Multiobjective workforce scheduling with combined safety, productivity, and job satisfaction consideration

Abstract

For industrial workers who are exposed to hazardous work conditions, their daily hazard exposures can be alleviated by appropriately rotating them among various jobs within a workday. To effectively implement job rotation, workers’ daily work schedules must be generated. This research studies the multiobjective ergonomic workforce scheduling problem (MO-EWSP) that is intended to generate safe daily rotating work schedules for workers such that their daily hazard exposures do not exceed a daily permissible exposure limit. Three problem objectives are considered: (1) minimizing manpower cost, (2) maximizing productivity, and (3) minimizing job dissatisfaction. The criteria used in the three objectives are number of workers, total worker-job fit score, and total dissatisfied worker-job and worker-partner assignment, respectively. Workers are heterogeneous with respective to the work ability, skill level, job preference, and partner preference. Jobs that are being considered have different operation work schedules as well as numbers of required operators. The problem solution consists of the number of workers for job rotation, their daily rotating work schedules, total worker-job fit score, and total number of dissatisfied worker-job and worker-partner assignment.In this research, the MO-EWSP is solved preemptively and nonpreemptively using both optimization and metaheuristic approaches. For the former solution approach, two mathematical models are developed, namely, preemptive mixed integer linear programming model (P-MILP) and nonpreemptive goal programming model (N-GP). For the latter approach, two genetic algorithms (GAs) are developed, namely, preemptive multiobjective GA (P-MOGA) and nonpreemptive multiobjective GA (N-MOGA). A numerical example is generated to illustrate the use of mathematical programming and GA approaches. Firstly, the problem is solved to optimality using the P-MILP and N-GP models alternately. Next, the same problem is solved using the P-MOGA and N-MOGA approaches. At the end of the last generation, a set of good solutions are obtained. It is shown that the preemptive and nonpreemptive considerations can yield the same targets if the scale weights, relative importance weights, targets, and objective functions are suitably defined. That is, the objective values in all objectives when they are solved preemptively are the same as those when solved nonpreemptively. Nevertheless, different daily work schedules are obtained. Additionally, a computation experiment is conducted with 18 hypothetical test problems. The P-MILP model can solve the test problems to optimality with a success rate of 83%. The N-GP, P-MOGA, and N-MOGA can solve all test problems in less than 1% of the computation time required by the P-MILP model. The N-GP requires the shortest computation time since it solves multiple objectives at once. However, it provides only one optimum solution that is the same as the P-MILP model. The P-MOGA approach can provide a set of good solutions at the end of the evolution process. For the N-MOGA approach, the overall average percent deviation from the P-MILP solution is larger than the P-MOGA approach. However, the N-MOGA approach provides more flexibility to the scheduler in adjusting the priority order of objectives as opposed to having only one priority order with the preemptive consideration.