Proof of Concept

Industrial Compliance

Demonstrating two AI approaches for low-cost industrial emissions monitoring using only free, public data. No physical sensors required.

Proof-of-concept scope: All data on this page is either (a) real ambient measurements from public APIs used as a stand-in for reference instruments, (b) mathematically simulated sensor readings based on documented error models, or (c) real published research data from non-Kenyan sources used to validate the modelling approach. No proprietary sensor data has been used. Each facility page clearly states its data provenance.

Module A

AI Sensor Calibration

Cheap optical PM sensors (PMS5003/PMS7003 class, ~$15) suffer predictable biases: humidity over-read, baseline drift, and read noise. A RandomForest model trained during a short co-location window with a reference instrument can correct these errors, recovering near-reference accuracy for continuous fence-line monitoring.

Module B

Predictive PEMS

Certified Emission Monitoring Systems (CEMS) analysers cost upward of $50,000. This module demonstrates that an XGBoost model trained on cheap process variables (temperature, pressure, humidity) can predict CO and NOX concentrations with high accuracy, validated on real published gas turbine data from a Turkish plant (UCI repository).