Article
Engineering, Electrical & Electronic
Chun Fai Lui, Yiqi Liu, Min Xie
Summary: This article proposes a data-driven dynamic soft sensor modeling method based on SBiLSTM structure, which extracts and utilizes nonlinear dynamic latent information from both process variables and quality variables to achieve efficient quality estimation.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Automation & Control Systems
Qiao Liu, Mingwei Jia, Zengliang Gao, Liangfeng Xu, Yi Liu
Summary: One typical challenge in constructing accurate soft sensors in the process industries is the presence of various noise and outliers in industrial process data. Inspired by the effectiveness of correntropy in tackling non-Gaussian noise, this study proposes a maximum correntropy criterion-based LSTM neural network, MCC-LSTM, to develop a reliable soft sensor model for quality prediction. By adopting the objective function centered on a Gaussian kernel, the MCC-LSTM assigns relatively smaller weights to outliers automatically, reducing their negative effects on the prediction and improving the performance for modeling process data with uncertainties.
CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS
(2022)
Article
Chemistry, Multidisciplinary
Zeyu Yang, Ruining Jia, Peiliang Wang, Le Yao, Bingbing Shen
Summary: Soft sensors are mathematical methods that describe the dependence of primary variables on secondary variables. This paper proposes a novel supervised attention-based bidirectional long short-term memory (SA-BiLSTM) model to handle nonlinear industrial process modeling with dynamic features. The SA-BiLSTM model introduces an attention mechanism to calculate the correlation between hidden features in each time step, thus avoiding the loss of important information. Additionally, this approach combines historical quality information and a moving window through a supervised strategy of quality variables to enhance the model's learning efficiency and prediction performance. Two real industrial examples demonstrate the superiority of the proposed method compared to conventional methods.
Article
Mathematics, Interdisciplinary Applications
Eunice Leung, King F. Ma, Nan Xie
Summary: In this paper, the bubble sound of different types of sparkling drinks is collected and a data driven model for bubble dynamics is constructed using experimental data. It is verified that the bubble dynamics of sparkling drinks are nonlinear and chaotic. The proposed PI-LSTM neural network, guided by the derived physical principle on bubble pressure, shows improved accuracy in modeling the bubble sound data compared to standard predictors.
CHAOS SOLITONS & FRACTALS
(2023)
Article
Computer Science, Artificial Intelligence
Paulo H. Marrocos, Igor G. I. Iwakiri, Marcio A. F. Martins, Alirio E. Rodrigues, Jose M. Loureiro, Ana M. Ribeiro, Idelfonso B. R. Nogueira
Summary: This article discusses the issue of evaluating unmeasurable quantities, particularly in the pharmaceutical production field. It proposes a new Deep Artificial Intelligence structure called the Improved Quasi-Virtual Analyzer, which serves as an online soft sensor to provide information about the main properties of a Simulated Moving Bed chromatographic unit. The structure demonstrates robust predictive capabilities and outperforms traditional artificial network structures.
APPLIED SOFT COMPUTING
(2022)
Article
Engineering, Environmental
Boyan Yu, Ching Kwek Pooi, Kar Ming Tan, Shujuan Huang, Xueqing Shi, How Yong Ng
Summary: Commercial instrumentation for measurement of various wastewater treatment processes parameters is costly and time-consuming. This study developed LSTM-based soft-sensors to monitor and forecast crucial performance parameters in a two-staged A/O process for wastewater treatment, using lower-cost sensors dataset. The proposed LSTM-based soft-sensors outperformed MLR-based soft-sensors in prediction accuracy, indicating their potential to lower the capital cost for sensor installation in WWTPs.
JOURNAL OF WATER PROCESS ENGINEERING
(2023)
Article
Automation & Control Systems
Xuan Hu, Qianqian Yu, Yongming Han, Zhiwei Chen, Zhiqiang Geng
Summary: A novel complex-valued long short-term memory (CVLSTM) integrating the variational mode decomposition (VMD) (CVLSTM-VMD) is proposed for soft sensor, which can achieve high prediction accuracies in industrial process data. The CVLSTM-VMD is applied in the polypropylene and the PTA processes, and the results show its effectiveness. This model can play a vital role in product quality control and new product development.
JOURNAL OF PROCESS CONTROL
(2023)
Article
Engineering, Manufacturing
Mahvash A. Moradi, Sayed Alireza Sadrossadat, Vali Derhami
Summary: This article presents a new macromodeling approach for nonlinear electronic components and circuits based on LSTM neural network, which offers a more efficient training process compared to conventional RNN. Numerical results and practical examples demonstrate that the proposed method is more effective in modeling components and packages than traditional techniques.
IEEE TRANSACTIONS ON COMPONENTS PACKAGING AND MANUFACTURING TECHNOLOGY
(2021)
Article
Computer Science, Interdisciplinary Applications
Tao Li, Teng Wu
Summary: In this study, a long short-term memory (LSTM) network is used as a reduced-order model for the nonlinear aerodynamic forces on a bridge deck. The LSTM network is employed to generate force inputs to a three-dimensional finite element model (FEM) of the long-span bridge. Both general knowledge and domain knowledge are leveraged to customize the LSTM cell and network architecture, and a fast-training algorithm is developed to improve the training efficiency. The computational efficiency is further improved by adopting convolution-based numerical integration in the FEM modeling.
COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING
(2023)
Article
Engineering, Civil
Hanlin Yin, Fandu Wang, Xiuwei Zhang, Yanning Zhang, Jiaojiao Chen, Runliang Xia, Jin Jin
Summary: This paper proposes a novel LSTM-SS framework for rainfall-runoff modeling, which utilizes sequential information and shows superior performance on a large-scale dataset compared to benchmark models.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Zekun Xu, Jun Chen, Jiaxu Shen, Mengjie Xiang
Summary: This paper proposed a recursive LSTM network for predicting nonlinear structural seismic responses with lower computational cost compared to traditional methods, demonstrating good accuracy and generalization capability.
ENGINEERING STRUCTURES
(2022)
Article
Optics
Thomas Adler, Manuel Erhard, Mario Krenn, Johannes Brandstetter, Johannes Kofler, Sepp Hochreiter
Summary: This study demonstrates how machine learning can model experiments in quantum physics, especially focusing on complex quantum states with multiple particles. The research shows that machine learning models can provide significant improvement over random search, particularly by using LSTM neural networks to successfully predict output state characteristics for quantum setups.
Article
Automation & Control Systems
Ping Wang, Yichao Yin, Xiaogang Deng, Yingchun Bo, Weiming Shao
Summary: This study proposes a semi-supervised ESN method assisted by temporal-spatial graph regularization for constructing soft sensor models. The method integrates labeled and unlabeled samples to enhance the performance of ESN models and incorporates a temporal-spatial graph to capture the dynamic characteristics of the data.
Article
Automation & Control Systems
Wei Xie, Jiesheng Wang, Cheng Xing, Shasha Guo, Mengwei Guo, Lingfeng Zhu
Summary: The article proposes a method for soft-sensor modeling using variational autoencoder bidirectional LSTM based on batch training to address the issue of critical data information being discarded by LSTM during training. The effectiveness of the proposed method is verified through simulation experiments on the grinding and classifying process, with L2 regularization introduced to prevent overfitting.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Automation & Control Systems
Chao Sun, Yuxuan Zhang, Gaolu Huang, Lin Liu, Xiaochen Hao
Summary: A soft sensor model based on L/S-ConvGRU is proposed to predict the specific surface area of cement. The model introduces parameters L and S to change the linear constraint relationship and designs long-term memory enhancement pathway and short-term dependence pathway for feature extraction. Experimental results show that the model has high precision and better generalization capability.
