Towards the True Hybrid: Physics-Informed Trainable Models for Prognostics and Health Management
Explore the source record for details and available documents.
Physics-Informed Machine Learning↗
SEARCH · Engineering Papers
Results for “Machine learning in physics”
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.
Development and Application of NASA SPoRT’s DustTracker-AI Model for Real-Time Identification and Tracking of Dust in Geostationary Satellite Imagery
The NASA Short-term Prediction Research and Transition (SPoRT) Center developed the DustTracker-AI model for identifying and tracking dust in NASA/NOAA Geostationary Operational Environmental Satellite (GOES) imagery in a real-time framework. A training dataset consisting of day and night dust cases was gathered over the southwestern consisting of 115 distinct images and over a million dust pixels and 256 million no dust pixels. The dataset was separated into training (60%), testing (20%), and validation (20%). A simple random forest machine learning model was developed originally to overcome the problem of night-time dust detection and has been expanded to a comprehensive day/night model for dust identification and tracking. This physically-based machine-learning approach uses NASA/NOAA GOES-16 Advanced Baseline Imager infrared imagery as inputs to the model. The model probability of dust output achieves an Area-Under-Curve (AUC) of 0.97 with a standard deviation of 0.04 for dust cases. For images with dust present, the model correctly labels 85% of dust pixels for all dust images in the validation data set. In conjunction with developing the machine-learning model, the NASA Short-term Prediction Research and Transition Center (SPoRT) partnered with NOAA National Weather Service forecast offices to evaluate the model for utility in weather forecasting operations during the 2021 and 2023 late winter-spring seasons. Preliminary evaluation has indicated the majority of forecasters described the DustTracker-AI probabilities as having added confidence to interpreting the Dust RGB and other satellite products to objectively assess the dust extent and trends and increased the amount of time the dust plume could be tracked into the night as compared to use of the Dust RGB. More recently, SPoRT tested small scale events associated with thunderstorm outflow and burn scars to determine the model’s ability to capture local events. This presentation highlights design of the model, validation/evaluation of model performance, and example cases collected during end user product assessments.
Connor H Welch↗
Opinion: Aerosol Remote Sensing Over the Next Twenty Years
More than two decades ago, aerosol remote sensing underwent a revolution with the launch of the Terra and Aqua satellites. Advancement continued via additional launches carrying new passive and active sensors. Capable of retrieving parameters characterizing aerosol loading, rudimentary particle properties and in some cases aerosol layer height, the satellite view of Earth’s aerosol system came into focus.The modeling communities have made similar advances. Now the efforts have continued long enough that we can see developing trends in both remote sensing and modeling communities, allowing us to speculate about the future and how the community will approach aerosol remote sensing twenty years from now. We anticipate technology that will replace today’s standard multi-wavelength radiometers with hyperspectral and/or polarimetry all viewing in multiple angles.These will be supported by advanced active sensors with the ability to measure profiles of aerosol extinction in addition to backscatter. The result will be greater insight into aerosol particle properties. Algorithms will move from being primarily physically-based to include an increasing degree of Machine Learning methods, but physically-based techniques will not go extinct. However, the practice of applying algorithms to a single sensor will be in decline. Retrieval algorithms will encompass multiple sensors and all available ground measurements into a unifying framework, and these inverted products will be ingested directly into assimilation systems, becoming “cyborgs”: half observations, half model. In twenty years we will see a true democratization in space with nations large and small, private organizations and commercial entities of all sizes launching space sensors. With this increasing amount of data and aerosol products available, there will be a lot of bad data. User communities will organize to set standards and the large national space agencies will lead the effort to maintain quality by deploying and maintaining validation ground networks and focused field experiments. Through it all, interest will remain high in the global aerosol system and how that system affects climate, clouds, precipitation and dynamics, air quality, the environment and public health, transport of pathogens and fertilization of ecosystems, and how these processes are adapting to a changing climate.
aerosol remote sensing↗
TPSAS-NF1676L-35322-DND
This talk discusses emerging methods that seek to fuse and integrate physics-based modeling with machine learning. With the recent rise of machine learning and artificial intelligence, there has been a huge surge in data-driven approaches to solve computational science and engineering problems. However, neglecting a priori knowledge of established physical laws and relying solely on data-driven methods can yield unreliable, less interpretable, and/or non-physical results, especially when data is sparse or predictions are required outside of the training data domain. This two part talk presents two distinct approaches for accelerating predictions with machine learning that are grounded and constrained by relevant physics and their application to problems at NASA.
Julian Cuevas Paniagua↗
Introduction to Analysis Methods for Big Earth Data
Big Earth Data are too big to be tractable to simple data inspection and require models to make sense of all the data. Useful models for Big Earth Data may be physical, statistical, or machine learning based. In many cases, hybrid models combine attributes of two or more of these types.
parallel processing (computers)↗
Applicability of Loads Estimation Techniques Using Sparse Acceleration Sensor Data to Spacecraft Structural Health Monitoring
The use of structural health monitoring systems on spacecraft structures can play a crucial role in ensuring the safety, reliability, and longevity of the structure by gathering and analyzing onboard sensor data. Of specific importance is monitoring for excessive loading at critical interfaces as any off-nominal structural excitations experienced by spacecraft structures can cause early unpredicted high structural life consumption or damage. The availability and cost of flight-certified sensors along with the size of spacecraft structures and allowable payload mass drives the need for a method to estimate loads using sparsely-located sensors. Numerous approaches such as physics-based, statistical learning, and physics-enhanced statistical learning algorithms have gained popularity among structural prognostics applications. However, developing noise-robust prediction models to assess loads and structural life predictions from a sparse multi-sensor data acquisition system can be a challenging task. This paper discusses the evaluation of physics-based versus machine-learning algorithms for predicting loads and structural life at mission critical locations on the spacecraft structure using a finite element loads analysis with the application of simulated noise and noise reduction techniques. To estimate the loads from accelerations, the physics-based algorithm leverages a loads transformation matrix from a Craig-Bampton reduced finite element model. A System Equivalent Reduction Expansion Process (SEREP) and a pseudo-inverse approach are considered to expand from the onboard sensor degrees of freedom to the Craig-Bampton model degrees of freedom. The machine learning algorithm provides a data driven solution/mapping of the sensor accelerations to the loads at the mission critical locations using a high dimensionality analysis. Although these strategies produce comparable loads prediction without noise, the limitations of these strategies with incorporating simulated noise and noise reduction techniques with low signal to noise ratio signals are evaluated. The study demonstrates the immense potential of statistical learning algorithms for sparse structural prognostic models and enhancing signal denoising techniques. These findings also highlight the need for noise-resilient prognostic models and low-noise data acquisition systems onboard spacecraft structures.
Spacecraft Structural Health Monitoring↗
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↗
Hybrid Approaches to 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.
Systems Health Management↗
Hybrid Model Based Approaches for Systems Health Management and Prognostics
This is a previously approved and published presentation. To accurately predict the future state of any system, it is required to possess knowledge of its current health state and future operational conditions. Present 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 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.
Prognostics↗
Use of TEMPO as a Proxy for Hyperspectral Geostationary Ocean Color Measurements from the GeoXO OCX Instrument: Harnessing Machine Learning and Principal Component Techniques for Atmospheric and Glint Correction
Retrievals of ocean color from space are important for better understanding the ocean ecosystem. The launch of atmospheric geostationary hyperspectral sensors such as TEMPO, provides a unique opportunity to examine the diurnal variability in ocean ecology. While TEMPO does not have as high spatial resolution or full spectral coverage as planned coastal ocean sensors such as the Geosynchronous Littoral Imaging and Monitoring Radiometer (GLIMR) or GeoXO Ocean Color instrument (OCX), its hourly measurements provide coverage of regions such as Lake Erie and the Gulf of Mexico at spatial scales of approximately 5 km. These data can be useful for testing new algorithms. We will apply our newly developed machine learning based atmospheric correction approach for ocean color retrievals to TEMPO data. Our approach begins by decomposing measured radiances from hyperspectral sensors into spectral features that describe the scattering and absorption of the atmosphere as well as the underlying surface reflectance. The coefficients of the principal components are then used to train a neural network to predict ocean color properties derived from collocated MODIS/VIIRS physically-based retrievals. This machine learning approach does not rely on radiative transfer modeling, and the use of MODIS/VIIRS data for training accounts for possible calibration b in hyperspectral data. Previously, we applied our approach using blue and UV wavelengths with the Ozone Monitoring Instrument (OMI) and TROPOspheric Monitoring Instrument (TROPOMI) to show that it can estimate ocean color properties in less-than-ideal conditions such as lightly to moderately clouded conditions as well as sun glint and thus improve the spatial coverage of ocean color measurements. TEMPO provides an opportunity to improve on this approach since it will provide collocated measurements at green and red wavelengths that were not available from OMI and TROPOMI and are important particularly for coastal waters. Additionally, our technique can be applied early in the mission and has potential to demonstrate the value of near real time ocean color products that are important for monitoring of harmful algae blooms and other oceanic phenomena.
Zachary Fasnacht↗
Materials Informatics at NASA GRC: Machine Learning Surrogate Modeling, Data Management, and Integrated Toolsets for Establishing/Maintaining the Digital Thread
Integrated Computational Materials Engineering (ICME) has recently received widespread attention due to its promises in reducing dependence on physical testing for engineering design by relying on simulation, reducing both time and cost to market for various applications. ICME however requires validated multiscale material models, which heavily depend on available test data with full material and test pedigree, including material processing, test and measurement equipment, raw data collection, and analysis methodology and results that is findable and usable, along with integrated, efficient toolsets for effectively passing information across various length and time scales across such models. At the NASA Glenn Research Center under the Transformational Tools and Technologies Project, significant recent efforts have been directed towards establishing the required cyberinfrastructure to enable optimized ICME processes and the design of “fit-for-purpose” materials to achieve the goals outlined in the NASA Vision 2040 report. Such efforts include development of multiscale physics-based material models, which can be used to train highly efficient surrogate machine learning models, development of best practices and infrastructure for effective, traceable materials information management, and development of toolsets that integrate with physics-based codes, machine learning models, and an information management system to enable high throughput of materials data collection and analysis, establishment of digital twins and the digital thread, and automation of the ICME design process for material optimization.
Machine Learning↗
Proceedings of the Workshop on Change of Representation and Problem Reformulation
The proceedings of the third Workshop on Change of representation and Problem Reformulation is presented. In contrast to the first two workshops, this workshop was focused on analytic or knowledge-based approaches, as opposed to statistical or empirical approaches called 'constructive induction'. The organizing committee believes that there is a potential for combining analytic and inductive approaches at a future date. However, it became apparent at the previous two workshops that the communities pursuing these different approaches are currently interested in largely non-overlapping issues. The constructive induction community has been holding its own workshops, principally in conjunction with the machine learning conference. While this workshop is more focused on analytic approaches, the organizing committee has made an effort to include more application domains. We have greatly expanded from the origins in the machine learning community. Participants in this workshop come from the full spectrum of AI application domains including planning, qualitative physics, software engineering, knowledge representation, and machine learning.
Lowry, Michael R.↗
Predicting Fiber Failure of Plain Weave Fabric with Recursive Multiscale Micromechanics
Recent advances in the development of machine learning (ML) algorithms have enabled the creation of predictive models that can improve decision making, decrease computational cost, and improve efficiency in a variety of fields. As an organization begins to develop and implement such models, the data used in the training, validation, and testing of machine learning models, the model parameters, and the use cases or limitations of the models must be properly stored to ensure models are both fully traceable and used correctly. In the context of predicting material behavior, advances in computationally intense, physics-based, modeling of material behavior at various length scales, and the emergence of Integrated Computational Materials Engineering (ICME) have driven the need for developing data-driven surrogate models of the physics-based simulation tools using machine learning (ML) techniques. Surrogate model development allows for accurate material behavior prediction at a fraction of the cost of its physics-based counterpart, allowing for multiscale simulations of real-world applications, further enabling the ability to design fit-for-purpose materials for a reasonable computational investment. However, training such models requires extensive data, and thus effective data management is necessary to reach the full potential that ML can offer to material design and ICME. This paper proposes a generalized, robust schema that allows organizations to store both real (experimental) and virtual (simulation) data used to train machine learning models and the defining model parameters and architectures. The developed schema allows for various types of data inputs and outputs, including single point values, time-series data, and images that can be used in for various types of machine learning models while following outlined best practices for effective data management. An effective schema for machine learning data and models can help prevent the recreation of virtual/real training data and surrogate models, can help reduce the time to create new models similar to existing ones by offering a starting point in the hyperparameter determination stages, minimize resources devoted to verification and validation (V&V) and certification of models, and ensure that data and surrogate models are not misused due to full traceability of both the data and ML model. It also allows organizations access to models that have already been developed, such that they can be used in the design of new materials, enabling the overall goals of ICME.
Failure↗
An Extensible Perturbed Parameter Ensemble for the Community Atmosphere Model Version 6
This paper documents the methodology and preliminary results from a Perturbed Parameter Ensemble (PPE) technique, where multiple parameters are varied simultaneously and the parameter values are determined with Latin hypercube sampling. This is done with the Community Atmosphere Model version 6 (CAM6), the atmospheric component of the Community Earth System Model version 2 (CESM2). We apply the PPE method to CESM2-CAM6 to understand climate sensitivity to atmospheric physics parameters. The initial simulations vary 45 parameters in the microphysics, convection, turbulence and aerosol schemes with 263 ensemble members. These atmospheric parameters are typically the most uncertain in many climate models. Control simulations are analyzed and targeted simulations to understand climate forcing due to aerosols and fast climate feedbacks. The use of various emulators is explored in the multi- dimensional space mapping input parameters to output metrics. Parameter impacts on various model outputs, such as radiation, cloud and aerosol properties are evaluated. Machine learning is also used to probe optimal parameter values against observations. Our findings show that using PPE is a valuable tool for climate uncertainty analysis. Furthermore, by varying many parameters simultaneously, we find that many different combinations of parameter values can produce results consistent with observations, and thus careful analysis of tuning is important. The CESM2-CAM6 PPE is publicly available, and extensible to other configurations to address questions of other model processes in the atmosphere and other model components (e.g. coupling to the land surface).
Machine learning↗
Using Machine-Learning Methods and Expert Prediction Probabilities to Forecast Solar Flares
It has long been known that studying connection between solar flares and properties of magnetic field in active regions is very important for understanding the flare physics and developing space weather forecasts. The Helioseismic and Magnetic Imager onboard the Solar Dynamics Observatory (SDO/HMI) obtains tremendous amounts of magnetic field data products. However the operational NOAA Space Weather Prediction Center (SWPC) forecasts of solar flares still represent prediction probabilities issued by the experts. In this research we investigate the possibilities to enhance the daily operational flare forecasts performed at the SWPC by developing a synergy of the expert predictions and physics-based criteria, and by employing machine-learning methods. Among the physics-based criteria we consider the descriptors of the Polarity Inversion Line (PIL) and Space weather HMI Active Region Patches (SHARP), and derive from them daily characteristics of the entire Sun. We also consider the daily descriptors of the GOES Soft X-Ray (SXR) 1-8 Angstroms flux such as the flare history of the previous days and averaged X-Ray flux. We estimate the effectiveness in separation of flaring and non-flaring cases for each characteristic, as well as for the expert prediction probabilities, and find that some PIL, SHARP and SXR descriptors are as effective as the expert prediction probabilities and should be considered to issue the flare forecast. Finally, we train and test several Machine-Learning classification algorithms (Support Vector Classifiers with various kernel functions, k-Nearest Neighbor Classifier, Random Forest Classifier, and Neural Networks) using the most effective descriptors and expert prediction probabilities, and compare the obtained predictions with the current SWPC forecasts.
Machine-Learning↗
Parametric Mechanism Design Through Numerical Optimization and Physics Simulation
Design-Build-Test approaches for developing spaceflight hardware are prohibitively time and cost intensive and often lead to suboptimal mechanism designs. Approaches that couple machine learning and high-fidelity physics simulation could eliminate the need for hardware prototyping and dramatically accelerate the engineering design cycle, ultimately reducing cost. This work presents a modular NASA-developed toolchain to optimize hardware mechanisms in a virtual environment using numerical optimization and multi-body physics simulation. The toolchain enables multi-objective optimization, generates parametric CAD files that can be further post-processed by an end user, and can be expanded to optimize full systems and non-mechanical parameters such as feedback control variables. We demonstrate the toolchain through an independently verifiable design problem that optimizes wheel radius to achieve a desired linear velocity in a rigid-body physics environment when the wheel rotates at a constant angular speed, and then post-process the parametric CAD file of the optimal design generated by the tool before ultimately manufacturing it via 3D printing. We end with a discussion of how the toolchain can incorporate other analysis tools, including finite element analysis, computational fluid dynamics, and granular media simulations.
Optimization↗
Towards the True Hybrid: Physics-Informed Trainable Models for Prognostics and Health Management
Explore the source record for details and available documents.
physics-informed machine learning↗
Identification of Flux Rope Orientation via Neural Networks
Geomagnetic disturbance forecasting is based on the identification of solar wind structures and accurate determination of their magnetic field orientation. For nowcasting activities, this is currently a tedious and manual process. Focusing on the main driver of geomagnetic disturbances, the twisted internal magnetic field of interplanetary coronal mass ejections (ICMEs), we explore a convolutional neural network’s (CNN) ability to predict the embedded magnetic flux rope’s orientation once it has been identified from in situ solar wind observations. Our work uses CNNs trained with magnetic field vectors from analytical flux rope data. The simulated flux ropes span many possible spacecraft trajectories and flux rope orientations. We train CNNs first with full duration flux ropes and then again with partial duration flux ropes. The former provides us with a baseline of how well CNNs can predict flux rope orientation while the latter provides insights into real-time forecasting by exploring how accuracy is affected by percentage of flux rope observed. The process of casting the physics problem as a machine learning problem is discussed as well as the impacts of different factors on prediction accuracy such as flux rope fluctuations and different neural network topologies. Finally, results from evaluating the trained network against observed ICMEs from Wind during 1995–2015 are presented.
Thomas Narock↗