Cause Prediction of Product Defects Using Event Causality

Abstract

For improving product safet y and reliability, possible defects are analyzed in advance using FMEA, FTA, and so on. However, whether such methods can predict defects depends on the knowledge and experience of engineers. In this paper, we propose a defect cause analysis method that can automatically derive possible causes without relying on the experience of engineers, and can extract causes difficult to predict. First, events and their causal relationships are modeled with Petri nets. In addition, an assembly model is introduced to represent the propagation of events between parts. Next, a method to predict the causes of product defects using the causal relationship model and the assembly model is proposed. In the method, a model including not only the functions of a product but also its manufacturing process is created at first. Then, past event causal data related to events in the product model are combined into the model. After that, a defect cause prediction model could be generated by reversing the relationship between cause and effect. By using this model, a cause event of a product defect can be traced step by step. Thus, possible underlying causes of the product defect are automatically derived. Finally, we confirmed the usefulness of our proposed method by implementing a system and predicting the causes of a defect of a soldering iron.