Compression Strength Prediction of RSC Corrugated Fiberboard Boxes by Artificial Neural Network using Multiple Materials and Design Parameters

Abstract

Importance of the work: An accurate compression strength prediction model of RSC corrugated boxes could be used as an effective packaging design tool for industry. 

Objectives: This research applied the material and design parameters for RSC corrugated box strength prediction using artificial backpropagation neural network modelling (BPN).

Materials and Methods:  Total of 17 material and design factors from 630 commercially corrugated box samples in Thailand were recorded as input parameters along with their box compression test (BCT) values as output parameter. Data was randomly grouped as 80:10:10 during the model development for training set, testing set and validating set, respectively.

Results: The backpropagation neural network BPN17-13-1 model (inputs: hidden layers: output) provided the highest prediction performance with R² of 0.982 compared to calculations from simplified McKee’s formula that having R² of 0.737. With regards to the material parameters, flute 2 as well as grammage of the liner 2 and medium 2 had a higher influence toward the predicted BCT (p ≤ 0.05) compared to flute 1 and the grammage of the middle layer and outer liner. Fiber composition showed a lesser influence level than the grammage factor. For the packaging design, the contribution to BCT of height, length, and width of the boxes were 9.94%, 5.62% and 1.64 % respectively. Further, printing area contributed more toward BCT than printing position.    

Main finding: The developed compression strength prediction model was more accurate than the simplified McKee’s formula applied in the industry.  Important material parameters were type of flute and grammage. Moreover, important packaging design parameters were box height, box length and printing area.