Article
Physics, Applied
Yi Liu, Mingkai Zheng, Kaixin Liu, Yuan Yao, Stefano Sfarra
Summary: In this paper, a thermographic data analysis method named TriMap thermography with convolutional autoencoder (CAE) is proposed to improve defect detectability. The method uses CAE to reduce noise and enhance the quality of thermograms, and then applies the TriMap algorithm to extract features from the enhanced data. The effectiveness of the proposed method is validated through testing on a carbon fiber-reinforced polymer specimen.
JOURNAL OF APPLIED PHYSICS
(2022)
Article
Polymer Science
Kaixin Liu, Fumin Wang, Yuxiang He, Yi Liu, Jianguo Yang, Yuan Yao
Summary: In this article, a novel generative manifold learning thermography (GMLT) method is proposed for defect detection and evaluation of composites. The method utilizes spectral normalized generative adversarial networks as an image augmentation strategy, to learn the thermal image distribution and generate virtual images to enrich the dataset. Manifold learning is employed for unsupervised dimensionality reduction, and partial least squares regression is used for defect visualization. Probability density maps and quantitative metrics are introduced to evaluate and explain the defect detection performance. Experimental results demonstrate the superiority of GMLT compared to other methods.
Article
Automation & Control Systems
Kaixin Liu, Mingkai Zheng, Yi Liu, Jianguo Yang, Yuan Yao
Summary: In this article, a deep autoencoder thermography (DAT) method is proposed for detecting subsurface defects in composite materials. The multilayer network structure of DAT can handle nonlinear temperature profiles, and the output of the intermediate hidden layer is visualized to highlight defects. The layer-by-layer feature visualization reveals how the model extracts defect features.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Engineering, Chemical
Mingwei Jia, Junhao Hu, Yi Liu, Zengliang Gao, Yuan Yao
Summary: Faults in the process industry can be diagnosed using various data-driven methods, but little attention has been given to the physical consistency of model prediction logic. To address this, we propose a graph learning fault diagnosis framework that combines graphs with process physics. Our framework focuses on knowledge embedding and explanation, and includes components such as a topology graph, self-attention mechanism, graph convolution, graph pooling, and a gating mechanism. We also use a graph explainer to assess the physical consistency of the model's prediction logic. The feasibility of our method is demonstrated using the Tennessee Eastman process.
INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH
(2023)
Article
Automation & Control Systems
Mingwei Jia, Danya Xu, Tao Yang, Yi Liu, Yuan Yao
Summary: In this study, a soft sensor based on a graph convolutional network is developed to model the nonlinear time-varying characteristics of the process industry. The focus is on obtaining localized spatial-temporal correlations to understand the intricate interactions among variables. The model is trained with regularization terms and learns distinctive localized spatial-temporal correlations in an end-to-end manner, capturing both the localized spatial-temporal correlations and time-series properties.
JOURNAL OF PROCESS CONTROL
(2023)
Article
Chemistry, Analytical
Yi Liu, Yuxin Jiang, Zengliang Gao, Kaixin Liu, Yuan Yao
Summary: To ensure safe and efficient operation of packed columns, we proposed a machine vision approach using a convolutional neural network (CNN) to detect flooding in real time. Real-time images of the packed column were captured and analyzed with a CNN model trained on a dataset of recorded images. The results showed that the proposed method provides a real-time pre-alarm approach for detecting flooding events.
Article
Materials Science, Multidisciplinary
Yi-Ting Tsai, Yu-Kai Huang, Zhen-Feng Jiang, Yuan Yao, Pei-Hsuan Lo, Yu-Chiang Chao, Bi-Hsuan Lin, Chun Che Lin
Summary: This study modified the partial inversion of a spinel structure through cation substitution, resulting in the creation of MgGa2O4 with a local structure. Increasing the concentration of Sn led to a decrease in the vibrational mode of GaO4 and a retention of the vibrational mode of MgO4. The long SWIR emission achieved through Sn substitution enabled the development of an SWIR light-emitting diode that improved the accuracy of an artificial intelligence based image recognition system.
ACS MATERIALS LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Yun Dai, Chao Yang, Yi Liu, Yuan Yao
Summary: This study develops a sample selection strategy for active learning called latent-enhanced variational adversarial AL (LVAAL) to improve quality prediction performance with limited labeled data. The LVAAL method uses a minimax game to explore the latent representation of the original data and deceives the adversarial network into predicting all samples as labeled. Then, Gaussian process regression (GPR) is used for prediction. The experimental results show that LVAAL outperforms existing AL strategies.
IEEE SENSORS JOURNAL
(2023)
Article
Automation & Control Systems
Jian-Guo Wang, Rui Chen, Xiang-Yun Ye, Yuan Yao, Zhong-Tao Xie, Shi-Wei Ma, Li-Lan Liu
Summary: This study proposes a novel root cause diagnosis framework based on Granger causality analysis, which utilizes both normal data and disturbance data to obtain compact and useful disturbance-related information.
JOURNAL OF PROCESS CONTROL
(2023)
Article
Engineering, Chemical
Jian-Guo Wang, Rui Chen, Jing-Ru Su, Hui-Min Shao, Yuan Yao, Shi-Wei Ma, Li-Lan Liu
Summary: In this study, the use of kernel multivariate Granger causality (KMGC) test as an analytical tool for diagnosing the root cause of plant-wide oscillations is proposed. The kernel function allows for easier processing of nonlinear data by mapping it to a higher-dimensional space. The combination of KMGC with a fuzzification function, creating fuzzy kernel multivariate Granger causality (FKMGC), improves computational efficiency and accurately identifies the root cause of oscillations. FKMGC reduces analysis time by 95% compared to Gaussian kernel-based algorithms.
JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS
(2023)
Article
Instruments & Instrumentation
Yi Liu, Fumin Wang, Kaixin Liu, Miranda Mostacci, Yuan Yao, Stefano Sfarra
Summary: Infrared thermography is a cost-effective non-destructive evaluation technique that is critical for detecting defects in cultural heritage. However, there is a lack of in-depth studies on using infrared thermography for contemporary artworks. This study proposes a deep convolutional autoencoder thermography method for defect detection in contemporary artworks, which significantly improves accuracy compared to commonly used methods, as shown in a case study on a handmade replica of Picasso's 'La Bouteille de Suze'.
QUANTITATIVE INFRARED THERMOGRAPHY JOURNAL
(2023)
Article
Chemistry, Analytical
Yi Liu, Fumin Wang, Zhili Jiang, Stefano Sfarra, Kaixin Liu, Yuan Yao
Summary: Infrared thermography is a widely used nondestructive testing technique for artwork inspection. However, raw thermograms often have limitations in quantity and background noise. To overcome these challenges, this study proposes a defect inspection method for artwork based on principal component analysis, incorporating two deep learning approaches: spectral normalized generative adversarial network (SNGAN) and convolutional autoencoder (CAE). The SNGAN strategy focuses on augmenting thermal images, while the CAE strategy emphasizes enhancing their quality. Principal component thermography (PCT) is then used to analyze the processed data and improve defect detectability. The integration of the SNGAN strategy led to a 1.08% enhancement in signal-to-noise ratio, while the utilization of the CAE strategy resulted in an 8.73% improvement.
Article
Engineering, Electrical & Electronic
Junhua Zheng, Yangxuan Liu, Yi Liu, Beiping Hou, Yuan Yao, Le Zhou
Summary: This article investigates the problem of modeling regression with labeled and unlabeled data samples commonly found in industrial processes. By incorporating additional information on unlabeled data samples, a new semi-supervised latent factor analysis model is developed. The semi-supervised model can efficiently extract useful information for improving regression performance compared to purely supervised regression models. The proposed basic semi-supervised model has also been extended to a mixture form, capable of describing data from more complex processes. Two soft sensors are constructed based on the semi-supervised models for predictive modeling of key/mass variables. Three case studies are used to evaluate the performance of the proposed soft sensing methods.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Information Systems
Wei Liu, Beiping Hou, Yuan Yao, Le Zhou
Summary: A novel methodology combining signal filtering and feature extraction was proposed to enhance detection of subsurface defects in carbon fiber reinforced polymer materials. The DoGC-SPCT method showed superior performance in defect identification, signal to noise ratios for defects, and interpretability compared to other feature extraction methods in experiments on sample laminates.
Article
Engineering, Electrical & Electronic
Kaixin Liu, Qing Yu, Yi Liu, Jianguo Yang, Yuan Yao
Summary: This study proposed a convolutional graph thermography (CGT) method for subsurface defect detection in polymer composites, reducing noise and inhomogeneous background of thermal images for feature extraction to highlight defects visually.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
