A Novel Reconstruction Method for Irregularly Sampled Observation Sequences for Digital Twin

Applied Sciences, Год журнала: 2025, Номер 15(9), С. 4706 - 4706

Опубликована: Апрель 24, 2025

Various uncertainties such as communication delay, packet loss and disconnection in the Industrial Internet, well asynchronous sampling of sensors, can cause irregularity, sparsity, misalignment sequences, thereby seriously affect training prediction performance a digital twin model. Sequence reconstruction is an effective way to deal with above problems, but if measurement data become sparse or contain significant noise due electromagnetic interference, existing methods struggle achieve ideal results. Therefore, novel variational autoencoder model based on parallel reference network neural controlled differential equation (PRN-NCDE) proposed this article solve problem reconstructing irregular series under measurements high levels. First, multi-channel self-attention module established, which not only analyze position feature information sampled improve accuracy measurements, also effectively tackle irregularity observation sequence through mask mechanisms. Second, large levels, PRN established obtain features, are weighted fused features observed data. Third, we use NCDE construct decoder that combine control input system predict output values system. Finally, function constructed better train parameters This takes furnace boiler coal-fired power plant test object verify effectiveness fitting PRN-NCDE compared for Simulation results show estimation by more than 50% 70% recurrent network-NCDE (RNN-NCDE) different numbers 80% 60% network-NODE (RNN-NODE).

Язык: Английский

---

Advancing Healthcare Systems with Generative AI-Driven Digital Twins

Опубликована: Апрель 29, 2025

The healthcare sector is undergoing a digital transformation thanks to new technologies, with twinning and generative artificial intelligence (AI) leading the innovation. Digital twins, conceptualized originally as engineering or manufacturing tools, are increasingly finding their way sector, in response growing need for sophisticated virtual patient representations scope modeling several complex biological systems. Empowered by AI, they start replace static models, open gates into dynamic, predictive, prescriptive systems, enabling personalized delivery, disease modeling, surgical planning, drug discovery. This paper reviews combined potential of AI twin technologies domain. It delivers comprehensive view on present possible applications, benefits, opportunities technology while putting perspective challenges regarding data privacy, ethical, computational, design biases. By intertwining results from various studies companies, research thereby expounds realizing positive thrust capability twins influencing delivery toward more stringent, preventive medicine. identifies future directions crucial confronting current ensuring responsible deployment these systems across globe.

Язык: Английский

---

An Improved Framework for Predictive Maintenance in Industry 4.0 And 5.0 Using Synthetic Iot Sensor Data and Boosting Regressor For Oil and Gas Operations.

International Journal of Latest Technology in Engineering Management & Applied Science, Год журнала: 2025, Номер 14(4), С. 383 - 395

Опубликована: Май 7, 2025

Abstract: Predictive Maintenance (PdM) plays a pivotal role in Industry 4.0 and 5.0 by minimizing equipment downtime optimizing performance. However, limitations such as scarce fault data, data quality issues, model interpretability hinder its effectiveness. This study presents machine learning-based PdM framework tailored for Vortex Oil Gas Nigeria Ltd., leveraging synthetic sensor eXtreme Boost (XGBoost) regression to predict Remaining Useful Life (RUL) of industrial equipment. Using simulated from 50 machines over 300 operational cycles, the achieved strong performance metrics, with an RMSE 40.73 MAE 32.38. A four-layer system architecture—comprising acquisition, edge processing, cloud analytics, user interface—enabled real-time monitoring decision-making. The results underscore system’s capacity detect early failure trends support proactive maintenance, aligning goals intelligent, sustainable, human-centric operations. research contributes scalable, data-driven solution suitable environments limited real-world data.

Язык: Английский

---

Integrating Deep Learning Into Mechanical Engineering: A Systematic Review of Applications and Educational Implications

Computer Applications in Engineering Education, Год журнала: 2025, Номер 33(3)

Опубликована: Май 1, 2025

ABSTRACT Deep learning (DL) is reshaping mechanical engineering by offering advanced capabilities for solving complex problems, particularly in fault diagnosis, predictive maintenance, and materials science. While conventional machine physics‐based approaches remain prevalent, DL models provide superior performance terms of accuracy, automation, adaptability. This systematic review investigates trends applications within from 2015 to 2024. An initial search using the query “deep AND engineering” across seven major databases—Google Scholar, Web Science, IEEE Xplore, ERIC, Science Direct, Compendex, Wiley Online Library—yielded 149 articles. After applying exclusion criteria (published before 2014, non‐English, short or work‐in‐progress papers, not and/or focus, conceptual papers), 49 studies were selected in‐depth analysis. The results indicate that improve prediction accuracy 10%–35% over traditional techniques various applications, including detection rotating machinery microstructural analysis engineering. Despite notable gains, challenges persist related data availability, computational intensity, model interpretability. highlights importance addressing these limitations recommends future research efforts toward improving generalization, incorporating explainable AI techniques, optimizing deployment under limited‐data scenarios. Furthermore, integration with Industry 4.0 technologies—such as IoT, digital twins, cyber‐physical systems—presents a promising direction real‐time, intelligent decision‐making systems. serves comprehensive resource researchers practitioners seeking apply advance methods contexts.

Язык: Английский

---

Enhancing Large Language Models by Integrating Domain Knowledge based Data-Driven Small Models for Anomaly Diagnosis and Trouble Shooting in Wastewater Treatment Plants

Journal of environmental chemical engineering, Год журнала: 2025, Номер unknown, С. 117379 - 117379

Опубликована: Май 1, 2025

Язык: Английский

---