AI and Simulation-Based Techniques for the Assessment of Supply Chain Logistic Performanc

The effectiveness of logistic network design and management for complex and geographically distributed production systems can be measured in terms of direct logistic costs and in terms of supply chain production performance. The management of transportation logistics, for instance, involves difficult trade-offs among capacity utilization, transportation costs, and production variability often leading to the identification of multiple logistic solutions. This paper defines and compares three different modeling approaches to systematically assess each identified logistic alternative in terms of actual transportation costs and expected production losses. The first modeling approach examined in the paper is a mathematical model which provides the statistical basis for estimating costs and risks of production losses in simple application cases. The second model is a stochastic, discrete event simulation model of bulk maritime transportation specifically designed to capture the dynamic interactions between the logistic network and the production facilities. The third one is an AI-based model implemented as a modular architecture of Artificial Neural Networks (ANNs). In such an architecture each network establishes a correlation between the logistic variables relevant to a specific sub-problem and the corresponding supply chain costs. Preliminary testing of the three models shows the relative effectiveness and flexibility of the ANN-based model; it also shows that good approximation levels may be attained when either the mathematical model or the simulation model are used to generate accurate ANN training data sets for each transportation/production sub-problem