Article
Engineering, Marine
Jisu Lim, Minjoo Choi, Seungjae Lee
Summary: Dynamic analysis is powerful but time-consuming for mooring system design. In this study, we proposed a fast convergence Bayesian optimization algorithm (BOA) that updated the objective function as more data points were obtained. Compared with genetic algorithm (GA), which used a pre-trained surrogate model, BOA achieved a 50% reduction in maximum tension for an initial mooring system. However, GA required 20 times more computation time due to the training of the surrogate model.
JOURNAL OF MARINE SCIENCE AND ENGINEERING
(2023)
Article
Automation & Control Systems
Di Wu, G. Gary Wang
Summary: To reduce computational costs in engineering design, a novel metamodel called causal artificial neural network (causal-ANN) is developed in this paper, which leverages cause-effect relations and intermediate variables to train sub-networks and improve accuracy. By utilizing the structure of the causal-ANN and Bayesian Networks theory, attractive design subspaces can be identified.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Lei Wang, Zhengchao Liu
Summary: This paper introduces a new data-driven product design evaluation method, which establishes a multi-stage evaluation indicator model and an improved artificial neural network combined with the PSO-Adam optimization algorithm to achieve fast and accurate design evaluation. Experimental results demonstrate that the proposed method can help designers comprehensively consider design parameters and conduct effective evaluation.
APPLIED SOFT COMPUTING
(2021)
Article
Engineering, Mechanical
Ali Usman, Saad Arif, Ahmed Hassan Raja, Reijo Kouhia, Andreas Almqvist, Marcus Liwicki
Summary: In this study, a data-driven approach utilizing machine learning techniques was proposed to optimize the composition of a hybrid oil by adding ceramic and carbon-based nanoparticles. The performance of various ML models was evaluated, and artificial neural network (ANN) showed the best prediction performance. The results revealed that the composition of the optimized hybrid oil greatly influenced the lubrication mechanism, and the coefficient of friction was significantly reduced with the addition of optimal concentrations of ceramic and carbon-based nanoparticles. This research work has potential applications in designing hybrid nano lubricants for achieving optimized tribological performance in different lubrication regimes.
Article
Mechanics
Hau T. Mai, Seunghye Lee, Donghyun Kim, Jaewook Lee, Joowon Kang, Jaehong Lee
Summary: This paper proposes a robust deep neural network-based parameterization framework to directly solve the optimum design for geometrically nonlinear trusses subject to displacement constraints. The integration of DNN into Bayesian optimization allows for finding the best optimum structural weight, which is further optimized through hyperparameter optimization. The experimental results demonstrate that this approach can overcome the drawbacks of machine learning applications in computational mechanics.
EUROPEAN JOURNAL OF MECHANICS A-SOLIDS
(2023)
Article
Thermodynamics
J. Graca Gomes, H. J. Xu, Q. Yang, C. Y. Zhao
Summary: This study proposes a novel optimization model for determining the most cost-efficient renewable power capacity mix in autonomous microgrids, with results showing optimal configuration and a lower levelized cost of electricity compared to diesel alternatives while ensuring minimum land area occupation. Sensitivity analysis is also conducted to examine the impact of variables on the system's cost and efficiency.
Article
Chemistry, Physical
Majid Saidi, Masha Yousefi, Mehran Minbashi, Fatemeh Arab Ameri
Summary: Catalytic upgrading of 4-methylanisole using Pt/g-Al2O3 catalyst was studied, showing major products like toluene, phenol derivatives, and cyclohexanone; artificial neural network and design of experiment were utilized to predict conversion rate, product selectivity, and reactions network; optimization of experimental data through response surface methodology and artificial neural network model demonstrated high prediction accuracy and potential for large-scale production of fuels from renewable resources.
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2021)
Review
Pharmacology & Pharmacy
Ming Gao, Sibo Liu, Jianan Chen, Keith C. Gordon, Fang Tian, Cushla M. McGoverin
Summary: Drug development is a time-consuming process with high failure rates, where pharmaceutical formulation development plays a crucial role in linking new chemical entities to clinical trials. Artificial intelligence and Raman spectroscopy have the potential to accelerate formulation development and provide new pathways for high-quality data gathering.
INTERNATIONAL JOURNAL OF PHARMACEUTICS
(2021)
Article
Nanoscience & Nanotechnology
Rene Rebollo, Feras Oyoun, Yohann Corvis, Mazen M. El-Hammadi, Bruno Saubamea, Karine Andrieux, Nathalie Mignet, Khair Alhareth
Summary: This study utilized a microfluidic system to fabricate liposomes with different molar ratios, and optimized the process and conditions using design of experiments (DoE) principles. An artificial neural network (ANN) model was constructed to predict the size and polydispersity index (PDI) of the liposomes. The results demonstrated the feasibility of producing optimized liposomes with high reproducibility using microfluidic manufacturing processes, DoE, and Artificial Intelligence (AI), enabling faster development and clinical transfer of nanobased medications.
ACS APPLIED MATERIALS & INTERFACES
(2022)
Article
Engineering, Chemical
Nasser Zouli
Summary: This paper proposes a novel approach using neural networks to predict the effectiveness of a hybrid solar-powered desalination system and verifies it using simulation results. The RSO-RBLSTM method is utilized to forecast the performance of the hybrid desalination process, and the proposed approach shows effective performance when powered by the solar system according to various metrics.
Article
Mechanics
Xiaoyang Liu, Jian Qin, Kai Zhao, Carol A. Featherston, David Kennedy, Yucai Jing, Guotao Yang
Summary: This paper proposes an efficient method for minimum weight optimization of composite laminates using artificial neural network (ANN) based surrogate models. By predicting the buckling loads of the laminates using ANN, the need for time-consuming buckling evaluations in the optimization process is eliminated. The use of lamination parameters and other dimensional inputs significantly reduces the number of required models and computational cost.
COMPOSITE STRUCTURES
(2023)
Article
Construction & Building Technology
Chuan-Hsuan Lin, Yaw-Shyan Tsay
Summary: Recent studies have developed a novel daylight prediction model by converting design models into intermediary features as input parameters, allowing for a wider range of applicability. The proposed model demonstrated good performance in predicting daylight performance of different types of facades and showed potential for significant time savings in daylighting evaluation compared to traditional simulation methods.
BUILDING AND ENVIRONMENT
(2021)
Article
Thermodynamics
Yicong Li, Chunyu Shi, Wei Liu, Zhichun Liu
Summary: A numerical investigation was conducted to explore the heat transfer and flow characteristics of a welded plate heat exchanger with chevron sinusoidal corrugated plates in this paper. The effects of chevron angle, height of corrugations, and pitch of corrugations on the thermal-hydraulic performance were studied. Artificial neural networks and multi-objective genetic algorithm were used to obtain optimization solutions. The results showed that the comprehensive performance of the plate with a chevron angle of 45 degrees was the best, and a set of optimal parameters was obtained.
INTERNATIONAL COMMUNICATIONS IN HEAT AND MASS TRANSFER
(2023)
Review
Chemistry, Physical
Beng Wei Chong, Rokiah Othman, Ramadhansyah Putra Jaya, Mohd Rosli Mohd Hasan, Andrei Victor Sandu, Marcin Nabialek, Bartlomiej Jez, Pawel Pietrusiewicz, Dariusz Kwiatkowski, Przemyslaw Postawa, Mohd Mustafa Al Bakri Abdullah
Summary: Concrete mix design and performance determination involve not only engineering, but also mathematical and statistical approaches. The study of concrete mechanical properties considers various factors, and design of experiments has become crucial in simplifying the work and providing accurate predictions. Common methods such as ANOVA, regression, Taguchi method, Response Surface Methodology, and Artificial Neural Network are used for optimizing data collection and analysis.
Article
Chemistry, Physical
Egor S. Rodionov, Victor V. Pogorelko, Victor G. Lupanov, Polina N. Mayer, Alexander E. Mayer
Summary: Current progress in numerical simulations and machine learning allows for the identification of parameters in plasticity models using complex loading conditions. A combined experimental-numerical approach is developed and applied to the study of cold-rolled OFHC copper. Profiled projectiles are proposed for the Taylor impact problem for the first time for material characterization, which allows for large plastic deformations with high strain rates. The optimized numerical model is successfully validated using experimental data.