable predictions of the vibrational formation entropy of point defects for $\mathcal{O}(N)$ computational cost. The present study investigates the limits of data driven surrogate models in accuracy and applicability for vibrational properties. We propose an improvement of the accuracy by extending the fitting capacity of the model by increasing the dimension of the descriptor space. This is achieved by using a nonlinear relation between descriptors---target observables and when it is possible by including physical relevant information of the underlying energy landscape. The nonlinear extension is used to learn the formation entropy of defects with or without applied strain while including physical information, such as the minimum-saddle point sequences employed for the migration of point defects, is a key ingredient of transition state theory rate approximations. We find excellent predictive power after augmenting the dimensionality of the descriptor space, as demonstrated on large defect databases in $\ensuremath{\alpha}$-iron and amorphous silicon based on semiempirical force fields. The current linear surrogate models are used to investigate the correlation between migration entropy and energy. Our approaches reproduce the Meyer-Neldel compensation law observed from direct calculations in amorphous Si systems. Moreover, the same abstract descriptor space representation for entropy and energy is then used for the statistical correlation analysis. For linear surrogate models, we show that the energy-entropy statistical correlations can be reinterpreted in descriptor space. This provides a simple statistical criterion for the marginal interpretation of the compensation law. More generally, the present work shows how linear surrogate models can accelerate high-throughput workflows, aid the construction of mesoscale material models, and provide new avenues for correlation analysis.

Machine-learning-aided design optimization of internal flow channel cross-sections

• Artificial neural network-based models are developed for prediction of nusselt number and friction factor. • The predictions apply for fully developed laminar flow in constant cross-section ducts of different cross section shapes. • The machine learning (ML) model predictions are validated and compared against numerical simulations. • Use of ML models in shape optimization of channel cross section for different objective functions is demonstrated. • Guidelines for usage of the developed ML models are discussed. The design optimization of various thermal management components such as cold plates, heat sinks, and heat exchangers relies on accurate prediction of flow heat transfer and pressure drop. During the iterative design process, the heat transfer and pressure drop is typically either computed numerically or obtained using geometry-specific correlations for Nusselt number and friction factor. Numerical approaches are accurate for evaluation of a single design but become computationally expensive if many design iterations are required (such as during formal optimization processes). Correlation-based approaches restrict the design space to a specific set of geometries for which correlations are available. Surrogate models for the Nusselt number and friction factor, which are more universally applicable to all geometries than traditional correlations, would enable flexible and computationally inexpensive design optimization. The current work develops machine-learning-based surrogate models for predicting the Nusselt number and friction factor under fully developed internal flow in channels of arbitrary cross section and demonstrates use of these models for optimization of the cross-sectional channel shape. The predictive performance and generality of the machine learning surrogate models is first verified on various shapes outside the training dataset, and then the models are used in the design optimization of flow cross sections based on performance metrics that weigh both heat transfer and pressure drop. The optimization process leads to novel shapes outside the training data, and so numerical simulations are carried out on these optimized shapes to compare with the surrogate model predictions and show their performance is at least as good as that of shapes with known correlations available. A three-lobed shape was found to reduce friction factor, whereas a pentagon with rounded corners and an ice cream cone-shaped duct, both found using different performance metrics. Although the ML model predictions lose accuracy outside the training set for these novel shapes, the predictions follow the correct trends with parametric variations of the shape and therefore successfully direct the search toward optimized shapes.

Surrogate role of machine learning in motor-drive optimization for more-electric aircraft applications

Motor drives form an essential part of the electric compressors, pumps, braking and actuation systems in the More-Electric Aircraft (MEA). In this paper, the application of Machine Learning (ML) in motor-drive design and optimization process is investigated. The general idea of using ML is to train surrogate models for the optimization. This training process is based on sample data collected from detailed simulation or experiment of motor drives. However, the Surrogate Role (SR) of ML may vary for different applications. This paper first introduces the principles of ML and then proposes two SRs (direct mapping approach and correction approach) of the ML in a motor-drive optimization process. Two different cases are given for the method comparison and validation of ML SRs. The first case is using the sample data from experiments to train the ML surrogate models. For the second case, the joint-simulation data is utilized for a multi-objective motor-drive optimization problem. It is found that both surrogate roles of ML can provide a good mapping model for the cases and in the second case, three feasible design schemes of ML are proposed and validated for the two SRs. Regarding the time consumption in optimizaiton, the proposed ML models can give one motor-drive design point up to 0.044 s while it takes more than 1.5 mins for the used simulation-based models.

Improving predictions of evapotranspiration by integrating multi-source observations and land surface model

Accurate estimation of evapotranspiration (ET) is essential for understanding terrestrial energy, water, and carbon cycles. This study proposes a hybrid model integrating in-situ and remote sensing-derived soil moisture (SM) observations and remote sensing leaf area index (LAI) with the Noah-MP model. The ensemble Kalman filter (EnKF) approach updates the leaf biomass and specific leaf area (SLA) by assimilating the remotely sensed LAI. A machine learning (ML) surrogate model is used to integrate multi-site SM profile observations and remote sensing SM products to estimate the three-layer SM. An iterative coupling of two parts implements the hybrid model: optimization of leaf biomass and SLA by assimilation of LAI in the Noah-MP model and simulation of three-layer SM in the ML surrogate model. The performance of the hybrid model is evaluated in the Heihe River Basin (HRB) in northwest China. The estimated ET from the hybrid model is compared with observations from the large aperture scintillometer (LAS) at the Arou, Daman, and Sidaoqiao sites and up-scaled watershed ET over the HRB. The findings indicate that the hybrid model performs well in vegetated areas but underestimates ET in extreme arid deserts. The three-site unbiased root mean squared errors (ubRMSEs) of ET estimates from the hybrid model are 29.06%, 42.76%, and 50.00% lower than Noah-MP at the Arou, Daman, and Sidaoqiao sites, respectively. The coupling of data assimilation (DA) and ML methods can improve vegetation dynamics and SM transport estimation in the Noah-MP model. The hybrid model can take advantage of DA and ML methods and integrate multi-source observations to improve the accuracy of ET estimation. The results also indicate that the ET predictions are more sensitive to root zone SM (0–40 cm) over croplands, grasslands, and shrublands, while the ET simulations are more affected by deeper rooting depths SM (0–100 cm) and groundwater over forests.

Prediction of grain structure after thermomechanical processing of U-10Mo alloy using sensitivity analysis and machine learning surrogate model

Hot rolling and annealing are critical intermediate steps for controlling microstructures and thickness variations when fabricating uranium alloyed with 10% molybdenum (U-10Mo), which is highly relevant to worldwide nuclear non-proliferation efforts. This work proposes a machine-learning surrogate model combined with sensitivity analysis to identify and predict U-10Mo microstructure development during thermomechanical processing. Over 200 simulations were collected using physics-based microstructure models covering a wide range of thermomechanical processing routes and initial alloy grain features. Based on the sensitivity analysis, we determined that an increase in rolling reduction percentage at each processing pass has the strongest effect in reducing the grain size. Multi-pass rolling and annealing can significantly improve recrystallization regardless of the reduction percentage. With a volume fraction below 2%, uranium carbide particles were found to have marginal effects on the average grain size and distribution. The proposed stratified stacking ensemble surrogate predicts the U-10Mo grain size with a mean square error four times smaller than a standard single deep neural network. At the same time, with a significant speedup (1000×) compared to the physics-based model, the machine learning surrogate shows good potential for U-10Mo fabrication process optimization.

Surrogate-assisted evolutionary multi-objective optimisation applied to a pressure swing adsorption system

The complexity of chemical plant systems (CPS) makes optimising their design and operation challenging tasks. This complexity also results in analytical and numerical simulation models of these systems having high computational costs. Research demonstrates the benefits of using machine learning models as surrogates or substitutes for these computationally expensive simulation models during CPS optimisation. This paper presents the results of our study, extending recent research into optimising chemical plant design and operation. The study explored the original surrogate-assisted genetic algorithms (SA-GA) in more complex variants of the plant design and operation optimisation problem. The more complex plant design variants include additional parallel and feedback components. The study also proposes a novel multivariate extension, surrogate-assisted NSGA (SA-NSGA), to the original univariate SA-GA algorithm. The study evaluated the SA-NSGA extension on the popular pressure swing adsorption (PSA) system. This paper outlines our extensive experimentation, comparing various meta-heuristic optimisation techniques and numerous machine learning models as surrogates. The results in both more complex plant design variants and the PSA case show the suitability of genetic algorithms combined with surrogate models as an optimisation framework for CPS design and operation in single and multi-objective scenarios. The analysis further confirms that combining a genetic algorithm framework with machine learning surrogate models as a substitute for long-running simulation models yields significant computational efficiency improvements, 1.7–1.84 times speedup for the increased complexity examples and a 2.7 times speedup for the pressure swing adsorption system. The discussion successfully concludes that surrogate-assisted evolutionary algorithms can be scaled to increasingly complex CPS with parallel and feedback components.

Data-driven surrogate model with latent data assimilation: Application to wildfire forecasting

The large and catastrophic wildfires have been increasing across the globe in the recent decade, highlighting the importance of simulating and forecasting fire dynamics in near real-time. This is extremely challenging due to the complexities of physical models and geographical features. Running physics-based simulations for large wildfire events in near real-time are computationally expensive, if not infeasible. In this work, we develop and test a novel data-model integration scheme for fire progression forecasting, that combines Reduced-order modelling, recurrent neural networks (Long-Short-Term Memory), data assimilation, and error covariance tuning. The Reduced-order modelling and the machine learning surrogate model ensure the efficiency of the proposed approach while the data assimilation enables the system to adjust the simulation with observations. We applied this algorithm to simulate and forecast three recent large wildfire events in California from 2017 to 2020. The deep-learning-based surrogate model runs around 1000 times faster than the Cellular Automata simulation which is used to generate training data-sets. The daily fire perimeters derived from satellite observation are used as observation data in Latent Assimilation to adjust the fire forecasting in near real-time. An error covariance tuning algorithm is also performed in the reduced space to estimate prior simulation and observation errors. The evolution of the averaged relative root mean square error (R-RMSE) shows that data assimilation and covariance tuning reduce the RMSE by about 50% and considerably improves the forecasting accuracy. As a first attempt at a reduced order wildfire spread forecasting, our exploratory work showed the potential of data-driven machine learning models to speed up fire forecasting for various applications.

Conservation laws in a neural network architecture: enforcing the atom balance of a Julia-based photochemical model (v0.2.0)

Abstract. Models of atmospheric phenomena provide insight into climate, air quality, and meteorology and provide a mechanism for understanding the effect of future emissions scenarios. To accurately represent atmospheric phenomena, these models consume vast quantities of computational resources. Machine learning (ML) techniques such as neural networks have the potential to emulate computationally intensive components of these models to reduce their computational burden. However, such ML surrogate models may lead to nonphysical predictions that are difficult to uncover. Here we present a neural network architecture that enforces conservation laws to numerical precision. Instead of simply predicting properties of interest, a physically interpretable hidden layer within the network predicts fluxes between properties which are subsequently related to the properties of interest. This approach is readily generalizable to physical processes where flux continuity is an essential governing equation. As an example application, we demonstrate our approach on a neural network surrogate model of photochemistry, trained to emulate a reference model that simulates formation and reaction of ozone. We design a physics-constrained neural network surrogate model of photochemistry using this approach and find that it conserves atoms as they flow between molecules while outperforming two other neural network architectures in terms of accuracy, physical consistency, and non-negativity of concentrations.

Predicting combined tidal and pluvial flood inundation using a machine learning surrogate model

Flooding increases in recent years, in particular for coastal communities facing sea level rise, have brought renewed attention to real-time, street-scale flood forecasting. Such flood models using conventional physics-based modeling approaches are often unrealistic for real-time decision support use cases due to their long model runtime. Machine learning offers an alternative strategy whereby a surrogate model can be trained to mimic relationships present within the physics-based model and, after training, can run in seconds rather than hours. This study used the Random Forest (RF) algorithm to emulate a 1D/2D physics-based model simulating surface water depths in an urban coastal watershed in Norfolk, Virginia. Environmental features from a selected set of pluvial and tidal flood events and topographic information of the roadway were the input variables to train the surrogate model. Results show the potential for the surrogate model to predict flood extent and depth for both pluvial and tidal flood events. Furthermore, the surrogate model can differentiate between flooding locations dominated by pluvial or tidal flooding or impacted by both flooding mechanisms. Flood reports from the mobile app Waze were used for model validation and show 90% agreement with flooding locations from the surrogate model. Finally, feature importance methods were investigated to interpret the performance of the RF models and understand the contribution of different physical features to localized flooding.

Accurate Machine-learning Atmospheric Retrieval via a Neural-network Surrogate Model for Radiative Transfer

Atmospheric retrieval determines the properties of an atmosphere based on its measured spectrum. The low signal-to-noise ratios of exoplanet observations require a Bayesian approach to determine posterior probability distributions of each model parameter, given observed spectra. This inference is computationally expensive, as it requires many executions of a costly radiative transfer (RT) simulation for each set of sampled model parameters. Machine learning (ML) has recently been shown to provide a significant reduction in runtime for retrievals, mainly by training inverse ML models that predict parameter distributions, given observed spectra, albeit with reduced posterior accuracy. Here we present a novel approach to retrieval by training a forward ML surrogate model that predicts spectra given model parameters, providing a fast approximate RT simulation that can be used in a conventional Bayesian retrieval framework without significant loss of accuracy. We demonstrate our method on the emission spectrum of HD 189733 b and find good agreement with a traditional retrieval from the Bayesian Atmospheric Radiative Transfer (BART) code (Bhattacharyya coefficients of 0.9843–0.9972, with a mean of 0.9925, between 1D marginalized posteriors). This accuracy comes while still offering significant speed enhancements over traditional RT, albeit not as much as ML methods with lower posterior accuracy. Our method is ∼9× faster per parallel chain than BART when run on an AMD EPYC 7402P central processing unit (CPU). Neural-network computation using an NVIDIA Titan Xp graphics processing unit is 90×–180× faster per chain than BART on that CPU.

Experimental validation of an adsorbent-agnostic artificial neural network (ANN) framework for the design and optimization of cyclic adsorption processes

The efficacy of an adsorbent agnostic machine-learning surrogate model for rapid design and optimization of a Skarstrom cycle vacuum swing adsorption (VSA) process is experimentally validated. The surrogate model is trained to predict the process performance using adsorbent features that include hypothetical Langmuir adsorption isotherm parameters, particle density, porosity and bed voidage, and process variables such as pressure, step duration and feed velocity. The training data was generated from a detailed process model for 20,000 unique combinations of the training variables. The model shows high accuracy of Radj2>0.99 for predicting key performance parameters such as product purity, recovery and productivity. The ability of this surrogate to predict the experimental performance for the purification of O2 from air on two adsorbents, namely 13X and LiX zeolites, was studied. Two separate multi-objective optimization studies, to maximize purity and recovery, and to maximize productivity and purity were performed. For these optimization studies, the volumetrically measured isotherms of N2 and O2 were used as inputs to the surrogate model. Note that these isotherms were not a part of the dataset used to train the model. Nine points were chosen from the Parteo curves and the corresponding decision variables were used as set-points in a two-column lab-scale rig. The average difference between the calculated and experimentally measured purity, recovery and productivity was 3%, 5% and 9%, respectively. This study provides the necessary confidence to use surrogate-based process models for adsorbent screening and adsorption process optimization.

Building a machine learning surrogate model for wildfire activities within a global Earth system model

Abstract. Wildfire is an important ecosystem process, influencing land biogeophysical and biogeochemical dynamics and atmospheric composition. Fire-driven loss of vegetation cover, for example, directly modifies the surface energy budget as a consequence of changing albedo, surface roughness, and partitioning of sensible and latent heat fluxes. Carbon dioxide and methane emitted by fires contribute to a positive atmospheric forcing, whereas emissions of carbonaceous aerosols may contribute to surface cooling. Process-based modeling of wildfires in Earth system land models is challenging due to limited understanding of human, climate, and ecosystem controls on fire counts, fire size, and burned area. Integration of mechanistic wildfire models within Earth system models requires careful parameter calibration, which is computationally expensive and subject to equifinality. To explore alternative approaches, we present a deep neural network (DNN) scheme that surrogates the process-based wildfire model with the Energy Exascale Earth System Model (E3SM) interface. The DNN wildfire model accurately simulates observed burned area with over 90 % higher accuracy with a large reduction in parameterization time compared with the current process-based wildfire model. The surrogate wildfire model successfully captured the observed monthly regional burned area during validation period 2011 to 2015 (coefficient of determination, R2=0.93). Since the DNN wildfire model has the same input and output requirements as the E3SM process-based wildfire model, our results demonstrate the applicability of machine learning for high accuracy and efficient large-scale land model development and predictions.

Visual Parameter Space Analysis for Optimizing the Quality of Industrial Nonwovens.

Technical textiles, in particular, nonwovens used, for example, in medical masks, have become increasingly important in our daily lives. The quality of these textiles depends on the manufacturing process parameters that cannot be easily optimized in live settings. In this article, we present a visual analytics framework that enables interactive parameter space exploration and parameter optimization in industrial production processes of nonwovens. Therefore, we survey analysis strategies used in optimizing industrial production processes of nonwovens and support them in our tool. To enable real-time interaction, we augment the digital twin with a machine learning surrogate model for rapid quality computations. In addition, we integrate mechanisms for sensitivity analysis that ensure consistent product quality under mild parameter changes. In our case study, we explore the finding of optimal parameter sets, investigate the input-output relationship between parameters, and conduct a sensitivity analysis to find settings that result in robust quality.

Prediction of the outflow temperature of large-scale hydropower using theory-guided machine learning surrogate models of a high-fidelity hydrodynamics model

Stratified water intake facilities are important pieces of engineering infrastructure that monitor the outflow temperature conditions of hydropower projects. The outflow temperature can significantly impact the downstream eco-environment and normal functioning of aquatic organisms in streams. However, due to the lack of scientific and effective management tools for stratified water intake and the complexity of the hydrodynamics associated with hydropower generation, both the research community and operation sectors are making great efforts to explore new tools and technology to better monitor, predict and control stratified intake facilities. In this study, surrogate models of the Environmental Fluid Dynamics Code (EFDC) model based on the theory-guided machine learning (TGML) paradigm are constructed. These models are applied to outflow temperature prediction for the Jinping-I Hydropower Plant in China. The results show that 1) the prediction precision of the high-fidelity hydrodynamic and water quality EFDC model is successfully emulated by a TGML model based on a long short-term memory (LSTM) algorithm; and 2) the TGML model accuracy, based on the mean absolute error (MAE) value obtained using the LSTM algorithm, can reach 0.228–0.269 °C, the prediction time is less than 2 s, the period of high-precision prediction is 6–13 days, and the model can guide the operation of stratified intake facilities in practice.

A Beam-Splitting Bianisotropic Metasurface Designed by Optimization and Machine Learning

Electromagnetic metasurfaces have attracted significant interest recently due to their low profile and advantageous applications. Practically, many metasurface designs start with a set of constraints for the radiated far-field, such as main-beam direction(s) and side lobe levels, and end with a non-uniform physical structure for the surface. This problem is quite challenging, since the required tangential field transformations are not completely known when only constraints are placed on the scattered fields. Hence, the required surface properties cannot be solved for analytically. Moreover, the translation of the desired surface properties to the physical unit cells can be time-consuming and difficult, as it is often a one-to-many mapping in a large solution space. Here, we divide the inverse design process into two steps: a macroscopic and microscopic design step. In the former, we use an iterative optimization process to find the surface properties that radiate a far-field pattern that complies with specified constraints. This iterative process exploits non-radiating currents to ensure a passive and lossless design. In the microscopic step, these optimized surface properties are realized with physical unit cells using machine learning surrogate models. The effectiveness of this end-to-end synthesis process is demonstrated through measurement results of a beam-splitting prototype.

Ensemble latent assimilation with deep learning surrogate model: application to drop interaction in a microfluidics device.

A major challenge in the field of microfluidics is to predict and control drop interactions. This work develops an image-based data-driven model to forecast drop dynamics based on experiments performed on a microfluidics device. Reduced-order modelling techniques are applied to compress the recorded images into low-dimensional spaces and alleviate the computational cost. Recurrent neural networks are then employed to build a surrogate model of drop interactions by learning the dynamics of compressed variables in the reduced-order space. The surrogate model is integrated with real-time observations using data assimilation. In this paper we developed an ensemble-based latent assimilation algorithm scheme which shows an improvement in terms of accuracy with respect to the previous approaches. This work demonstrates the possibility to create a reliable data-driven model enabling a high fidelity prediction of drop interactions in microfluidics device. The performance of the developed system is evaluated against experimental data (i.e., recorded videos), which are excluded from the training of the surrogate model. The developed scheme is general and can be applied to other dynamical systems.

A comparison of machine learning surrogate models for net present value prediction from well placement binary data

Net Present Value (NPV) is an important indicator to guide investment decisions. In oil production planning, NPV is employed to evaluate and select among different production strategies. However, NPV estimation requires computational costly numerical simulations. So, evaluating as many production strategies as is desirable may be prohibitive. Therefore, one can only evaluate a small part of the search space, decreasing the chance of finding a near-optimal production strategy. To speed up the searching process, a much faster, but error-prone, surrogate model is used to approximate the simulator output. Data-driven surrogate modeling depends on both: 1) building a simple model to reproduce the quality of a high-fidelity model, while 2) considering a large volume of data to build it. In this work, we address the well placement optimization task by considering a binary data representation, indicating the presence or absence of a given well in a production strategy. We show the possibility of predicting the NPV from binary data, thus reducing data dimension and model complexity. Specifically, we compare six machine learning regression algorithms to predict the NPV. The simulations conducted in a benchmark case, based on a real field, showed that some regression algorithms can be used as a surrogate model to the simulator to efficiently perform well placement optimization considering binary data. The best results were obtained with Multi-Layer Perceptron, whose estimations covered a wide range of NPV with a small and constant error.

Airborne hyperspectral imaging of nitrogen deficiency on crop traits and yield of maize by machine learning and radiative transfer modeling

Nitrogen is an essential nutrient that directly affects plant photosynthesis, crop yield, and biomass production for bioenergy crops, but excessive application of nitrogen fertilizers can cause environmental degradation. To achieve sustainable nitrogen fertilizer management for precision agriculture, there is an urgent need for nondestructive and high spatial resolution monitoring of crop nitrogen and its allocation to photosynthetic proteins as that changes over time. Here, we used visible to shortwave infrared (400–2400 nm) airborne hyperspectral imaging with high spatial (0.5 m) and spectral (3–5 nm) resolutions to accurately estimate critical crop traits, i.e., nitrogen, chlorophyll, and photosynthetic capacity (CO2-saturated photosynthesis rate, Vmax,27), at leaf and canopy scales, and to assess nitrogen deficiency on crop yield. We conducted three airborne campaigns over a maize (Zea mays L.) field during the growing season of 2019. Physically based soil-canopy Radiative Transfer Modeling (RTM) and data-driven approaches i.e. Partial-Least Squares Regression (PLSR) were used to retrieve crop traits from hyperspectral reflectance, with ground truth of leaf nitrogen, chlorophyll, Vmax,27, Leaf Area Index (LAI), and harvested grain yield. To improve computational efficiency of RTMs, Random Forest (RF) was used to mimic RTM simulations to generate machine learning surrogate models RTM-RF. The results show that prior knowledge of soil background and leaf angle distribution can significantly reduce the ill-posed RTM retrieval. RTM-RF achieved a high accuracy to predict leaf chlorophyll content (R2 = 0.73) and LAI (R2 = 0.75). Meanwhile, PLSR exhibited better accuracy to predict leaf chlorophyll content (R2 = 0.79), nitrogen concentration (R2 = 0.83), nitrogen content (R2 = 0.77), and Vmax,27 (R2 = 0.69) but required measured traits for model training. We also found that canopy structure signals can enhance the use of spectral data to predict nitrogen related photosynthetic traits, as combining RTM-RF LAI and PLSR leaf traits well predicted canopy-level traits (leaf traits × LAI) including canopy chlorophyll (R2 = 0.80), nitrogen (R2 = 0.85) and Vmax,27 (R2 = 0.82). Compared to leaf traits, we further found that canopy-level photosynthetic traits, particularly canopy Vmax,27, have higher correlation with maize grain yield. This study highlights the potential for synergistic use of process-based and data-driven approaches of hyperspectral imaging to quantify crop traits that facilitate precision agricultural management to secure food and bioenergy production.

Machine learning reinforced microstructure-sensitive prediction of material property closures

This study addresses a machine learning (ML)-reinforced strategy to build both linear and non-linear property closures for metallic materials. A property closure is a closed space of material properties that contains all possible values of the closure variables. The material properties of metals are significantly dependent on the underlying microstructure texture. Here, the polycrystalline material is expressed by the Orientation Distribution Function (ODF) that relates to the volume densities of the crystallographic orientations. Theoretically, the property closures of volume-averaged material properties can be derived using single-crystal microstructure solutions; however, this theory is not valid for non-linear properties. Therefore, we use an ML-reinforced strategy to generate both linear and non-linear material property closures using the Linear Regression (LR) and Artificial Neural Network (ANN) method with Bayesian Regularization. The closures for material properties, such as the elastic stiffness parameters and critical buckling load, are generated for Titanium, Magnesium, and Aluminum. The outcomes of the ML surrogate models for these properties are compared to each other. The results demonstrate that the ANN model with Bayesian regularization is capable of predicting both linear and non-linear material properties with almost 100% accuracy. However, the linear regression algorithm is found to be not as accurate as of the Bayesian inference for the non-linear property even though it provides similar accuracy as ANN for the linear property. Therefore, ANN with Bayesian regularization is utilized for predicting the property closure of critical buckling load, which is a non-linear property.