SEARCH · Engineering Papers

Results for “predictive”

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.

SourceResearch connectionFrom yearThrough year

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

At least 1,171 records · Page 65

All sources NASA NTRS DOE OSTI

NASA NTRSDate not supplied

NASA/ONERA Collaboration on Small Hovering Rotor Broadband Noise Prediction using Lattice-Boltzmann and Structured Navier-Stokes Solvers

This work compares two lattice-Boltzmann method solvers, PowerFLOW and ProLB, and two structured Navier-Stokes solvers, OVERFLOW2 and FAST, used by NASA and ONERA, respectively, for the broadband noise prediction of an ideally twisted rotor as part of Implementing Arrangement number FR-0685-0, ‘Comparing Computational Fluid Dynamics Solvers for Broadband Noise Prediction.’ Predicted results are evaluated against measured data from both smooth and rough blade sets acquired in the Small Hover Anechoic Chamber at the NASA Langley Research Center. Aerodynamic thrust predictions are seen to agree more favorably with the rough-blade measurements, whereas torque predictions agree better with the smooth-blade measurements. A tonal noise comparison shows better agreement to the measured data with the two structured Navier-Stokes solvers than with the two lattice-Boltzmann solvers, which is thought to be caused by the different geometric discretization associated with each solver paradigm. Broadband noise comparisons show that both lattice-Boltzmann method solvers trend well with the smooth-blade measurements, with the exception of an overprediction by ProLB between 4 kHz and 15 kHz. OVERFLOW2 is seen to capture the measured nondeterministic tonal content between 3 kHz and 8 kHz on a narrowband spectral basis and FAST agrees well with the rough blades on a one-third octave band basis.

Christopher S. Thurman↗

NASA NTRSDate not supplied

Variation in Predicted Orbital Lifetime Due to Launch Year

We present trends in predicted orbital lifetimes of CubeSats based not only on orbital parameters, but also launch year and the Area-to-Mass (AtM) ratio of the CubeSat. Determining the orbital lifetime variation of CubeSats in low-Earth orbit (LEO) is an important aspect of mission planning because of two competing factors: (1) the maximum orbital lifetime for orbital debris mitigation requirements, and (2) the minimum orbit duration necessary to accomplish the spacecraft mission requirements. The orbital lifetime is a function of orbital parameters, the AtM ratio of the CubeSat, and date of orbit insertion. Solar flux varies with time, peaking and declining across the 11-year solar cycle and affecting the amount of atmospheric drag on the CubeSat. This results in large variations in orbital lifetime dependent on the mission's launch date. We calculated the variation of orbital lifetime for multiple commonly used 1U to 6U CubeSat mission types across the two upcoming solar cycles. For any given AtM ratio and altitude combination in this analysis, the predicted orbital lifetime varies up to a factor of five due to orbit insertion occurring in a different year. Some examples of orbital lifetime spreads are 5 months to 2 years, and 1.5 to 7.5 years. While orbital lifetimes correlate to the solar cycle, the phasing of the maximum values varies based on a given AtM ratio and altitude combination. Different combinations of these two factors will result in the maximum predicted orbital lifetime occurring at different launch years throughout the solar cycle. Therefore, there is not a specific year within a solar cycle which can be used to calculate the maximum predicted orbital lifetime for all CubeSats. Since CubeSats are typically flown as a rideshare payload on a launch vehicle, mission planners must account for launch date variation in their orbital lifetime predictions. We recommend calculating orbital lifetime for a range of dates to allow for risk planning due to launch date slips, and other mission planning best practices.

CubeSat↗

NASA NTRSDate not supplied

Using Long-Short Term Memory Models to Predict Solar Active Regions Emergence

We train Long Short-Term Memory (LSTM) models that predict the formation of active regions (ARs), the main source of eruptive solar activity. Using the Doppler shift velocity, the continuum intensity and the magnetic field full-disk maps from SDO/HMI we have created time-series datasets of acoustic power and magnetic flux which are used to train Long Short-Term Memory (LSTM) models on predicting decreases in continuum intensity 12 hours in advance. Testing of the models' performance was done on data from 5 ARs, unseen from the model during training. The model predicted the emergence of AR11726, AR13165 and AR13179, 10, 29 and 5 hours in advance, and variations of this model achieved average RMSE values of 0.11 for both active and quiet parts of the solar disc, showing the ability of the model to capture acoustic power anomalies and predict continuum intensity variations. This work sets the foundations for the very first ML-aided prediction of solar ARs.

SMD↗

NASA NTRSDate not supplied

EVA Planning: Using Neutral Buoyancy Laboratory (NBL) Training to Predict in-Flight Energy Expenditure

Metabolic rate (“met rate”) is the amount of energy expended over a period of time and is influenced by many factors including body composition, level of physical activity, resting metabolic rate, sex, age, and food intake. Met rate is measured during Extravehicular Activity (EVA) training at the Neutral Buoyancy Laboratory (NBL) and during in-flight EVAs through indirect calorimetry, calculating energy expenditure from respiratory measurements of O 2 consumption and/or CO 2 production. During Extravehicular Activity (EVA) planning, metabolic cost is important to consider and is used to inform EVA duration based on spacesuit consumables associated with life support systems. Currently, NBL and previous ISS EVA met rate data for specified crewmembers are utilized to predict in-flight EVA metabolic costs based on a proposed EVA timeline. Timeline data collected during training is used to relate met rates to specific EVA activities, which are in turn assigned to more generalized EVA task categories, categorizing by both task type and restraint type. EVA task categories include EVA Setup/Cleanup, Worksite Setup/Cleanup, Cable Routing, Bolts, Fluid Connectors, Electrical Connectors, R&R Work, Miscellaneous Work, Incapacitated Crew Rescue (being rescued or performing), Assisted Crew Rescue (being assisted or performing), and Translation. Restraint types consist of Free-Float, Body Restraint Tether (BRT), Articulating Portable Foot Restraint (APFR), and Space Station Remote Manipulator System (SSRMS). From a crewmember’s historical data, individualized 10th, 50th and 90th percentile met rate estimates are generated for each task category and used to estimate the proposed EVA timeline metabolic cost. In-flight metabolic data (“As-Executed”) from recent ISS US EVAs 85-88 (totaling eight EVA crewmember met rates) was compared with their predicted metabolic cost (“As-Planned”) to evaluate the accuracy of the current met rate estimation method. Across the four EVAs, As-Executed Cumulative EVA Total Metabolic Cost (M = 5893.78 BTU, SD = 816.60) was not significantly different compared to As-Planned Cumulative EVA Total Metabolic Cost (M = 6106.60 BTU, SD = 826.62; t(7) = 0.751 , p = .477). Though not a significant difference, generally, As-Planned total estimates were slightly higher than As-Executed total metabolic cost. Relative Error for Cumulative EVA Total Metabolic Cost ranged from -33% to 14.7%, depending on the crewmember and EVA. When comparing As-Planned to A-Executed EVA task categories for Bolts, Electrical Connectors, EVA Cleanup, EVA Setup, Miscellaneous Work, Translation, Worksite Cleanup, and Worksite Setup during these EVAs, no significant differences were observed, however, there was a significant difference in As-Planned (M = 783.71 BTU, SD = 408.88) compared to As-Executed (M = 608.87 BTU, SD = 425.29) metabolic cost for the task category of Repair-and-Replace (R&R) Work (t(17) = 3.21 , p = .005). Looking closer within the R&R Work task category, As-Executed R&R Work with Free-Float restraint type (M = 706.17 BTU, SD = 352.18) was significantly less than As-Planned R&R Work with Free-Float restraint type (M= 887.96 BTU, SD = 381.42; t(12) = 2.59, p < .024). As-Executed R&R Work with SSRMS Restraint type (M = 355.89 BTU, SD = 534.65) was not significantly different from As-Planned values (M = 512.69 BTU, SD = 383.34; t(4) = 1.88, p = 0.132). These findings suggest that the energy expended performing R&R Work (Free-Float) is lower in flight than predicted. Accurate predictions of the metabolic cost of EVA are essential for planning and executing successful ISS EVAs. Overall, the current met rate prediction method is similar to actual in-flight values, slightly erring on the side of overestimation. Future work includes analysis of more historical in-flight EVA data to increase the power of the analysis, evaluating the NBL-ISS met rate conversion factor between NBL and ISS tasks, as well as exploring methods of substitution when crewmembers are missing prior task category data.

Lauren Cox↗

NASA NTRS2024

Soil Salinity Level Assessment and Prediction Integrating UAV-borne Hyperspectral Imaging and Machine Learning Algorithms to Combat Desertification

In response to the ongoing global food crisis, the United Nations has identified “Zero Hunger” as one of its Sustainable Development Goals. A central contributor to the crisis is the process in which agricultural lands go through desertification. Research has shown a direct correlation between soil salinity and desertification - increased salinity levels indicate a higher risk for desertification. Furthermore, researchers have explored various techniques to map soil salinity, but these methods are oftentimes inefficient and don’t address future salinity predictions. To improve desertification monitoring, soil salinity can be observed via hyperspectral imaging on unmanned aerial vehicles (UAVs) to predict the risk of agricultural desertification using artificial intelligence (AI) and machine learning (ML) techniques. A significant gap exists in past research that applies ML and imaging techniques to soil salinity: convolutional neural networks (CNNs) and regression models are rarely leveraged together, despite the efficiency and accuracy of these models. To compensate for this gap, the proposed system leverages the use of these AI and ML models to improve soil assessment and prediction techniques. This approach involves three steps - data collection, image analysis, and future prediction. Using hyperspectral cameras on UAVs to collect the data from the region, a trained CNN model will output estimated soil salinity levels at a specific time. The estimations will then be analyzed by a regression model to assess the accuracy of future soil salinity predictions. The proposed system will identify regions at risk of desertification to help farmers mitigate agricultural loss, in turn helping alleviate the food crisis.

UAV systems↗

NASA NTRSDate not supplied

Jet Noise Modeling Using an Acoustic Analogy: From Theoretical Formulation to Practical Noise Predictions

In this presentation an overview of the development of an acoustic analogy formulation, and its implementation into a reduced-order prediction method for turbulent jet mixing noise, will be given. The definition and rationale of an acoustic analogy approach to the prediction of flow-generated noise will be provided. A high-level description of the mathematical processes involved will be outlined in the context of a particular formulation developed at NASA Glenn Research Center. The modelling approximations needed to implement the general theoretical formulation into a practical noise prediction tool will be discussed. Sample results obtained using a noise prediction code based on the NASA Glenn formulation, and comparisons with experimental data, will be presented for a series of turbulent jet test cases. A view on the prospects for continued use of acoustic analogy-based methods for the prediction of flow-generated noise and the assessment of noise-reduction technologies will conclude the presentation.

Jet noise↗

NASA NTRSDate not supplied

USM3D-ME Contributions to the 5th AIAA High Lift Prediction Workshop

This paper presents the results of Reynolds-averaged Navier-Stokes (RANS) simulations conducted by NASA’s flow solver, mixed-element USM3D (USM3D-ME), for the 5th AIAA High-Lift Prediction Workshop. As part of the Fixed-Grid RANS Technology Focus Group (TFG), these simulations were performed to assess the accuracy and efficiency of the USM3D-ME solutions in predicting high-lift flows. The High-Lift Common Research Model (CRM-HL) served as the primary geometry. Several CRM-HL configurations were used for three case studies: a verification study (Case 1), a configuration buildup study (Case 2), and a Reynolds-number variation study (Case 3). Overall, USM3D-ME RANS results aligned with the solutions selected by the Fixed-Grid RANS TFG and available wind tunnel data Simulations for Cases 1 and Configuration 2.1 achieved machine-zero residual convergence, with aerodynamic coefficients converging to steady-state values. However, Configurations 2.2-2.4 and Case 3 encountered iterative- and grid-convergence challenges, particularly at high angles of attack. Compared with the experimental data available for Configurations 2.2-2.4, close agreement was demonstrated at low angles of attack. However, for angles of attack approaching the maximum lift conditions, the predicted lift coefficient and pitching moment deviated from experimental values. The drag-coefficient predictions were in a relatively good agreement, however, slight overpredictions were observed at the highest angle of attack corresponding to the maximum-lift condition. Although iterative convergence for Configurations 2.2-2.4 at high angles of attack remains a persistent challenge, averaging aerodynamic coefficients over the last 5000 iterations yielded satisfactory agreement with the available wind tunnel experimental data. During the workshop, the lack of iterative convergence was attributed to the vortex structures emanating from the slat brackets. To investigate this issue further, post-workshop simulations were conducted on Configuration 2.2. In one study, RANS simulations were performed on a simplified geometry with the slat brackets removed. The second study focused on performing URANS simulations on the original Configuration 2.2 geometry. Preliminary results from both studies are presented and compared with wind tunnel data for Configuration 2.2. Consistent with the findings of other participants in the Fixed-Grid RANS TFG, this study emphasizes the necessity for further exploration and advancement in RANS technology for predicting high-lift flows.

CFD↗

NASA NTRSDate not supplied

USM3D-ME Contributions to the 5th AIAA High Lift Prediction Workshop

This paper presents the results of Reynolds-averaged Navier-Stokes (RANS) simulations conducted by the NASA flow solver, mixed-element USM3D (USM3D-ME), for the 5th AIAA High-Lift Prediction Workshop. As part of the Fixed-Grid RANS Technology Focus Group (TFG), these simulations were performed to assess the accuracy and efficiency of the USM3D-ME solutions in predicting high-lift flows. The High-Lift Common Research Model (CRM-HL) served as the primary geometry. Several CRM-HL configurations were used for three case studies: a verification study (Case 1), a configuration buildup study (Case 2), and a Reynolds-number variation study (Case 3). Overall, USM3D-ME RANS results aligned with the solutions selected by the Fixed-Grid RANS TFG and available wind tunnel data. Simulations for Case 1 and Configuration 2.1 achieved machine-zero residual convergence, with aerodynamic coefficients converging to steady-state values. However, Configurations 2.2-2.4 and Case 3 encountered iterative- and grid-convergence challenges, particularly at high angles of attack. Compared with the experimental data available for Configurations 2.2-2.4, close agreement was demonstrated at low angles of attack. However, for angles of attack approaching the maximum lift conditions, the predicted lift coefficient and pitching moment deviated from experimental values. The drag-coefficient predictions were in relatively good agreement, however, slight overpredictions were observed at the highest angle of attack corresponding to the maximum-lift condition. Although iterative convergence for Configurations 2.2-2.4 at high angles of attack remains a persistent challenge, averaging aerodynamic coefficients over the last 5000 iterations yielded satisfactory agreement with the available wind tunnel experimental data. During the workshop, the lack of iterative convergence was attributed to the vortex structures emanating from the slat brackets. To investigate this issue further, post-workshop simulations were conducted on Configuration 2.2. In one study, RANS simulations were performed on a simplified geometry with the slat brackets removed. The second study focused on performing unsteady RANS (URANS) simulations on the original Configuration 2.2 geometry. Preliminary results from both studies are presented and compared with wind tunnel data for Configuration 2.2. Consistent with the findings of other participants in the Fixed-Grid RANS TFG, this study emphasizes the necessity for further exploration and advancement in RANS technology for predicting high-lift flows.

Aerodynamics↗

NASA NTRSDate not supplied

Predicting Team Functioning in Long Term Space Missions Using Acoustic and Linguistic Measures

Maintaining optimal team functioning is critical for long-duration space exploration missions, yet traditional monitoring methods, such as self-reports and wearable sensors, often impose operational burdens or suffer from bias. This paper investigates a non-intrusive speech-based artificial intelligence (AI) framework to predict degradations in team functioning using data from the Human Exploration Research Analog (HERA) of the U.S. National Aeronautics and Space Administration (NASA). Using acoustic features, linguistic descriptors, and semantic embeddings, we evaluate static non-linear and temporal machine learning models to predict both objective (task accuracy) and subjective (self-reported efficacy and cohesion) team functioning outcomes. Results indicate that temporal models outperform static approaches, with prediction of objective task accuracy in Team Interaction Battery (TIB) improving from near chance to 71%. Self-reported outcomes, including team efficacy and cohesion, are predicted more reliably than task performance, achieving balanced accuracies of up to 85.56% and 78.12%, respectively, and are found to be most strongly associated with acoustic features. In a second interdependent task, the MMSEV–EVA, accuracies of up to 78% are achieved using temporal models with acoustic features. Furthermore, incorporating just 1–2 days of team-specific historical data systematically improved performance, and acoustic markers from informal pre-task interactions provided modest predictive gains. Finally, while automated preprocessing yielded viable accuracy, humancorrected data provided moderate performance gains, though transcription error rates did not significantly correlate with model performance. These findings highlight the potential of speech as a passive, high-fidelity monitoring tool for autonomous habitats.

Temporal modeling↗

DOE OSTI2020

On Performance Prediction of Big Data Transfer in High-performance Networks

Big data generated by large-scale scientific and industrial applications need to be transferred between different geographical locations for remote storage, processing, and analysis. High-speed dedicated connections provisioned in High-performance Networks (HPNs) are increasingly utilized to carry out such big data transfer. HPN management highly relies on an important capability of performance (mainly throughput) prediction to reserve sufficient bandwidth and meanwhile avoid over-provisioning that may result in unnecessary resource waste. This capability is critical to improving the resource (mainly bandwidth) utilization of dedicated connections and meeting various user requests for data transfer. Conventional methods conduct performance prediction by fitting prior observed transfer history with predefined loss functions, without considering unobservable latent factors such as competing loads on end hosts. Such latent factors also have a significant impact on the application-level data transfer performance, which may result in an inaccurate prediction model. In this paper, we first investigate the impact of latent factors and propose a clustering-based method to eliminate their negative impact on performance prediction. We then develop a robust machine learning-based performance predictor by: i) incorporating the proposed latent factor elimination method into data preprocessing, and ii) adopting a customized domain guided loss function. Extensive experimental results show that our predictor achieves significantly higher prediction accuracy than several other state-of-the-art methods.

Liu, Wuji↗

DOE OSTI2020

Dynamic Queue Prediction at Signalized Intersections with Fusing Sensory Information and Connected Vehicles

Queue length is one of the most essential metrics for indicating traffic conditions at signalized intersections. Most existing studies mainly focus on queue length estimation in a retrieving way which cannot effectively predict the future state of vehicle in queue. This paper proposes an innovative dynamic queue prediction model in a vehicle-infrastructure cooperative way. The proposed dynamic queue predictor can customize the queue prediction using the trajectories of preceding vehicles from radar detection with varying information from loop detectors and connected vehicles (CVs). The dynamic queue prediction accuracy can be further improved with at least one present CV in the same cycle. With the accurate dynamic queue prediction, we present a case study of Queue-Aware Eco-Approach and Departure (QEAD) to provide an online optimal vehicle trajectory considering both the signal status and preceding vehicle's future status in queue. The proposed framework and algorithms are of pragmatic significance and one of its advanced features is the ability to work under low penetration rate of CVs. The test using Next Generation SIMulation (NGSIM) dataset and a case study in VISSIM microscopic simulation show promising results.

Ye, Fei↗

DOE OSTI2020

Controlling accesses to a branch prediction unit for sequences of fetch groups

An electronic device handles accesses of a branch prediction functional block when executing instructions in program code. The electronic device includes a processor having the branch prediction functional block that provides branch prediction information for control transfer instructions (CTIs) in the program code and a minimum predictor use (MPU) functional block. The MPU functional block determines, based on a record associated with a given fetch group of instructions, that a specified number of subsequent fetch groups of instructions that were previously determined to include no CTIs or conditional CTIs that were not taken are to be fetched for execution in sequence following the given fetch group. The MPU functional block then, when each of the specified number of the subsequent fetch groups is fetched and prepared for execution, prevents corresponding accesses of the branch prediction functional block for acquiring branch prediction information for instructions in that subsequent fetch group.

Agrawal, Varun↗

DOE OSTI2021

Towards Efficient Uncertainty estimation in deep learning for robust energy prediction in crystal materials

In material science, recent studies have started to explore the potential of using deep learning to improve property prediction from high-fidelity simulations, e.g, density functional theory (DFT). However, the design spaces are sometimes too large and intractable to sample completely. This results in a critical question that is how to evaluate the confidence and robustness of the prediction. In this paper, we propose an efficient approach to estimate uncertainty in deep learning using a single forward pass and then apply it for robust prediction of the total energy in crystal lattice structures. Our approach is built upon the deep kernel learning (DKL) that originally introduces to leverage the expressiveness of deep neural networks as input with a probabilistic prediction of Gaussian processes (GPs) as output. Existing DKL methods have difficulties in the accuracy of predictive uncertainty, training stability, and scaling to large datasets, which lead to significant barriers in real-world applications. We propose to address these challenges by using an inducing point approximate GP in feature space combined with spectral normalization as a regularization. We finally demonstrate our robust performance on an artificial example and a real-world application from materials chemistry.

Bi, Sirui↗

DOE OSTI2021

AGR-5/6/7 Irradiation As Run Predictions Using PARFUME

The PARticle FUel ModEl (PARFUME), a fuel performance modeling code used for high-temperature gas-cooled reactors, was used to model the Advanced Gas Reactor (AGR)-5/6/7 irradiation test using as-run physics and thermal data. The AGR-5/6/7 irradiation test consists of the combined fifth, sixth, and seventh planned irradiations of the AGR Fuel Development and Qualification Program. The AGR-5/6/7 test train is a multi-capsule, instrumented experiment that is designed for irradiation in the 133.4-mm diameter northeast flux trap position of the Advanced Test Reactor (ATR) at Idaho National Laboratory. Each capsule contains compacts filled with uranium oxycarbide unaltered fuel particles. This report documents the calculations performed to predict the failure probability of tristructural isotropic (TRISO)-coated fuel particles during the AGR-5/6/7 experiment. In addition, this report documents the calculated fission product release fraction from the fuel. The calculations include modeling of the AGR 5/6/7 irradiation that occurred from February 2018 to July 2020 over nine ATR cycles, including six normal cycles and three power axial locator mechanism cycles, for a total of approximately 376 effective full power days (EFPD). The irradiation conditions and material properties of the AGR-5/6/7 test predicted zero fuel particle failures in Capsules 1, 3, and 4. Fuel particle failures were predicted in two of the compacts in Capsule 2 and one particle failure is predicted in each one of the compacts in Capsule 5. All compacts that exhibited fuel particle failures predicted by PARFUME were caused by localized stress concentrations in the silicon carbide (SiC) layer caused by cracking in the inner pyrolytic (IPyC) layer. In addition, shrinkage of the buffer and IPyC layer during irradiation resulted in formation of a buffer-IPyC gap. Compacts with a lower irradiation temperature and fluence experienced the smallest buffer-IPyC gap formation. Conversely, higher irradiated temperature compacts with a high fluence experienced the largest buffer-IPyC gap formation. Compact 3-6-3 experienced the largest buffer IPyC gap formation of just under 21.7 µm. The release fraction of fission products silver (Ag), cesium (Cs), and strontium (Sr) vary depending on capsule location and irradiation temperature. The maximum release fraction of Ag occurs in Capsule 3, reaching up to 59.5% for the TRISO fuel particles (compact 3-6-3). The release fraction of the other two fission products, Cs and Sr, are much smaller. A maximum Cs release fraction of 1.1% occurred in compact 3-4-3 and 4.4% for Sr in compact 3-6-3.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

DOE OSTI2022

Multimodel ensemble predictions of precipitation using bayesian neural networks

Multimodel ensembling improves predictions and considers model uncertainties. In this study, we present a Bayesian Neural Network (BNN) ensemble approach for large-scale precipitation predictions based on a set of climate models. BNN infers spatiotemporally varying model weights and biases through the calibration against observations. This ensemble scheme of BNN sufficiently leverages individual model skill for accurate predictions as well as provides interpretability about which models contribute more to the ensemble prediction at which locations and times to inform model development. Additionally, BNN accurately quantifies epistemic uncertainty to avoid overconfident projections. We demonstrate BNN’s superior prediction performance to three state-of-the-art ensemble approaches and discuss its interpretability and uncertainty quantification.

Fan, Patrick↗

DOE OSTI2022

Multimodel Ensemble Predictions of Precipitation using Bayesian Neural Networks

Fan, Patrick↗

DOE OSTI2022

Physiochemical Machine Learning Models Predict Operational Lifetimes of CH3NH3PbI3 Perovskite Solar Cells

Halide perovskites are promising photovoltaic (PV) materials with the potential to lower the cost of electricity and greatly expand the penetration of PV if they can demonstrate long-term stability under illumination in the presence of moisture and oxygen. The solar cell service lifetime as quantified by the T80 (the time required for the power conversion efficiency to drop to 80% of its starting value) is a useful metric to assess stability. The T80 for utility, commercial, or residential PV systems needs to be several decades in order to yield low-cost electricity, and thus it is not practical to directly measure the T80. It would be useful if T80 could be predicted from the initial dynamics of a solar cell’s performance, but until now no models have been developed to forecast T80. In this work, we report the development of machine learning models to predict T80 of ITO/NiOx/CH3NH3PbI3/C60/BCP/Ag solar cells operating at maximum power point under 1-sun equivalent photon flux in air at varying temperatures and relative humidities. Efficiency losses are driven by short-circuit current and fill factor, indicating that chemical decomposition of the perovskite is a major contributor to degradation. Spatial patterns evident from in situ dark field optical microscopy suggest that the electric field gradient at device edges plays a significant role in perovskite decomposition, along with photochemical reactions with O2 and H2O. Models are trained using a menu of features from three distinct categories: (i) features based on measurements of the initial rates of change of device parameters, (ii) features based on the ambient conditions during operation (temperature, & partial pressure of H2O), and (iii) features based on underlying physics and chemistry. We show that a theory-based physiochemical feature derived from a model of the chemical reaction kinetics of the rate of degradation of the CH3NH3PbI3 is particularly valuable for prediction. This physiochemical feature was selected as the first or second most dominant feature in the best performing models. With a dataset consisting of 45 accelerated degradation experiments with T80 that range over a factor of 30, the model predicts T80 with an accuracy of about 40% (|predicted T80 - observed T80| / observed T80) on samples not used in training. This hybrid ML approach should be effective when applied to other compositions, device architectures, and advanced packaging schemes.

14 SOLAR ENERGY↗

DOE OSTI2022

Physiochemical Machine Learning Models Predict Operational Lifetimes of CH3NH3PbI3 Perovskite Solar Cells

14 SOLAR ENERGY↗