Hybrid flow management - multi objective optimisation

Lorenzo Lassig, Fabio Mazzer, Marino Nicolich, Carlo Poloni (University of Trieste)

CHALLENGE - Currently, management and investigation of manufacturing systems is a very important aspect for any industry. The inaccurate or defective definition of a manufacturing system can affect the real capacity of a company to stay in the market. Several techniques to support the decision making have been developed in order to evaluate the manufacturing production strategies and the system itself. The present work considers both assembly and process Hybrid Flow-Shop lines as a case study.  

SOLUTION -  In this paper, the aim is solving a HFS scheduling problem by minimizing the makespan and inventory cost through a multi-objective-hybrid-metaheuristic approach, which combine genetic algorithm and variable neighbourhood search. The systems discussed here are two Hybrid Flow Shop lines with different objectives and constraints. One is an assembly and testing plant of large mechanical products, whereas the second is a processing production plant. The approach is to connect an optimisation cycle to a model that simulates the production using a discrete event simulation (DES) with WITNESS 14. The concept of the approach used to solve the problem is inserting variable inputs into a model that simulates the process flow. This model gives an output, which is compared with the main objective. This information is sent to the Genetic Algorithm in modeFRONTIER. In the process flow test case, objectives include maximizing the IF=ProcessTime/Lead Time, as well as objective function such as limiting the production unbalance, maximizing the throughput, limiting the buffer capacities.

BENEFITS - The results the system gives are substantially better than before. It strengthens the idea that this method has great potential in improving a production system and decision making by using “what if” situations. The preferred solution is from cluster 0 since it has the highest IF and throughput rate. The cycle time for each optimisation run varies between them. The first cycle of optimisation including the post-processing time lasted 3-5 days depending on the case study. The second cycle of optimisation lasted only 2-3 hours. This underlines the fast speed of convergence of the simplex algorithm.