Characterization of biomass in Nepal used for energy purpose using near infrared spectroscopy

การกำหนดลักษณะของชีวมวลในประเทศเนปาลซึ่งใช้เพื่อวัตถุประสงค์ด้านพลังงานโดยใช้สเปกโทรสโกปีอินฟราเรดย่านใกล้

Name: Bijendra Shrestha

LCSH: Near infrared spectroscopy

LCSH: Spectrum analysis

Abstract: Biomass, a carbon-neutral and readily available renewable resource, offers a sustainable alternative to fossil fuels, contributing to energy savings and environmental protection. It accounts for two-thirds of global renewable energy utilization, with applications ranging from household cooking and heating to electricity generation, transportation, and industrial use. Despite its widespread adoption, inefficient consumption of traditional biomass remains prevalent in both household and industrial settings. Additionally, biomass is commonly traded by volume or mass rather than its actual energy properties. Consequently, rapid, reliable, and non-destructive evaluation of biomass, tailored to its specific energy characteristics, and effective management and utilization through appropriate energy technologies are crucial for sustainably meeting the energy demands of rural and industrial sectors, as well as traders.This thesis develops, compares, and selects the best-performing model based on partial least squares regression (PLSR) for predicting the energy properties of fast-growing trees and agricultural residues in Nepal. The study employs Fourier Transform (FT) near-infrared spectroscopy (NIRS) within a wavenumber range of 3595 to 12,489 cm-1 as an alternative method. The target energy properties include higher heating value (HHV) obtained from a bomb calorimeter, ultimate analysis parameters (carbon (C), hydrogen (H), nitrogen (N), and oxygen (O) in weight percentages (wt.%)) from a CHNO/S elemental analyzer, and proximate analysis parameters (moisture content (MC), volatile matter (VM), fixed carbon (FC), and ash content (A) in percentage (%)) from a thermogravimetric analyzer. Additionally, the study aims to predict combustion characteristics, such as the ignition index (Di), burnout index (Di), combustion performance index (Si), and flammability index (Ci), using data from a thermogravimetric analyzer. This study also seeks to demonstrate that the developed models are ready for industrial application, particularly in terms of HHV Different PLSR models were developed using various preprocessing approaches: no preprocessing, traditional preprocessing, multi-preprocessing in 5-range and 3-range techniques. A genetic algorithm (GA), and a successive projection algorithm (SPA) were used for featured wavenumber selection. A novel multi-preprocessing approach was employed, assuming that using multiple preprocessing methods across the entire NIR wavenumber range would enhance model performance. Furthermore, the study analyzed the impact of different biomass types, specifically wood and non-wood species, on model development through scatter plot analysis. This investigation, the first of its kind, seeks to validate the readiness of the developed models for industrial application, especially in terms of HHV. In this study, the best model was selected based on its high coefficient of determination in the validation set (R2p), low root mean square error of prediction (RMSEP), and the high ratio of prediction to deviation (RPD). The highest-performing model for HHV was derived from ground biomass using GA-PLSR with the first derivative, achieving an R2p of 0.9574, RMSEP of 170.3282 J/g, RPD of 4.9, indicating suitability for a wide range of applications including research. For elemental composition, the best models for c and FI were developed from ground biomass using GA-PLSR, achieving Revalues of 0.7217 and 0.7678, RMSEP of 170.3282 wt.% and 0.1434 wt.%, and RPD of 1.9 and 2.1, respectively; however, these models are primarily useful only for rough screening. The N model, developed using MP-PLSR with a 5 range technique from ground biomass, showed an R2p of 0.8410, RMSEP of 0.0973 wt.%, and RPD of 2.7, making it reliable for broader applications but should be used with caution. In contrast, the model for o from chipped biomass using the same MP-PLSR technique yielded R2p, RMSEP, and RPD values of 0.7150, 1.3088 wt.%, and 1.9 respectively, making it suitable only for preliminary screening. For proximate analysis parameters, the models for MC and FC in chip biomass show satisfactory performance, making them cautiously applicable in various applications, including research The optimal models for MC and FC, constructed using GA-PLSR with the second derivative and Full-PLSR with a constant offset, achieved high R2p values of 0.8654 and 0.8773, low RMSEP values of 0.85% and 2.12%, and high RPD values of 2.9 and 3.0, respectively, indicating their applicative capabilities. In contrast, for VM, the best model was derived using MP-PLSR with a 5-range technique from chip biomass, yielding an R2p of 0.7937, an RMSEP of 2.47%, and an RPD of 2.2, making it suitable only for rough screening. Similarly, the best model for ash content was obtained from ground biomass using Full-PLSR with the first derivative, resulting in an R2p of 0.7983, an RMSEP of 1.26%, and an RPD of 2.2, indicating suitability for rough screening as well. For combustion performance indices (Di, Df, Si, and Ci), the best models were all derived using ground biomass, although with varying preprocessing techniques. The model for Di, developed using MP-PLSR with a 5-range technique, achieved an R2p of 0.6782, an RMSEP of 0.3879 wt.%.min-4, and an RPD of 1.8, suitable only for rough screening. In contrast, the Df model, utilizing GA-PLSR with the first derivative, secured an R2p of 0.8426, an RMSEP of 0.4968 wt.%.min-4, and an RPD of 2.5, making it appropriate for broader applications, including research. For Si, the optimal model used Full-PLSR with MSC, achieving an R2p of 0.8808, an RMSEP of 0.1566 wt.%2.min'2.℃'3, and an RPD of 3.1, rendering it suitable for most applications, including research. Lastly, for Ci, the optimal model was obtained using MP-PLSR with a 5-range technique, resulting in an R2p of 0.7204, an RMSEP of 0.3610 wt.%.min-1.°C-2, and an RPD of 1.8, making it applicable only for rough screening. Based on model performance, the selected PLSR-based model-derived from ground biomass for FHHV (J/g), wt.% N, Df, Si, and from chip biomass for MC and FC—offers a rapid, reliable, and non-destructive alternative method for assessing biomass properties for energy usage using NIRS. This model is suitable for use in biomass trading and is ready for industrial application. Flowever, to develop a more comprehensive global model, continuous enhancement is necessary. This involves incorporating a larger number of representative biomass samples, consistently validating with unknown samples, and exploring more suitable machine learning algorithms to ensure accurate predictions of biomass energy properties across all parameters .

King Mongkut's Institute of Technology Ladkrabang. Central Library

Address: BANGKOK

Email: Lifelong@kmitl.ac.th

Name: Panmanas Sirisomboon

Role: Thesis Advisor

Name: Jetsada Posom

Role: Thesis Co-Advisor

Email : panmanas.si@kmitl.ac.th

Name: Prasad Shrestha

Role: Thesis Co-Advisor

Created: 2024

Modified: 2024-11-08

Issued: 2024-11-08

วิทยานิพนธ์/Thesis

application/pdf

eng

DegreeName: Doctor of Engineering

Level: Doctoral Degree

Descipline: Food and Agricultural Intelligence Engineering

Grantor: King Mongkut's Institute of Technology Ladkrabang

©copyrights King Mongkut's Institute of Technology Ladkrabang

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	Bijendra Shrestha.pdf	10.25 MB	1	2025-01-09 09:39:31

ใช้เวลา

0.023422 วินาที