Scholarlink

<p>The impulse injection Residence Time Distribution (RTD) method uses sensors at the inlet and outlet of the industrial-scale pulp digester to measure values that are recorded on the Data Acquisition System (DAS). The data collected on DAS is imperfect due to random measurement errors, sampling intervals, and premature termination of the tracer test. Manual pre-processing of DAS data requires replacing missing and null values, alphanumeric values, special characters, etc. After these corrections, the hydrodynamics by fitting RTD models such as the Axial Dispersion Model (ADM) and the Tank-in-Series with Back-Mixing Model (TIBM) are assessed. The repeated correction on DAS data is labor-intensive, tedious, and prone to errors and generates inaccurate results. The proposed work uses Machine Learning (ML) to eliminate these errors and optimize data. The DAS data is treated for background, radioactive decay, starting point, and tail correction, and <i>E(t)</i> curves are generated for RTD analysis. The work therefore evaluates the predictive ability of different machine learning models in estimating the degree of fit between the traditional computational ADM and TIBM model with the ML-predicted model. Researchers utilize Linear Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN), Generalized Additive Model (GAM), Decision Tree (DT), XGBoost, LightGBM, and AdaBoost to develop ML predictive models. The present findings show a thorough comparison of different ML strategies for forecasting the degree of agreement between experimental models and ML models. The evaluation metrics were obtained using R-squared scores for comparison. The RF followed by XGBoost ML models outperformed the other selected algorithms. Thus, the presented work demonstrates the effectiveness of ML in predictive modelling for RTD data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Residence time distribution data analysis and model prediction using machine learning for industrial scale pulp digester

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Sharad Saxena,
  • Avinash Chandra,
  • Vibhuti Sharma,
  • Anu Bajaj

摘要

The impulse injection Residence Time Distribution (RTD) method uses sensors at the inlet and outlet of the industrial-scale pulp digester to measure values that are recorded on the Data Acquisition System (DAS). The data collected on DAS is imperfect due to random measurement errors, sampling intervals, and premature termination of the tracer test. Manual pre-processing of DAS data requires replacing missing and null values, alphanumeric values, special characters, etc. After these corrections, the hydrodynamics by fitting RTD models such as the Axial Dispersion Model (ADM) and the Tank-in-Series with Back-Mixing Model (TIBM) are assessed. The repeated correction on DAS data is labor-intensive, tedious, and prone to errors and generates inaccurate results. The proposed work uses Machine Learning (ML) to eliminate these errors and optimize data. The DAS data is treated for background, radioactive decay, starting point, and tail correction, and E(t) curves are generated for RTD analysis. The work therefore evaluates the predictive ability of different machine learning models in estimating the degree of fit between the traditional computational ADM and TIBM model with the ML-predicted model. Researchers utilize Linear Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN), Generalized Additive Model (GAM), Decision Tree (DT), XGBoost, LightGBM, and AdaBoost to develop ML predictive models. The present findings show a thorough comparison of different ML strategies for forecasting the degree of agreement between experimental models and ML models. The evaluation metrics were obtained using R-squared scores for comparison. The RF followed by XGBoost ML models outperformed the other selected algorithms. Thus, the presented work demonstrates the effectiveness of ML in predictive modelling for RTD data.