This repository provides an Excel implementation of artificial neural network (ANN) models developed to predict fiber attenuation in the melt-blowing process.
The models are presented in the article:
Formoso, I. Optimizing Melt-Blowing Nozzles for Small-Diameter Fibers: An Artificial Neural Network Framework Fibers and Polymers https://doi.org/10.1007/s12221-025-00919-y
The spreadsheet allows users to evaluate the trained neural networks and estimate fiber attenuation for different combinations of operating conditions and nozzle geometries.
The melt-blowing nozzle configuration and the main geometric parameters used to construct the dimensionless variables are illustrated below.
The melt-blowing process produces micro- and nanofibers by attenuating a molten polymer stream using high-velocity hot air. Predicting the resulting fiber diameter requires capturing the complex interaction between operating conditions and nozzle geometry.
To address this challenge, multilayer perceptron (MLP) neural networks were trained using dimensionless parameters that represent the governing physical mechanisms of the process.
This repository provides a transparent implementation of the trained ANN models, including:
- input normalization parameters
- network weights and biases
- activation functions
- forward propagation calculations
This allows the models to be reproduced or implemented in other computational environments.
The Excel spreadsheet contains two trained ANN models:
Predicts fiber attenuation in hot-melt adhesive (HMA) melt-blowing systems.
Architecture:
4 → 1 → 1 → 1 Activation functions: tansig → tansig → purelin
Predicts fiber attenuation in polypropylene (PP) melt-blowing systems.
Architecture:
4 → 2 → 1 Activation functions: tansig → purelin
Both models predict the fiber drawing ratio
d′ = d_f / L_Rf
which represents the ratio between the final fiber diameter and the polymer exit diameter.
The ANN models use four dimensionless parameters:
m′ = ṁ_a / ṁ_f
T′ = (T_f + T_a) / (2 T_amb)
g₁ = tan(ψ) L_Rf / D
g₂ = (L_Rf / L_R)² (1 / N_a/f)
These parameters capture the combined effects of operating conditions and nozzle geometry on fiber attenuation.
-
Open the Excel file.
-
Select the desired model:
- Net_1 for HMA systems
- Net_2 for PP systems
-
Enter the input variables in the yellow cells using physical values.
-
The spreadsheet automatically performs:
- normalization of inputs to [-1,1]
- ANN forward propagation
- output denormalization
- The predicted fiber drawing ratio is displayed in the results section.
The Excel file contains the following sheets:
Nomenclature Definition of all variables and symbols.
Activation_functions Transfer functions used by the neural network.
Net_weights_biases Trained ANN weights and neuron biases.
Net_1 Implementation of the HMA ANN model.
Net_2 Implementation of the PP ANN model.
The spreadsheet provides a fully transparent implementation of the trained neural networks. All elements required to reproduce the models are explicitly included:
- normalization limits
- neuron weights
- biases
- activation functions
- network architecture
This allows researchers to implement the models in other environments such as MATLAB, Python, or other numerical tools.
Ignacio Formoso
If you use this model in academic work, please cite the associated article:
Formoso, I. Optimizing Melt-Blowing Nozzles for Small-Diameter Fibers: An Artificial Neural Network Framework Fibers and Polymers https://doi.org/10.1007/s12221-025-00919-y
This repository is released under the MIT License.