Indoor Climate Change: Understanding the Air We Breathe Using Data (Part 2)
Ask anyone “Did you know that rising carbon dioxide (CO2) levels can cause climate change?” and you will most likely receive a “yes”, accompanied by strong passions or opinions. However, if you ask that same person, “What is the optimal CO2 level for the air you breathe, either outside or in your home or office?” you will probably receive a shrug. In Part 1 of this series, the importance of CO2 measurements to human health, the history and importance of measuring outdoor CO2, as well as the complexity of CO2 measurements in indoor environments, were discussed. Part 2 explores the type of insights that indoor CO2 measurements can provide using machine learning (ML).
Data-Based Insights Using CO2 Measurement Data
Determining CO2 sensor accuracy has been the goal of sensor users for years. Confirming calibration or determining if a CO2 sensors was corrupted or poisoned (developed a layer of oxide coating due to chemical exposure) is a genuine concern. To overcome these challenges, we used machine learning and big data techniques on the data produced from the mass-deployment of our CO2 sensors. We did this not by implementing the ML algorithm at the sensor level, but by getting the data into the cloud and using ML for design and analysis.
For example, temporal drift of CO2 sensors are often seen. The key question a user will want to know, especially when the changes look odd, is if these changes are due to seasonal changes, environmental changes specific to sensor placement, or can it be caused by a physical sensor error . By aggregating data from multiple sensors in both similar and distinct locations, one could use modern machine learning and big data statistical techniques to identify and isolate such drifts, and then adapt as appropriate.
The machine learning techniques discussed here were developed for Infineon’s XENSIV™ PAS CO2 sensor during its development. When hundreds of prototype sensors are deployed in the field, a lot of data (tens of Gigabytes of data over months) is created. It is hard to manually scroll through raw data to identify and isolate anomalies worth exploring. To simplify the process, automated data-based techniques were developed to continuously improve the quality of sensors.
This data-based insight can change the nature of the CO2 sensing problem, which is easiest explained through example. Today, advancements in modern commercial buildings and homes to improve the efficiency and reduce the power consumption of heating ventilating and air conditioning (HVAC) systems have resulted in excellent sealing and insufficient air cycling. Sensing CO2 provides the means to avoid the unsatisfactory levels and the resulting impact on people’s performance and wellbeing, which was shown in Part 1 of this blog. In addition, by reliably measuring the CO2 level of an indoor space, and using this as a proxy to measure air circulation, the risk of transmitting viruses (like COVID-19 or influenza) can be actively managed with an active HVAC system.
Historically, the concern for regulating the HVAC system only involved temperature. While CO2 measurements and CO2 modeling have provided insights into global warming, the same situation has not occurred for the indoor environment. Previously, when measurements were made, the data was acted upon, not accumulated, and put in perspective compared to the local environment. With data collection, one of the insights provided is the visualization of the difference in CO2 levels over time in a building by days of the week.
For example, running a machine learning algorithm in a test room showed that between noon and 2 pm, the CO2 level was way above the outside baseline. Investigation revealed that the room was next to the cafeteria and with COVID-19 distancing in effect, people would get their food and sit in a conference room without social distancing. As a result, the CO2 level increased dramatically during this timeframe. This shows that data analysis can reveal a potential problem. In fact, the difference between indoor and outdoor air is a measure of air quality. If the differential is too high, it means the inside air is more humid and has more CO2, which could indicate the potential for carrying airborne diseases.
In the third and final part of this series, we will discuss specific measurements and the conclusions that have been drawn from them.