Time for Temperature Tests
In Part 1, Jeff explained how to use the Melexis MLX90614 IR temperature sensor to perform non-contact body temperature monitoring. Here, in Part 2, he dives into actual temperature readings using the system, and highlights some of the important external factors that can affect the readings.
Part 1 of this article series (Circuit Cellar 374, September 2021 ) was about using a Melexis MLX90614 infrared (IR) temperature sensor for non-contact monitoring of body temperature. This sensor is available in 3V and 5V varieties, with several FOV (field of view) package options. I purchased the 3V medical version (ESF), which offers a 5-degree FOV, and the less expensive DAA version, which has a 90-degree FOV (Figure 1). Note: For the rest of this article, I’ll be referring to the ESF as the “-I” version and the DAA as the “-A” version, using the part number designations from the MLX90614’s datasheet .
The medical version (-I) give 0.2˚C accuracy within the normal range of body temperatures. The IR sensor has additional electronics that sample, filter and convert the sensor’s thermopile output directly into Kelvin (K) temperatures. Some conversion parameters are available from the SMBus (I2C-like) registered device. This two-pin communication bus is used to interface with the device’s internal registers. While the device can be instructed to provide temperature output via optional PWM output, I used the 2-wire interface to read registers (temperature) directly.
A sensor element is a series combination of thermocouples or dissimilar metal junctions that create a temperature sensitive micro-voltage producing junction. Multiple thermocouples create a thermopile, where the combination of series voltages are great enough to measure accurately for their temperature varying properties.
I encourage you to read Part 1  to understand the sensor and how its registers are used. Last month’s conclusion included reading all available EEPROM and RAM registers and storing them in
registerArray. Three temperatures are available on this device—ambient (thermopile internal side) temperature, and two IR (thermopile external side) temperatures. The thermopile’s external surface is exposed to IR, and its internal surface remains at ambient temperature. The temperature difference between the “cool” and “hot” sides of the thermopile allows calculation of an external temperature.
The Kelvin temperature scale is based around absolute zero, the point where all matter is solid, at the point of lowest contained energy. Matter has three states—solid, liquid and gas. The Kelvin scale has three points of interest, absolute zero (minimum solid temperature, for all matter), the point where a solid changes to a liquid and the point where a liquid changes to a gas. For instance, water’s melting point (solid to liquid) is +273.15K and its boiling point (liquid to gas) is +373.15K. Absolute zero doesn’t come up very often under normal activity but the freezing and boiling points of water are used often. It makes more sense to have a useful temperature scale where the freezing point was considered 0 degrees and the boiling point 100 degrees. The Celsius scale does this. It is just a Kelvin temperature offset by +273.15.
All this theoretical measurement is fine but our world has variables that make this theoretical perfection a little less than perfect. What are some of these variables? We already know that the sensor has less expensive models, but there are other variables outside the sensor itself. In addition to the different versions of the sensor, two major variables affect the sensor’s performance. They are distance (environment) and emissivity (object surface).
Let’s start with the variable that we have most control over, distance. For now, we’ll leave the emissivity at the default value of 1, a “black body” (see discussion under Emissivity later). My initial tests (first experiment) were taking my own temperature with a sensor at various distances from my forehead, using both the -A (FOV = 90 degrees) and -I (FOV = 5 degrees) sensors.
I began by determining a reference temperature, using an oral thermometer (Figure 2), which has an accuracy of ±0.2°F and a resolution of 0.1°F. The average of 10 readings (lowest = 96.7°F, highest = 97.5°F) was 97.0°F. Note that oral temperatures can vary by more than a degree, depending on where in the mouth you position the thermometer. While my temperature was a bit lower than “normal” 98.6°F, “normal” actually varies for individuals, between 97°F and 99°F.
Next, I took a reading with the -A sensor. With an FOV of 90 degrees, the area of measurement has a diameter of twice the distance to the sensor. So, for an area 1” in diameter, we must be 0.5” away from the sensor. If we move much further away from the sensor, the area of measurement becomes larger than the forehead, and we will be sensing the average temperature of an area that might include the nose, hair and eye glasses. Figure 3 shows the test data for distances of 0.5″ and 6″. This sensor’s readings averaged 94°F (Figure 3a) and 84°F (Figure 3b), respectively.
The -I sensor, with an FOV of 5 degrees, allows the subject to be further away from the sensor, without loss of accuracy. The -I version also has two overlapping sensors. Based on the device’s (TO-39) tab, you can orient the device so the sensors are either vertical or horizontal. You could use the temperature difference between these to help point the sensor, assuming proper placement would create equal temperatures. I did not adjust the sensor position for this test. I just used the higher of the two readings. This time, the tests were at distances of 6″, 12″ and 24″. For the three distances shown in Figure 4, I obtained average readings of 95°F, 95°F and 93°F, respectively.
With the emissivity at the default value of 1, readings for both sensors were a few degrees lower than the oral reference out to optimum distance, but began to fall beyond this point. The advantage here is that of “maximum distance” and not necessarily “accuracy.”
Emissivity is defined as the ratio of the energy radiated from a material’s surface to that radiated from a perfect emitter, known as a “black body”, at the same temperature and wavelength, and under the same viewing conditions. It is a dimensionless number between 0 (for a perfect reflector) and 1 (for a perfect emitter). A “black body emits 100% of its temperature, while a “white body” emits 0%. Most objects fall into the “gray body” area, effectively emitting some part of its temperature.
An object with emissivity <1 does not emit 100% of its actual temperature. The human body has a very high emissivity, above 0.95. Even so, this can vary from person to person, depending on skin type. Our skin surface radiates our body temperature, but this surface may also reflect radiation from other objects. The reflected radiation combines with the emitted radiation, affecting what is measured. The emissivity register allows the measured value to be adjusted, to compensate for a surface’s “grayness.”
All emissivity tests were performed with the 5-degree FOV sensor. In the second experiment, I took temperature readings using a constant sensor distance and subject, varying only the value in the emissivity register. As mentioned earlier, the sensor defaults to an emissivity of 1. The value for this is 0xFFFF. You might think a value of 1 to 10 or 1 to 100 might be sufficient here, but by using this large value, we can adjust by tiny increments. Using 1 to 100 would allow control to a resolution of 0.01. Here we have resolution of 1/65,535 or approximately 0.00002 change per bit.
As the emissivity is lowered, the compensation raises the presented temperature. As shown in Figure 5, the presented temperature has been compensated based on the emissivity register. At some point, the temperature crosses the reference temperature of 97.0°F. This occurred at an emissivity of about 0.96, which is within the expected emissivity of 0.95 to 0.99.
In the third experiment I took temperature readings of a wall with various skin tone color swatches, using a constant sensor distance. Behr neutral paint swatches were applied to the wall, and the temperature of the swatches were allowed to reach room temperature. I expected there to be some measurable difference between the light and dark pigments.
To my surprise I found each swatch to be the same temperature within the variations of multiple temperature samples. I guess I should have expected this, based on the swatches having the same matte surface. Hmm. Maybe it’s not the surface pigment that would affect the emissivity, but the surface itself. This observation led me to design a fourth experiment.
I decided to add one more experiment while I had everything in place. This fourth experiment, which consisted of four different tests, looked at “surface preparation”—how might the surface of the skin, or substances on the skin affect the emissivity? I compared washed skin, free from external substances, with substances that might appear naturally or artificially on the skin. Each test measured the forehead temperature with one side cleansed and the other side coated with some substance. After each test, the entire forehead was cleansed with soap and water, and then with rubbing alcohol. All tests were performed with the emissivity register held constant. The resultant graphs (Figure 6) were printed using the same temperature ranges for ease of comparison.
THE FOUR TESTS
First, since my skin is naturally oily, I started by cleansing just half of my forehead. The surface was allowed to return to temperature to negate any temperature changes from the washing. Previous tests were all performed without any skin cleansing. The emissivity was adjusted based on my natural oily skin. The following tests were all performed with that emissivity register held constant. Figure 6a shows that the temperature of the washed skin was 2°F lower than that of the unwashed skin, with the clean surface less able to emit the actual temperature.
The second test was designed to show the effect of perspiration on skin temperature. Most perspiration is made up of water with hints of sodium, potassium, calcium and magnesium. I didn’t want to raise my body temperature by exercising, so after a skin cleansing process I applied a bit of salt water to one side of my forehead. I hoped this test would show that the evaporation of perspiration lowers temperature. As shown in Figure 6b, the evaporation of salt water lowered the temperature by about 6°F.
(Click graphs to enlarge)
This final experiment involved various characteristics of a forehead’s surface on emissivity. All tests were done without any changes to the emissivity register. (a) Oily skin compared to washed and dried skin. (b) Dry, cleansed skin compared to skin wet with a salt solution representing perspiration. (c) The effect of cosmetic foundation. (d) Comparison of clean, dry skin with skin to which sunscreen has been applied (Y-axis: °F, X-axis: Sample #).
In the third test, makeup was applied to the skin. I wish to state here that I had to ask my wife for help here, since it had been a while since I used any makeup for Cosplay, other than for Halloween. I used a cosmetic (liquid) foundation on half my forehead to see if this affected emissivity. Figure 6c shows that the makeup caused a small loss in emissivity (2°F). This led me to wonder about other chemicals that might commonly be applied to the skin.
The fourth test evaluated the effect of sunscreen on emissivity. SPF (sun protection factor) is a value attached to a substance that indicates how well it protects against harmful UV rays. With proper use, the SPF value indicates how many times longer it would take to burn your skin, than if you had not used any at all. While at opposite ends of the visual spectrum, UV protection might actually block IR. As shown in Figure 6d, the sunscreen did in fact make it more difficult for the heat to radiate from the skin’s surface. The minerals titanium dioxide and zinc oxide are the main active ingredients in many sunblock products. Whether this result is from those or other substances in the sunscreen is unknown.
WHAT DOES ALL THIS MEAN?
The technology to measure body temperature without physical contact is important for preventing the transmission of diseases spread by close contact. When an infectious state can be diagnosed by an elevated temperature, non-contact temperature sensing is a relatively low-cost and immediate technique that can be implemented by anyone with little training. It has been proven that social distancing directly affects the spread of many viruses. Awareness of an infection alerts individuals to be morally cautious when interacting with other people.
The Melexis MLX90614 IR temperature sensor can be used to achieve a high level of accuracy without direct contact. The high-FOV sensors have the same accuracy as the low-FOV counterparts, but lack the optics to achieve distances of greater than about 0.5″. Since IR is invisible (outside our vision range), the sensor must be aimed with some accuracy. Some instruments incorporate a low-level laser (Class II, <1mW) light source that has been pre-focused to the sensor’s “hot” spot, to help aim the sensor at an appropriate area.
In Part 1 of this project, I demonstrated how to communicate with the sensor using an SMBus (I2C) interface. Sensor registers allow a user to configure various adjustable parameters and to perform accurate temperature readings. EEPROM registers hold values between power cycles. Optimal sensor-to-subject distance of the sensor is based on the sensor optics, which must be chosen during sensor selection. As a sensor’s FOV is reduced, its optimal distance to target increases.
Although these sensors can be used for reading the temperatures of any surface, Part 2 examines the skin temperature of a human. While the normal internal body temperature of a human is 98.6°F, skin temperature is lower and depends on where the temperature is taken. For the forehead, you can expect it will be approximately 1°F lower.
Other factors also come into play. For example, women have a wider range of normal temperatures measured orally (91.8°F to 100.6°F) compared to men (96.3°F to 99.9°F) . In general, when the body rests (sleep), temperatures are at their lowest (-0.5°F). During daytime activities, it is above average (+0.5°F). You can see that these natural factors will affect a particular body’s temperature.
When reading an IR surface temperature, the surface itself, can also affect the radiation of IR. Emissivity is a surface’s relative ability to emit heat by radiation. An object with a low emissivity (aluminum foil, <0.1) does not radiate much energy, whereas an object with a high emissivity (skin, 0.95 to 0.99) radiates most of its energy. Note that skin has a rather wide range of emissivity. The radiated energy seen by the IR sensor is therefore lower than the actual temperature. Adjusting the emissivity register can artificially increase the reported temperature, based on the known emissivity of the material.
This project suggests that substances on skin may have an additional effect on emissivity. The measurements performed on my forehead showed substantial differences between clean, dry skin and skin treated with natural and artificial substances. While these tests were far from controlled laboratory studies, they had some interesting results. I found that oily skin had a higher emissivity than clean dry skin, whereas perspiration on skin had substantially lower emissivity (due to evaporation). Substances such as cosmetics and sunscreen also lowered emissivity.
With temperatures of over 100.4°F considered feverish. There is only a 2°F difference between normal and fevered temperature. It is pretty clear to me that while the technology is quite accurate, differences among individuals can have such wide emissivity variations that an accurate temperature of any one individual may be suspect. If my normal temperature were on the high side, my reported temperature could very well be a false positive for a fever. Or, if my normal temperature were low and I had a fever, wearing sunscreen could cause my reported temperature to be a false negative for fever.
Considering the results of these experiments, it is clear that while body temperature measurement with an IR sensor is a good method of detecting fever, true body temperature must be measured internally. All that said, I think most people can tell whether they are feverish or not without taking a temperature. It is really their responsibility to take appropriate actions to keep the rest of us safe from infection.
Temperature can be measured automatically in various ways. Large companies and healthcare facilities presently require anyone entering a facility to undergo a temperature check. This operation has been automated using a combination of instruments. The use of facial recognition, in a very basic sense, can detect a face and eyes within frame of a camera. This allows the system to prompt a subject into the appropriate position so that an IR sensor is pointing at the individual’s forehead. The temperature is then reported, and the person is allowed entry or turned away.
Other methods include the use of thermography or IR cameras. Color is normally assigned to temperature, so one can visualize temperature differences by color. Sensitivity to the difference between normal and fever is presently not as good as through IR sensors, however. For now, I place this in the sci-fi realm, but it shows promise for sensing moving multitudes in the future.
Because of my results I believe that the only way to improve the accuracy of automated systems is to keep track of employee temperatures daily, so a person’s sample can be compared to his or her past samples to determine a normal value. This walks a fine line between what information is private and what is public. How do we keep data used for public safety out of the wrong hands? Check out the study “Investigation of the Impact of Infrared Sensors on Core Body Temperature Monitoring by Comparing Measurement Sites” at  for additional information.
 “How Gun-Style IR Thermometers Work (Part 1)” Circuit Cellar 374, September 2021
 MLX90614 Data Sheet
 Human Body Temperature (Table of normal temperatures for men and women, measured by different techniques)
 Chen, H.-Y.; Chen, A.; Chen, C. “Investigation of the Impact of Infrared Sensors on Core Body Temperature Monitoring by Comparing Measurement Sites.” Sensors 2020, 20, 2885.
Melexis | www.melexis.com
PUBLISHED IN CIRCUIT CELLAR MAGAZINE • OCTOBER 2021 #375 – Get a PDF of the issueSponsor this Article