Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “deep learning methods”

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.

At least 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
NASA NTRSDate not supplied

Communicating Metrics of Land Surface Temperature Variability Using Multi-sensor Machine Learning

Land surface temperature (LST) is a key climate observable used to detect changes in the Earth’s surface energy budget that influence carbon and water cycles. Land surface temperature exhibits strong diurnal variability, which geostationary satellites can observe at scale thanks to their temporal resolution. Due to anthropogenic climate and land use changes, the surface energy balance has been considerably modified and may be described by changes in diurnal temperature range and extremes. Using high performance computing and datasets from the NASA Earth Exchange, we exploit co-located, co-temporal observations from low-earth orbit (LEO) and geostationary (GEO) sensors to develop a deep learning-based method for LEO-to-GEO algorithm emulation. Our model is trained to predict MODIS Terra LST from GOES-16 thermal bands and achieves validation error <2K. Application of the model to unseen times of day (observed by MODIS Aqua) and a new GEO sensor (Himawari-8) observing an unseen spatial domain, demonstrate the generalization of the deep learning model across space, time and spectra. Further, time series clustering approaches are examined with the objective of identifying key indicators of change in diurnal cycling and extremes on a continental scale. Communicating LST variability observed by geostationary satellites can have impacts in multiple disciplines, from understanding of snow, vegetation and soil dynamics, to recognizing trends in heat events relevant to human health.

Kate Duffy↗
NASA NTRSDate not supplied

LEO Sensor to GEO Sensor Algorithm Transfer Models for Land Surface Temperature

Land surface temperature (LST) is a key climate observable used to detect changes in the Earth’s surface energy budget that influence carbon and water cycles. Land surface temperature exhibits strong diurnal variability, which geostationary satellites can observe at scale thanks to their temporal resolution. Due to anthropogenic climate and land use changes, the surface energy balance has been considerably modified and may be described by changes in diurnal temperature range and extremes. Using high performance computing and datasets from the NASA Earth Exchange, we exploit co-located, co-temporal observations from low-earth orbit (LEO) and geostationary (GEO) sensors to develop a deep learning-based method for LEO-to-GEO algorithm emulation. Our model is trained to predict MODIS Terra LST from GOES-16 thermal bands and achieves validation error <2K. Application of the model to unseen times of day (observed by MODIS Aqua) and a new GEO sensor (Himawari-8) observing an unseen spatial domain, demonstrate the generalization of the deep learning model across space, time and spectra. Communicating diurnal LST variability observed by geostationary satellites can have impacts in multiple disciplines, from understanding of snow, vegetation and soil dynamics, to recognizing trends in heat events relevant to human health.

Kate Marie Duffy↗
NASA NTRSDate not supplied

Parameterization of Vertical Cloud Distribution from C3M and MERRA Data Using ML Method

Clouds play a key role in regulating the hydrological cycle and the Earth's radiative energy budget. However, global climate models (GCMs) with a horizontal grid spacing on the order of 100 km have limitations in representing sub-grid cloud dynamics with spatial scales on the order of 1 km, leading to potential uncertainties in cloud radiative feedback on the global scale. In our research, we will leverage the capabilities of Deep Machine Learning (DML) methods to construct parameterizations of sub-grid volumetric cloud fraction (VCF), which is the frequency of occurrence on a grid volume accumulated in the horizontal and vertical directions. Our investigation delves into the intricate relationship between VCF obtained from the NASA CALIPSO-CloudSat-CERES-MODIS (CCCM) satellite observation data and 3-D MERRA-2 reanalysis meteorological profiling data (e.g., wind, relative humidity, temperature). Through a comprehensive one-year data training utilizing the Sequence to Sequence DML method, we have successfully disentangled the complicated cloud formation dynamics across diverse meteorological conditions through a day-to-day analysis framework. Preliminary findings reveal promising statistical agreements in geographical and vertical distributions and seasonal variations of volumetric cloud fraction between ML prediction and satellite measurements. These results underscore the aptitude of our DML model to discern underlying cloud physical processes and accurately represent sub-grid cloud formation dynamics. Additionally, we have also employed trained neural network to analyze uncertainties arising from errors in meteorological data, further enhancing the robustness of our VCF parameterization.

Shan Zeng↗
NASA NTRS2024

AI Foundation Model for Heliophysics: Applications, Design, and Implementation

Deep learning-based methods have been widely researched in the areas of language and vision, demonstrating their capacity to understand long sequences of data and their usefulness in numerous helio-physics applications. Foundation models (FMs), which are pre-trained on a large-scale datasets, form the basis fora variety of downstream tasks. These models, especially those based on trans-formers in vision and language, show exceptional potential for adapting to a wide range of downstream applications. In this paper, we provide our perspective on the criteria for designing a FM for heliophysic and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset. We believe that this is the first study to design a foundation model in the domain of heliophysics.

Sujit Roy↗
NASA NTRS2022

Bayesian Deep Learning for Segmentation for Autonomous Safe Planetary Landing

Hazard detection is critical for enabling autonomous landing on planetary surfaces. Current state-of-the-art methods leverage traditional computer vision approaches to automate the identification of safe terrain from input digital elevation models (DEMs). However, performance for these methods can degrade for input DEMs with increased sensor noise. In the last decade, deep learning techniques have been developed for various applications. Nevertheless, their applicability to safety-critical space missions has often been limited due to concerns regarding their outputs’ reliability. In response to these limitations, this paper proposes an application of the Bayesian deep learning segmentation method for hazard detection. The developed approach enables reliable, safe landing site detection by i) generating simultaneously a safety prediction map and its uncertainty map via Bayesian deep learning and semantic segmentation, and ii) using the uncertainty map to filter out the uncertain pixels in the prediction map so that the safe site identification is performed only based on the certain pixels (i.e., pixels for which the model is certain about its safety prediction). Experiments are presented with simulated data based on a Mars HiRISE digital terrain model by varying uncertainty threshold and noise levels to demonstrate the performance of the proposed approach.

Kento Tomita↗
NASA NTRS2021

Automated Data Accountability for Missions in Mars Rover Data

As the Mars Curiosity Rover transmits data to the JPL Ground Data System (GDS), it frequently observes data loss and corruption, requiring re-transmits from the rover and Ground Data System Analysts (GDSA) to monitor the downlink process. As new missions are launched, the GDSA team redistributes analysts to these new missions, causing shortages in previous missions. The GDSA team can significantly benefit from the automation and optimization of the downlink process of telemetry data. In fact, there is a need for a better understanding of why the data is corrupted, so that the GDSA team can best determine the root cause of the issues in the GDS. This paper presents machine learning and deep learning based approaches to automate and optimize the detection of data loss. We first created a pipeline to automatically accumulate data from the telemetry databases (MAROS, Telemetry Data Storage, and GDS Elastic Search Database) in the downlink process. With our newly created datasets, we perform feature selection to supplement the GDSA understanding of the downlink process and provide supplemental analysis on the importance of different features. We implement various machine learning and deep learning based models, including support vector machines, ensemble methods, and deep neural networks and evaluate their accuracies in identifying whether a downlink process is complete or incomplete. We utilize fast hyperparameter optimization methods that allow our models to quickly be re-trained, allowing them to quickly be tuned and optimized on daily incoming data in real time. This hyperparameter optimization also allows our methods to be quickly integrated into other JPL missions. Our results show that our best-performing machine learning and deep learning based models outperform the existing GDSA detection software by 6 accuracy points and can aid analysts by providing insights into the data accountability problem. Since these various machine learning and deep learning approaches vary significantly in interpretability, we provide a discussion on the tradeoffs between their performance and trustworthiness in helping detect issues in data transmission.

Divsalar, Dariush↗
NASA NTRSDate not supplied

The Application of Artificial Intelligence Deep Learning to Visually Identify Micrometeoroid and Orbital Debris Impacts

Recent advances in Artificial Intelligence (AI) are changing the World. Novel approaches to training AI systems have led to dramatic reductions in the amount of time required. Training an AI system could take years and teams of people using traditional methods, but with the advancements of Deep Learning (DL) models this training can now be accomplished by an individual in a matter of minutes. The development of “fast AI” libraries has delivered AI to essentially everyone. Democratization of AI power has inspired many to revisit past problems that will benefit from DL approaches. For example, the application of AI has improved detection of breast cancer by 20% compared to traditional detection methods. Computer vision and machine learning are being used to identify soil deficiencies and provide planting recommendations to farmers. Success stories like these and many others have provided inspiration to see if AI can help improve one of our needed capabilities – that of visually identifying micrometeoroid and orbital debris (MMOD) impact damage to spacecraft from images of the spacecraft exterior. The need to visually locate and characterize spacecraft MMOD impact damage has been present since the early days of space travel. This is often done by either having a crew member take photographs of the spacecraft through a window using a hand-held camera or ground personnel directing externally-mounted cameras. The photographs are then transmitted back to Earth for visual analysis. This method of MMOD damage inspection works well and has been used on various spacecraft including the Space Shuttle and the International Space Station (ISS). One of the issues with the current method that we believe AI could improve is the speed and possibly the accuracy in identifying MMOD impacts. Note that detecting MMOD impacts in images can be very difficult. The visual appearance of an MMOD impact can change dramatically with lighting conditions, size of impact, depth of penetration, material types, surface waviness, fabric coverings, camera & lens, distance to surface, spacecraft orientation, analyst experience, and many other factors. Currently, this takes a team of highly-experienced specialists in both the fields of Image Analysis and MMOD impacts. This paper documents our initial research in training an AI DL model using the fast-AI library to identify actual and simulated MMOD impacts and perforations into exposed flat surfaces. While we recognize that this initial goal seems modest, it must be noted that what we have done would have taken teams of individuals and years of training just ten years ago. Our long-term goal is to add complexity and use-cases to the DL model being trained to expand the capabilities of this model so that it can be used to identify MMOD impacts on all types of spacecraft surfaces.

Cameron M Collins↗
NASA NTRS2020

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↗
NASA NTRS2020

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↗
NASA NTRS2024

Celebrating 10 Years of the Sub-Seasonal to Seasonal Prediction Project and Looking to the Future

The conference clearly demonstrated the increasing interest and growth of the scientific community working on the development and application of sub-seasonal to seasonal prediction since the start of the World Weather Research Programme (WWRP)/World Climate Research Programme (WCRP) sub-seasonal to seasonal (S2S) prediction project in 2013. The conference, which was held at the University of Reading (United Kingdom), was organized into three main themes as briefly summarized below, with eleven invited talks, 74 oral contributed talks, and 101 posters. The conference also included a two-hour breakout session, wherein eight groups discussed the current state and prospect for S2S prediction, and an early career researcher event. A summary of these discussions and recommendations is presented below. The conference web page (https://research.reading.ac.uk/s2s-summit2023/) is archived at the University of Reading. Introductory comments by representatives of the World Meteorological Organization (WMO) WWRP and WCRP emphasized the importance of the weather–climate linkage, targeted by S2S forecasts (from 2 weeks to a season ahead), addressing the challenges of creating “end-to-end” forecasts that encompass the entire climate-services chain from the prediction science and forecast, to the development and issuing of forecast products tailored to informing user-decisions. They also emphasized the efficacy of multi-model ensemble efforts and databases to foster collaborations internationally and between operational centres and academia. Although the WWRP/WCRP S2S project comes to an end in 2023, S2S prediction will remain an important focus for WWRP and WCRP. In WWRP, a new project called SAGE (Sub-seasonal to seasonal predictions for Agriculture and Environment) will start in 2024. Another important legacy of the S2S project will be the maintenance of the S2S database (Vitart et al. 2017) and the establishment of a WMO Lead Center for sub-seasonal prediction multi-model ensemble (LC-SSPMME) which will provide real-time multi-model S2S climate information. In two keynote presentations, Prof. Brian Hoskins (University of Reading) and Dr. Gilbert Brunet (Australian Bureau of Meteorology) discussed the potential of S2S predictability and the ongoing journey for understanding and improving these predictions. This conference was a sequel to the International Conference on Sub-seasonal to Seasonal Prediction (Robertson et al., 2014) which took place in College Park (Maryland, USA) in February 2014 to celebrate the start of the WWRP/WCRP S2S project, and to WCRP and WWRP conferences in Boulder, USA, in 2018 (Merryfield et al., 2020). A significant development compared to the previous S2S conferences was the large number of presentations on research to operation (R2O) and S2S applications and on the use of artificial intelligence and machine learning (AI/ML) methods for S2S prediction. Some of these methods provide empirical S2S forecasts which are competitive with state-of-the-art dynamical models. Other presentations demonstrated that AI/ML can provide alternative calibration of dynamical model outputs to traditional methods. Several talks and posters highlighted the increasing use of AI/ML, including deep learning, in S2S forecast post-processing and using AI to identify higher flow-dependent skill. Finally, some presentations demonstrated the value of AI/ML methods for a better understanding of S2S sources of predictability and attribution of extreme events.

S. J. Woolnough↗
NASA NTRS2017

Advances in Medical Analytics Solutions for Autonomous Medical Operations on Long-Duration Missions

A review will be presented on the progress made under STMDGame Changing Development Program Funding towards the development of a Medical Decision Support System for augmenting crew capabilities during long-duration missions, such as Mars Transit. To create an MDSS, initial work requires acquiring images and developing models that analyze and assess the features in such medical biosensor images that support medical assessment of pathologies. For FY17, the project has focused on ultrasound images towards cardiac pathologies: namely, evaluation and assessment of pericardial effusion identification and discrimination from related pneumothorax and even bladder-induced infections that cause inflammation around the heart. This identification is substantially changed due to uncertainty due to conditions of fluid behavior under space-microgravity. This talk will present and discuss the work-to-date in this Project, recognizing conditions under which various machine learning technologies, deep-learning via convolutional neural nets, and statistical learning methods for feature identification and classification can be employed and conditioned to graphical format in preparation for attachment to an inference engine that eventually creates decision support recommendations to remote crew in a triage setting.

Medical Decision Support Systems↗
NASA NTRS2018

Future of Big Earth Data Analytics

The state of the art of Big Earth Data Analytics can be expected to evolve rapidly in the coming years. The forces driving evolution come from both growth in the data and advancement in the field of data analytics. In the data area, advances in sensor instrumentation and platform miniaturization are increasing both data resolution and coverage, resulting in enormous growth in data Volume. Increases in temporal resolution in particular also generate demands for higher data Velocity. At the same time, the proliferation of instruments and the platforms on which they reside is increasing the Variety of datasets. The Variety increase in turn leads to questions about the Veracity of the data. In the algorithm area, powerful machine learning methods are coming to the fore, particularly Deep Neural Networks. These are powerful at detecting interesting features in the data, integrating many different measurements (i.e., data fusion), and classification problems. However, they are still challenging when seeking explanations of how natural or socio-economic phenomena work using Earth Observations. Thus, classical analysis techniques will remain relevant when the emphasis is on forming or testing explanations, as well as to support interactive data exploration.

Lynnes, Christopher↗
NASA NTRSDate not supplied

TPSAS-NF1676L-35370-DND

The above anvil cirrus plume (AACP) is a weather phenomenon that signifies an intense tropopause penetrating updraft which can inject cirrus clouds several kilometers into the stratosphere. Storms that have such intense updrafts are often supercells which generate severe weather such as tornadoes, high winds, and hail. In addition, AACPs moisten the stratosphere and influence the Earth's radiative balance. Though an AACP can be identified by the human eye, no automated AACP detection methods currently exist. Lack of detection inhibits understanding of where and how often AACPs occur, and how these storms influence stratospheric air composition. Previous work involved synthesis of multiple remote sensing and severe weather report/warning data sources to identify AACPs in Geostationary Operational Environmental Satellite system (GOES) satellite imagery and better understand their weather impacts (Bedka et al. (Wea. Forecasting, 2018)). This current study demonstrates an automated AACP identification method based on the application of a deep learning segmentation model known as a U-net. This study documents the development of a U-net model capable of identifying emergent AACPs using only satellite infrared (IR) and visible reflected sunlight imagery. The performance of a U-net is quantitatively benchmarked with human AACP identifications and qualitatively assessed through animations of detections generated from GOES-16 1-minute temporal resolution imagery.

Charles Liles↗
NASA NTRS2019

Machine Learning for the Zwicky Transient Facility

The Zwicky Transient Facility is a large optical survey in multiple filters producing hundreds of thousands of transient alerts per night. We describe here various machine learning (ML) implementations and plans to make the maximal use of the large data set by taking advantage of the temporal nature of the data, and further combining it with other data sets. We start with the initial steps of separating bogus candidates from real ones, separating stars and galaxies, and go on to the classification of real objects into various classes. Besides the usual methods (e.g., based on features extracted from light curves) we also describe early plans for alternate methods including the use of domain adaptation, and deep learning. In a similar fashion we describe efforts to detect fast moving asteroids. We also describe the use of the Zooniverse platform for helping with classifications through the creation of training samples, and active learning. Finally we mention the synergistic aspects of ZTF and LSST from the ML perspective.

Ashish Mahabal↗
NASA NTRSDate not supplied

Hybrid Model Based Approaches for Systems Health Management and Prognostics

To facilitate and solve the prediction problem, awareness of the current health state of the system is key, since it is necessary to perform condition-based predictions. To accurately predict the future state of any system, it is required to possess knowledge of its current health state and future operational conditions. Latest achievements of data-driven algorithms in regression of complex nonlinear functions and classification tasks have generated a growing interest in artificial intelligence for industrial applications. Complex multi-physics models as well as digital twins, once purely built on physics and corresponding simplified lumped parameter iterations, can now benefit from machine learning algorithms to mitigate the lack of understanding of some complex behavior. Given models of the current and future system behavior, a general approach of model-based prognostics can solve the prediction problem and further decision making. In principle, data driven approaches can replace expensive experimental test-setups as well as reduce the number of simulations needed to explore, e.g., the parametric space of a multi-parameter model. Nonetheless, the limitations of pure data-driven methods came to light rather quickly, at least for some industries. In many industrial applications, data acquisition is costly, and the volume of data that can be collected does not satisfy the requirements for an effective model training and cross-validation. Therefore, some recent works in the area of machine learning is focusing on blending physics with data-driven algorithms, thus mitigating the drawbacks of the two approaches and emphasizing respective advantages. Partial physical knowledge of the problem can aid the learning process by “guiding” the algorithm towards efficient solutions that satisfy the physics driving the system behavior. The result is a hybrid modeling approach combining physical knowledge as well data driven methods to develop a unified hybrid approach. A hybrid framework for fusing information from physics-based performance models along with deep learning algorithms for prognostics of complex safety critical systems is presented. In this framework, physics-based performance models infer unobservable model parameters related to the system's components health solving a calibration problem in the deep learning approach.

Hybrid Modeling↗
NASA NTRSDate not supplied

Machine-Learned Committor Functions for Reactive Molecular Dynamics

Reactive molecular dynamics (MD) is a powerful tool for atomistic-scale modeling of a diverse range of chemical processes. However, scaling these simulations to large systems and long times scales remains a challenge because of the complexity of the potential energy function required. The authors previously developed a heuristic approach, called REACTER, that incorporates reactivity in MD simulations in a less general but much more computationally efficient manner. REACTER uses standard, fixed valence force fields as the underlying potentialenergy surface for describing all interatomic interactions but adds a procedure for enforcing user-defined reactions that occur when certain geometric constraints on relative atomic positions are satisfied. Further, these bonding changes can be accepted or rejected with a probability related tothe local thermal energy. This work seeks to generalize this approach by replacing the set of user defined geometric constraints and energetic criteria with a committor function that specifies the probability of a reaction occurring on the basis of the local atomic configuration. The committor function is a useful mathematical tool for modeling rare events but, unfortunately, is very difficult to compute for realistic systems in a general way. This work describes a method for approximating the committor function using a machine learning approach, specifically a deep neural network trained with data from reactive MD and DFT-based dynamics simulations. This network is coupled to the existing REACTER protocol, as implemented in the LAMMPS MD package, and used to make on-the-fly predictions of reaction probabilities without the more extensive user input previously required. The new method is demonstrated using the polymerization of polystyrene as a case study. Although very dependent on the quality and quantity of training data, machine-learned committor functions show promise as a method for incorporating reaction probability from higher level calculations into highly scalable MD simulations.

polymer simulations↗
NASA NTRSDate not supplied

Enhancing Neural Network Explainability with Variational Autoencoders

Machine intelligence has been used to tackle increasingly complex problems and deep learning solutions are at the forefront of tackling these problems. In general, these architectures have a great number of parameters that are methodically updated in training. The vast number and complexity of deep neural networks makes it very difficult to decipher the inner workings of the neurons and layers that make up the network. This paper posits that trustworthiness and trust in autonomous systems are increased through eXplainable Artificial Intelligence (XAI) and presents a method that enhances the explainability and understanding of a neural network decision. We leverage variational autoencoders to produce human interpretable features from complex data sets. We show that the explainable features can then be used for machine learning applications. Explainability inspires trust in autonomous systems that use deep learning, which is necessary for safety critical systems.

Loc Tran↗