Though box compression strength (BCS) is commonly used as a performance criterion for shipping containers, the state-of-the-art BCS estimation produces results within a broad range of values. In this study we implemented a new approach, artificial neural networks (ANN), to explore how much data may be needed for an ANN to reasonably predict compression strength, and how the ANN approach performs while facing variation that adversely impacts other modeling methodologies. An ANN model can be built by comprehensively adjusting four modeling factors that interact with each other to influence model accuracy and can be optimized by minimizing model MSE. Using both data available from the literature and a “synthetic” data set using idealized data based on the McKee equation, we find that model estimation accuracy remains limited by the uncertainty or error in the input parameters combined with uncertainty from the ANN process itself, and we produce an estimate for this impact. The population size to build an ANN model that can reasonably estimate BCS has been identified based on different data sets in this study.Packaging design plays a crucial role in ensuring the protective performance of packages. Various factors must be considered to ensure package strength during the packaging design process. Understanding the relative importance of each influencing factor or design feature provides valuable insights for optimizing packaging material utilization. However, current methods such as testing and finite element analysis have limitations in evaluating the relative significance of these parameters. In response to these challenges, in this research, we applied different methods to comprehensively evaluate the relative importance of different packaging design features on a given packaging property. Using BCS as a representative packaging property, the relative importance of up to six BCS features (Edge Crush Test (ECT), Perimeter, Thickness, Depth, and Flexural Stiffness in both the machine and cross-machine directions) were evaluated. Four distinct ANN methods were employed - Connection weights method, Gradient-based method, Permutation method, and SHAP values. These techniques were applied to two datasets: one comprising "synthetic" data based on the McKee formula and the other representing real-world scenarios. The reliability of these methods was assessed. Different input feature importance (FI) scores obtained from the four methods have been calculated and compared with theoretical BCS FI derived from the McKee formula. The BCS feature ranking result given by the synthetic data is verified by the theoretical feature importance ranking indicated by the McKee formula. Although box depth is considered to have zero importance in the McKee formula, the BCS feature importance ranking from the real dataset highlights its significance, aligning with buckling theory. The study gives an insight into the BCS feature importance evaluation using ANN approach and guides packaging design material and cost saving. The ultimate objective of this research is to develop a comprehensive ANN model for predicting Box Compression Strength (BCS). To achieve this, we utilized a dataset encompassing a wide range of box dimensions commonly encountered in industrial applications. After applying multiple optimization methods to determine the optimal number of hidden neurons and further identifying the key factor values influencing the model, a generalized ANN model was trained. The trained ANN model can predict the BCS commonly used in the industrial applicable level with an error of 9.51%. The primary factor contributing to the high BCS error is the presence of boundary data points and the small sample size of the current data set. One possible strategy to improve ANN prediction accuracy is to continually expand the current dataset sample size using available resources. In essence, this study serves as a roadmap for forthcoming research endeavors seeking to leverage ANN techniques to tackle challenges and provide solutions within the corrugated industry.
Read
