Federated Learning for the Maritime and Offshore Industries: Training Machine Learning Models Without Data Sharing
Gaspard Ducamp, Peter Kim, Eivind Ruth · 2025
In the fast-evolving field of machine learning, data privacy and security are critical concerns, especially in sensitive and heavily regulated sectors such as offshore energy and maritime operations. Traditional centralized machine learning approaches require consolidating all data in a single location for model training, which can introduce substantial privacy risks and often conflict with regulatory requirements. Federated Learning (FL) allows data to remain decentralized while enabling collaborative and individual model development in industry. FL is particularly valuable for two key use cases. First, multiple companies with a shared interest in developing a generalized model can collaborate through FL while reducing data privacy concerns. Second, a single company with distributed data—whether due to regulatory restrictions or logistical challenges—can leverage FL to train a unified model across its datasets without centralizing data on one server. By keeping data localized, FL enhances privacy, regulatory compliance, and operational flexibility, making it well-suited for the needs of the maritime and offshore industries. Popularized by Google, FL not only lowers privacy concerns but also strengthens model robustness and accuracy by drawing on diverse, distributed data sources. This capability is particularly valuable in offshore and maritime applications, where varied data from different installations, vessels, and sea conditions can inform more reliable predictive insights. In this paper, we discuss the challenges and opportunities of applying Federated Learning to the offshore and maritime industries. We present an overview of different applications, including examples from recent studies and toy cases that demonstrate FL's potential for these industries. We illustrate the value of FL further through the case of a container ship navigating various seas, where we aim to predict the vessel's performance under different conditions. Additionally, we describe a working procedure using an open-source package to implement FL, tested in a real-world setting with participants retaining data locally. A server hosted by a cloud provider was used to coordinate the Federated Learning process. This example demonstrates the strength of Federated Learning in developing generalized and robust models—a crucial need in the offshore and maritime sectors, where diverse and adaptable models are essential to address complex and variable real-world conditions. By maintaining data locally for each participant, FL emerges as a promising approach for developing regulatory-compliant, data-driven solutions for critical maritime applications.