A Digital Twin Model for an Educational Turbocharger Demonstrator

The contemporary concept of digital twin has a vast technical scope, making it one of the most relevant components of cyber- physical production systems. In practice, creating a digital twin can happen at different points in the lifecycle of an automated system. Besides, the digital twin can have multiple final uses, including simulation, forecasting, fault detection, and fault recognition. Probably, one of the most labor-intensive activities on creating a functional digital twin is to accurately deploy a sampling system to collect operational data from the process that can contribute effectively to its final use. Such activity demands experimentation, engineering work, and continuous improvement.<br><br>This paper presents the analysis and synthesis of principles required to create a fault detection and classification digital twin application through machine learning techniques. The principles are demonstrated in a compact educational turbocharger prototype, built by the University College Dublin, and deployed at the Aalto Factory of the Future (AFoF) learning factory. The turbocharger demonstrator carries a data acquisition and storage system, with adaptable environmental and process sensors interconnected to a wireless Industrial Internet of Things platform. The software architecture and deployment workflow that facilitated the experimentation on creating the digital twin in the AFoF learning factory is presented, with comparison of different classifiers as the main contribution of this work. On preliminary tests, the system detected and classified three possible faults caused by controlled muffler obstruction conditions.<br><br>From another perspective, the demonstrator system aims to be an affordable alternative with potential to become a platform for teaching and education to technology students, e.g., on the digital twin and Industrial Internet of Things concepts. Both the physical turbocharger and its digital twin, including the predictive fault detection model associated with it, are developed for educational use at the AFoF learning factory.