The engineering story behind sparse drain monitoring, with the assumptions left visible.
IMERG_2024_08_11
Network at rest. Four proposed sensor locations are online.
Hydraulic states are not independent. Network connectivity and shared rainfall forcing correlate them, so a handful of well-chosen locations can recover the dominant patterns that describe the rest.
Five stages that keep observed, physics-derived, simulated, and computed information separate. They narrow left to right, and the panels narrow with them.
254 features
Real infrastructure
GMDA OneMap geometry and drain attributes anchor the network in Gurugram.
369 × 38
Physics
EPA SWMM routes historical IMERG rainfall through the normalised drainage model, producing peak depth at every node for every scenario.
Φ, 38 × k
Sparse sensing
SVD captures the dominant hydraulic patterns; pivoted QR ranks the node rows that form a numerically stable sampled basis.
4 rows
Reconstruction
Four deployable readings estimate the peak-depth state at the remaining 34 modelled points.
Top 3
Fault localisation
A Random Forest maps sparse hydraulic signatures to candidate conduits and reports real probabilities.
38
modelled nodes
SVD + QR
reduced basis
4
proposed sensors
34
states estimated
State feature
Peak node depth
Interpretable hydraulic response
Energy retained
95%
Reduced-order dynamics
Sensor budget
4
Against 75 random placements
Deployment framing
Proposed
Not live municipal sensors
OSDosed · Smart India Hackathon 2026
GMDA · IMERG · SWMM · SVD/QR