This paper deals with the production plan for the foaming process, the core part of the refrigerator manufacturing process. In accordance with this change, the refrigerator manufacturing process has also been converted into the mixed-model production system and it is necessary to optimize the production release pattern for the foaming process. The pattern optimization is to create a mixed-model combination which can minimize the number of setup operations and maintain mixed-model production. The existing method is a simple heuristic that depends on the demand priority. Its disadvantages are low mixed-model configuration rate and high setup frequency. Therefore, demand partitioning occurs frequently. In this study, we introduce the tolerance concept and propose a new pattern optimization algorithm based the large neighborhood search (LNS). The proposed algorithm was applied to a refrigerator plant and it was found that mixed-model configuration rate can be improved without demand partitioning.