Salt, moisture, and heat: how a KES 4,500 ESP32 thermal camera can find building defects before they become structural problems

A building on Mombasa Island, Nyali, or the Diani beach strip ages differently from a building in Nairobi. The difference is the marine environment. Salt-laden air — carried inland by the southeast monsoon from April to October — deposits chlorides on exposed concrete surfaces. These chlorides penetrate through micro-cracks in the concrete cover and reach the reinforcing steel. Once chlorides reach the steel at sufficient concentration, corrosion begins. The steel expands as it rusts, cracking the concrete from the inside. The first visible sign — a rust stain or a hairline crack — appears months or years after the corrosion started.

At the same time, the coastal climate imposes two moisture regimes. During the long rains, driving rain saturates external walls that have not been properly sealed. During the dry season, high humidity keeps internal spaces damp. Moisture trapped in wall cavities and roof spaces supports mould growth, degrades finishes, and accelerates the corrosion of metal fixings and fasteners.

The third mechanism is thermal. Coastal buildings — particularly older apartment blocks and villas with uninsulated concrete roofs — absorb solar radiation during the day and radiate heat into the interior at night. The thermal expansion and contraction cycle stresses materials, opens joints, and creates pathways for moisture ingress. The occupants feel it as a hot room and a high air-conditioning bill. The building feels it as cumulative structural stress.

The problem with all three mechanisms is that they are invisible until they are expensive. By the time a rust stain appears on the wall, the rebar behind it has already lost cross-section. By the time a damp patch appears on the ceiling, the roof membrane has been leaking for months. Thermal cameras — sensors that see temperature differences rather than visible light — can detect these problems before they become visible.

A thermal camera captures infrared radiation — heat — rather than visible light. It produces an image where each pixel represents a temperature rather than a colour. On a building, a thermal image reveals patterns that are invisible to the naked eye.

Moisture in a wall or ceiling evaporates more slowly than the surrounding dry material, creating a cool spot that the thermal camera sees as a temperature anomaly. An electrical connection that is loose or overloaded generates heat — a hotspot that the thermal camera sees long before a fuse blows or a fire starts. A gap in wall insulation or a failed window seal shows as a thermal bridge — a line or patch where heat is transferring faster than through the surrounding structure. An AC unit with a blocked condenser coil runs hotter than it should — the thermal camera sees the elevated temperature before the unit fails.

Until recently, thermal cameras were expensive professional instruments costing upwards of KES 50,000. The AMG8833 thermal sensor — an 8x8 pixel infrared array — changed that. It costs approximately KES 2,500 and connects directly to an ESP32 microcontroller. Combined with an ESP32-CAM module for a visible-light reference image, the total hardware cost is under KES 4,500.

The resolution is low — 64 pixels cannot produce a detailed thermal image. But 64 pixels are sufficient to detect temperature anomalies at the scale of a wall section, a ceiling panel, an electrical panel, or an AC condenser. A machine learning model running on the ESP32 compares the thermal pattern against a baseline and flags anomalies — no human needs to interpret the thermal data, and no cloud processing is required.

The traditional thermal inspection workflow requires a trained thermographer: someone who understands emissivity, reflected temperature, and thermal patterns specific to building materials. The thermographer takes thermal images, interprets them, and writes a report. This is a specialised professional service — valuable, but not affordable or practical for routine building monitoring.

Edge AI changes the workflow from skilled interpretation to automated anomaly detection. The ESP32 runs a TinyML model that has been trained on baseline thermal patterns for the specific building — the normal temperature distribution across a south-facing wall at 2 pm in August, the normal temperature of the main electrical panel under load, the normal cooling pattern of the AC condenser during operation.

The device captures thermal readings on a schedule — daily, weekly, or monthly. The TinyML model compares each reading against the baseline. If the reading is within normal variation, nothing happens. If the reading deviates — a persistent cool spot on a wall that was previously uniform, indicating moisture ingress; a hotspot on an electrical connection that was previously ambient, indicating a failing contact — the device flags an alert.

No thermal imaging expertise is required to operate the system, because the system does not produce images for human interpretation — it produces alerts based on deviation from baseline. The building owner or maintenance manager receives a notification: 'South wall, third floor, moisture anomaly detected — recommend inspection'. The thermal expertise is embedded in the model, not in the operator.

The hardware: an ESP32 development board (KES 800), an AMG8833 thermal sensor (KES 2,500), and a small solar panel or battery pack (KES 1,200). Total hardware: approximately KES 4,500 per monitoring point. A typical apartment building might deploy three to five monitoring points — one for the electrical room, one for the roof, and one for each exposed facade. Total system cost for the building: KES 15,000-25,000.

The software: Edge Impulse or TensorFlow Lite Micro for the TinyML model. Both are free and open-source. The model training requires a laptop and a few weeks of baseline thermal readings to establish normal patterns. A university engineering student or a local electronics technician can set up and calibrate the system.

What this prevents: a single undetected moisture ingress event that leads to rebar corrosion and concrete spalling can cost KES 100,000-500,000 to repair on a coastal building, depending on the extent and location of the damage. A single electrical fire in a distribution panel — the most common cause of building fires in Mombasa — can cost lives and millions in damage. A single AC condenser failure during the hot season costs KES 30,000-80,000 to replace and weeks of discomfort for occupants.

The economics are straightforward: the hardware costs less than one unplanned repair. The system pays for itself the first time it catches a problem before it becomes a failure.

Preventive thermal inspection of buildings is standard practice in markets where construction quality risk is priced into insurance and maintenance contracts. Dubai mandates thermal inspection of building envelopes as part of its green building regulations. Singapore requires periodic thermal scanning of electrical installations in commercial buildings. Facilities management contracts in Europe and North America increasingly include thermal monitoring as a standard scope item.

These markets use professional-grade thermal cameras costing $5,000-20,000 and trained thermographers charging professional rates. The same quality of anomaly detection — not the same image resolution, but the same ability to detect moisture, thermal bridging, and electrical hotspots — is now available on hardware costing under $50.

The application to Kenya's coastal buildings is direct and specific. A hotel in Diani that wants to reduce maintenance costs and avoid guest complaints about damp rooms. An apartment block in Nyali where the owners' corporation wants to catch envelope defects before they become special levy items. A commercial building on Moi Avenue where the facilities manager wants to prevent electrical failures that close the building for days.

REDM's construction management and MEP engineering services cover the professional side — specification review, installation inspection, snagging, and handover inspection. The edge AI thermal monitoring layer extends this coverage into the operational life of the building — the years after handover when maintenance decisions determine whether the building lasts 20 years or 50.