| Observational data from production lines are collected as process monitoring activities. Observational data can provide valuable information about the dynamic nature and underlying mechanisms of manufacturing processes. However, these precious data are usually not fully used to identify the fundamental causes of output variations. In this research, multiple regression analysis of observational data is proposed as an effective method for the continuous improvement of manufacturing processes in full-scale production, as multiple regression can handle observational data and analyze the relationships between independent variables and a dependent variable.; Observational data analysis was used successfully to clarify the nature of processes and improve their control schemes in two case studies: process improvement of a casting process of chill iron and a molding process of flexible polyurethane foam. For the successful analysis of observational data, it is essential to understand the limitations of observational data: high collinearity, important hidden variables, narrow ranges of some important variables, excessive influence of extreme points, and serially correlated errors. The problems caused by these limitations, effective diagnostic methods, and relevant remedial actions are discussed. To prevent systematically the problems caused by the limitations, a method is developed for continuous improvement of process control using regression analysis of observational data. This method emphasizes three important concepts: use of a cross-functional team, collection and analysis of multiple observational data sets with continuous improvement efforts, and use of a step-by-step guideline to circumvent the potential pitfalls of observational data analysis. The two case studies provide good insight for this generalization of observational data analysis. Following this approach, substantial process improvements were achieved in the two specific cases investigated. |
