C Rajendran

A novel genetic algorithm (GA) is proposed in this paper for solving the machine-part cell formation problem in the presence of alternative process plans. Parent chromosomes with number of genes equal to the number of parts are generated based on the correlation value calculated using a ranking index. Crossover is carried out on the 60% of the parent chromosomes and mutation is carried out at the weakly correlated part (gene) in the chromosomes. Performance of the algorithm was tested using 20 test instances from the literature. The proposed genetic algorithm is superior in terms of solution quality for 10% of the total test instances and equal to the best solution achieved by the other algorithms for the remaining 90% of the test instances.

This paper aims to demystify the possibility of using Hamann, Yule (value ranges from-1 to +1) and Jaccard (value ranges from 0 to +1) similarity measures for the machine-part cell formation (MPCF) using the CARI heuristic[17] that uses correlation coefficient (value ranges from-1 to +1) as the similarity measure. It has been found that grouping efficacy (GE) achieved by CARI heuristic while using Hamann and Yule as similarity measure is less for 71.42% and 51% of the dataset respectively compared to the GE achieved while using correlation coefficient as similarity measure. Jaccard similarity measure has been found unsuitable for MPCF using CARI heuristic due to the formation of single large cluster for all the benchmark dataset. CARI heuristic in its current version produces machine-part cells with high GE, only while using correlation coefficient as similarity measure whereas while using the other similarity measures, it produces machine-part cells with low GE or single large cluster is created. To overcome this issue, a normalizing procedure is appended to the current version of CARI heuristic, thus making it capable of producing machine-part cells with any of the similarity measures. Computational performance of the proposed version of CARI heuristic was tested using 35 dataset. The proposed method resulted in increasing GE for 42.84% and 74.27% of the dataset for Hamann and Jaccard similarity measures respectively