ERI: Enhancing Privacy-Aware System Failure Prediction through Integrated Longitudinal and Survival Data Modeling

Accurately predicting failures before they occur can transform industrial operations by reducing downtime, improving safety, and optimizing maintenance strategies. However, achieving this requires overcoming key challenges, including handling irregularly sampled data, integrating sparse yet critical failure-time information, and addressing privacy concerns in data-sensitive applications. This Engineering Research Initiation (ERI) award supports research that aims to develop a scalable predictive modeling framework that leverages both high-dimensional sensor signals and failure event data for online system failure prognosis while preserving industrial data privacy. By combining longitudinal analysis, survival modeling, and decentralized learning methodologies, this approach looks to enhance predictive accuracy and enable robust decision-making in industrial settings. If successful, the project will significantly advance data-driven monitoring, prognostics, and contribute to the emerging personalized predictive maintenance activities. Beyond research, this project intends to enrich undergraduate education by integrating real-world applications into classroom learning, providing students with hands-on opportunities to engage in model development, simulation, and testing in predictive maintenance and reliability. Additionally, the research will offer K-12 students opportunities to engage with cutting-edge technologies, fostering a new generation of innovators ready to tackle future challenges.

The project seeks to advance industrial prognostics by developing a new federated survival analysis tool, enabling robust failure predictions from heterogeneous sensor data without compromising industrial data privacy. The research focuses on three key objectives: 1) reimagining survival analysis through federated learning to enhance prediction accuracy by leveraging decentralized data sources, 2) facilitating adaptive failure predictions that dynamically adjusts to changing system conditions using Bayesian theory and multivariate functional principal component analysis, and 3) rethinking data fusion in federated survival analysis to seamlessly integrate heterogeneous data sources, thus ensuring a holistic view of system health. Research completed in association with this project could be critical for industries where timely failure forecasting can drastically reduce downtime, improve operational efficiency, and enhance safety. The scalable, adaptable models that will be developed in this research can be applied beyond industrial settings, with potential impacts across diverse fields such as healthcare, transportation and energy. The interdisciplinary nature of the project will looks to drive innovation and foster collaboration between academia and industry, advancing the development of intelligent, privacy-preserving systems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Award Number: 2501643
Principal Investigator: Yuxin Wen
Funds Obligated: $199,106
State: CA