Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “benchmark dataset”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

835 records · Page 47

Industrial Geospatial Analysis Tool for Energy Evaluation (IGATE-E)

IGATE-E is an energy analysis tool for industrial energy evaluation. The tool applies statistical modeling to multiple publicly available datasets and provides information at the geospatial resolution of zip code using bottom up approaches. Within each zip code, the current version of the tool estimates electrical energy consumption of manufacturing industries based on each type of industries using DOE s Industrial Assessment Center database (IAC-DB) and DOE s Energy Information Administration Manufacturing Energy Consumption Survey database (EIA-MECS DB), in addition to other commercially available databases such as the Manufacturing News database (MNI, Inc.). Ongoing and future work include adding modules for the predictions of fuel energy consumption streams, manufacturing process steps energy consumption, major energy intensive processes (EIPs) within each industry type among other metrics of interest. The tool provides validation against DOE s EIA-MECS state level energy estimations and permits several statistical examinations. IGATE-E is intended to be a decision support and planning tool to a wide spectrum of energy analysts, researchers, government organizations, private consultants, industry partners, and alike.

Energy Analysis Tool↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-ofthe-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Hesham ElAbd↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-of-the-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Deep-learning↗

DECOVALEX-2023: Task C Final Report

The Full-scale Emplacement (FE) heater experiment at the Mont Terri Underground Rock Laboratory (URL) was designed and conducted by Nagra to replicate an emplacement tunnel of Nagra’s reference repository design at 1:1 scale. Alongside testing the technical feasibility of constructing disposal tunnels, emplacing waste containers in the tunnels and then backfilling them, the main goals of the FE experiment are (1) to obtain a better understanding of the coupled effects of induced thermo-hydro-mechanical (THM) processes that may occur and (2) to validate existing coupled THM models (Müller et al., 2017). A key aspect of ensuring safety for repositories located in low-permeability rock involves minimizing any damage to the rock itself, thereby preserving its integrity and promoting a stable environment Amongst a number of processes that could damage the rock is the increase in pore pressure due to thermal loading caused by heat emitted from the waste. To reduce the potential damage of the rock, it is important to analyse the evolution of heat over time due to the heat load of the containers and assess possible consequences by coupled THM models. The aim of Task C of DECOVALEX-2023 was to build 3D numerical models of the FE experiment, focussing in particular on the heating induced pore pressure change in the Opalinus Clay. Data from a large number of sensors were available from the FE experiment for model comparison. These sensors measured temperature and relative humidity in the bentonite around the heaters, and temperature, pressure and displacement/strain in the surrounding Opalinus clay. Data were available from the start of excavation (April 2012) up to August 2020 for most sensors (more than 5 years from the start of heating in December 2014). To fulfil the overall aim of the task, the work was broken down into a number of steps, starting with simpler models to build confidence in each team’s approach and then moving to more complex models that better represent the FE experiment. Step 0 consisted of 2D benchmark models, gradually increasing the number of processes that are represented from thermal (T) only models in Step 0a, to coupled thermal hydraulic (TH) models in Step 0b with a representation of changing porosity, to coupled thermo-hydro-mechanical (THM) models in Step 0c, where porosity changes are calculated by the mechanical model. A detailed specification of processes, parameters, initial and boundary conditions was provided for this step, with the ambition that all teams would work towards close agreement in their model results, thus building confidence in the model implementations. vi It was not straightforward to achieve agreement between the teams, so additional steps (Step 0b2, 0b3, 0c2, 0c3) were added along with derivation of some analytical solutions against which the models could be compared. The reasons for the differences between teams were investigated and found to be caused primarily by different conceptual model assumptions (including temperature dependence of the thermal expansion of water), different model formulations (including porosity evolution) and differences in modelled domain sizes, boundary conditions and grid discretisation. This demonstrates that comparisons between multiple modelling teams and/or comparison with analytical results and experimental data are highly beneficial in providing an indication of uncertainty in model predictions. At the conclusion of Step 0, almost all teams had achieved a close agreement in model results and those that had not achieved an agreement knew the reason for this. Step 1 moved from 2D models to 3D models of the FE experiment without adding technical features like shotcrete or EDZ, and only considering the heating phase. Initially the 3D model was tightly specified to continue to build confidence in the model implementations (Step 1a). The results of Step 1a were compared to the data from the FE-experiment without the teams seeing the data. The teams were then provided with a sub-set of the data from the FE-experiment and invited to consider how best to use the large dataset for model comparison (Step 1b). Teams were then asked to use the data provided to calibrate their models, only changing material property values rather than adding features or processes to their models (Step 1c). In Step 1, teams were asked to only model the heating phase of the experiment, so pressure in the Opalinus Clay was reported as change in pressure since the initial conditions were specified rather than modelled. The change from 2D to 3D models was accompanied by an increase in the dispersion of results between the teams. Some of this was resolved during the task, but some remained and is potentially due to model discretisation. Calibration of parameters was useful in improving the fit of the models to the data but the remaining differences indicated that the models were missing features or processes. In Step 2, the teams were asked to update their models with additional features and processes as well as calibrating parameters to try and improve the fit of the models to the data. Teams were encouraged to represent ventilation of the open FE tunnel prior to backfilling with heaters and bentonite and in Step 2, the absolute pressure in the Opalinus Clay was compared between the teams. Teams took different approaches, but there was consideration of adding shotcrete and an EDZ into the model, representing stress change during excavation and different approaches to modelling ventilation of the FE tunnel. Overall, the documented results showed a very good agreement for temperature. The results for porewater pressure evolution showed a significant improvement for most teams compared to Step 1c with a good agreement to the measurements for several teams whereas some teams overpredicted the pressure increase and others overpredicted the drainage effect especially for the sensors close to the heater. Step 3 was an opportunity for teams to use the models developed in Step 1 and Step 2 to make predictions about the temperature and pressure changes that will be expected at the FE experiment over the next few years in light of the planned changes in thermal output of the heaters.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

The Collection 6 'dark-target' MODIS Aerosol Products

Aerosol retrieval algorithms are applied to Moderate resolution Imaging Spectroradiometer (MODIS) sensors on both Terra and Aqua, creating two streams of decade-plus aerosol information. Products of aerosol optical depth (AOD) and aerosol size are used for many applications, but the primary concern is that these global products are comprehensive and consistent enough for use in climate studies. One of our major customers is the international modeling comparison study known as AEROCOM, which relies on the MODIS data as a benchmark. In order to keep up with the needs of AEROCOM and other MODIS data users, while utilizing new science and tools, we have improved the algorithms and products. The code, and the associated products, will be known as Collection 6 (C6). While not a major overhaul from the previous Collection 5 (C5) version, there are enough changes that there are significant impacts to the products and their interpretation. In its entirety, the C6 algorithm is comprised of three sub-algorithms for retrieving aerosol properties over different surfaces: These include the dark-target DT algorithms to retrieve over (1) ocean and (2) vegetated-dark-soiled land, plus the (3) Deep Blue (DB) algorithm, originally developed to retrieve over desert-arid land. Focusing on the two DT algorithms, we have updated assumptions for central wavelengths, Rayleigh optical depths and gas (H2O, O3, CO2, etc.) absorption corrections, while relaxing the solar zenith angle limit (up to 84) to increase pole-ward coverage. For DT-land, we have updated the cloud mask to allow heavy smoke retrievals, fine-tuned the assignments for aerosol type as function of season location, corrected bugs in the Quality Assurance (QA) logic, and added diagnostic parameters such as topographic altitude. For DT-ocean, improvements include a revised cloud mask for thin-cirrus detection, inclusion of wind speed dependence in the retrieval, updates to logic of QA Confidence flag (QAC) assignment, and additions of important diagnostic information. At the same time as we have introduced algorithm changes, we have also accounted for upstream changes including: new instrument calibration, revised land-sea masking, and changed cloud masking. Upstream changes also impact the coverage and global statistics of the retrieved AOD. Although our responsibility is to the DT code and products, we have also added a product that merges DT and DB product over semi-arid land surfaces to provide a more gap-free dataset, primarily for visualization purposes. Preliminary validation shows that compared to surface-based sunphotometer data, the C6, Level 2 (along swath) DT-products compare at least as well as those from C5. C6 will include new diagnostic information about clouds in the aerosol field, including an aerosol cloud mask at 500 m resolution, and calculations of the distance to the nearest cloud from clear pixels. Finally, we have revised the strategy for aggregating and averaging the Level 2 (swath) data to become Level 3 (gridded) data. All together, the changes to the DT algorithms will result in reduced global AOD (by 0.02) over ocean and increased AOD (by 0.02) over land, along with changes in spatial coverage. Changes in calibration will have more impact to Terras time series, especially over land. This will result in a significant reduction in artificial differences in the Terra and Aqua datasets, and will stabilize the MODIS data as a target for AEROCOM studie

Aerosol retrieval algorithms↗

Pulsed Thermal Tomography Nondestructive Examination of Additively Manufactured Reactor Materials and Components (Final Technical Report)

Metal Additive Manufacturing (AM) is a promising method for cost-efficient fabrication of complex shape structures for applications in harsh environment, such as in a nuclear reactor. However, internal defects (pores) occur in high-strength AM alloys, which are manufactured with Laser Powder Bed Fusion (LPBF) AM method. Pulsed Infrared Thermography (PIT) is an efficient nondestructive evaluation (NDE) method to examine actual structures, because this method offers one-sided non-contact measurements, and fast processing of large sample areas. However, imaging of material defects, particularly defects with sizes at microscopic level, is challenging. In this report, we benchmark the performance of several Unsupervised Learning (UL) algorithms designed to enhance imaging of microscopic defects in metals with PIT. UL aims to learn the latent principal patterns (dictionaries) in PIT data to detect defects with minimal human supervision. Performance of Independent Component Analysis (ICA), Sparse Coding (SC), Principal Component Analysis (PCA) and Exploratory Factor Analysis (EFA) was compared using F-score, UL model training time and defects reconstruction time. We obtained the average F-score of 0.75, and a highest F-score of 0.89 for the EFA algorithm. Overall, EFA outperforms other UL algorithms considered in this study. In another approach, we investigate Thermal Tomography (TT), which is a computational method for reconstruction of depth profile of internal material defects from PIT nondestructive evaluation (NDE). TT algorithm obtains depth reconstructions of thermal effusivity, which has been shown to provide visualization of subsurface internals defects in metals. In many applications, one needs to determine the defect shape and orientation from reconstructed effusivity images. Interpretation of TT images is non-trivial because of blurring, which increases with depth due to heat diffusion-based nature of image formation. We have developed a deep learning convolutional neural network (CNN) to classify size and orientation of subsurface material defects in TT images. CNN was trained with TT images produced with computer simulations of 2D metallic structures (thin plates) containing elliptical subsurface voids. Performance of CNN was investigated using test TT images developed with computer simulations of plates containing elliptical defects, and defects with shape imported from scanning electron microscopy (SEM) images. CNN demonstrated the ability to classify radii and angular orientation of elliptical defects in previously unseen test TT images. We have also demonstrated that CNN trained on TT images of elliptical defects is capable of classifying shape and orientation of irregular defects. Training the CNN on irregular defect shapes instead of on elliptical shapes would make the resulting classifications more descriptive of actual defect shapes. However, this requires a much higher volume of SEM images of material defects, which are difficult to obtain because of random occurrence of defects in LPBF. To address this challenge, we developed a generative adversarial network (GAN) to augment the existing dataset of SEM defect images. The GAN model is demonstrated to create novel yet realistic defect shapes that can be used as input for simulated PTT images to train CNN. We also investigate several approaches based on Gaussian Random Circle and Bezier Curves for constructing parametric models of irregular-shape defects.

36 MATERIALS SCIENCE↗

Brain‐age prediction: Systematic evaluation of site effects, and sample age range and size

Abstract Structural neuroimaging data have been used to compute an estimate of the biological age of the brain (brain‐age) which has been associated with other biologically and behaviorally meaningful measures of brain development and aging. The ongoing research interest in brain‐age has highlighted the need for robust and publicly available brain‐age models pre‐trained on data from large samples of healthy individuals. To address this need we have previously released a developmental brain‐age model. Here we expand this work to develop, empirically validate, and disseminate a pre‐trained brain‐age model to cover most of the human lifespan. To achieve this, we selected the best‐performing model after systematically examining the impact of seven site harmonization strategies, age range, and sample size on brain‐age prediction in a discovery sample of brain morphometric measures from 35,683 healthy individuals (age range: 5–90 years; 53.59% female). The pre‐trained models were tested for cross‐dataset generalizability in an independent sample comprising 2101 healthy individuals (age range: 8–80 years; 55.35% female) and for longitudinal consistency in a further sample comprising 377 healthy individuals (age range: 9–25 years; 49.87% female). This empirical examination yielded the following findings: (1) the accuracy of age prediction from morphometry data was higher when no site harmonization was applied; (2) dividing the discovery sample into two age‐bins (5–40 and 40–90 years) provided a better balance between model accuracy and explained age variance than other alternatives; (3) model accuracy for brain‐age prediction plateaued at a sample size exceeding 1600 participants. These findings have been incorporated into CentileBrain ( https://centilebrain.org/#/brainAGE2 ), an open‐science, web‐based platform for individualized neuroimaging metrics.

60 APPLIED LIFE SCIENCES↗