Presented is a novel way to combine snapshot compressive imaging and lateral shearing interferometry in order to capture the spatio-spectral phase of an ultrashort laser pulse in a single shot. A deep unrolling algorithm is utilized for snapshot compressive imaging reconstruction due to its parameter efficiency and superior speed relative to other methods, potentially allowing for online reconstruction. The algorithm’s regularization term is represented using a neural network with 3D convolutional layers to exploit the spatio-spectral correlations that exist in laser wavefronts. Compressed sensing is not typically applied to modulated signals, but we demonstrate its success here. Furthermore, we train a neural network to predict the wavefronts from a lateral shearing interferogram in terms of Zernike polynomials, which again increases the speed of our technique without sacrificing fidelity. This method is supported with simulation-based results. While applied to the example of lateral shearing interferometry, the methods presented here are generally applicable to a wide range of signals, including Shack–Hartmann-type sensors. The results may be of interest beyond the context of laser wavefront characterization, including within quantitative phase imaging.

Cold rolling involves large deformation of the workpiece leading to temperature increase due to plastic deformation. This process is highly nonlinear and leads to large computation times to fully model the process. This paper describes the use of dimension-reduced neural networks (DR-NNs) for predicting temperature changes due to plastic deformation in a two-stage cold rolling process. The main objective of these models is to reduce computational demand, error, and uncertainty in predictions. Material properties, feed velocity, sheet dimensions, and friction models are introduced as inputs for the dimensionality reduction. Different linear and nonlinear dimensionality reduction methods reduce the input space to a smaller set of principal components. The principal components are fed as inputs to the neural networks for predicting the output temperature change. The DR-NNs are compared against a standalone neural network and show improvements in terms of lower computational time and prediction uncertainty.

The past 50 yr of advances in weed recognition technologies have poised site-specific weed control (SSWC) on the cusp of requisite performance for large-scale production systems. The technology offers improved management of diverse weed morphology over highly variable background environments. SSWC enables the use of nonselective weed control options, such as lasers and electrical weeding, as feasible in-crop selective alternatives to herbicides by targeting individual weeds. This review looks at the progress made over this half-century of research and its implications for future weed recognition and control efforts; summarizing advances in computer vision techniques and the most recent deep convolutional neural network (CNN) approaches to weed recognition. The first use of CNNs for plant identification in 2015 began an era of rapid improvement in algorithm performance on larger and more diverse datasets. These performance gains and subsequent research have shown that the variability of large-scale cropping systems is best managed by deep learning for in-crop weed recognition. The benefits of deep learning and improved accessibility to open-source software and hardware tools has been evident in the adoption of these tools by weed researchers and the increased popularity of CNN-based weed recognition research. The field of machine learning holds substantial promise for weed control, especially the implementation of truly integrated weed management strategies. Whereas previous approaches sought to reduce environmental variability or manage it with advanced algorithms, research in deep learning architectures suggests that large-scale, multi-modal approaches are the future for weed recognition.

Although robustness is an important consideration to guarantee the performance of designs under deviation, systems are often engineered by evaluating their performance exclusively at nominal conditions. Robustness is sometimes evaluated a posteriori through a sensitivity analysis, which does not guarantee optimality in terms of robustness. This article introduces an automated design framework based on multiobjective optimisation to evaluate robustness as an additional competing objective. Robustness is computed as a sampled hypervolume of imposed geometrical and operational deviations from the nominal point. In order to address the high number of additional evaluations needed to compute robustness, artificial neutral networks are used to generate fast and accurate surrogates of high-fidelity models. The identification of their hyperparameters is formulated as an optimisation problem. In the frame of a case study, the developed methodology was applied to the design of a small-scale turbocompressor. Robustness was included as an objective to be maximised alongside nominal efficiency and mass-flow range between surge and choke. An experimentally validated 1D radial turbocompressor meanline model was used to generate the training data. The optimisation results suggest a clear competition between efficiency, range and robustness, while the use of neural networks led to a speed-up by four orders of magnitude compared to the 1D code.

This paper discusses the development of an experimental software prototype that uses surrogate models for predicting the monthly energy consumption of urban-scale community design scenarios in real time. The surrogate models were prepared by training artificial neural networks on datasets of urban form and monthly energy consumption values of all zip codes in San Diego county. The surrogate models were then used as the simulation engine of a generative urban design tool, which generates hypothetical communities in San Diego following the county's existing urban typologies and then estimates the monthly energy consumption value of each generated design option. This paper and developed software prototype is part of a larger research project that evaluates the energy performance of community microgrids via their urban spatial configurations. This prototype takes the first step in introducing a new set of tools for architects and urban designers with the goal of engaging them in the development process of community microgrids.

The integration of Artificial Neural Networks (ANNs) and Feature Extraction (FE) in the context of the Sample- Partitioning Adaptive Reduced Chemistry approach was investigated in this work, to increase the on-the-fly classification accuracy for very large thermochemical states. The proposed methodology was firstly compared with an on-the-fly classifier based on the Principal Component Analysis reconstruction error, as well as with a standard ANN (s-ANN) classifier, operating on the full thermochemical space, for the adaptive simulation of a steady laminar flame fed with a nitrogen-diluted stream of n-heptane in air. The numerical simulations were carried out with a kinetic mechanism accounting for 172 species and 6,067 reactions, which includes the chemistry of Polycyclic Aromatic Hydrocarbons (PAHs) up to C $ {}_{20} $ . Among all the aforementioned classifiers, the one exploiting the combination of an FE step with ANN proved to be more efficient for the classification of high-dimensional spaces, leading to a higher speed-up factor and a higher accuracy of the adaptive simulation in the description of the PAH and soot-precursor chemistry. Finally, the investigation of the classifier’s performances was also extended to flames with different boundary conditions with respect to the training one, obtained imposing a higher Reynolds number or time-dependent sinusoidal perturbations. Satisfying results were observed on all the test flames.

This work presents a methodology for the generation of novel 3D objects resembling wireframes of building types. These result from the reconstruction of interpolated locations within the learnt distribution of variational autoencoders (VAEs), a deep generative machine learning model based on neural networks. The data set used features a scheme for geometry representation based on a ‘connectivity map’ that is especially suited to express the wireframe objects that compose it. Additionally, the input samples are generated through ‘parametric augmentation’, a strategy proposed in this study that creates coherent variations among data by enabling a set of parameters to alter representative features on a given building type. In the experiments that are described in this paper, more than 150 k input samples belonging to two building types have been processed during the training of a VAE model. The main contribution of this paper has been to explore parametric augmentation for the generation of large data sets of 3D geometries, showcasing its problems and limitations in the context of neural networks and VAEs. Results show that the generation of interpolated hybrid geometries is a challenging task. Despite the difficulty of the endeavour, promising advances are presented.

A first attempt is made to use recently developed, non-conventional Artificial Neural Network (ANN) models with Multilayer Perceptron (MLP), Radial Basis Function (RBF) and Adaptive Neuro-Fuzzy Interference System (ANFIS) architectures to predict the fuel flow rate of a commercial aircraft using real data obtained from Flight Data Records (FDRs) of the cruise, climb and descent phases. The training of the architectures with a single hidden layer is performed by utilising the Delta-Bar-Delta (DBD), Conjugate Gradient (CG) and Quickprop (QP) algorithms. The optimum network topologies are sought by varying the number of processing elements in the hidden layer of the networks using a trial-and-error method. An evaluation of the approximate fuel intake values against the ideal fuel intake data from the FDRs indicates a good fit for all three ANN models. Thus, more accurate fuel intake estimations can be obtained by applying the RBF-ANN model during the climb and descent flight stages, whereas the MLP-ANN model is more effective for the cruise phase. The best accuracy obtained in terms of the linear correlation coefficient is 0.99988, 0.91946 and 0.95252 for the climb, cruise and descent phase, respectively.

In laser-pointing-related applications, when only the centroid of a laser spot is considered, then the position and angular errors of the laser beam are often coupled together. In this study, the decoupling of the position and angular errors is achieved from one single spot image by utilizing a neural network technique. In particular, the successful application of the neural network technique relies on novel experimental procedures, including using an appropriate small-focal-length lens and tilting the detector, to physically enlarge the contrast of different spots. This technique, with the corresponding new system design, may prove to be instructive in the future design of laser-pointing-related systems.

The enzymatic hydrolysis of milk proteins yield final products with improved properties and reduced allergenicity. The degree of hydrolysis (DH) influences both technological (e.g., solubility, water binding capacity) and biological (e.g., angiotensin-converting enzyme (ACE) inhibition, antioxidation) properties of the resulting hydrolysate. Phenomenological models are unable to reproduce the complexity of enzymatic reactions in dairy systems. However, empirical approaches offer high predictability and can be easily transposed to different substrates and enzymes. In this work, the DH of goat milk protein by subtilisin and trypsin was modelled by feedforward artificial neural networks (ANN). To this end, we produced a set of protein hydrolysates, employing various reaction temperatures and enzyme/substrate ratios, based on an experimental design. The time evolution of the DH was monitored and processed to generate the ANN models. Extensive hydrolysis is desirable because a high DH enhances some bioactivities in the final hydrolysate, such as antioxidant or antihypertensive. The optimization of both ANN models led to a maximal DH of 23·47% at 56·4 °C and enzyme–substrate ratio of 5% for subtilisin, while hydrolysis with trypsin reached a maximum of 21·3% at 35 °C and an enzyme–substrate ratio of 4%.

Seed longevity is influenced by many factors, a widely discussed one of which is the seed lipid content and fatty acid composition. Here, linear and non-linear regressions based on machine learning were applied to analyse germinability and seed composition of a set of 42 oilseed rape (Brassica napus L.) accessions grown under the same single environment and at the same time following a period of up to 31 years storage at 7°C. Mean viability was halved after 27.0 years of storage, but this figure concealed a major influence of genotype. There was also wide variation with respect to fatty acid composition, particularly with respect to oleic, α-linolenic, eicosenoic and erucic acid. Linear regression (rL) revealed significant correlation coefficients between normal seedling appearance and the content of α-linolenic acid (+0.52) and total oil (+0.59). Multivariate regression using artificial neural networks including a radial basis function (RBF), a multilayer perceptron (MLP) and a partial least square (PLS) recognized underlying structures and revealed high significant correlation coefficients (rM) for oil content (+0.87), eicosenoic acid (+0.75), stearic acid (+0.73) and lignoceric acid (+0.97). Oil content or a combination of oleic, α-linolenic, arachidic, eicosenoic and eicosadienoic acids and glucosinolates resulted in highest model fitting parameters R2 of 0.90 and 0.88, respectively. In addition, the glucosinolate content, predominantly in the Brassicaceae family and ranging from 4.6 to 79.5 µM, was negatively correlated with viability (rL = ‒0.43). Summarizing, oil content, some fatty acids and glucosinolates contribute to variations in average half-life (15.2 to 50.7 years) of oilseed rape seeds. In contrast to linear regression, multivariate regression using artificial neural networks revealed high associations for combinations of parameters including underestimated minor fatty acids such as arachidic, stearic and eicosadienoic acids. This indicates that genetic and seed composition factors contribute to seed longevity. In addition, multivariate regressions might be a successful approach to predict seed viability based on fatty acids and seed oil content.

Benchmarking function modeling and representation approaches requires a direct comparison, including the inferencing support by the different approaches. To this end, this paper explores the value of a representation by comparing the ability of a representation to support reasoning based on varying amounts of information stored in the representational components of a function structure: vocabulary, grammar, and topology. This is done by classifying the previously developed functional pruning rules into vocabulary, grammatical, and topological classes and applying them to function structures available from an external design repository. The original and pruned function structures of electromechanical devices are then evaluated for how accurately market values can be predicted using the graph complexity connectivity method. The accuracy is found to be inversely related to the amount of information and level of detail. Applying the topological rule does not significantly impact the predictive power of the models, while applying the vocabulary rules and the grammar rules reduces the accuracy of the predictions. Finally, the least predictive model set is that which had all rules applied. In this manner, the value of a representation to predict or answer questions is quantified.

Specifications of the substrates are among the most important and problematic parameters that still do not have proper models in the design procedures of metasurfaces. In this paper, a new fast and exact algorithm based on artificial neural networks (ANNs) is presented, which makes it possible to design frequency-selective surfaces (FSSs) on various kinds of standard substrates. Also for the first time, designing FSSs on uniaxial anisotropic substrates can be easily done in short time and without any optimization algorithms. During this paper, first equivalent geometry approach (EGA) is demonstrated as a new method of preparation the ANNs. Then EGA is used to train geometry transformation ANNs. The advantage of this approach is to reduce the size of training datasets by about 98% and prevent from superfluous simulations. Hence, the time needed for training of the networks is much less than before. Numerical results are used to show that the required time for developing FSSs is <200 ms on average, and errors are <2%. For the final validation, a prototype sample of FSS is fabricated on the RO4003 substrate with 20 mil thickness. Both analytical and experimental results confirm the correctness of the predicted values.