PREDICTION OF PACKAGING LIFE-CYCLE DESIGN PERFORMANCE Lerpong Jarupan , Northeastern University, MA 02115, (617) 373-7635, l.jarupan@coe.neu.edu Sagar V. Kamarthi*, Northeastern University, MA 02115, (617) 373-3070, sagar@coe.neu.edu Surendra M. Gupta , Northeastern University, MA 02115, (617) 373-4846, gupta@neu.edu (* Corresponding author) ABSTRACT We develop a back-propagation neural network (BPN) to predict the life-cycle design performance for transport packaging. The BPN is constructed and trained on the packaging design attributes to detect hidden relationships among historical or pre-existing life-cycle design data to predict a new concept design through supervised learning, by minimizing the squared difference between the actual and the predicted life-cycle design performance. To this end, the designer could use the predicted life-cycle design in a trade-off analysis and concept selection for a potential packaging design. A case example is used to illustrate the methodology. INTRODUCTION Life-cycle design for transport packaging uses a decision-making methodology during the conceptual stage, by considering the packaging performance, environmental impairment and cost requirements [5]. Presently, life-cycle inventory and cost-analysis tools applie d to packaging products offer guidelines for achieving better environmental design and management. However, these approaches can be expensive, time-consuming and labor intensive, and somewhat prohibitive from a modeling viewpoint because diverse and numerous ideas and quality information in the conceptual design phase may be difficult during multi-dimensional multi-attribute trade-offs. In this paper, an artificial neural network (ANN) technique is employed to make better predictions for life- cycle design performance for transport packaging at the conceptual stage. A major advantage of ANN over other analytical tools is that ANN attempts to fit curves through data without utilizing a predetermined function with free parameters, resulting in quick data generation and transfer function with reasonable accuracy. The popular back-propagation (BP) learning algorithm is used to develop a robust system. This algorithm is based on learning capability of known information to predict life-cycle design performance for transport packaging. The BP neural network is constructed and trained on the packaging design attributes to detect hidden relationships among historical or pre-existing life-cycle design data to predict a new concept design through supervised learning, by minimizing the squared difference between the actual and the predicted life-cycle design. A few attempts have been made in applying neural network for product designs. Hsiao and Huang [3] constructed a BP neural network to analyze the relationships between product forms and adjective image words at the design stage. Seo et al. [13] [14] developed an approximate method for providing the preliminary life cycle cost during conceptual design. However, to the best of our knowledge, while there are few studies that are conducted on cushioning-type packaging, none of the studies use ANN for life-cycle design of transport packaging. Zhang et al . [16] and Zhang and Fuh [17] proposed a BP neural network for estimating packaging costs. Lye et al . [11] proposed a design methodology for the design of protective packaging buffer configurations. In a series of papers, Liang et al . [6] [7] [8], and Liang and Zhou [9] modeled a BP neural network to identify nonlinear characteristics in cushioning type packaging. A combination approach between a fuzzy-adaptive BP and genetic algorithm is developed in their model to