OPTIMASI PENGAMBILAN PESANAN MENGGUNAKAN MODEL REGRESI MULTILEVEL DAN ALGORITMA METAHEURISTIK PADA CENTRAL PARTS DEPO PT TOYOTA-ASTRA MOTOR

• Main Author: Izzah Ramadhan, Thariq
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/53635
• Institution: Institut Teknologi Bandung

The SPLD Division of PT Toyota-Astra Motor is responsible for distributing Toyota service parts throughout Indonesia by managing a distribution network originating from the Central Parts Depo (CPD). In December 2019-February 2020, the percentage of on-time delivery truck departures from CPD was still below the target. The delays were mostly caused by the unsynchronized order-picking process between picking operators. This is because the current pick batch formulation and assignment process does not consider picking vehicle capacities, different pace between pickers, and workload balancing. Therefore, this study develops an optimization model for order-picking synchronization with the objective function of minimizing the makespan time of picking all daily orders received by CPD. This study considers five multiple linear regression model alternatives to predict the picking batch execution time. The alternative of the multilevel regression model for each prediction zone by considering individual operators as a random effect has significantly better performance than other regression model alternatives and is used as a sub-model of the optimization model. Overall, the developed optimization model can be declared as valid and robust. This optimization problem is classified as NP, so a metaheuristic algorithm is needed to solve the optimization model efficiently. The initial solutions resulting from the earliest start date (ESD) constructive algorithm are improved with three metaheuristic algorithm alternatives, namely: variable neighborhood descent (VND), variable neighborhood search (VNS), and list-based simulated annealing (LBSA). Based on the test results, there is no most superior algorithm, the LBSA algorithm is superior in scenarios with tight precedence limits, while the VNS algorithm is superior in scenarios with the least number of pickers. In general, the three improvement algorithms can improve the solution with an average makespan time improvement of 7.21% from the initial solution. The solution can also provide an average increase in picking device capacity utilization by 67.4% and an increase in picker's productivity of 62.9% from the actual condition. However, the solution resulting from the constructive algorithm of ESD always has a better average lead time than the solution resulting from the improvement algorithms. Therefore, the ESD constructive algorithm is used to implement the solution, as the basis of the order-picking simulation tool.