This study presents a methodology for determining the optimal placement of seismic sensors along the height of buildings to minimize the uncertainty in reconstructing structural responses at non-instrumented floors. Due to the extensive benefits of instrumentation—from model validation to damage detection and structural health monitoring—the number of instrumented structures is steadily increasing. However, to keep installation and maintenance costs within a reasonable range, structures are often instrumented sparsely. The response at non-instrumented locations is typically estimated using deterministic or probabilistic model-based, data-driven, or hybrid methods. Specifically, the authors recently proposed a method that combines a deterministic beam model with Gaussian Process Regression (GPR) to estimate responses at non-instrumented floors of an instrumented building. The present paper proposes a methodology to determine optimal sensor locations that minimize the uncertainty associated with this response estimation. This work is a sequel to a previous study that was limited to stationary excitation and extends the method to seismic excitations. The methodology is first verified through a numerical example and then applied to two real instrumented buildings. The results demonstrate that an average 40% reduction in uncertainty is achievable when sensors are positioned according to the proposed optimization approach, in comparison with a random distribution of sensors. Between the two real-life cases studied in this paper, the level of reduction in the response uncertainty is around 10% for the 52-story building because the existing sensors are almost uniformly distributed, while it is around 80% for the 73-story building because the existing sensors are distributed to measure the localized behavior of the building.
Ghahari et al. (2026) studied this question.