Sensing body dehydration

, & -December 13, 2017

While dehydration is usually associated with humans involved in physical exertion, it is also common in sedentary environments, such as office workplaces or college classrooms. This apparently “imperceptible” dehydration is an important issue and has many implications regarding health and loss of productivity. We here consider various methods of sensing hydration level, and present our own approach.

Dehydration has a wide range of adverse effects on human physiology. Dehydration levels of as little as 2% body weight have been observed to cause noticeable physical performance decrease. Mild dehydration can also impair cognitive functions including short term memory, concentration, simple arithmetic, and motor skills. It is also empirically known that lack of water can increase irritability and propensity for headaches. Dehydration has long-term effects on gastrointestinal, kidney, and heart function, contributing to constipation, chronic kidney disease, and coronary heart disease. In combination, these symptoms can lead to major health degradation, as well as loss of productivity in the workplace or at school/college. While it is nearly impossible to estimate the exact contribution of dehydration toward occupational health hazards, the total economic burden for occupational injury and illness in 2007 in the United States has been estimated to be approximately $250 billion.


Figure 1  As an example of the enormity of the problem, the average American drinks 2.5 glasses a day whereas the prescribed quantity is 8 glasses.

Because thirst is not an accurate gauge of dehydration, the problem can be especially deceptive in less-active individuals. Ordinary white-collar workers, students, and other less-active groups can often be too occupied by their schedules to actively think about their hydration status. A wearable device for evaluating the user’s hydration status and alerting the individual would be ideal for such a scenario.

Many designs for wearable sensors have been developed with the aim of providing real-time information about potentially alarming changes in the body, including hydration monitoring. Dehydration sensors, such as the device developed by the Javey group from the University of California, Berkeley utilize electrochemical sensors to test the contents of the user’s perspiration. Another device, by Koh et al., uses colorimetric sensors and optical image analysis of captured sweat samples to quantitate water loss. Other clinical hydration assessment biomarkers include urine osmolality, urine specific gravity, tear production, and mucus wetness. Most of these methods are not suitable to be adapted to non-clinical, personal consumer-grade sensors; thus the need for a noninvasive, real-time, hydration sensor system that does not require perspiration.

Existing methods

Current methods for detecting dehydration are based on the analysis of different characteristics of the skin and body fluids such as sweat and blood. Some existing approaches of detecting dehydration include:

  1. Detection of changes in mechanical properties of the skin: Elasticity and texture are two properties that can indicate the approximate state of hydration of the body. For example, the “pinch test” is a commonly-used skin elasticity test to quickly determine if a person is dehydrated. The drawback of this approach is the lack of a defined threshold value for the mechanical property to distinguish hydrated and dehydrated states.
  2. Interstitial fluid measurement: Interstitial fluid volume drops with dehydration and can be measured using microscopic needles. However, interstitial fluid drop could be a sign of other conditions such as low blood pressure. Hence exclusively identifying dehydration as the cause is a challenge for this technique.
  3. Light-based detection: Non-invasive light patterns cast through the skin with a laser can measure changes in blood glucose that happen with decreasing water volume. As with other biological indicators, the concern here is in filtering out other confounding bodily conditions that can induce glucose increase/decrease (diabetes, nutrition, etc.).
  4. Sweat analysis: The technique most commonly used to measure dehydration is chemical analysis of sweat. Mineral content (sodium, potassium) decreases with dehydration. Conductivity of sweat varies with sodium concentration and can be an indirect measurement of dehydration. Another indication of dehydration is the density of sweat which has a direct relation to water volume in the body.
  5. pH level of skin: Dehydrated skin has a pH level similar to that of water and a device that can identify this condition should have adequate sensitivity to detect the transition from acidic (normal skin) to slightly basic(dehydrated). Alkaline skin can also be induced by natural dryness or eczema which makes it more challenging to identify dehydration as the sole cause.


While all are viable methods of determining dehydration, the approaches described above have limitations in that many cannot be easily adapted to a wearable form-factor. Other solutions suffer from confounding factors where the measurements may be influenced by values that are not related to hydration.

This project tries to address sedentary dehydration through bio-impedance analysis. Measurement of body impedance is a technique that has the advantage of being directly related to the total volume of water in the body (Total Body Water) which is equivalent to the hydration state of the body.


Theoretical background

Total body water (TBW) provides an estimate of the water content of the body as a percentage of total weight. The relative hydration state of the body is related to TBW which can be estimated by bioelectrical impedance analysis. Our design aims to emulate this function. Empirical observation correlates the hydration of the body inversely to the impedance measured. A model used by Deurenberg relates TBW volume with height in centimeters (H), weight in kilograms (W), age in years (A), sex (S, 1 for male, 0 for female), and impedance at 50 kHz (Z50):

   TBW = 6.53 + 0.3674·H2/Z50 + 0.1753·W - 0.11·A + 2.83·S             (1)

Bioimpedance is measured by introducing a signal of a particular frequency or multiple frequencies through electrodes placed on the skin. 

Figure 2  Voltage output measurement circuit for skin impedance

Hydration correlates positively with the magnitude of the voltage response of the skin impedance circuit. By monitoring voltage decrease over time, it is possible to detect dehydration. This calibrate-and-compare method is especially useful because impedance values can vary widely from person to person.

The system considers the impedance of the skin at 50 kHz. It the most commonly accepted testing frequency in bioimpedance analysis due to its good sensitivity to changes in body composition as well as smaller likelihood of interfering with biological signals. The detection range of the device corresponds to the TBW levels typically found in humans (roughly 55%). Using the Deurenberg model and assuming a height of 160 cm and weight of 70 kg, |Zskin| is estimated to be 300 Ω. This value varies by ±30 Ω as the hydration state of the body changes. Zload is chosen to limit the current below human perceptual thresholds (less than 1mA).


System description and specifications

The goal of developing a wrist-mounted device set a number of constraints on the design. The immediate limitation was size. It had to be roughly the size and weight of a small watch. Operation lifetime should be significant enough so that frequent battery replacement is not necessary. Enclosure design also had to be robust so that natural everyday movement would be comfortable and not hindered, while at the same time maintaining adequate contact with the skin.

Figure 3  Conceptual block diagram

The system can be categorized into three major components: AC signal generation, impedance reading, and dehydration indication (Fig. 3). The Atmel ATtiny85 microprocessor was selected for its small size and low power draw. In the earliest prototype, the processor used a look-up table of sine wave values and wrote to a digital to analog converter (MCP4901 8-bit DAC) via SPI. This produced a clean 3V sine wave but because of software limitations, the device was unable to produce sufficiently high frequency signals. In later revisions, the microprocessor alone was used to produce a square wave and it was not until adjustments were made to the timing registers that the function generated was of sufficient frequency. The most optimum design would likely involve IC units specifically made for sine wave generation.

The output voltage of the skin-load divider is read by the ATtiny85 itself with the onboard 10-bit analog to digital converter. Upon system start up and/or reset, calibration data is taken and stored in the microprocessor. The AC signal is generated and a running average algorithm detects the wave peaks to determine magnitude of the output response. Threshold logic determines if the hydration has dropped significantly enough for the user to be considered dehydrated. This polling process is cycled periodically with a sleep cycle in which the device enters low power mode to preserve battery life.

The load resistor is set to 3kΩ to prevent too much current from travelling through the body. Two LEDs are also present on the device. A yellow indicator LED is on whenever the device is receiving power and active. A red indicator LED is switched on by the microcontroller whenever dehydration is detected.

Figure 4  Circuit board and schematic

The enclosure was created by forming a watch-shaped model and creating a mold from it. The electronics were placed inside the mold and the device was cast using polydimethylsiloxane (PDMS) polymer (Fig.5). This protects the electronics from shock and liquids as well as providing a good contact with the skin. The LED indicators are clearly visible through the clear material. Also visible are the copper strip contacts used to interface the electronics with the skin. A mass production design would most likely abandon the difficult casting process and utilize a plastic casing.

Figure 5  Device cast in silicone


Experimental results: Validation of concept with skin equivalent circuit

The skin impedance equivalent circuit shown in Fig.2 was created using resistors and capacitors corresponding to the hydrated and dehydrated states. The hydrated state equivalent was considered to have a nominal normalized skin resistance Rskin = 284 Ω and capacitance Cskin = 86 nF. The values for dehydrated skin are Rskin = 298 Ω and capacitance Cskin = 80 nF. Using discrete resistors and capacitors with these values, an emulated circuit response could be measured as shown in Fig.6. This serves as a comparison to impedance measurements taken through the human skin.

Figure 6  Voltage response of experimental circuit implemented with resistors and capacitors corresponding to hydrated and dehydrated skin impedance. Load resistance used was 3kΩ, and input sine wave, 3Vp-p.

Figure 7  Voltage response for frequency sweep in range 1-50 kHz with 3kΩ load and input sine wave signal 3Vp-p.

Actual skin impedance measurements were made by taping copper contacts and EKG pads to the skin and applying a sine wave (3Vp-p) by connecting the output of a sine wave generator to the pad/contact. The output voltage was measured across a load of 3kΩ and observed on an oscilloscope.

Fig.7 shows the comparison between simulation results using PSPICE and the actual skin impedance measurements. There is a 10% difference at 50 kHz between the two sets of measurements both for dehydrated and hydrated states, which is expected given the body-specific variations in Rskin and Cskin. The stop band slope is steeper in the experimental response specifically below 2kHz. This may be due to higher actual skin capacitance variations in that frequency range.

Figure 8  Comparison of TBW calculated with Deurenberg model using predicted and experimentally determined skin impedance values.


As shown in Fig.8, the predicted and experimentally determined values of TBW follow a similar trend for male and female subjects with a maximum difference of 12%. This indicates a validation of the experiments with the Deurenberg model.

Figure 9  Variation of hydrated state of individuals over time.

Observations on the change in state of the body from dehydrated to hydrated were made over a period of two hours. Fig.9 plots the variation for a male and a female over this duration. Water intake of 600 ml 30 minutes after the start of the observation interval was adequate to induce a detectable change in skin impedance 1.5 hours into the interval.

Discussion and future work

The data showed great consistency between expected and experimental values. Transition thresholds between dehydrated and hydrated states are similar for both sexes according to our experiments. This consistency is advantageous for our device as there is no gender-specific tuning required for calibration. The dehydrated and hydrated state exhibited a significant enough variation for the 10-bit ADC to differentiate though future upgrades are still an option. And after many different algorithm and threshold optimizations, the device was able to distinguish simulated dehydrated skin from simulated hydrated skin with an accuracy close to 100%. This is all with components that are easily small enough to be assembled into a small form factor wearable around the wrist. Battery life was less than desired, with an average replacement interval of 3-4 days. However, this was without power-saving techniques implemented.

The future of the dehydration sensor can be described as three major improvements: accuracy, comprehensiveness, and robustness. The first iteration of the device showed good accuracy for relatively controlled, sedentary environments. Feedback from other parties have demonstrated very positive interest, especially in fields such as assisted living and airline in-flight services. Being able to detect dehydration in various other environments such as rapidly changing climates or in the presence of sweat, would greatly increase the target audience as well as improve the reliability of the data. These improvements would entail better algorithms, more complex or compound sensors, and a stronger enclosure.

A goal of the project from its inception was to create a small, lightweight, and elegant solution to detecting dehydration. Because the first iteration of the device has largely proved it possible, a very viable future path is to incorporate the system into an existing health monitoring device already on the market. These devices already contain multiple different types of sensors and already benefit from strong, robust enclosures and the powerful processors embedded in the health monitors will be more than sufficient to run far more complex algorithms. Integrating the two devices would increase the power of the dehydration sensing using an already existing system.

 Download µC code

Related articles:

  • A. Koh, D. Kang, Y. Xue, S. Lee, R. Pielak, & J. Kim, “ A soft, wearable microfluidic device for the capture, storage, and colorimetric sensing of sweat,” Science Translational Medicine,2016, 8(366)
  • W.Gao, S.Emaminejad, H.Nyein, S.Challa, K.Chen, & A.Peck, ”Fully integrated wearable sensor arrays for multiplexed in situ perspiration analysis,” Nature, 2016,529(7587), 509-514.
  • H.Morel & M.Jaffrin,”A bridge from bioimpedance spectroscopy to 50 kHz bioimpedance analysis: application to total body water measurements,” Physiological Measurement,2008, 29(6), S465-S478.
  • Z.Vosika, G.Lazovic et al,”Fractional Calculus Model of Electrical Impedance Applied to Human Skin,”PLOSOne, 2013,Vol.8(4)



Vidhya Balaji is a graduate student in the Electrical Engineering department of University of Washington, Seattle. She is a member of IEEE Electron Devices Society and is currently researching sensor systems in biomedical applications.

Yi-hsin (Chris) Jong is an undergraduate Electrical Engineering student at the University of Washington. His interests and experiences span the various fields of Electrical Engineering with special interest in medical application.

Jonathan Chen is a graduate of the UW B. S. Bioengineering program with research interests in biosensors and diagnostic monitoring. Since 2017, he has been pursuing opportunities in the biotechnology industry.



Loading comments...

Write a Comment

To comment please Log In