Artificial Intelligence and Model Order Reduction for Optimal Sensor Placement: A Critical Review of Methods
Optimal sensor placement determines which measurements can support reconstruction, identification, diagnosis, or control under practical limits on instrumentation. This critical review examines methodological developments in machine learning, Artificial Intelligence, and model order reduction from 2016 to October 2026, giving priority to papers that establish or substantially extend selection principles. A common measurement and task formulation connects sparse discriminant sensing, reduced-basis reconstruction, pivoted orthogonal–triangular matrix factorization (QR), surrogate-assisted optimization, and reinforcement learning. A flowchart shows how these approaches can combine representations, objectives, and selection procedures. The analysis shows how representation error, observation noise, target sensitivity, and physical constraints affect the optimal layout. Comparative tables separate the contributions of the objective, representation, selector, and estimator. Applications illustrate these differences in physical systems. Simulation, retrospective subsets of measured data, and implemented sparse networks provide different levels of validation. Research priorities include calibrated uncertainty under deployment shift, feasible layouts, benchmarks that isolate placement gains from estimator capacity, and physical validation under realistic failure and life cycle costs.
Authors
- Chady Ghnatios (ORCID: https://orcid.org/0000-0001-7100-3813)
Institutions
- University of North Florida (US)
Publication Details
- Journal
- Sensors
- Published
- 2026-11-19
- DOI
- https://doi.org/10.3390/s26227309
- Primary Topic
- Model Reduction and Neural Networks
- Type
- article
- Field-Weighted Citation Impact
- 0.00