Experiment Log
A chronological, dated record of every Vriksh Vani research trial, sensor calibration event, and benchtop experiment — with conditions, outcomes, and raw data links where available.
BME688 MOX Gas Resistance Baseline Calibration Under Humidity Gradient
Hypothesis
BME688 raw gas resistance measurements vary significantly with ambient humidity and require SHT41 compensation before biological signal attribution.
Experimental Conditions
- ·Ambient RH range: 28%–65%
- ·Temperature: 24°C +/- 0.5°C (controlled)
- ·Duration: 48 hours continuous logging
- ·Hardware: BME688 + SHT41 + ESP32-S3 DevKit
Outcome
Gas resistance ranged from 480–590 kOhm across humidity gradient. Strong RH correlation confirmed (R2=0.94). Compensation model calibrated. Baseline drift: approx 2.1 kOhm per % RH.
FLIR Lepton 3.5 Thermal Core CPU Heat Leakage Isolation
Hypothesis
ESP32-S3 CPU heat dissipation (0.8W) will contaminate FLIR thermal measurements without a dedicated physical isolation barrier.
Experimental Conditions
- ·ESP32-S3 at 100% CPU load (inference task)
- ·FLIR Lepton 3.5 mounted 22mm above compute PCB
- ·Two conditions: with and without ceramic fiber barrier
- ·Environmental: 22°C ambient, still air
Outcome
Without barrier: +2.8°C CPU heat leakage into LWIR path confirmed. With 2.5mm ceramic fiber barrier: zero measurable thermal contamination. Design requirement confirmed.
Monstera deliciosa Leaf Surface Thermal Shift Under VPD Elevation
Hypothesis
Elevating ambient VPD from 0.8 kPa to 2.8 kPa will induce stomatal constriction, measurable as a positive leaf surface temperature shift relative to ambient air.
Experimental Conditions
- ·Species: Monstera deliciosa (mature, healthy, acclimated)
- ·VPD elevated by reducing ambient humidity via silica gel
- ·Duration: 5-hour trial, 1-hour VPD intervals
- ·Thermal measurement: modelled simulation — physical FLIR not yet available
- ·Control: empty pot with moistened substrate
Outcome
ILLUSTRATIVE MODELLED RESULT: Simulated +1.4°C positive leaf delta at VPD=2.8 kPa (180 minutes before visible wilting). Physical FLIR hardware required for experimental validation. Null hypothesis cannot yet be rejected.
ESP32-S3 Local INT8 TinyML Inference Latency Benchmark
Hypothesis
A quantized INT8 TFLite Micro model fitting within 200KB Flash can run inference on ESP32-S3 in under 100ms without dedicated NPU hardware.
Experimental Conditions
- ·Model: 3-class INT8 quantized TFLite Micro
- ·Model footprint: 184KB Flash, 62KB SRAM
- ·Hardware: ESP32-S3-DevKitC-1 at 240MHz
- ·Input: 5-feature sensor fusion vector
- ·100 sequential inference runs, averaged
Outcome
Mean inference latency: 38ms per frame. P95 latency: 42ms. Zero external network calls. Model accuracy on held-out synthetic dataset: TBD — requires real sensor data.
Root Zone Gas Resistance Under Progressive Substrate Moisture Deficit
Hypothesis
BME688 MOX gas resistance shifts systematically under root zone moisture deficit beyond humidity-corrected baseline, suggesting root zone biological activity changes.
Experimental Conditions
- ·Three groups: test plant, control plant, blank substrate
- ·Progressive drying over 30 days from field capacity to 30% moisture
- ·BME688 readings every 30 minutes
- ·SHT41 compensation logging every 5 minutes
- ·Gravimetric soil moisture every 6 hours
Outcome
PENDING — experiment not yet executed.
Full NIH-01 Prototype Sensor Fusion Validation Trial
Hypothesis
Integrated NIH-01 prototype combining FLIR + BME688 + SHT41 + ESP32-S3 can produce stable, reproducible multi-sensor telemetry across 72-hour continuous indoor plant monitoring trial.
Experimental Conditions
- ·Requires: NIH-01 ceramic prototype assembly
- ·Requires: FLIR Lepton 3.5 hardware
- ·Requires: custom PCB design and fabrication
- ·Target: multiple plant species, 3 environmental conditions
Outcome
PENDING — awaiting Phase 05 hardware assembly.