Archives of Computational Methods in Engineering最新文献

Directly obtaining the dynamic values of the unsteady aerodynamics at large angle of attack by either the CFD or experimental technologies to present further analysis should pay great costs. Therefore, the unsteady aerodynamic modeling based on a few calculations or experimental data has been established and developed. This study mainly discusses the development and challenges of unsteady aerodynamic modeling of aircraft at high angle of attack, investigates the accuracy, efficiency, and future development of the conventional and modern intelligent models divided according to the established physical basis. The conventional methods have been built on valuating changing law of either the macroscopic aerodynamic performance or microscopic flow separating characteristics, which is mainly composed of linear/nonlinear aerodynamic derivative model, integrated models, differential models, aerodynamic incremental model and angular rate model. The intelligent methods are represented by fuzzy logic, support vector machines and shallow / deep neural network models, all of which are proposed by training the sample data based on various intelligence algorithms. Compared to the conventional aerodynamic models, these intelligent models have strong generalization ability and high predication efficiency. However, they are poorly interpretable due to the lack of physical basis on the dynamic flow fields. In general, the future unsteady aerodynamic models should be developed by focusing on the intelligently characterization of physical meaning of the nonlinear dynamic flow fields to improve the predication accuracy and efficiency on the complex aerodynamic forces/moments, and the applications in aircraft design and flight dynamics.

Hidden object detection has attracted a lot of attention recently due to its importance in security surveillance and other real-world applications. It is considered one of the most challenging tasks in computer vision. Thanks to deep learning for playing a significant role in the rapid technical evolution in this field over the past decade. This article presents a roadmap of hidden object detection, starting from its insightful evolution in 1984, and extensively reviews the technical evolution and shifts in detection approaches. To the best of our knowledge, this is the first ever review work carried out in this field. Various aspects related to hidden object detection have been discussed, including basic building blocks of the detection system, historical milestone detectors, detection datasets, challenges, pre-processing techniques, modern state-of-the-art detection frameworks, and the various evaluation metrics used to assess the detection performance. Towards the end, the paper emphasizes on some unanswered research concerns and possible future prospects in the field of hidden object detection. This review paper aims to serve as a valuable resource for researchers, practitioners, and enthusiasts seeking a thorough understanding of the concepts, advancements, and challenges in this dynamic area of computer vision as hidden object detection continues to have an impact on a variety of interdisciplinary fields of research.

Generative Adversarial Networks (GANs) constitute an advanced category of deep learning models that have significantly transformed the domain of generative modelling. They demonstrate a profound capability to produce realistic and high-quality synthetic data across various domains. Recently, GANs have also emerged as a powerful and innovative approach in medical image processing. Numerous scholarly investigations consistently highlight the superiority of GAN-based methodologies in this context. The generation of realistic synthetic images aims to advance segmentation precision, augment image quality, and facilitate multimodal analysis. These enhancements significantly bolster the analytical capabilities of medical professionals, leading to more precise diagnostic evaluations and the formulation of personalized treatment plans, thereby contributing to improved patient prognosis. In this work, we rigorously review the latest advancements in the application of Generative Adversarial Networks (GANs) within the domain of medical imaging, encompassing research published between 2018 and 2024. The corpus of literature selected for this review is derived from the most relevant and authoritative databases, including Elsevier, Springer, IEEE Xplore, and Google Scholar, among others. This review rigorously evaluates scholarly publications employing Generative Adversarial Networks (GANs) for the synthesis and generation of medical images, segmentation of medical imaging data, image-to-image translation in medical contexts, and denoising or reconstruction of medical imagery. The findings of this review present a thorough synthesis of contemporary applications of Generative Adversarial Networks (GANs) in the domain of medical imaging. This investigation serves as a prospective reference in the realm of GAN utilization for medical image processing, offering guidance and insights for current and future research endeavors.

Due to the absence of definitive treatment for Alzheimer’s disease (AD), slowing its development is essential. Accurately predicting the conversion of mild cognitive impairment (MCI) -a potential early stage of AD- to AD is challenging due to the subtle distinctions between individuals who will develop AD and those who will not. As an increasing body of evidence in the literature suggests, advanced magnetic resonance imaging (MRI) scans, coupled with high-performance computing techniques and novel deep learning techniques, have revolutionized the ability to predict MCI to AD conversion. This study systematically reviewed the publications from 2013 to 2023 (July) to investigate the contribution of deep learning in predicting the MCI conversion to AD, concentrating on the MRI data (structural or functional) and clinical information. The search conducted across seven different databases yielded a total of 2273 studies. Out of these, 78 relevant studies were included, which were thoroughly reviewed, and their essential details and findings were extracted. Furthermore, this study comprehensively explores the challenges associated with predicting the conversion from MCI to AD using deep learning methods with MRI data. Also, it identifies potential solutions to address these challenges. The research field of predicting MCI to AD conversion from MRI data using deep learning techniques is constantly evolving. There is an increasing focus on employing explainable approaches to improve transparency in the analysis process. The paper concludes with an overview of future perspectives and recommends conducting further studies in MCI to AD conversion prediction using deep learning methods.

Renewable energy forecasting, such as Wind and Solar forecasting, is becoming more critical as the demand for clean energy increases. Thus, it is crucial to enhance the accuracy of wind power predictions to ensure electrical energy system’s efficient, reliable, and safe operation. Research on wind forecasting has increased dramatically over the past 10 years due to the success of Artificial Intelligence (AI) technologies like machine learning and deep learning. Despite their potential, AI approaches are fraught with uncertainties. It remains unclear how certain factors may influence the accuracy of AI algorithm predictions. This study reviews AI applications in Wind energy forecasting, aiming to provide an analysis of (1) AI-based structures and optimizers for Wind forecasting, (2) forecast performance evaluation for Deterministic and Probabilistic techniques.

The large intestine is an important part of the human digestive system and the target of many diseases. The biomechanical properties of the large intestine are closely related to the proper functioning of its mechanical behavior. Therefore, the study of its biomechanical properties can help to better understand the key factors of lesions and provide a theoretical basis for the research and application of disease treatment, artificial anal sphincter, and other related medical devices. Physiologic structure of the large intestine is the basis for the study of its biomechanical properties. Constitutive models are commonly used to describe the biomechanical properties of soft tissues and to provide a mathematical characterization of the effects of biological components on the behavior of materials. The studies and results of biomechanical experiments provide a basis for determining the material parameters and the causes of functional damage to the material. In this paper, the biomechanical properties of the large intestine are investigated from two aspects: active properties and passive properties, and three perspectives: physiological structure, constitutive models, and biomechanical experiments. The paper concludes that their effective combination is an important comprehensive method to study the biomechanical properties of the large intestine. It provides a new perspective for the study of biomechanical properties of the large intestine, and provides a theoretical basis for future related biomechanical studies and the development of related medical devices.

Ant colony optimization (ACO) algorithm is one of the most popular swarm-based algorithms inspired by the behavior of an ant colony to find the shortest path for food. The multi-objective ACO (MOACO) is a modified variant of ACO introduced to deal with multi-objective optimization problems (MOPs). The MOACO is seeking to find a set of solutions that achieve trade-offs between the different objectives, which help the decision-makers select the most appreciated solution according to their preferences. Recently, a large number of MOACO research works have been published in the literature, reaching 384 research papers according to the SCOPUS database. In this review paper, 189 different research works of MOACOs published in only scientific journals are considered. Through this research, researchers will gain insights into the expansion of MOACO, the theoretical foundations of MOPs and the MOACO algorithm, various MOACO variants documented in existing literature will be reviewed, and the specific application domains where MOACO has been implemented will be summarized. The critical discussion of the MOACO advantages and limitations is analyzed to provide better insight into the main research gaps in this domain. Finally, the conclusion and some possible future research directions of MOACO are also given in this work.

Alzheimer’s disease (AD) represents a growing global health concern, emphasizing the urgent need for early detection and intervention strategies. This review article aims to provide a comprehensive analysis of the pivotal role that machine learning (ML) and advanced imaging techniques play in the early identification of AD. The study delves into an extensive array of methodologies, encompassing visual biomarkers, datasets, imaging modalities, and evaluation metrics essential for AD detection. The investigation encompasses diverse ML techniques, starting with pre-processing steps and extending to various data types, feature extractor models, and both conventional and deep learning algorithms. Highlighting the significance of Convolutional Neural Networks, Autoencoders, and Transfer Learning, the review assesses their efficacy in AD diagnosis. The findings underscore the intricate challenges and opportunities within the realm of AD detection. Notably, the integration of ML and imaging methods yields promising results in distinguishing AD patterns from healthy brain states. Robust models and algorithms demonstrated notable accuracy, sensitivity, and specificity when confronted with diverse datasets, thus paving the way for early AD identification. In conclusion, this review advocates for the pivotal role of ML and advanced imaging techniques in revolutionizing AD diagnosis. The amalgamation of these approaches holds immense potential for unveiling AD at its nascent stages, enabling timely therapeutic interventions and personalized patient care. Ultimately, this synthesis of biomedical research signifies a transformative leap towards addressing the pressing global issue of Alzheimer’s disease.

Restoration of damaged artwork is an important task to preserve the culture and history of humankind. Restoration of damaged artwork is a delicate, complex, and irreversible process that requires preserving the artist’s style and semantics while removing the damages from the artwork. Digital restoration of artworks can guide artists in physically restoring artworks. This paper groups the virtual artwork restoration methods into various categories: image processing, machine learning, encoder-decoder neural networks, and generative adversarial network-based methods. This paper discusses and analyses different restoration methods’ underlying merits and demerits. The category-wise review of various artwork restoration methods reveals that the generative adversarial network-based methods have attracted the attention of researchers in recent years for restoring damaged artworks. This paper describes datasets used for training and testing of artwork restoration methods and discusses various metrics used for performance evaluation of the artwork restoration methods. This paper compares the restoration results of various methods quantitatively using performance evaluation metrics and qualitatively using visual inspection of the results. Further, the paper also identifies research gaps, challenges, and future directions for research in this field. The proposed review aims to provide researchers with an important reference for working in the artwork restoration field.