Physics-informed machine learning

Karniadakis, George Em; Kevrekidis, Ioannis G.; Lu, Lu; Perdikaris, Paris; Wang, Sifan; Liu, Yang

doi:10.1038/s42254-021-00314-5

Cited by 3,194 publications

(1,488 citation statements)

References 127 publications

Supporting

Mentioning

1,203

Contrasting

Unclassified

Order By: Relevance

“…The results of this study are in strong agreement with those observed by Meng et al (2021) and Wei (2021) that each found that AI was highly effective at predicting hurricane-induced SWHs. However, although contemporary applications of AI in the forecasting of both in mean and extreme (i.e., TC-forced) waves states have relied traditionally on singular inputs of SWH (Ali and Prasad, 2019;Zhao and Wang, 2018;Zhou et al, 2021a, b), a growing body of literature have demonstrated that the addition of other variables such as wind speed (as done here), wind direction and other variables improves forecast effectiveness (Kaloop et al, 2020;Zubier, 2020;Raj and Brown, 2021;Wang et al, 2021). Uncertainties in variable selection have also stimulated research into how to best identify predictors for the SWH or other predictands (Li and Liu, 2020;.…”

Section: Discussionmentioning

confidence: 83%

“…Accompanying increased scrutiny in building LSTM training datasets to improve predictions, the usage of physicsbased/informed/infused versions of LSTM and other artificial intelligence and machine learning algorithms (Karniadakis et al, 2021;Zhang et al, 2021) may help to bridge the gap in forecasting efficacy between physics-based third-generation numerical wave models such as WaveWatch III or SWAN. Crucially, this will ensure that forecasting remains significantly computationally cheaper than the usage of wave models.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Forecasting Hurricane-forced Significant Wave Heights using the Long Short-Term Memory Network in the Caribbean Sea

Bethel

Sun

2021

Preprint

View full text Add to dashboard Cite

Abstract. A Long Short-Term Memory (LSTM) neural network is proposed to predict hurricane-forced significant wave heights (SWH) in the Caribbean Sea (CS) based on a dataset of 20 CS, Gulf of Mexico, and Western Atlantic hurricane events collected from 10 buoys from 2010–2020. SWH nowcasting and forecasting are initiated using LSTM on 0-, 3-, 6-, 9-, and 12-hour horizons. Through examining study cases Hurricanes Dorian (2019), Sandy (2012), and Igor (2010), results illustrate that the model is well suited to forecast hurricane-forced wave heights. Forecasts are highly accurate with regard to observations. For example, Hurricane Dorian nowcasts had correlation (R), root mean square error (RMSE), and mean absolute percentage error (MAPE) values of 0.99, 0.16 m, and 2.6 %, respectively. Similarly, on the 3-, 6-, 9-, and 12-hour forecasts, results produced R (RMSE; MAPE) values of 0.95 (0.51 m; 7.99 %), 0.92 (0.74 m; 10.83 %), 0.85 (1 m; 13.13 %), and 0.84 (1.24 m; 14.82 %), respectively. However, the model also consistently over-predicted the maximum observed SWHs. To improve models results, additional research should be geared towards improving single-point LSTM neural network training datasets by considering hurricane track and identifying the hurricane quadrant in which buoy observations are made.

show abstract

Section: Discussionmentioning

confidence: 83%

Section: Discussionmentioning

confidence: 99%

Forecasting Hurricane-forced Significant Wave Heights using the Long Short-Term Memory Network in the Caribbean Sea

Bethel

Sun

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…This finding (differences between pure ML and physics-informed ML) is worth discussing. The project of adding physical constraints to ML is an active area of research across most fields of science and engineering (Karniadakis et al, 2021), including hydrology (e.g., Zhao et al, 2019;Jiang et al, 2020;Frame et al, 2020). It is important to understand that there is only one type of situation in which adding any type of constraint (physically-based or otherwise) to a data-driven model can add value: if constraints help optimization.…”

Section: Conclusion and Discussionmentioning

confidence: 99%

Deep learning rainfall-runoff predictions of extreme events

Frame

Kratzert

Klotz

et al. 2021

Preprint

View full text Add to dashboard Cite

Abstract. The most accurate rainfall-runoff predictions are currently based on deep learning. There is a concern among hydrologists that data-driven models based on deep learning may not be reliable in extrapolation or for predicting extreme events. This study tests that hypothesis using Long Short-Term Memory networks (LSTMs) and an LSTM variant that is architecturally constrained to conserve mass. The LSTM (and the mass-conserving LSTM variant) remained relatively accurate in predicting extreme (high return-period) events compared to both a conceptual model (the Sacramento Model) and a process-based model (US National Water Model), even when extreme events were not included in the training period. Adding mass balance constraints to the data-driven model (LSTM) reduced model skill during extreme events.

show abstract

“…In developing a hybrid PIML model, the physicsbased model can be used to provide the domain-expertise based checks and balances by penalising any physically inconsistent outputs. Additional discussions on the subject are available elsewhere, e.g., [72][73][74].…”

Section: Extracting Additional Scientific Value From ML Models Using a 'Grey Box Big Data' Approachmentioning

confidence: 99%

Modelling of Microstructure Formation in Metal Additive Manufacturing: Recent Progress, Research Gaps and Perspectives

Gunasegaram

Steinbach

2021

Metals

View full text Add to dashboard Cite

Microstructures encountered in the various metal additive manufacturing (AM) processes are unique because these form under rapid solidification conditions not frequently experienced elsewhere. Some of these highly nonequilibrium microstructures are subject to self-tempering or even forced to undergo recrystallisation when extra energy is supplied in the form of heat as adjacent layers are deposited. Further complexity arises from the fact that the same microstructure may be attained via more than one route—since many permutations and combinations available in terms of AM process parameters give rise to multiple phase transformation pathways. There are additional difficulties in obtaining insights into the underlying phenomena. For instance, the unstable, rapid and dynamic nature of the powder-based AM processes and the microscopic scale of the melt pool behaviour make it difficult to gather crucial information through in-situ observations of the process. Therefore, it is unsurprising that many of the mechanisms responsible for the final microstructures—including defects—found in AM parts are yet to be fully understood. Fortunately, however, computational modelling provides a means for recreating these processes in the virtual domain for testing theories—thereby discovering and rationalising the potential influences of various process parameters on microstructure formation mechanisms. In what is expected to be fertile ground for research and development for some time to come, modelling and experimental efforts that go hand in glove are likely to provide the fastest route to uncovering the unique and complex physical phenomena that determine metal AM microstructures. In this short Editorial, we summarise the status quo and identify research opportunities for modelling microstructures in AM. The vital role that will be played by machine learning (ML) models is also discussed.

show abstract

Physics-informed machine learning

Cited by 3,194 publications

References 127 publications

Forecasting Hurricane-forced Significant Wave Heights using the Long Short-Term Memory Network in the Caribbean Sea

Forecasting Hurricane-forced Significant Wave Heights using the Long Short-Term Memory Network in the Caribbean Sea

Deep learning rainfall-runoff predictions of extreme events

Modelling of Microstructure Formation in Metal Additive Manufacturing: Recent Progress, Research Gaps and Perspectives

Contact Info

Product

Resources

About