<p>This paper presents the development of a predictive modeling and sensitivity analysis of an industrial dehydration plant located in El Llano, Mexico, with a capacity to process up to two tons of fresh fruit. The proposed methodology allows estimating the percentage influence of the input variables on the output variable and evaluating the efficiency of the thermal process. This process involves measuring variables inside the drying chamber at different heights. Subsequently, we employed machine learning algorithms, including artificial neural networks (ANNs), support vector machines (SVMs), Gaussian process regression (GPR), and multiple linear regression (MLR), to estimate mass loss. The optimal model was identified by applying statistical metrics: mean absolute error (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">$MAE$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="italic">MAE</mi> </mrow> </math></EquationSource> </InlineEquation>), mean-squared error (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$MSE$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="italic">MSE</mi> </mrow> </math></EquationSource> </InlineEquation>), root-mean-squared error (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">$RMSE$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="italic">RMSE</mi> </mrow> </math></EquationSource> </InlineEquation>), and coefficient of determination (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">${R}^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mrow> <mi>R</mi> </mrow> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>). Finally, Probability Analysis of the Wild Number (PAWN) was used to determine the contribution of each input variable. The experimental results demonstrated temperature fluctuations at the upper and lower levels, with a range of ± 7 °C. The GPR model demonstrated superior performance in terms of statistical metrics (<InlineEquation ID="IEq5"> <EquationSource Format="TEX">${R}^{2}=0.99; MAE=0.05\%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mrow> <mi>R</mi> </mrow> <mn>2</mn> </msup> <mo>=</mo> <mn>0.99</mn> <mo>;</mo> <mi>M</mi> <mi>A</mi> <mi>E</mi> <mo>=</mo> <mn>0.05</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>) and correlation plots. The PAWN analysis indicated that drying time is the most significant factor, with an impact of 72.2%. This methodology is intended to be a valuable tool to optimize and improve the operational performance of dehydration plants, promoting Industry 4.0 with a focus on sustainability.</p>

Predictive modeling and sensitivity analysis to optimize the efficiency and operation conditions of the drying process

期刊论文
发表日期：

2025年11月25日

查看全文

A. J. Hernandez-Bautista,
A. J. Cetina‑Quiñones,
Sarah Messina,
M. A. Escalante Soberanis,
Luis J. Ricalde,
C. Acosta,
A. Bassam

摘要

This paper presents the development of a predictive modeling and sensitivity analysis of an industrial dehydration plant located in El Llano, Mexico, with a capacity to process up to two tons of fresh fruit. The proposed methodology allows estimating the percentage influence of the input variables on the output variable and evaluating the efficiency of the thermal process. This process involves measuring variables inside the drying chamber at different heights. Subsequently, we employed machine learning algorithms, including artificial neural networks (ANNs), support vector machines (SVMs), Gaussian process regression (GPR), and multiple linear regression (MLR), to estimate mass loss. The optimal model was identified by applying statistical metrics: mean absolute error ( $MAE$ $MAE$ ), mean-squared error ( $MSE$ $MSE$ ), root-mean-squared error ( $RMSE$ $RMSE$ ), and coefficient of determination ( ${R}^{2}$ $R^{2}$ ). Finally, Probability Analysis of the Wild Number (PAWN) was used to determine the contribution of each input variable. The experimental results demonstrated temperature fluctuations at the upper and lower levels, with a range of ± 7 °C. The GPR model demonstrated superior performance in terms of statistical metrics ( ${R}^{2}=0.99; MAE=0.05\%$ $R^{2} = 0.99; M A E = 0.05 %$ ) and correlation plots. The PAWN analysis indicated that drying time is the most significant factor, with an impact of 72.2%. This methodology is intended to be a valuable tool to optimize and improve the operational performance of dehydration plants, promoting Industry 4.0 with a focus on sustainability.