Implementation and Application of Decision-Tree-Based MSQIM : Quality Data Mining through the Multi-Stage Quality Information Model (MSQIM) (2nd Report)

Abstract

Detecting causal factors of quality defects is a crucial step in improving manufacturing quality in a multi-stage production system. In order to streamline this task with the help of a manufacturing database, this paper presents an explorative data analysis approach called the multi-stage quality information model (MSQIM) based on decision tree analysis. This approach first defines a subset of the manufacturing database for each single process step of the production system, which contains only the process data that have been collected until the process step. Then, it utilizes decision tree analysis to extract quality information from each defined database subset. This results in plural decision tree models corresponding to the process steps of the production system. The approach finally traces how quality information varies along the process steps, both quantitatively and qualitatively. How the quality information amount changes makes it possible to identify the process steps that require further focus, even when sufficient potential factors cannot be specified from the given database alone. Further, a leaf node transition graph is introduced to interpret the relationships between consecutive decision tree models, and is found to be effective to guide hypothesis generation about causal factors of defects and defect-causing mechanisms. An industrial example shows the power of this approach.