Determining pre-assembly resequencing buffer content in automotive assembly lines

In mixed model assembly lines, smooth operation of the line depends on adherence to the scheduled sequence which is determined according to production constraints and customer demand. However, the scheduled sequence is scrambled due to intentional and unintentional sequence alterations. A resequencing buffer between paint and final assembly is located to restore the altered sequence. Restoring the altered sequence requires three distinct operations of this buffer: (i) Changing the positions of vehicles (i.e., resequencing), (ii) replacing spare vehicles with paint defective vehicles, (iii) lastly inserting spare vehicles to final assembly from the buffer when restoring the altered sequence is not possible by resequencing due to paint defects and limited buffer capacity. In this thesis, a two-stage stochastic programming model which considers the stochastic nature of paint defect occurrences is developed. In the first stage of the model, optimal number of model-color types placed into resequencing buffer to restore the scheduled sequence is determined. In the second stage after defect occurrences, the assembly entrance sequences of the vehicles are decided. Since the solution of the problem depends on the resequencing buffer type, two different models for automated storage and retrieval system (AS/RS) and mix-bank are built. The developed two-stage stochastic program is solved by sample average approximation (SAA) algorithm to and the optimal number of model-color types to be placed in the buffer is found. Also a numerical study is performed to investigate the problem parameters to the solution such as paint defect rate, capacity of buffer, adherence ratio of vehicles entering paint shop to scheduled sequence. Since vehicle resequencing due to the paint defects requires instant decision making, another heuristic rule based model to resequence vehicles easily by car manufacturers is developed. The proposed heuristic model performs as good as mathematical model. Lastly, to solve the large scale problems for both AS/RS and mix-bank resequencing buffers the purposed model is enhanced with genetic algorithm.