Identification of significant variables using random forest, in a process of injection moulding: A case study of yield reduction analysis in changing plastic injection moulds for auto parts products

This paper describes the application of a data mining method for the identification of significant variables in a manufacturing process. A practical case is analysed, where is needed a reduction of yield due to a production delay. This case is developed in an auto parts injection moulding company. The problem lies in the assembly and disassembly process of the moulds, which is a step prior to the injection of the pieces. When there is a delay in this process, yield and delivery to the customer are affected. The purpose of this analysis is to find the root causes of the delays and take actions in this regard. To reach the goal, first, the data is extracted, then prepared, afterwards analysed and finally interpreted. In the data analysis, the algorithm "random forest" is used to find the most significant factors. The efficiency of the algorithm is evaluated with the Area under the Receiver Operating Characteristic (AUROC) method. It was determined that time, machines, and operators are the main causes of the delays with almost 70% accuracy, in addition, these results were compared with the possible root causes provided by production management.