Development of an AI-enabled Low Cost Real Time Mine Subsidence Monitoring, Prediction and Early Warning System for Underground Coal Mines in India
Ministry of Coal · Disaster Management · Hardware
The rare hardware statement with a genuinely student-scale bill of materials, a near-complete spec and a physically honest demo — just be candid that you are showing deformation detection rather than validated subsidence prediction, because the sponsor has told everyone what to build and execution will decide it.
What it actually is
When coal is mined underground the ground above it slowly sinks, and that can crack houses, roads and farmland with no warning. Right now it is checked by surveyors visiting periodically, which usually means the damage is found after it happens. The ask is a cheap network of sensors spread over the surface that watches the ground continuously and raises an alarm early.
What to build
A wireless surface mesh of low-cost sensor nodes over an underground panel, each node carrying the sensing the statement names — tilt and inclination, vibration, displacement or stretch, crack initiation, with optional positioning — running on ESP32-class hardware with solar or long-life battery power, linked by a LoRa or Zigbee mesh so nodes relay through each other without cellular coverage, feeding a server that tracks inter-node relative distance and tilt drift over time, runs anomaly detection over the deformation field to flag abnormal patterns and estimate progression, renders a live GIS deformation map with risk zones, and dispatches SMS and app alerts, with local buffering and periodic cloud sync.
Smallest thing that wins the room
Lay the mesh across a sand tray, lower one section of the base to simulate ground settlement, and watch tilt drift propagate node to node through the mesh into a deformation map that shades red and fires an alert.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 88% of the 226 · #27 of 226 by expected field
Few teams are likely to go here. The best odds on the board come from statements like this.
Why: central ministry statements sat below the average; hardware halves the field a software statement gets.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
4/5An unusual combination — hardware that is actually affordable, a specification precise enough that you cannot misread it, a physically honest demo, and a hardware category that thins the field — with the main drag being that the sponsor has pre-announced the intended innovation so most entries will look alike.
Feasibility
4/5Uniquely among hardware statements this one specifies its own affordable bill of materials — the description names Arduino, ESP32 and Raspberry Pi, and MPU6050 tilt sensors, accelerometers, strain-based crack sensors and LoRa mesh radios together cost a few hundred rupees per node, so a genuine multi-node network is within a student budget.
Innovation scope
3/5The architecture is handed to you in unusual detail — the mesh topology, the sensor types, the hardware platforms and even the intended differentiator are all named — so your originality is confined to the deformation analytics and the node power and packaging design.
Clarity
5/5This reads closer to a design document than a problem statement: it specifies the sensing modalities, the network protocols, the hardware platforms, the four detection targets, the five AI tasks and the full output stack, and even leaves in the author's own note identifying the intended innovation hook.
Effort
HeavyMultiple sensor nodes to build and calibrate identically, mesh networking firmware with power management, a deformation analytics layer, GIS visualisation, alerting and offline sync is six workstreams, and node-to-node calibration consistency is a slow, unglamorous grind teams underestimate.
Demo-ability
EasyGround tilt is one of the few geotechnical phenomena you can genuinely reproduce on a table — a sand tray with a lowering section produces real deformation that your real sensors really detect, so the demo is honest rather than staged.
In its favour
- Green flag: The statement names the hardware platforms and calls the solution student prototype friendly, which is a rare and explicit signal that the sponsor expects something buildable rather than aspirational
- Green flag: Per-node cost of a few hundred rupees means you can build eight or ten nodes and demonstrate a genuine mesh rather than two devices calling themselves a network
- Green flag: Ground tilt is physically reproducible on a bench, so your sensors detect real deformation and the demo does not rely on simulation
- Green flag: The specification is detailed enough that requirements ambiguity — the usual cause of a wasted first day — simply does not arise here
Against it
- Red flag: No dataset of real subsidence deformation signatures exists for you to train on, so the prediction and severity estimation half is learned from tilted sand and must be presented as detection with trending rather than validated forecasting
- Red flag: The description literally contains the author's own note naming 'Wireless Surface Mesh Network for Real Time Subsidence Detection' as the innovation hook, which means every team reads the same intended answer and submissions will converge
- Red flag: Real subsidence develops over months at millimetre scale while your bench test moves centimetres in seconds, so sensor resolution and drift over long deployments are questions your demo cannot address
- Red flag: Node-to-node calibration consistency across ten identical units is tedious and is where uncalibrated meshes produce nonsense deformation fields that look like signal
What you will be writing
- ESP32 + MPU6050 tilt and inclination nodes
- LoRa mesh via Meshtastic or RadioHead
- strain gauge crack initiation sensing
- isolation forest / autoencoder deformation anomaly detection
- Leaflet GIS deformation heatmap
- solar node power with deep sleep duty cycling
- Mine safety and geotechnics
- Wireless sensor networks
- Structural deformation monitoring
Prior art to read before you start
wireless mesh sensor network deployment · ground deformation and tilt monitoring · low-cost IoT early warning system
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.