Non-skin-contact wearable for personalised thermal comfort HVAC feedback

Product · HVAC & Energy

Product · InnDex 18 · Evidence provided · High specification risk

Wrist-worn non-contact thermal sensor with ML-predicted HVAC feedback for personalised comfort.

A wearable device measures individual skin temperature and humidity contactlessly, inputs data into a random-forest model to predict thermal sensation, and generates real-time HVAC adjustment recommendations. It addresses occupant thermal comfort variability—a major source of HVAC inefficiency and dissatisfaction—by personalising setpoints to individual physiology rather than space averages. Lab testing (29 subjects, single office) showed 17% accuracy gain over PMV and 98% prediction in controlled conditions.

This wrist-worn device measures individual skin temperature and humidity without contact, feeds the data into a random-forest model to predict thermal sensation, and generates real-time HVAC adjustment recommendations — addressing the well-documented mismatch between zone-averaged setpoints and individual occupant comfort. Lab testing with 29 subjects in a single office showed 17% accuracy improvement over the PMV model and 98% prediction under controlled conditions. One piece of evidence supports these lab-scale figures. The central risk is generalisation: a model trained on 29 people in one room and one season is not a robust basis for deployment across diverse building types, climates, occupant demographics, clothing, and activity levels. The practical dependency on continuous voluntary device wear is also a high bar — adoption friction in real occupancy scenarios is fundamentally different from a research participation setting. Beyond the model, there is no defined pathway for integrating wearable-generated setpoint recommendations into existing BMS or HVAC controls, which is a non-trivial engineering task. Continuous biometric data streaming raises occupant consent and data privacy obligations (GDPR, CCPA) that are not addressed in the research. There is no commercial product and no evidence of a transition from laboratory research to a viable product. A specifier should file this as a directionally interesting concept — personalised comfort is a real gap — while recognising that it would need substantial additional validation and a credible commercial form before it could be specified on a project.

Strengths

• Captures individual thermal sensation without skin contact, reducing adoption friction vs. intrusive sensors
• Personalised HVAC feedback addresses known gap between zone-based control and occupant comfort variance
• ML model potentially adaptable across populations if generalisation is solved
• Real-time adjustment capability could reduce thermal cycling and associated energy waste

Considerations

• Requires continuous device adoption and wear by occupants—no infrastructure change, high behavioural dependency (high)
• Non-contact measurement adds cost and complexity vs. standard thermostats; limited evidence of durability/calibration drift (moderate)
• No quantified energy savings reported; unclear if personalised feedback reduces total building energy or just shifts discomfort (high)
• Single-building, single-season validation; unclear performance across climates, building types, or HVAC architectures (high)

Risks

• ML model trained on 29 subjects in lab; severe risk of poor generalisation to diverse occupant populations, age, clothing, activity levels in real buildings (critical)
• No integration pathway with existing BMS or HVAC systems defined; likely requires retrofit controls and coordination logic (high)
• Privacy and data security not addressed; continuous biometric streaming from wearable raises occupant consent and regulatory risk (GDPR, CCPA) (high)
• Beluga-whale optimisation algorithm is unconventional and not peer-validated in thermal comfort literature; reproducibility and robustness unclear (moderate)
• No commercial product exists; no evidence of transition from research to viable business model or supply chain (moderate)

Performance

• Thermal comfort prediction accuracy (in-sample): >98% (controlled lab)
• Accuracy improvement over PMV baseline: 17.04%
• Study cohort: 29 subjects, single office room
• Claimed comfort improvement (discovery layer — unverified): 70% (not corroborated)

Reality check

Academic paper (ScienceDirect) reports controlled lab results only: n=29, single room, no field trial or post-deployment data. Claimed 98% accuracy is within controlled conditions and not generalizable. The 70% comfort improvement cited in discovery layer is not supported by the paper (which reports 17% accuracy improvement over PMV, not comfort outcome). No independent energy-savings data. No commercialised product, pilot deployment, or user feedback found. Beluga-whale algorithm optimisation is a novelty that lacks peer validation.

#wearable #thermal_comfort #machine_learning #personalisation #indoor_environment

Source