(429b) Woodchip Moisture Content Estimation Using Short-Range Iot Wi-Fi for the Pulp & Paper Industry | AIChE

(429b) Woodchip Moisture Content Estimation Using Short-Range Iot Wi-Fi for the Pulp & Paper Industry


Suthar, K. - Presenter, Auburn University
He, Q. P., Auburn University
Jiang, Z., AC-PABE
Wang, J., Auburn University
The pulp and paper industry is the third-largest consumer of energy in the US industrial sector thereby leading to tremendous opportunities to improve its energy efficiency and productivity. The pulping process, which converts woodchips into pulp by displacing lignin from cellulose fibers, is one of the most important operations in a pulp and paper mill. For the pulping process, it is important to have real-time knowledge about the moisture content (MC) of the woodchips so that the process can be optimized and/or controlled correspondingly to achieve satisfactory product quality while minimizing the consumption of energy and chemicals. There have been attempts to develop real-time woodchip MC sensors, such as capacitance, resistance, microwave and NIR-based sensors. But these techniques have not been adopted by the industry due to unsatisfactory performance and/or high maintenance requirements that can hardly be met in a manufacturing environment. Also, some of the accurate methods are often lab-based that cannot be implemented online because of time constraint [1]. Due to these limitations, at present, the moisture level is measured only a few times throughout the year in the pulp and paper industry.

To address these limitations, we propose a non-destructive, economic, and robust approach using industrial Internet-of-Things (IIoT) based Wi-Fi and employ channel state information (CSI) to estimate MC in woodchips. We extract CSI by modifying the open-source device drivers for Intel Wi-Fi link 5300 Network Interface Card based on CSITool [2]. CSI contains information about the channel in the form of individual data sub-carriers capturing indoor channel features such as the effect of scattering, fading, and power decay with distance. We hypothesize that when a Wi-Fi signal passes through wood chips, the water or MC in woodchips would cause a corresponding change in the Wi-Fi signal which is then captured in CSI. The goal is to develop a data-driven model to extract the relationship between the change in CSI and the woodchip MC. CSI has been successfully used for moisture classification in wheat [3]. However, compared to wheat moisture classification, woodchip moisture estimation is much more challenging due to the significant heterogeneity in the shape, structure, and arrangement of the woodchips. For example, the woodchip arrangement in the container is expected to have a significant impact on the CSI data. In addition, CSI data have only been used in classifications, which is much easier than regression modeling.

In this work, we propose a regression-based approach to estimate the moisture content in wood chips. To the best of our knowledge, this is the first work to address the continuous estimation of MC using CSI data. Our goal is to develop a robust model that is insensitive to the heterogeneity of the woodchips, particularly the confounding factors such as woodchip shape and the effect of shuffling, to achieve a consistent and accurate MC estimation. While the IIOT devices used are small, low-cost, and rugged to stand for harsh environment, they do have their limitations such as the raw CSI data are very noisy, which can lead to low performing models when directly fed to machine learning algorithms. To address this challenge, our previously developed statistics pattern analysis (SPA) framework [4][5] is used to extract robust and predictive features from raw CSI data. The SPA-based features are then used to develop robust machine learning models for continuous MC estimation in woodchips. In other words, SPA turns messy big data into so-called smart data, which can then be used for accurate sensing. The machine learning techniques studied in this work include artificial neural networks (ANN), support vector machines (SVM), and Gaussian process regression (GPR). We conducted systematic study of these nonlinear regression approaches, and demonstrated that, when combined with SPA, provide a potential solution for real-time woodchip MC estimation for the pulp and paper industry. Specifically, the experimental validation results show highly accurate prediction of more than 20 MC levels in a wide range. In addition to the excellent performance, the proposed approach is also economic and non-destructive, making it a potential technology for industrial implementation in the pulp and paper industry.

[1] R. Govett, A PRACTICAL GUIDE FOR THE DETERMINATION OF MOISTURE CONTENT OF WOODY BIOMASS A Practical Handbook of Basic Information, Definitions, Calculations, Practices and Procedures for Purchasers and Suppliers of Woody Biomass.

[2] D. Halperin, W. Hu, A. Sheth, and D. Wetherall, Tool Release: Gathering 802.11n Traces with Channel State Information, ACM SIGCOMM Comput. Commun. Rev., 2011.

[3] W. Yang, X. Wang, S. Cao, H. Wang, and S. Mao, Multi-class wheat moisture detection with 5GHz Wi-Fi: A deep LSTM approach, in Proceedings - International Conference on Computer Communications and Networks, ICCCN, 2018.

[4] Q. P. He and J. Wang, Statistics pattern analysis: A new process monitoring framework and its application to semiconductor batch processes, AIChE J., 2011.

[5] J. Wang and Q. P. He, Multivariate Statistical Process Monitoring Based on Statistics Pattern Analysis, Ind. Eng. Chem. Res., 2010.