Abstract green illustration of a building wireframe with sensor nodes and a petri dish

AI in the built environment: sensors, digital twins and why mould still needs a lab

Healthy buildingsBy Dr Cameron Jones, PhD, environmental microbiologist, Biological Health Services, Melbourne

Buildings are getting smarter. Sensors log humidity every few minutes, "digital twins" mirror whole buildings in software, and machine learning promises to flag damp before mould ever appears. As a microbiologist, I think this is one of the most hopeful shifts in healthy architecture in years. But there's one thing a sensor can't do: tell you what is actually growing. Here's where the technology stands in 2026, and where a laboratory still has the final word.

Key takeaways

  • Digital twins bring together building models, Internet of Things (IoT) sensors, data analytics and smart controls. Using them for occupant health is a growing research area.
  • Humidity and temperature data can be turned into a "mould index" that estimates the risk of mould growth.
  • Australia's building code and the WELL standard now build moisture and mould control into design.
  • Sensors measure conditions, not organisms. Confirming whether mould is present, and which kinds, still takes a lab.

From smart thermostats to digital twins

A digital twin is a live, data-fed model of a real building. A 2023 review describes how digital twins combine building information modelling (BIM), IoT sensing, data analytics and smart building control. It also notes that people spend about 90% of their time indoors, yet very few digital twin studies had focused on the health of the indoor environment [1]. A 2024 analysis of the research literature mapped how digital twins and machine learning are being combined to improve indoor environmental quality, energy efficiency and occupant comfort [2].

For healthy architecture, the promise is simple. Instead of discovering a problem when someone notices a musty smell, the building tells you conditions are drifting into the danger zone.

Predicting mould before it appears

Turning humidity data into a mould index

Mould needs moisture. Relative humidity above about 70–80% is known to create mould-favourable conditions [3]. One research team placed sensors in the living rooms and bedrooms of 219 homes, logging humidity and temperature every 3–5 minutes for a year. They fed the data into the VTT mould model, which produces a mould index, and compared the results with what occupants reported about visible mould and mouldy smells. The index was linked to those reports, and it worked best when the model was set to be more sensitive than the original laboratory version [3].

Machine learning speeds up design checks

Engineers check wall designs by simulating how moisture moves through them. A 2023 study trained a random forest model on these simulations so it could run many more scenarios, cutting the time needed to assess mould risk in external walls [4].

Predicting damp across thousands of homes

A 2025 study used 2,073 inspection records from an English housing association, combined with energy performance data, to predict which homes were damp. The best model was 63.6% accurate on a balanced dataset and 79.3% on the full dataset. Heating cost, energy consumption and wall energy efficiency were the strongest predictors [5]. That's helpful for deciding where to inspect first, but it's a long way from certainty about any single home.

The rules are catching up

Australia's building code

Under NCC 2022, the condensation management verification methods set a limit: a mould index of 3 must not be exceeded. Many exhaust fans must now vent outside rather than into the roof space [6].

WELL v2 standard

WELL v2 includes indoor air monitors measuring at least three of PM2.5 or PM10, CO₂, carbon monoxide, ozone, nitrogen dioxide, total VOCs and formaldehyde. It also covers ultraviolet treatment of cooling coils and drain pans, humidity control and moisture management [7].

Notice what's missing from that monitoring list: mould itself. None of those parameters tells you whether fungi are growing.

Your sensor says it's damp. Is it mould?

A lab-analysed settle plate test answers the question a humidity reading can't. It shows which moulds are settling in each room you test.

What sensors can't tell you

Sensors are only as good as their accuracy. In a 2024 study, 87 low-cost indoor air monitors were placed next to professional reference instruments. The carbon dioxide, temperature and humidity sensors met their specifications. The total VOC sensors deviated by up to 79% from the reference readings, and the PM2.5 sensors had an error margin of ±27%, compared with the ±15% claimed [8].

Even a perfectly accurate humidity sensor only reports the conditions that favour mould. It can't see a colony behind a wardrobe, and it can't tell Cladosporium from Aspergillus or Penicillium. A mould index is a risk estimate, not a finding.

Why lab confirmation still matters

The smartest approach uses both. Let sensors and AI tell you where to look, then confirm what's there with a laboratory test.

  • Real organisms, not proxies. Our kits use potato dextrose agar (PDA) settle plates. Only living, culturable mould grows on them.
  • Consistent sampling. Each plate is opened for 45 minutes, following the kit's step-by-step instructions.
  • Identified, not just counted. We identify the moulds to genus where possible, and your report includes an estimated CFU/m³ calculated with the published Omeliansky formula.
  • A certified lab. Plates are analysed by Biological Health Services, which holds ISO 9001:2015 certification (AUS-Q-900149) and an ISO/IEC 17025:2017 Certificate of Compliance (AUS-TL-17025-149).

If your smart home monitor keeps flagging high humidity in a bedroom, or a building manager's dashboard points to a problem wing, a settle plate test turns that warning into evidence. Read more about the science on our Our Method page.

Please note: This article is general information, not medical or building advice. If you have asthma, follow your asthma action plan and speak with your doctor. For building defects, talk to a qualified building professional.

Data tells you where to look. A lab tells you what's there.

Confirm what's in your indoor air with a lab-analysed settle plate test, sent from Melbourne.

Renting? Use your results with our Mould & Your Rental guides when you raise mould with your landlord.

References

  1. Cai J, Chen J, Hu Y, Li S, He Q (2023). Digital twin for healthy indoor environment: a vision for the post-pandemic era. Frontiers of Engineering Management 10(2):300–318. https://doi.org/10.1007/s42524-022-0244-y
  2. Karatzas S, Papageorgiou G, Lazari V, et al. (2024). A text analytic framework for gaining insights on the integration of digital twins and machine learning for optimizing indoor building environmental performance. Developments in the Built Environment 18:100386. https://doi.org/10.1016/j.dibe.2024.100386
  3. Menneer T, Mueller M, Sharpe RA, Townley S (2022). Modelling mould growth in domestic environments using relative humidity and temperature. Building and Environment 208:108583. https://doi.org/10.1016/j.buildenv.2021.108583
  4. Bayat Pour M, Niklewski J, Naghibi A, Frühwald Hansson E (2023). Robust probabilistic modelling of mould growth in building envelopes using random forests machine learning algorithm. Building and Environment 243:110703. https://doi.org/10.1016/j.buildenv.2023.110703
  5. Aziz G, Hardy A (2025). A predictive model for damp risk in English housing with explainable AI. Scientific Reports 15:12658. https://doi.org/10.1038/s41598-025-96396-7
  6. Australian Building Codes Board (2022). Improved condensation management in NCC 2022. abcb.gov.au
  7. International WELL Building Institute. WELL Building Standard v2 (features A08 Air Quality Monitoring and Awareness, A14 Microbe and Mold Control, T07 Humidity Control, W07 Moisture Management). v2.wellcertified.com
  8. Manu S, Rysanek A (2024). A co-location study of 87 low-cost environmental monitors: assessing outliers, variability, and uncertainty. Buildings 14(9):2965. https://doi.org/10.3390/buildings14092965
Back to blog