An Ensemble Learning based Hierarchical Multi-label Classification Approach to Identify Impacts of Engineering Changes

In the process of complex products development, design decisions of products are constantly changed to improve quality or functionality, reduce costs, respond to statutory constraints or implement wishes from customers, with respect to new functionality. The changes on already released design decisions are known as Engineering Changes (EC). Due to interdependency, changes on one component may cause changes on another, and this will spread along the product structure. Therefore, to completely identify the affected components of engineering changes is a major challenge. This paper presents a novel approach to use properties of components as target variables and applying the predicted properties to locate the EC affected components in product structure. We create a hierarchy of the properties and divide the label space into separate communities. A stacked multi-label classifier is trained in each community, the result is obtained by union of assigned labels from different communities. Finally, the predicted labels are adjusted by incorporating the hierarchical relation. Experiments conducted on real-world industrial EC dataset with mixed data types. Results demonstrated that, the ensemble framework in our approach is more efficient and effective than our baseline models and has achieved superior performance on real industrial engineering change data.