American Scientist

In any office, home, or shared space, there’s almost always someone who’s too cold, someone who’s too hot, and someone who doesn’t know what the fuss around the thermostat is all about.

Even though the variables that determine a comfortable temperature are all too human, building environments typically use static temperatures—a range of 20 to 23.8 degrees Celsius (68.5 to 75 degrees Fahrenheit) in winter and 23.8 to 26.9 (75 to 80.5) in summer—to program and run office heating, ventilation, and air conditioning (HVAC) systems. As a result, people often feel too hot or too cold, despite how much energy heating and cooling systems expend in an effort to create comfort.

Regulating an Artificial Environment 

In the United States and worldwide, HVAC systems account for about 50 percent of the total energy required to operate commercial and residential buildings. Commercial buildings alone account for 22 percent of the United States’ national energy consumption. Yet this high energy use rarely results in its goal of universal satisfaction with a building’s temperature.

Because HVAC systems are programmed to always deliver predetermined levels of heating or cooling, they do not take into account, or adjust for, human data or personal comfort preferences. Further, because these systems do not adjust for variable factors—such as heat from direct sunlight—in a space, the air temperature is not uniform and people can experience varied thermal sensations.

Performance, Health, and Comfort

People would be more comfortable if building HVAC systems could respond in real time to their varying comfort levels and their fluctuations throughout the day. A person’s satisfaction with their environment, and specifically their thermal comfort level, has a direct effect on their health, well-being, and performance.

And, considering that people in the United States and Canada spend an average of 90 percent of their time indoors in the summer and 97 percent indoors during the winter, this relationship can have significant, far-reaching effects. Employee performance in indoor offices can be greatly affected by the environment’s perceived thermal condition. An employee who is personally comfortable in their thermal environment is likely to have a reduced number of complaints, decreased absenteeism, increased productivity, and improved motivation.

An uncomfortable thermal environment, however, leads to low job satisfaction, negatively impacts job performance, decreases self-estimated performance, and may cause difficulties concentrating. High room temperatures can lead to increased reports of symptoms of sick building syndrome (a feeling of ill health that occurs in the occupants of a building, such as eye, nose, and throat irritation) as well as increased mental workload to complete cognitive tasks.

And although factors such as motivation can offset a negative thermal environment’s effect on task performance, dealing with that thermal stress can still deplete a person’s overall neural resources. In the long term, that continuous high mental workload can be detrimental to overall health. Controlling the thermal environment through the HVAC system by adjusting thermostat set points is one of the simplest ways to achieve an optimal workplace environment.

Different Comfort Levels

Most often, building owners find out how well their heating and cooling systems are operating by asking how comfortable the occupants are. However, everyone has an ideal temperature that may change at any given time based on all sorts of factors, including their age, gender, physical activity level, apparel, and even current stress level. These differing comfort levels create a complex problem: how to determine the most comfortable thermal environment for the greatest number of people, and how to successfully regulate it.

The predicted mean vote (PMV) model, created by Danish environmental engineer and expert in thermal comfort Povl Ole Fanger, assesses thermal comfort using four environmental factors (air temperature, mean radiant temperature, air velocity, and relative humidity) and two human factors (metabolic rate and clothing insulation). The PMV model has become an international standard and is used to evaluate occupant indoor thermal comfort on a seven-point thermal sensation scale, with -3 being cold, 0 being neutral, and +3 being hot.

This model doesn’t account for all factors, however, and its major drawback is that human factors are assumed to be the same across all people—it doesn’t take personal variations into consideration. Even when exposed to the same indoor environment and workload, people may still have diverse sensations and preferences because of their own personal factors. Time of day and seasons influence this comfort level as well: People entering a cool room in the summer may initially feel comfortable but end up feeling too cold after a while.

Assessing Thermal Comfort

Improving a thermal environment first requires an understanding of both the current situation and relevant information on the occupants of the space to determine how it could be improved, which means occupants’ thermal comfort must be accurately assessed. There are two main ways to achieve this measurement: occupant self-reports and comfort assessments through human physiological responses.

Some researchers have proposed asking occupants to self-report their comfort level or vote on what they feel the temperature should be. Using a phone app or website, building occupants select whether they’re too hot or too cold, and what would make them more comfortable. An algorithm then analyzes the answers and calculates a temperature estimated to be the most acceptable to most people. However, for it to work well, self-reporting requires near-constant input from people who are supposed to be working—and it still doesn’t factor in whether someone who is uncomfortable could help themselves by putting on or taking off a sweater. It also doesn’t take into account how people’s bodies experience temperature, which is closely tied to how cool or warm they prefer their environment to be.

Another option to assess thermal comfort is to measure physiological responses from human bodies. A body’s physiological responses—such as skin temperature, respiration, and heart rate—are correlated to its comfort level and surrounding thermal sensations. As a result, it’s possible to gather occupants’ comfort status without inconveniently requesting self-reports or votes. Among these physiological responses, skin temperature is an effective predictor of thermal comfort.

Skin temperature can be measured using wearable devices, usually through either direct skin contact or infrared thermometers placed very close to the skin. Direct skin contact measurements are collected with contact thermocouples. These are accurate and easy to use, but can be intrusive because they must always be in contact with the person’s skin. Because temperature gradients on the hand, wrist, and upper arm are good indicators of thermal sensation, contact thermocouples are often placed in wristbands. Infrared thermometers are more difficult to use because they must be placed very close to the skin to obtain accurate readings, which means that although good choices exist, such as those built into eyeglass frames, there are fewer less-intrusive options.

In previous research, our group used a combination of these devices by placing multiple temperature sensors around an office, having occupants wear wristbands that monitored skin temperature and heart rate, and requesting occupants to self-report their comfort level. We found that adding data about how people’s bodies were reacting made the algorithm more accurate at calculating the room temperature at which people occupying a given space would feel most comfortable. However, this combination of data-gathering techniques still suffered from the previously mentioned challenges, which leads to a third option: thermal cameras.

Thermal cameras create images using infrared radiation and can be used to remotely sense occupant body temperature. They have a longer and more flexible working range but generally are less accurate than contact thermocouples and infrared thermometers. They also provide a full-frame image of thermographic measurements that can be used to determine temperature readings at each pixel location.

Although it can be helpful to incorporate additional information, such as heart rate and activity level, from other data collection methods into the comfort prediction model to improve its accuracy, skin temperature is often the most effective and easy-to-use quantitative data.

Data from Occupant Bodies

The human body maintains its core temperature at around 37 degrees Celsius (98.6 degrees Fahrenheit) through thermoregulatory control of blood flow to the skin’s surface. Because skin temperature is directly affected by changes in blood flow, these changes in temperature are often used to estimate thermal comfort and sensation.

If an occupant is under heat stress, their body will react: Vasodilation—the widening of blood vessels—increases blood flow to the skin’s surface, which allows the body to dissipate excess internal heat and increases the skin’s temperature. If the same person is too cold in their environment, their body will react: Vasoconstriction decreases blood flow to limit heat loss, which results in a cooler skin temperature.

Measuring people's comfort in an environment and then automating HVAC systems to adjust as needed would be groundbreaking for many high-occupancy spaces.

• Facebook
• Twitter

The human face has a higher density of blood vessels than other skin surfaces, which leads to a larger variation in skin temperature in response to changes in the body or environment. This feature makes faces the best location for consistent, noninvasive skin temperature measurement. One experiment found that a variation of 5 degrees Celsius (9 degrees Fahrenheit) in room temperature shows a statistically significant impact on facial skin temperature.

The ideal facial regions to measure skin temperature are those that show the overall largest variations when under thermal stress, which are the ears, nose, and cheeks. Under heat stress, these features have large temperature variations. Under cold stress, ears have the largest variation, followed by the nose, cheeks, and forehead. While a person is in a cooling or heating phase, cheeks show the highest variation.

However, some facial regions, such as the forehead and mouth, are more sensitive to cold stress, which shows that the significant features used to predict a person’s thermal comfort may vary under hot and cold conditions. In addition to being sensitive to temperature changes, faces are ideal for remote monitoring because they often are not covered by clothing, so any emitted infrared energy can be directly measured using a thermal camera. Also, algorithms can detect human faces well, and can be written to easily locate specific regions of interest—such as the ears, nose, and cheeks—within images for data analysis.

Automating Temperature Changes

To accurately collect and process the data, each data point first must be detected correctly. A machine learning–based object detection algorithm detects the existence of certain features—such as edges or changes in texture—in an image to identify first faces and then specific facial features.

Once faces have been correctly identified, facial skin temperature measurements are processed. Thermal cameras capture images and temperature readings of occupant faces. A set of algorithms then divides the images into different facial regions, extracts temperature readings based on the regions of interest, discards the images to alleviate privacy concerns, processes the data, and determines the best thermal change to stay within statistically acceptable comfort levels for most people.

Comfort prediction models then take what are called linear regression models, which use skin temperature as the independent variable and the overall thermal sensation as the dependent variable. These data are combined with machine-learning methods that complete the regression and classification analysis on the data; common methods used to classify thermal comfort are support vector machines (which classify new data based on learned categories) and decision trees (which guide the response path through a framework of possible outcomes based on the input).

An HVAC control framework then uses these algorithms, machine-learning methods, and human data to set the building temperature controls. This process is called a human-in-the-loop system, meaning it requires both human input and machine intelligence to create the machine-learning model. In this specific instance, the human input is the physiological responses to thermal sensations and behavioral data, then the system’s allowance for human-based adjustments to the HVAC system.

A collective decision algorithm determines the most optimal temperature set point by maximizing the PMV index, or group comfort, score. Then an ensemble classification algorithm called Random Forest is used to produce better prediction results of classifying objects by averaging a large collection of decision trees. These personalized prediction models only need to be trained once and then will continuously predict each occupant’s comfort level. A control script then continually executes the decision algorithm to connect the data to the physical HVAC system. All of this culminates in an automated, user- centric, data-driven system that will consistently monitor occupant skin temperature and adjust the environment’s temperature to ensure each person is as comfortable as possible.

This method is most effective in multi-occupancy spaces, such as open-plan offices, meeting rooms, and theaters. It can accommodate, and account for, differences in temperature between people in different areas of a room, whether they are standing, sitting, or moving around. And it can adjust on the fly without requiring active human feedback. In tests of this system, people complained less about feeling uncomfortably warm or cold, and occupant thermal comfort preference was predicted with 85 percent accuracy by using facial skin temperature measurements.

Measuring how comfortable people are in an environment and then automating HVAC systems to adjust as needed would be groundbreaking for many high-occupancy spaces. Especially exciting would be the successful integration of remote-sensing temperature measurements, which do not place any constraints on occupants’ activities. Integrating a type of occupant registration that would track people as they move through the space would further improve real-time data and, thus, occupants’ comfort level.

A thermally comfortable environment benefits everyone. Occupants experience improved satisfaction in their environment and overall improved health and well-being. Building owners benefit from occupants’ increased motivation and productivity, and they can often reap energy savings from the well-managed HVAC system.

This article is adapted and extended from a version that appeared in The Conversation, https://theconversation.com.