Answers are grounded in this paper and cite their sources.
In this paper, the fault-tolerant tracking problem is studied for parabolic PDE multi-agent systems with unknown distributed-parameter faults. A physics-informed neural network (PINN) reconstructs the states and faults from sparse interior measurements, and the learned control field is then applied without measurement updates. The goal is to relate the training loss of the PINN to the resulting tracking error. Under a rank condition on the sampled fault basis, the joint identifiability of the states and fault coefficients is established. Based on a Volterra observation equation, the independently evaluated losses are converted into fault and state error bounds, which yield an explicit tracking radius and transient over strongly connected directed graphs. These bounds provide conditions for comparing sensor placements, selecting a stopping loss and choosing the deployment duration. Numerical examples on a six-reactor network illustrate the results and identify the terms that dominate the a priori tracking radius.
No takes yet. Share an insight, caveat, or question.
Pan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: