KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 50

Some of the metrics are blocked by your
consent settings
A machine learning approach for predicting osmotic coefficients and deriving activity coefficients in alkyl ammonium salts
(2026-12-01)
Chawuthai, R.
;
Murathathunyaluk, S.
;
Saengsuradech, S.
;
Nukaew, A.
;
Simasatitkul, L.
Quaternary Ammonium Salts (Quats) have diverse applications across various domains. They are extensively used as phase-transfer catalysts (PTCs) in chemical reactions, facilitating the transfer of reactants between aqueous and organic phases. Their unique structure enables the formation of ion pairs, enhancing reaction rates at phase boundaries. This research develops a novel method for predicting Quats’ osmotic coefficients using Simplified Molecular Input Line Entry System (SMILES) notation and supervised machine learning. A comprehensive dataset of 1,654 data points from 52 distinct Quats was compiled. The structural characteristics were encoded using SMILES notation. The data was evaluated using random splitting and Leave-One-Group-Out (LOGO) validation to train seven machine learning algorithms. Gaussian Process (GP) emerged as the optimal algorithm. The GP model achieved a mean absolute percentage error (MAPE) of 5.29% and root mean square error (RMSE) of 0.034. Comparisons with Electrolyte-NRTL and Extended UNIQUAC models demonstrate that this data-driven approach offers competitive accuracy while enabling generalization to structurally similar compounds. This work marks a significant starting point for the machine learning-enhanced prediction of activity coefficients, with considerable potential for future refinement and application.
Some of the metrics are blocked by your
consent settings
1H NMR-based machine learning methods for rapid authentication and composition profiling of crude palm oil
(2026-09-01)
Nggofur, Abdul
;
Sueviriyapan, Natthapong
;
Nuntawong, Noppadon
;
Sutthiumporn, Ketsada
;
Sooknoi, Tawan
A rapid analytical workflow for determining geographical origin and predicting fatty acid composition of crude palm oil (CPO) was developed using 1H NMR, GC-FID, and machine learning. Analyzing CPO samples from Indonesia, Malaysia, the Philippines, and Thailand using unsupervised fingerprinting with principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), and uniform manifold approximation and projection (UMAP) revealed partial origin-based grouping. Supervised classification, validated via leave-one-out cross-validation (LOOCV) and uncertainty quantification (UQ), reliably discriminated the origins above random chance. Additionally, partial least squares regression (PLSR) accurately predicted oleic, linoleic and myristic acid levels measured by GC-FID, whereas the accuracy decreased for lauric, stearic and palmitic acids. PLSR reliability was rigorously validated using latent variable selection and permutation testing to rule out random correlations. Overall, this integrated 1H NMR and machine learning approach offers a rapid tool for CPO geographical traceability and compositional evaluation, demonstrating its potential for industrial quality control.
Some of the metrics are blocked by your
consent settings
Real-time interpretable and cluster-stratified lightGBM framework for high-precision concrete strength prediction and instantaneous mixture optimization
(2026-08-29)
Elsheikh, Ahmed
;
Hematibahar, Mohammad
;
Jueyendah, Sebghatullah
;
Aljarah, Abdelmalek H.
;
Martins, Carlos Humberto
This study presents a real-time, interpretable framework based on the light gradient boosting machine (LightGBM) algorithm for the accurate prediction and optimization of 28-day concrete compressive strength (Fc), validated using a dataset of 500 concrete mixtures. The proposed model was benchmarked against seven widely used regression algorithms, including linear regression (LR), ridge regression (RR), random forest (RF), K-nearest neighbors (KNN), support vector regression (SVR), decision tree (DT), and multivariate adaptive regression splines (MARS), to ensure a comprehensive comparative evaluation. The LightGBM model demonstrated superior predictive performance relative to the benchmark models, achieving an RMSE of 6.11 MPa and an R² of 0.951 during the initial evaluation. Model robustness and generalization capability were further verified using a 10 × 10 repeated k-fold cross-validation procedure, yielding stable results (R² = 0.940 ± 0.017; RMSE = 6.37 ± 0.49 MPa). To capture heterogeneity in mixture compositions, K-means clustering was applied to partition the dataset into four distinct mixture regimes, within which stratified LightGBM models further improved predictive accuracy, reducing RMSE to 3.7–5.1 MPa and achieving R² values exceeding 0.97. Model interpretability was enhanced through global and regime-specific SHAP (Shapley Additive Explanations) analyses, which provided transparent and physically consistent insights into feature contributions, consistently identifying cement as the dominant positive factor and water as the primary negative driver of CS. Furthermore, an interactive web-based prediction engine was developed to enable instantaneous strength prediction, real-time sensitivity analysis, 95% prediction interval estimation, and specification-driven mixture optimization with millisecond-level computational efficiency. Comprehensive diagnostic evaluations, including Taylor diagrams, residual control charts, calibration plots, and prediction-interval validation, confirmed the statistical reliability and practical applicability of the proposed framework. Overall, the developed LightGBM-based system provides an accurate, interpretable, and scalable decision-support tool for data-driven concrete mix design and performance optimization.
Some of the metrics are blocked by your
consent settings
Understanding drying processes of steam-blanched Wolffia globosa under microwave vacuum drying through data-driven and semi-theoretical modeling
(2026-06-15)
Pambudi, Suluh
;
Churat, Chutikan
;
Nitinarot, Manassanan
;
Buntreekanok, Withitpong
;
Saechua, Wanphut
Steam blanching and microwave vacuum drying (MVD) are increasingly applied to preserve heat-sensitive, protein-rich biomaterials such as Wolffia globosa . However, accurately modeling moisture removal under MVD remains challenging due to the highly nonlinear and stage-dependent nature of microwave-induced drying. This study investigates the combined effects of steam blanching pretreatment and microwave power (540, 720, and 900 W at 5 kPa) on the drying kinetics of W. globosa and evaluates the capability of machine-learning models to predict moisture evolution in comparison with conventional semi-theoretical models. The results indicated that steam-blanched samples consistently exhibited higher maximum drying rates across all power levels, increasing from 1.088 to 1.298 g<inf>w</inf>/(g<inf>dm</inf>·min) at 540 W and from 1.974 to 2.126 g<inf>w</inf>/(g<inf>dm</inf>·min) at 900 W. The pretreatment also reduced total drying time from 19.5 to 18.0 min at 540 W, while drying time remained unchanged at higher microwave powers. Among the evaluated semi-theoretical models, the Midilli equation provided the best fit to condition-specific experimental data (R2 = 0.9981–0.9994). However, the generalized Midilli model showed systematic underestimation at low moisture ratios across operating conditions (R2 = 0.9865, RMSE = 0.9990). In contrast, the k-nearest neighbors (k-NN) model demonstrated strong generalized predictive capability across multiple drying conditions, achieving R2 = 0.9798 and RMSE = 0.0481 on the testing dataset. Beyond improved predictive accuracy, the machine-learning framework effectively captured stage-dependent drying behavior patterns that conventional generalized semi-theoretical models failed to represent. These findings highlight the potential of data-driven approaches for modeling complex drying behavior under microwave vacuum conditions. The accurate prediction of moisture ratio may further support the optimization of MVD processes for W. globosa -based food ingredients.
Some of the metrics are blocked by your
consent settings
An operational machine learning framework for calibrating COSMIC radio occultation TEC to ground-based GNSS-derived TEC
(2026-03-15)
Okoh, Daniel
;
Habarulema, John Bosco
;
Nava, Bruno
;
Cesaroni, Claudio
;
Baki, Paul
The Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) provides global Radio Occultation (RO) measurements of ionospheric total electron content (TEC), but these values are systematically underestimated relative to ground-based Global Navigation Satellite System (GNSS)-derived TEC due to the exclusion of the plasmaspheric contribution. This study presents a machine learning calibration framework that transforms COSMIC TEC into GNSS-equivalent values. Using co-located COSMIC and GNSS observations from 2006 to 2025, we developed neural network models (ROTEC-A and ROTEC-B) trained on (19 and 22) input features respectively, including COSMIC profile parameters, spatiotemporal descriptors, and optionally, solar and geomagnetic activity indices. Results show that the calibration effectively mitigates systematic underestimation, reducing mean bias from 6.97 TECU (uncalibrated COSMIC) to near zero (0.02–0.03 TECU). The calibrated products also substantially reduce skewness in residuals, yielding nearly symmetric error distributions suitable for data assimilation. Across various latitudinal, local time, and seasonal sectors, mean absolute errors were reduced by 50–75%, with the best performance at mid-latitudes and slightly elevated errors in high-latitude and equatorial regions. Although, the inclusion of solar and geomagnetic indices yielded marginal improvements, statistical tests confirmed no significant advantage over the baseline model. The operationally oriented framework outputs calibrated GNSS-equivalent TEC in near real-time, providing enhanced ionospheric monitoring capability, especially over GNSS-sparse regions such as oceans and deserts. These results demonstrate the potential of COSMIC RO data, once calibrated, to serve as a reliable complement to GNSS observations for ionospheric research, space weather monitoring, and operational applications.
Some of the metrics are blocked by your
consent settings
Deep learning-based comparative evaluation of EEG, HRV, and EDA biomarkers for personal thermal comfort prediction
(2026-03-01)
Sahoh, Bukhoree
;
Wongsontham, Fatimah
;
Tipsavak, Apaporn
;
Chaithong, Paweena
;
Kliangkhlao, Mallika
Personal thermal comfort prediction significantly impacts health, well-being, and productivity, yet existing systems typically employ multiple physiological biomarkers without clear evidence for optimal single-modality solutions. This study presents a deep learning (DL)-based metrologically grounded framework to evaluate and compare three biomarkers—electroencephalography (EEG), heart rate variability (HRV), and electrodermal activity (EDA)—as traceable sensors under precisely controlled thermal conditions. We developed specialized signal preprocessing and feature engineering pipelines for each modality: 1) EEG spectral decomposition via Welch's power spectral density estimation across frequency bands (delta, theta, alpha, beta, and gamma); 2) HRV analysis through Fast Fourier Transform spectral estimation focusing on low-frequency to high-frequency component ratios; and 3) EDA feature extraction utilizing optimized filtering techniques to isolate skin conductance level and response characteristics. Three biomarker-specific DL architectures undergo Bayesian hyperparameter optimization, enabling equitable comparison of each modality's predictive performance. Results demonstrate that the HRV-based model achieves superior performance (F-measure = 0.95) while requiring minimal computational resources (0.78 MB memory footprint, 0.89 relative cost compared to the EEG baseline), establishing it as the optimal single-modality solution. These findings provide a paradigm for reproducible physiological measurement systems, advancing thermal comfort prediction by combining clinical-grade reliability with practical implementation benefits for clinical applications and next-generation indoor environmental control systems.
Some of the metrics are blocked by your
consent settings
Thermo-Mechanical Stress Prediction in Steel IPE Profiles under Asymmetric Thermal Loading: A Finite Element and XGBoost-Based Approach
(2026-03-01)
Shaik, Nagoor Basha
;
Derakhshan, Ali
;
Nasim, Maryam
;
Jongkittinarukorn, Kittiphong
Accurate prediction of thermally induced stresses in structural members remains a significant challenge in engineering, especially under complex real-world conditions. Traditional analytical and numerical methods, while robust, often struggle to capture the complicated relationship between uneven thermal loads and structural responses without significant computational effort. This study investigates the effect of asymmetric thermal loading on standard steel IPE profiles, which are widely employed in buildings and structures. These members, often exposed partially to outdoor conditions, experience uneven temperature distributions across their cross-sections, resulting in complex internal stress patterns. To simulate such scenarios, a range of thermal conditions is applied to beams and columns with varying geometries using the Finite Element Method (FEM) numerical analysis. The resulting stress components, including von Mises, axial, and shear stresses, are analyzed in detail. This study introduces a mixed approach that integrates FEM with eXtreme Gradient Boosting (XGBoost) to forecast thermal stresses in steel IPE profiles subjected to asymmetrical temperature gradients. The suggested technique, in contrast to traditional assessments that emphasize uniform heating, accounts for the interrelated impacts of irregular thermal exposures and geometric variations among IPE sections. The FEM database enabled the training of an improved XGBoost model that achieved exceptional accuracy (R² > 0.98) in predicting multiple stress components. The results highlight the critical role of cross-sectional geometry in stress development under thermal gradients and underscore the effectiveness of machine learning techniques in forecasting structural responses. This integration offers a quick, adaptable method for assessing thermal impacts in steel IPE structures, with considerable promise for design and real-time structural evaluation in industrial settings. This approach offers substantial benefits to the petroleum and broader oil and gas sectors, particularly in enhancing structural dependability under thermal and mechanical stresses.
Some of the metrics are blocked by your
consent settings
Understanding green house gases emission dynamics from forest fires in Thailand using predictive models
(2026-02-01)
Shahzad, Fahad
;
Mehmood, Kaleem
;
Anees, Shoaib Ahmad
;
Adnan, Muhammad
;
Hussain, Khadim
Forest fires are a major driver of carbon emissions, particularly in tropical regions where climate variability and land use practices intensify their frequency and impact. This study investigates the spatiotemporal trends and emission dynamics of forest fires across Thailand's three dominant vegetation types- Evergreen Broadleaf Forest (EBF), Deciduous Broadleaf Forest (DBF), and Grassland over three climatic seasons (Dry, Hot, and Wet) in the period 2001–2023. Using the Mann-Kendall trend test and Sen's Slope estimator, we observed significant declines in burnt area during the Dry season in EBF and Grasslands, with no consistent trend in DBF. Fire–vegetation interactions revealed seasonally specific effects: positive correlations between fire count and Net Primary Productivity (NPP) were detected in the Wet and the hot seasons in the case of DBF and Grasslands, respectively. Emission analysis showed that CO₂ was the dominant greenhouse gas released, with the Dry season contributing to most emissions, although Hot season emissions have increased over time. Machine learning models Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) explained over 78 % of the variance in CO₂ emissions on test data (R2 = 0.79 for RF, 0.78 for XGBoost), despite higher Root Mean Square Error (RMSE) values (∼550) on unseen data. The Shapley Additive Explanations (SHAP) analysis identified wind components and solar radiation as key predictive variables. Central, Northeastern, and Northern Thailand emerged as emission hotspots. These findings improve our understanding of emission dynamics from tropical fires and underscore the need for region-specific mitigation strategies to inform carbon inventories and climate policy.
Some of the metrics are blocked by your
consent settings
Efficient machine learning for strength prediction of ready-mix concrete production (prolonged mixing)
(2026-01-19)
Tuvayanond, Wiput
;
Kamchoom, Viroon
;
Prasittisopin, Lapyote
Purpose – This paper aims to clarify the efficient process of the machine learning algorithms implemented in the ready-mix concrete (RMC) onsite. It proposes innovative machine learning algorithms in terms of preciseness and computation time for the RMC strength prediction. Design/methodology/approach – This paper presents an investigation of five different machine learning algorithms, namely, multilinear regression, support vector regression, k-nearest neighbors, extreme gradient boosting (XGBOOST) and deep neural network (DNN), that can be used to predict the 28- and 56-day compressive strengths of nine mix designs and four mixing conditions. Two algorithms were designated for fitting the actual and predicted 28- and 56-day compressive strength data. Moreover, the 28-day compressive strength data were implemented to predict 56-day compressive strength. Findings – The efficacy of the compressive strength data was predicted by DNN and XGBOOST algorithms. The computation time of the XGBOOST algorithm was apparently faster than the DNN, offering it to be the most suitable strength prediction tool for RMC. Research limitations/implications – Since none has been practically adopted the machine learning for strength prediction for RMC, the scope of this work focuses on the commercially available algorithms. The adoption of the modified methods to fit with the RMC data should be determined thereafter. Practical implications – The selected algorithms offer efficient prediction for promoting sustainability to the RMC industries. The standard adopting such algorithms can be established, excluding the traditional labor testing. The manufacturers can implement research to introduce machine learning in the quality controcl process of their plants. Originality/value – Regarding literature review, machine learning has been assessed regarding the laboratory concrete mix design and concrete performance. A study conducted based on the on-site production and prolonged mixing parameters is lacking.
Some of the metrics are blocked by your
consent settings
Bayesian-informed DNN and ensemble learning for predicting soil water characteristic curves from easily measurable parameters
(2026-01-01)
Liu, Yifei
;
Ni, Junjun
;
Zhang, Fei
;
Kravchenko, Ekaterina
;
Kamchoom, Viroon
Soil-water characteristic curve (SWCC) represents one of the important properties describing the hydraulic characteristics of unsaturated soil, with extensive application value in geotechnical engineering, but the experimental process for obtaining SWCC is complex and time-consuming. This research proposes a prediction framework based on Bayesian-informed machine learning for SWCC applicable to different soil types, using easily measurable soil parameters. This model uses quantified particle size distribution, bulk density, and saturated water content as input features, and employs Bayesian-Markov Chain Monte Carlo methods to inversely derive Fredlund-Xing (FX) model parameters as output features. Deep Neural Network (DNN) and stacking models based on ensemble learning of five regressors were constructed to establish the prediction framework. Results show that both DNN and stacking models effectively capture complex nonlinear relationships between the three easily measured soil parameters and FX parameters, demonstrating good prediction accuracy. The stacked model shows better prediction results, with R2 exceeding 0.94 for all three parameters, outperforming the DNN model (R2 = 0.93). Through feature engineering and SHAP (Shapley Additive Explanations)-based feature sensitivity analysis, the relationships between input features and the three FX model parameters were physically interpreted, providing physical interpretability for the machine learning models. The prediction method offers a new approach for fast and accurate acquisition of SWCC, expanding the application of machine learning methods in the field of unsaturated soils.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results