In recent years, approximate methods or metaheuristics, have been used in several researches, due to efficiency for solving NP-Hard and NP-Complete problems using small computional times. Nevertheless, the metaheuristic resolution process depends on parameter settings. This task is considered an optimization task itself. In this paper, we propose a variation of bat algorithm that self-adapting its parameter settings according to the performance exhibited. Bat algorithm is a relatively modern metaheuristic inspired on the ecolocation behavior of bats that employ echoes to identify the objects in their surrounding area. To evaluate the algorithm performance, we resolve 90 instances of the manufactoring cell design problem taken from Boctor's experiments. We show interesting experimental results where the proposed approach achieves all optimal values.