Article
Computer Science, Artificial Intelligence
Mehmet Yamac, Ugur Akpinar, Erdem Sahin, Moncef Gabbouj, Serkan Kiranyaz
Summary: Efforts in compressive sensing (CS) literature can be categorized into finding a measurement matrix that preserves compressed information effectively and finding a reliable reconstruction algorithm. While traditional CS methods use random matrices and iterative optimizations, recent deep learning-based solutions accelerate recovery and improve accuracy. However, jointly learning the entire measurement matrix remains challenging. This work introduces a separable multi-linear learning method for the CS matrix, which improves performance compared to block-wise CS, especially at low measurement rates.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Engineering, Mechanical
Tanmoy Chatterjee, Alexander D. Shaw, Michael I. Friswell, Hamed Haddad Khodaparast
Summary: This study addresses the challenging issue of discovering governing laws for complex nonlinear structural dynamic systems using sparse Bayesian machine learning techniques. Two sparsity promoting ML algorithms based on relevance vector machines are employed and the performance of the proposed methods is validated through numerical examples and real datasets.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Thuong Nguyen Canh, Byeungwoo Jeon
Summary: This work introduces a novel sampling matrix RSRM, aiming to improve sensing and compressing efficiency while maintaining security. RSRM combines the advantages of frame-based and block-based sensing, achieving compressive measurements through random projection of multiple randomly sub-sampled signals, and satisfying the Restricted Isometry Property.
SIGNAL PROCESSING-IMAGE COMMUNICATION
(2021)
Article
Engineering, Electrical & Electronic
Moises J. Castro-Toscano, Julio C. Rodriguez-Quinonez, Oleg Sergiyenko, Wendy Flores-Fuentes, Luis Roberto Ramirez-Hernandez, Daniel Hernandez-Balbuena, Lars Lindner, Raul Rascon
Summary: This paper proposes a technical vision system (TVS) for structural behavior analysis using dynamic laser triangulation and k-Nearest Neighbor (k-NN) machine learning regression algorithm. The system was tested on real structures in controlled laboratory conditions, demonstrating practicality and reproducibility of the experimentation. The TVS prototype proved to be a reliable option for structural health monitoring tasks.
IEEE SENSORS JOURNAL
(2021)
Article
Computer Science, Information Systems
Shih-Wei Hu, Gang-Xuan Lin, Chun-Shien Lu
Summary: A learning-based method termed GPX-ADMM-Net is proposed for solving image compressive sensing problems, achieving high performance and adaptivity to measurement rates and cross-task tasks, such as other image inverse problems.
Article
Construction & Building Technology
Syed Farasat Ali Shah, Bing Chen, Muhammad Zahid, Muhammad Riaz Ahmad
Summary: Alkali activated material (AAM) or geopolymer has become a sustainable alternative to cement due to its low power consumption and greenhouse gas emissions, as well as good mechanical and durability features. However, developing AAM mixtures with desired properties is challenging due to the nature and diversity of available source materials. This study evaluates the performance of various machine learning models for predicting the compressive strength of one-part AAM binder, with XGBoost outperforming other algorithms. The use of SHapley Additive exPlanations (SHAP) helps interpret the predicted compressive strength and evaluate the effects of various parameters. The interpretable ML strategy used in this study aids in the production and performance tuning of durable and sustainable one-part AAMs for widespread applications.
CONSTRUCTION AND BUILDING MATERIALS
(2022)
Article
Engineering, Electrical & Electronic
Pengxia Wu, Julian Cheng
Summary: This paper investigates massive multiple-input multiple-output systems in frequency-division duplex mode. Sparse channel estimation techniques are employed to reduce the overhead of downlink channel state information (CSI) acquisition. The paper proposes novel data-driven solutions to design the measurement matrix, optimizing the channel estimation performance. The data-driven measurement matrices achieve more accurate reconstructions and lower measurement requirements compared to random matrices, leading to a higher achievable rate for CSI acquisition. The hybrid data-driven scheme combining the proposed measurement matrices and conventional sparse reconstruction algorithms achieves higher reconstruction accuracy compared to pure deep learning-based methods.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS
(2022)
Article
Computer Science, Interdisciplinary Applications
Huaming Tian, Yu Wang
Summary: A digital twin is created to continuously learn and improve model prediction in geotechnical projects using actual observation data. The challenges come from the spatial sparsity and spatiotemporal variations of the real geotechnical data. This study proposes a novel data-driven and physics-informed Bayesian learning framework to tackle these challenges and improve the model prediction.
COMPUTERS AND GEOTECHNICS
(2023)
Article
Green & Sustainable Science & Technology
Yushi Tian, Xu Yang, Nianhua Chen, Chunyan Li, Wulin Yang
Summary: This study introduces an innovative data-driven approach using multiple artificial intelligence techniques to enhance polysaccharide production. Machine learning models and data analysis methods are used to predict polysaccharide yield and optimize enzymatic parameter combinations.
ENVIRONMENTAL SCIENCE AND ECOTECHNOLOGY
(2024)
Article
Engineering, Mechanical
Xiaocen Wang, Jian Li, Dingpeng Wang, Xinjing Huang, Lin Liang, Zhifeng Tang, Zheng Fan, Yang Liu
Summary: A sparse UGW imaging algorithm based on compressive sensing and deep learning models is proposed to address the limitation of imaging quality due to the number of transducers in service. By using CS and deep learning models, the method can effectively reconstruct sparse detection signals acquired by a small number of transducers. Experimental and simulation results demonstrate the effectiveness of the proposed method in achieving high-quality imaging with limited transducers.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Engineering, Mechanical
Tongtong Yan, Dong Wang, Meimei Zheng, Changqing Shen, Tangbin Xia, Zhike Peng
Summary: This study proposes a physics-informed learning framework that combines weight-based sparse degradation modeling with entropy-based indicators for online incipient fault detection and diagnosis. The weak fault characteristics can be significantly enhanced by continuously updating the model weights. A family of entropy-based indicators is introduced for machine health monitoring, aiming to quantify the amplified fault characteristics revealed by the updated model weights for online incipient fault detection.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Ljubisa Stankovic
Summary: This paper discusses the uniqueness of signal reconstruction in compressive sensing, proposing a method to relax the coherence index condition and using the orthogonal matching pursuit approach for unique signal reconstruction, while improving the limit of sparsity.
Article
Engineering, Mechanical
Dong Wang, Bingchang Hou, Tongtong Yan, Changqing Shen, Zhike Peng
Summary: This article discusses the construction of a physically interpretable prototypical neural network and its correlation with physically interpretable fault features to support machine health conditions. It emphasizes the importance of physically interpretable weights in machine condition monitoring.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Energy & Fuels
Zhihua Deng, Qihong Chen, Liyan Zhang, Keliang Zhou, Yi Zong, Zhichao Fu, Hao Liu
Summary: The study introduces a data-driven sparse identification based on autoencoder method for establishing the model of fuel cell air supply system, aiming to enhance the precision of control system. The results show that the proposed method exhibits smaller errors under simulation data and real data, with reconstruction results perfectly aligning with the original data, providing new insights for system modeling studies.
Article
Computer Science, Interdisciplinary Applications
Pin Zhang, Zhen-Yu Yin, Brian Sheil
Summary: There is great potential for machine learning to improve constitutive modelling of geomaterials. However, a lack of interpretability and heavy reliance on big data has been a common criticism. This study proposes an interpretable data-driven approach for geotechnical modelling, incorporating prior knowledge and uncertainty. By adopting a multi-fidelity modelling framework, the impact of small datasets can be maximized. The results show that data-driven modelling with physical constraints performs robustly, even for extrapolation beyond the original dataset.
COMPUTERS AND GEOTECHNICS
(2023)
Article
Engineering, Multidisciplinary
Yuequan Bao, Zhiyi Tang, Hui Li, Yufeng Zhang
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2019)
Article
Construction & Building Technology
Zhiyi Tang, Zhicheng Chen, Yuequan Bao, Hui Li
STRUCTURAL CONTROL & HEALTH MONITORING
(2019)
Article
Engineering, Multidisciplinary
Yuequan Bao, Zhicheng Chen, Shiyin Wei, Yang Xu, Zhiyi Tang, Hui Li
Article
Engineering, Mechanical
Zhicheng Chen, Zhiyi Tang, Jiahui Chen, Hui Li
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2020)
Article
Engineering, Industrial
Zhengliang Xiang, Yuequan Bao, Zhiyi Tang, Hui Li
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2020)
Article
Engineering, Multidisciplinary
Zhiyi Tang, Yuequan Bao, Hui Li
Summary: In structural health monitoring, data quality is crucial for tasks such as structural damage identification. A convolutional neural network-based data recovery method is proposed to simultaneously recover multi-channel data and maximize the use of interrupted information. The method achieves good recovery results on synthetic data, field-test data, and seismic response monitoring data.
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2021)
Article
Construction & Building Technology
Dawei Liu, Zhiyi Tang, Yuequan Bao, Hui Li
Summary: This study proposes a machine-learning-based approach to identify the modal parameters of output-only data for structural health monitoring, utilizing the independence of modal responses and the principles of unsupervised learning. The designed neural network uses a complex loss function to extract modal responses from structural vibration data and constrain the training process. The approach is able to blindly extract modal information from system responses, as demonstrated through numerical examples and verification with actual dataset from a cable-stayed bridge.
STRUCTURAL CONTROL & HEALTH MONITORING
(2021)
Article
Chemistry, Multidisciplinary
Jiaxing Guo, Zhiyi Tang, Changxing Zhang, Wei Xu, Yonghong Wu
Summary: Structural health monitoring systems can continuously monitor the operational state of structures, generating a large amount of monitoring data. Identifying extreme events in the presence of faulty data is challenging. This study proposes a deep learning-based method with visual interpretability to identify seismic data under sensor faults interference, which effectively identifies seismic data mixed with various types of faulty data while providing good interpretability.
APPLIED SCIENCES-BASEL
(2023)