The use of data-driven modeling for digital twins (DTs) is growing in popularity. However, many models do not make it into production. Those that do quickly become outdated. Streamlined model update is still a major challenge. There is a need to establish methods and techniques for managing data-driven digital twins throughout their entire life cycle. Machine learning operations (MLOps) recently emerged as an effective means to foster the integration of Machine Learning (ML) models and their operational workflows. In this paper, we exploit MLOps for development and operation of data-driven digital twins. To validate our approach, a case study is conducted in which an ML model of a physical test rig is trained in accordance with the MLOps principles. The work aims to demonstrate how MLOps practices can contribute to overcoming issues related to scalability, accuracy, and adaptability in the context of digital twin training.