HFSS_GA_instances

With the rapid advancement of technology, competition in manufacturing environments has intensified, making scheduling a critical factor for firms striving to maintain their market position. In this study, the production system of a company that manufactures emergency lighting products was analyzed, and it was determined that the system corresponds to the Hybrid Flow Shop Scheduling (HFSS) problem class. Furthermore, the speed and quality of production are largely dependent on workers’ manual skills.

A multi-objective HFSS problem is addressed, incorporating sequence-dependent setup times, removal times, and learning effects, with the aim of minimizing the makespan and total tardiness. A mixed-integer linear programming model is proposed, and due to the complexity of the problem, three scalarization-based bi-objective genetic algorithm approaches are developed: Weighted Sum Method (WSM), Conic Scalarization Method (CSM), and Tchebycheff Scalarization Method (TSM). The performance of these algorithms is assessed using 21 benchmark instances of small, medium, and large sizes from the literature, evaluated with weighted distance to the ideal point and weighted distance to a reference point.

Since the HFSS configuration considered in this study simultaneously includes sequence-dependent setup times, removal times, and learning-effect-based processing times, no existing benchmark set fully matches these characteristics. Therefore, a new dataset was generated using uniform distributions, based on commonly used instance sizes in the literature.

The dataset is divided into three main groups—small, medium, and large. The data generation procedure follows the method of Naderi, Zandieh, and Shirazi (2009) for determining the number of jobs (n), number of stages (t), number of machines in each stage (mt), setup times (Sijtm), and processing times (Pitm). Accordingly:

Small-sized instances:
n = {4, 6, 8}, t = {2, 3, 4}
Medium and large-sized instances:
n = {20, 50, 80, 120}, t = {2, 4, 8}

The data were generated as follows:
Processing times: U[1, 99]
Setup times: U[1, 75]
Removal times: U[1, 30]
Number of machines per stage:
U[1, 2] for small-size sets
U[1, 4] for medium and large-size sets
The compressed folder “Data Sets” contains:
Small-Sized Instances (S1–S9)
Medium-Sized Instances (M1–M6)
Large-Sized Instances (L1–L6)
Real-times Instances