Machine learning and physics-driven modelling and simulation of multiphase systems

International Journal of Multiphase Flow, Год журнала: 2024, Номер 179, С. 104936 - 104936

Опубликована: Авг. 1, 2024

We highlight the work of a multi-university collaborative programme, PREMIERE (PREdictive Modelling with QuantIfication UncERtainty for MultiphasE Systems), which is at intersection multi-physics and machine learning, aiming to enhance predictive capabilities in complex multiphase flow systems across diverse length time scales. Our contributions encompass variety approaches, including Design Experiments nanoparticle synthesis optimisation, Generalised Latent Assimilation models drop coalescence prediction, Bayesian regularised artificial neural networks, eXtreme Gradient Boosting microdroplet formation sub-sampling based adversarial network predicting slug behaviour two-phase pipe flows. Additionally, we introduce generalised latent assimilation technique, Long Short-Term Memory networks sequence forecasting mixing performance stirred static mixers, active learning via optimisation recover model parameters high current density electrolysers, Gaussian process regression size distribution predictions sprays, acoustic emission signal inversion using gradient boosting machines characterise particle fluidised beds. also offer perspectives on development shape framework that leverages use multi-fidelity emulator. The results presented have applications chemical synthesis, microfluidics, product manufacturing, green hydrogen generation.

Язык: Английский

---

Explainable AI Chatbots Towards XAI ChatGPT: A Review

Heliyon, Год журнала: 2025, Номер 11(2), С. e42077 - e42077

Опубликована: Янв. 1, 2025

Advances in artificial intelligence (AI) have had a major impact on natural language processing (NLP), even more so with the emergence of large-scale models like ChatGPT. This paper aims to provide critical review explainable AI (XAI) methodologies for chatbots, particular focus Its main objectives are investigate applied methods that improve explainability identify challenges and limitations within them, explore future research directions. Such goals emphasize need transparency interpretability systems build trust users allow accountability. While integrating such interdisciplinary methods, as hybrid combining knowledge graphs ChatGPT, enhancing explainability, they also highlight industry needs user-centred design. will be followed by discussion balance between performance, then role human judgement, finally verifiable AI. These avenues through which insights can used guide development transparent, reliable efficient chatbots.

Язык: Английский

---

A novel hybrid group method of data handling and Levenberg Marquardt model for estimating total organic carbon in source rocks with explainable artificial intelligence

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 144, С. 110137 - 110137

Опубликована: Янв. 27, 2025

Язык: Английский

---

Dual-directional small-sampling deep-learning modelling on co-flowing microfluidic droplet generation

Chemical Engineering Journal, Год журнала: 2024, Номер 485, С. 149467 - 149467

Опубликована: Фев. 10, 2024

Язык: Английский

---

Additive-feature-attribution methods: A review on explainable artificial intelligence for fluid dynamics and heat transfer

International Journal of Heat and Fluid Flow, Год журнала: 2024, Номер 112, С. 109662 - 109662

Опубликована: Дек. 9, 2024

Язык: Английский

---

Explainable AI model for predicting equivalent viscous damping in dual frame–wall resilient system

Journal of Building Engineering, Год журнала: 2024, Номер 96, С. 110564 - 110564

Опубликована: Авг. 28, 2024

A prominent challenge in applying the direct displacement-based design (DDBD) method to proposed dual frame–wall lateral force-resisting system lies determining equivalent viscous damping ratio (EVDR). However, strong nonlinearity and complexity behind procedure lead limited choice, mostly trial error based on experience, explain predict EVDR context of traditional research. This study employs XGBoost unravel intricate relationships using over 5 million data points from nonlinear time-history (NLTH) analyses, encompassing various parameters including fundamental period, ductility, subsystem stiffness ratios, post-yielding ratios subsystems ground motion types. SHapley Additive exPlanations (SHAP) values consistently identify critical features relevant procedure. Comprehensive feature ablation tests further illuminate robustness susceptibility each model. Additionally, incorporation Local Interpretable Model-agnostic Explanations (LIME) for local interpretability provides insights into decision-making mechanisms inherent model's predictions. Both predicting results machine learning (ML) are also compared. Findings highlight relative importance present a refined prediction It underscores pivotal role model reinforcing confidence complex models advocates leveraging ML techniques enhance effectiveness efficiency DDBD structural design.

Язык: Английский

---

Optimizing Droplet Coalescence Dynamics in Microchannels: A Comprehensive Study Using Response Surface Methodology and Machine Learning Algorithms

Heliyon, Год журнала: 2025, Номер 11(1), С. e41510 - e41510

Опубликована: Янв. 1, 2025

Droplet coalescence in microchannels is a complex phenomenon influenced by various parameters such as droplet size, velocity, liquid surface tension, and droplet-droplet spacing. In this study, we thoroughly investigate the impact of these control on dynamics within sudden expansion microchannel using two distinct numerical methods. Initially, employ boundary element method to solve Brinkman integral equation, providing detailed insights into underlying physics coalescence. Furthermore, integrate Response Surface Methodology (RSM) effectively optimize dynamics, harnessing power machine learning algorithms. Our results showcase efficacy computational techniques enhancing experimental efficiency. Through rigorous evaluation utilizing Regression Coefficient Mean Absolute Error metrics, ascertain accuracy our estimations. findings highlight significant influence key parameters, specifically non-dimensional initial distance droplets (D), viscosity ratio ( μ ), Capillary number (Ca), width (w), identified final spacing (DD), velocity first (VFD), second (VBD), respectively. This comprehensive approach provides valuable phenomena offers robust framework for optimizing microfluidic systems. The most influential DD are values Ad D, while has lowest DD. channel width, whereas Ca have least velocity. comparison different algorithms indicates that best ones predicting DD, VFD, VBD function, SMOreg, Lazy-IBK, Meta-Bagging,

Язык: Английский

---

Modeling and analysis of droplet generation in microchannels using interpretable machine learning methods

Chemical Engineering Journal, Год журнала: 2025, Номер unknown, С. 161972 - 161972

Опубликована: Март 1, 2025

Язык: Английский

---

Striking a balance: navigating the trade-offs between predictive accuracy and interpretability in machine learning models

Electronic Research Archive, Год журнала: 2025, Номер 33(4), С. 2092 - 2117

Опубликована: Янв. 1, 2025

Язык: Английский

---

Explainable Machine Learning with Two-Layer Multi-Objective Optimization Algorithm Applied to Sealing Structure Design

Materials, Год журнала: 2025, Номер 18(10), С. 2307 - 2307

Опубликована: Май 15, 2025

This study addresses the challenge of optimizing seal structure design through a novel two-stage interpretable optimization framework. Focusing on O-ring waterproof performance under hyperelastic material behavior, this proposes double-layer method integrating explainable machine learning with hierarchical clustering algorithms. The key innovation lies in employing modified to categorize parameters into two groups: bolt preload and groove depth. enables dimensionality reduction while maintaining physical interpretability critical parameters. In first layer, systematic parameter screening are applied variable reduce database, six remaining data points that constitute one-seventh original data. second layer subsequently refines configurations using E-TOPSIS (Entropy Weight—Technique for Order Preference by Similarity Ideal Solution) optimization. All evaluations performed FEA (finite element analysis) considering nonlinear responses. optimal is depth 0.8 mm 80 N. experimental validation demonstrates efficiently identifies designs meeting IPX8 requirements, zero leakage observed both surfaces motor interiors. proposed methodology provides physically meaningful guidelines.

Язык: Английский

---