Development of An Integrated Product Data Model with Machine Learning Models and Datasets to Facilitate Part Identification and Detection of Assembly Products

To effectively use machine learning (ML), virtual ML applications that take into account aspects particular to an ML application domain are required. This paper presents a product data model that incorporates ML objects to assist ML applications for the part detection among components of a given assembly product during product development. The proposed product data model combines ML model and dataset objects during the ML life cycle, as well as item, technical document, product structure and engineering change objects over the product life cycle. The ML objects in the product data model are tightly connected with product structures and support production routings, which are the domain of ML applications in product development. Furthermore, they apply transfer learning approaches to repurpose previously learned ML models to create new ML models for engineering changed products. To evaluate the feasibility of the proposed product data model, a test application of the part detection ML is implemented utilizing an existing information system for product development and manual operations.