Michael Beer

Probabilistic safety assessment of adjacent floating bodies under extreme waves requires accurate estimation of structural failure probability, which is hampered by strongly nonlinear hydrodynamic coupling and prohibitive simulation costs. This study develops an active-learning framework that couples a high-fidelity computational fluid dynamics numerical wave tank with Gaussian-process surrogate modeling and wave-group parameterization, achieving converged failure-probability estimates with fewer simulations than direct Monte Carlo. Applied to a simplified two-body system under a 100-year return-period sea state, the framework reveals a nonmonotonic dependence of failure probability on the gap-to-breadth ratio G / D , with minimum risk at G / D = 0.5 and increasing risk at wider spacings due to progressive interbody phase decoupling. Sensitivity analyses across multiple failure thresholds confirm the robustness of this trend.

Near-fault strong-motion records often contain large-amplitude, long-period velocity pulses, and their identification remains sensitive to the choice of pulse criteria and labeling strategies. This paper presents an automated classification framework that couples iterative wavelet-based pulse/residual decomposition with multivariate shapelet learning to distinguish pulse-like and non-pulse-like ground motions and to further separate early- and late-arriving pulses. A consistently labeled ground-motion dataset is constructed by adopting the intersection of three established identification methods. Starting from velocity time histories, the iterative wavelet decomposition extracts pulse and residual components over 50 iterations, forming a compact 6 × 50 multivariate feature matrix to summarize amplitude- and energy-related pulse attributes. A metric-learning objective (triplet loss) jointly optimized with a classification objective is used to learn an embedding of candidate subsequences and select 20 representative multivariate shapelets. The shapelet-to-record distance features are then used to train standard machine-learning classifiers. Among these, a linear support vector machine achieves the best overall performance, yielding classification accuracies of 98.2% on the training set and 94.4% on the test set, while maintaining consistently strong sensitivity, specificity, F 1 -score, and Cohen’s κ . The selected shapelets provide interpretable physical signatures (e.g., residual-energy decay patterns in nonpulse records), while a physics-based dual-threshold criterion anchored by temporal energy centroids enables robust discrimination between early and late directivity pulses.

Soil slope stability is profoundly affected by the spatial variability of soil properties. Although random field is a widely used tool for characterizing spatial variability in slope reliability analysis, its application to a real slope at a specific site is often constrained. The challenge lies in the difficulty of properly determining random field parameters from sparse and limited geotechnical investigation (GI) data (e.g., boreholes and test data) at a specific site. To tackle this challenge, this study proposes a novel knowledge-informed machine learning framework for slope stability analysis. The proposed framework is conditional on limited site-specific GI data and bypasses a need for estimation of random field parameters. The exploited knowledge includes geometry and soil stratigraphy of a target slope, typical statistics of soil properties, and physics-based model of slope stability analysis. The knowledge is then collectively represented by a series of simulation slopes with different spatial variability. Stability of the target slope is predicted by machine learning of similar simulation slopes identified by site-specific GI data. The proposed framework is illustrated through a numerical example and a real case history, demonstrating its ability to deliver accurate and efficient stability analysis for real slopes with consideration of spatial variability using limited GI data.

Economic systems are complex, adaptive and interact with other systems. Their characteristics include stochastic and non-stationary behavior, complex interrelationships between elements, and small data samples. This makes it difficult to monitor and forecast their state. The purpose of this study is to describe and test a systematic approach to entropy monitoring and forecasting of the state of economic systems on real macroeconomic data. The article describes a vector entropy model that considers the main systemic patterns. The model is based on the decomposition of the differential entropy of a set of indicators into components—entropic measures of dynamism, randomness and self-organization. This model develops the extended version of the second law of thermodynamics proposed by Ilya Prigozhin for the entropy of open systems and allows it to be implemented in practice. An example of monitoring the German macroeconomic indicators system is considered. The results obtained generally correspond to the known estimates of the situation. Vector entropy modeling makes it possible to supplement well-known macro economical modeling and forecasting methods with a system-wide interpretation and assessment of the state of the analyzed processes and systems.