SENSING-00 / MODALITIES
Sensing through earth, concrete, and silence.
Four independent detection modalities — ground-penetrating radar, passive acoustic, seismic mesh, and sensor fusion — operating in parallel within a single deployable node. Each catches what the others miss.
GPR-01 / GROUND-PENETRATING RADAR
GPR: maps what acoustic cannot reach.
Ground-penetrating radar transmits electromagnetic pulses into the substrate and measures the time-of-flight of reflections from density boundaries — the interface between soil and void, concrete and tunnel cavity.
The fundamental tradeoff: higher frequency (1–2 GHz) provides centimeter-scale resolution but penetrates only 3–5m. Lower frequency (300–500 MHz) reaches 10–15m but resolves voids at 0.3m³ minimum. The Tarysar GPR module is field-tunable across this range, allowing the operator to optimize for the specific substrate and depth of interest.
In a running tunnel sensor network, GPR nodes activate on a 30-minute sweep cycle by default to conserve battery, with event-triggered active sweep when acoustic or seismic modalities log a classification event.
ACOUSTIC-02 / PASSIVE ARRAY
Passive acoustic: footfall, cadence, excavation.
The passive acoustic array covers 2 Hz – 20 kHz, encompassing both infrasound (tunneling machinery, heavy vehicles) and the full audible range (footfall, voice, equipment). Each node carries a multi-channel transducer array for directional signal processing.
On-device classification uses a lightweight neural architecture trained on labeled vibration signatures: human footfall cadence patterns, excavation tool vibration profiles, water flow signatures, and geological settling events. The goal is not perfect recall — it is minimum false positives.
Multi-node confirmation is required before an event is escalated. A classification event logged by a single node is held at status "pending" until a second node within acoustic range confirms. This two-node confirmation step reduces false positive escalation by an estimated 60–70% in field testing — internal data, not independently verified.
SEISMIC-03 / DISTRIBUTED ARRAY
Seismic mesh: structural movement at 0.0005g.
Seismic geophones in each node detect ground-coupled vibrations — the mechanical energy conducted through soil, rock, and concrete. At 0.0005g minimum sensitivity, the array can detect footfall cadence at 150m through tested substrates, and construction-scale excavation at significantly greater distances.
The key challenge in seismic detection is classification: distinguishing human-relevant signals from background geological activity, vehicle traffic on surface roads, water main vibration, and equipment cycles. The Tarysar seismic classifier segments 12 event categories, with dedicated models for tunnel-context environments versus open-ground deployment.
Mesh correlation — comparing seismic waveforms across multiple nodes simultaneously — provides directional velocity estimation for moving sources, which is not achievable from a single node.
FUSION-04 / COMBINED OUTPUT
Sensor fusion: confidence from disagreement.
No single modality is reliable alone in underground environments. Acoustic saturates on water drip and geological settling. GPR cannot reach targets below its penetration depth. Seismic generates false positives from surface road traffic and water main vibration. The fusion engine is where detection reliability emerges from the combination.
Each modality outputs a continuous confidence score and a category label. The fusion engine applies weighted Bayesian combination, with modality weights adapted for the substrate profile loaded at deployment. Escalation requires either (a) single-modality confidence above 0.92, or (b) two-modality corroboration above 0.75 each. Events that don't meet threshold are held as "pending" — not discarded, not escalated.
Alert latency from fusion output to operator notification: under 8 seconds at 4-node mesh in tested conditions. Internal test data — not independently verified. Not a certification claim.
Discuss technical fit for your scenario.
Substrate type, corridor geometry, threat velocity profile, and operational tempo all affect sensor configuration. Our engineers discuss specific deployment scenarios in pre-engagement conversations.