Taguchi experimental design for manufacturing process optimisation using historical data and a neural network process model

Abstract

Listen

Translate article

PurposeThe paper describes the methods of manufacturing process optimization, using Taguchi experimental design methods with historical process data, collected during normal production.Design/methodology/approachThe objectives are achieved with two separate techniques: the Retrospective Taguchi approach selects the designed experiment's data from a historical database, whilst in the Neural Network (NN) – Taguchi approach, this data is used to train a NN to estimate process response for the experimental settings. A case study illustrates both approaches, using real production data from an aerospace application.FindingsDetailed results are presented. Both techniques identified the important factor settings to ensure the process was improved. The case study shows that these techniques can be used to gain process understanding and identify significant factors.Research limitations/implicationsThe most significant limitation of these techniques relates to process data availability and quality. Current databases were not designed for process improvement, resulting in potential difficulties for the Taguchi experimentation; where available data does not explain all the variability in process outcomes.Practical implicationsManufacturers may use these techniques to optimise processes, without expensive and time‐consuming experimentation.Originality/valueThe paper describes novel approaches to data acquisition associated with Taguchi experimentation.