Understanding complex systems with their reduced model is one of the central roles in scientific activities. Although physics has greatly been developed with the physical insights of physicists, it is sometimes challenging to build a reduced model of such complex systems on the basis of insight alone. We propose a framework that can infer hidden conservation laws of a complex system from deep neural networks (DNNs) that have been trained with physical data of the system. The purpose of the proposed framework is not to analyze physical data with deep learning but to extract interpretable physical information from trained DNNs. With Noether's theorem and by an efficient sampling method, the proposed framework infers conservation laws by extracting the symmetries of dynamics from trained DNNs. The proposed framework is developed by deriving the relationship between a manifold structure of a time-series data set and the necessary conditions for Noether's theorem. The feasibility of the proposed framework has been verified in some primitive cases in which the conservation law is well known. We also apply the proposed framework to conservation law estimation for a more practical case, that is, a large-scale collective motion system in the metastable state, and we obtain a result consistent with that of a previous study.
31 January 2020 in Science and Technology of Advanced Materials
We develop an automatic peak fitting algorithm using the Bayesian information criterion (BIC) fitting method with confidence-interval estimation in spectral decomposition. First, spectral decomposition is carried out by adopting the Bayesian exchange Monte Carlo method for various artificial spectral data, and the confidence interval of fitting parameters is evaluated. From the results, an approximated model formula that expresses the confidence interval of parameters and the relationship between the peak-to-peak distance and the signal-to-noise ratio is derived. Next, for real spectral data, we compare the confidence interval of each peak parameter obtained using the Bayesian exchange Monte Carlo method with the confidence interval obtained from the BIC-fitting with the model selection function and the proposed approximated formula. We thus confirm that the parameter confidence intervals obtained using the two methods agree well. It is therefore possible to not only simply estimate the appropriate number of peaks by BIC-fitting but also obtain the confidence interval of fitting parameters. Graphical abstract
5 September 2022 in Science and Technology of Advanced Materials: Methods
Fracture surfaces provide various types of information about fracture. The fracture toughness , which represents the resistance to fracture, can be estimated using the three-dimensional (3D) information of a fracture surface, i.e. its roughness. However, this is time-consuming and expensive to obtain the 3D information of a fracture surface; thus, it is desirable to estimate from a two-dimensional (2D) image, which can be easily obtained. In recent years, methods of estimating a 3D structure from its 2D image using deep learning have been rapidly developed. In this study, we propose a framework for fractography that directly estimates from a 2D fracture surface image using deep neural networks (DNNs). Typically, image recognition using a DNN requires a tremendous amount of image data, which is difficult to acquire for fractography owing to the high experimental cost. To compensate for the limited number of data, in this study, we used the transfer learning (TL) method and constructed high-performance prediction models even with a small dataset by transferring machine learning models trained using other large datasets. We found that the regression model obtained using our proposed framework can predict in the range of approximately 1–5 [MPa ] with a standard deviation of the estimation error of approximately 0.37 [MPa ]. The present results demonstrate that the DNN trained with TL opens a new route for quantitative fractography by which parameters of fracture process can be estimated from a fracture surface even with a small dataset. GRAPHICAL ABSTRACT
1 January 2021 in Science and Technology of Advanced Materials: Methods
The alloying elements that control the creep rupture life of low carbon steel in the region of inherent creep strength were investigated by a data-driven model selection method. The experimental data published in NRIM Creep Data Sheet No. 7B was used. A model in which the Larson–Miller parameter is expressed as a linear sum of the logarithmic stress and the chemical composition of alloying elements was proposed on the basis of the analysis of the experimental data. There were 1023 ( ) candidate models containing one or more of the 10 alloying elements, and they were compared using two model selection methods: 10-fold cross validation (CV) and posterior probability based on Bayesian inference. The 10-fold CV suggested that Mo must be included in the model to improve the predictivity of the unknown data. A comparison among the Bayesian posterior probabilities of the candidate models showed that the Mo-only model has a predominantly high posterior probability of 79.76% compared with the other models. Furthermore, other element terms that co-occur with Mo evenly appeared in the top list after the Mo-only model, indicating that only Mo is crucial for increasing the posterior probability. From these results, it is concluded that the Mo-only model is the most appropriate as a data generation model. The selected Mo-only model was confirmed to have high predictivity of the creep life. Graphical abstract
13 September 2024 in Physica D: Nonlinear Phenomena
We present a Bayesian model selection framework to identify reduced dynamical equations describing the evolution of volume-integrated turbulence energy in magnetically confined plasmas. Using data from two-dimensional Hasegawa–Wakatani simulations with varying adiabatic parameters, we consider a candidate model with polynomial nonlinearities up to ninth order and a coupling term to zonal flow energy. All combinations of these terms are evaluated based on Bayesian model evidence. The optimal model includes only linear growth, quadratic nonlinear saturation, and zonal flow suppression. While subcritical turbulence was not observed in the present dataset, the selection framework successfully excluded redundant higher-order terms, confirming its robustness. Crucially, this approach enables data-driven identification of nonlinear saturation and turbulence-zonal flow coupling, which are often introduced heuristically in conventional models. These results suggest that the proposed framework can serve as a useful tool for constructing interpretable reduced models of plasma turbulence and for examining key mechanisms involved in nonlinear energy regulation.