A fully printed sensor with optical readout for real-time flow monitoring

In recent years, there has been a growing interest in the development of flexible thermal flow sensing devices due to their wide-ranging applications. In this study, we present the fabrication of a screen-printed flow sensor with optical readout on a 125 μm polyethylene terephthalate substrate in a three-layer configuration. The device comprises electrodes made from a commercial silver (Ag) ink, a heating area using a commercial carbon ink, and a thermochromic (TC) layer employing a commercial ink with a standard activation temperature of 31 °C. We designed a specialized experimental setup to evaluate the performance of the optical flow sensor under static and dynamic conditions. To analyze the device’s thermal response and performance across various flow conditions, we utilized a combination of electrical measurements and infrared (IR)-optical imaging techniques. The all-printed device operates on the basis of a thermodynamic cycle frequency, which activates the TC ink, causing it to blink at a frequency related to the flow passing over the sensor. The results of our preliminary testing are highly promising, as the sensor successfully demonstrated a clear relationship between flow and optical duty cycle. This innovative device offers a contactless, low-cost, easy-to-use flow detection method and holds significant potential for various practical applications.


Introduction
Flow detection plays a vital role in a number of applications.In the medical field some examples are sweat flow monitoring [1], blood flow and pressure [2,3].Bioinspired systems for flow detection have also been developed, from neuromast-inspired sensors [4,5] to micro-hairs and cilia arrays [6,7] and so on.For environmental monitoring, a typical example of flow sensing is for the estimation of wind speed [8,9].In unmanned aerial vehicles [10] and robotics [11][12][13] combination of such sensor arrays can be critical for localization and position calculations.Recent trends in research couple these systems with triboelectric generators, for harvesting energy from wind flow; it is important for these devices to have a low cut-in speed [14] and for the sensors to interfere with the harvesters' workings as little as possible.Additionally, various commercial-level applications, such as electronic cigarettes [15] incorporate flow sensing.
Flow sensors can be divided, based on their principle of operation, mainly into thermal, differential pressure, ultrasonic, piezoresistive and piezoelectric devices [16].Thermal flow sensors work by exploiting heat transfer caused by a working flow [17].They can operate in 'hot-wire' and 'hot-film' mode, where the flow magnitude is extracted by the power required to keep a heating element at constant temperature, or by the temperature difference caused by convection cooling when the heater is supplied with constant power.Another mode of operation is calorimetric mode, where profile asymmetries around the heater are detected through the surrounding temperature sensors, and the flow vector can be extracted.Finally, in time-of-flight sensors, a heat pulse travels from the heater to the detecting sensors via the working flow, and the flow can be correlated to the time required to travel this known distance.All these modes of operation share one common element: they provide an electrical output related to the measured flow.This requires somewhat complex readout circuitry, and in combination with the appropriate driving electronics, adds a non-negligible engineering overhead.Additional approaches for flow sensors include flow sensing with ultrasonic transducers, which utilize Doppler or transit time principles of operation [18], and pressure sensors for flow detection, where different approaches such as piezoelectric, piezoresistive [19] or capacitive can be employed.A special case is a differential pressure flow sensor, where flow can be extracted from the difference between the inlet and the outlet of a channel [20].
In recent years, there has been a significant focus on improving flow sensors in terms of accuracy, cost-effectiveness, scalability, and environmental sustainability.These efforts have focused on optimizing various fabrication techniques to meet the diverse requirements of different applications.Microfabrication techniques, including photolithography, etching, and bonding, have played a crucial role in developing miniaturized flow sensors on silicon substrates [21][22][23].These silicon-based flow sensors gained popularity in Microelectromechanical Systems applications due to their straightforward fabrication process, reproducibility, and excellent electrical and thermal properties.The use of microfabrication techniques has enabled the production of flow sensors with precise dimensions and integrated features, facilitating their broad application across multiple fields.Additionally, thin-film deposition is employed to deposit sensing materials, such as metal or oxide films, onto substrates using techniques such as sputtering [24,25] and vapor deposition [26].Thin-film flow sensors exhibit high sensitivity and are compatible with microelectronic processing, further enhancing their performance.Another notable technique that has gained significant attention is additive manufacturing, commonly known as 3D printing.This method has been extensively employed to fabricate complex flow sensor structures, offering exceptional design flexibility.Using additive manufacturing, customized sensors can be created with intricate geometries and integrated flow channels to meet specific application requirements [27][28][29].Finally, flow sensors can also be fabricated using printed electronics techniques, such as screen printing [30,31] or inkjet printing [32,33] or a combination of them.These methods offer advantages in terms of affordability and versatility, enabling the development of flow sensors with enhanced flexibility and adaptability for various application fields.
Sensors based on temperature sensitive paint and thermochromic liquid crystals (TLCs) have recently been presented and cover a wide range of applications [34].The major advantage of this family of sensors is the purely visual representation of a result, such as a sticker indicating hypothermia in infants [35].Typically, the thermochromic (TC) materials are available in either pure TLCs or microencapsulated TLCs, with pure TLCs being more optically clear and having high spatial resolution, and micro-encapsulated TLCs being more stable [36].These coatings can indicate temperature changes by either changing from one color to another, by going through a series of colors, or by changing the intensity of a single color.This mainly depends on the crystals employed in the paint.Most common categories are smectic, nematic, cholesteric, and nonsterol-based crystals (chiral-nematic).Advanced systems can accurately represent different temperature levels with different colors, providing irreversible [37] and even reversible effect in color change [38,39] with high limits.More complex systems incorporate a heater beneath the TC layer to exploit or control the pattern presented on the sensor; mechanical strain can be represented with color changes [40], and pressure as well [41].Typically the fabricated devices employ a colored substrate and a thermoactive surface; however other approaches have also been presented, as an array of TC coated wires and a heater nichrome wire placed in flow stream [42] where the wires are fixed to a sample holder.Other than that, most of the reported sensors comprise of coated surfaces that rapidly respond to flow-induced thermal disturbances.Stasiek present a very comprehensive review article with numerous approaches [43].Most of these works require the fluid under measurement to be of some specific temperature so the TC surface can react to it, and require extensive calibration with high-end camera systems [44].High spatial definition and detailed morphological information for convection can be extracted, but at a cost of a relatively demanding calibration process.More recent works have presented development and integration of such microsystem in microfluidics channels for effective and multi-channel temperature and flow monitoring [45].Nevertheless, such strategies on one hand involve complicated microfabrication techniques, and the measurement setup is not trivial.
In this work, we propose an innovative way for extracting flow via a TC layer; a heater, driven with a fixed frequency signal creates a heating-cooling cycle which is translated to a fixed blinking effect on TC ink; once flow passes over the device, this cycle is altered in a direct relation to flow rate due to convection cooling, thus changing the blinking timing.This, for a human eye, translates as 'higherfrequency blinking-higher flow rate'; by using a smartphone, we can quantify this relationship and extract flow rate based on the blinking duty cycle.The proposed system is mainly targeting low-cost applications such as disposable medical devices, water treatment plants, chemical dosing, steam flow measurement, liquid and gas flow in pipes, aquariums and aquaculture etc.This system is, to our knowledge, the first that exploits a pulse modulation method for direct, readout electronics-free, extraction of flow rate via correlation with electrical stimulation frequency.Additionally, we provide a base transfer function of the proposed sensor, which can easily be implemented in real time using a smartphone camera.

Electrical characterization-surface topography
Initially, I-V curve was extracted in the range −2-2 V (figure 2(a)).This measurement mainly refers to the carbon black layer between the Ag electrodes, considering that the TC material acts as an electrical insulator.Both of carbon black and Ag layers exhibit linear I-V behavior, with mean resistances of 891 Ω and 1.15 Ω, respectively.To determine the thickness of each layer, we used a Filmetrics Profilm3D (Unterhaching, Germany) and employed the mean of the line profile plateaus in conjunction with the standard deviation within each layer.The measured layer thicknesses were 4.54 ± 0.492 µm for Ag, 2.6 ± 0.012 µm for carbon black and 3.1 ± 0.166 µm for the TC layer (figures 2(b)-(d)).In this study, an automated step-height analysis feature provided by the software associated with the optical profiler was utilized to extract the thickness of each layer.
Although describing the surface roughness accurately requires numerous parameters, we conducted a roughness analysis in a specific area measuring 200 µm × 200 µm across all layers.It is important to note that we refrained from using leveling, flattening, or filtering techniques prior to the analysis.The software automatically established roughness parameters that correspond to the screen-printing mesh pattern; lack of uniformity does not affect the performance of the device, as long as the electrical conductivity of the electrode remains high.The carbon-black heater layer was printed relatively uniformly, with some small (<5 µm) agglomerations; identical pattern was observed for the TC ink.The observed spikes can be attributed to environmental particles, keeping in mind that the device was developed in room conditions and not in a clean room.

Thermal-flow characterization setup
For assessing the thermal-electrical performance, the device was placed onto a probe station.While two needle probes engaged the Ag electrodes, a highprecision FLIR A655SC infrared (IR) camera (resolution: 640 × 480, detector pitch: 17 µm) monitored the sample in parallel to a mid-range commercial smartphone mounted to a tripod stand, whose camera was used to record high resolution photographs (smartphone camera specifications: 48 MP, f/1.8 aperture AF + 16 MP, f/2.2 aperture FF + 2 MP, f/2.4 aperture FF + 2 M, f/2.4 aperture FF).A Keithley 2612 source meter was used for voltage supply, while current was monitored at the same time.An Alicat M-series mass flow controller (accuracy: ±0.6%, 0-500 SLPM) was used to monitor air flow.A photograph of the experimental setup is presented in figure 3.

Results and discussion
Initially, different power values were applied to the device (0-12 V, corresponding to 0-0.139 W), in order to assess required power for reaching the activating temperature of the TC ink.Red-Green-Blue (RGB) images (figure 4(a)), IR images (figure 4(b)) and grayscale images (figure 4(c)) were captured at each power step.The RGB images taken from the smartphone were converted to 8 bit grayscale using MATLAB R2021a built-in function rgb2gray, applied to the image as captured from the smartphone.Afterward, in each power step, the maximum temperature and maximum grayscale level for each image were recorded.That way, the powertemperature (figure 4(d)) and power-grayscale level profiles (figure 4(e)) were extracted; a linear powertemperature relationship was found, with a heating efficiency of 5.5 mW • C −1 .
Subsequently, the dynamic response of the device was evaluated, to extract the corresponding time constant.For this reason, a step voltage pulse of 11 V was applied to the heater, which corresponds to 139 mW  of input power, while the temperature of the device was continuously monitored.In this way the normalized curves for the device response in rise and fall phases were extracted under no flow conditions.The corresponding rise (τ r ) and fall (τ f ) times were extracted utilizing the typical decaying exponential function of a first-order system.A slightly faster response to the rising phase was extracted comparing to the fall (figures 5(a) and (b)) with the correspond time constants being τ r = 4.38 s and τ f = 5.22 s.This was expected, mainly because the heating is assisted with external energy, while cooling is natural.Afterwards, a flow sweep between 0 and 25 standard liters per minute (SLPM) of dry air was performed, with a step of 5 SLPM, while keeping the device under constant voltage.The flow was supplied by a dry laboratory air compressor and the outlet was kept at adequate distance from the device, as depicted in figure 5(c), to avoid turbulences.
Two different voltages were tested (3.3 V and 11 V) and the corresponding temperature dependence to the applied flow is presented in figures 5(d) and (e).In both cases, the applied power with zero flow results in a device temperature that is above the critical temperature of the TC ink (31 • C); therefore, the TC layer is transparent (active mode) in the initial stage.By increasing the flow, the temperature drops, and if the critical temperature of 31 • C is reached, then the TC layer becomes opaque for this value and below (inactive mode).The transition phenomenon is not abrupt; thus, the grey scale of the TC layer is gradually changing with the corresponding temperature between the two limits (fully transparent and fully opaque).This is the main principle for the proposed optical flow sensor, which aims to determine the flow visually.The selection of the initial device power is critical for the above-described principle of operation, since it should assure the transition from the active (TC is transparent) to inactive (TC is opaque) mode.In figure 5(d) the flow-induced temperature decreases and crosses the critical temperature of TC at about 8 SLPM flow rate, which means that below this flow value the device would appear active, while above 8 SLPM the device would appear as inactive.On the contrary in the case where the applied voltage is 11 V (figure 5(e)), the maximum flow value (25 SLPM) is not enough to reduce the temperature below 31 • C, thus the device is active withing the whole flow range.The aforementioned experiment highlights a challenge in visualizing a range of flow rates using a specific electrical signal as trigger.It should be noted that these measurements have been extracted under steady state conditions, for getting a coarse estimate on power requirements.In figure 5(f), both bias voltages have been plotted together after normalization, in order to remove the difference in temperature ranges, rearranging both graphs in [0, 1] range.The normalized temperature was extracted by the formula (equation ( 1)), where T min and T max are the minimum and maximum temperature respectively, and T is each temperature measurement.Figure 5(f) highlights the fact that the device behavior is constant in both cases regardless the value of the applied power, The above-mentioned operation utilizes constant voltage input which is slow, needs precise adjustment and cannot cover a wide range of flows.Alternatively, a non-symmetrical power pulse can be used, as denoted in figure 6(a).This pulse induces periodic temperature variations in the heater, as presented in figure 6(b).Under no flow, the pulse on-state duration has to be adequate for activating the device while the off-state time-duration must allow the device to cool below the TC (figure 6(b)).The specific temperature transitions result to a periodic blinking with fixed period, that is directly related to the input power pulses parameters (Vo, frequency and duty cycle) (figure 6(c)).The t on is the time-frame when the device appears to be active and the t off is the timeinterval within the device is inactive.Once flow is applied, two mechanisms are activated: first, the heating rate and the maximum temperature are slightly affected with an increase in the heat transfer coefficient for a given power, resulting in a decrease of the maximum temperature (figure 7); this difference can be neglected as the maximum temperature remains above the critical temperature.Additionally, higher flows assist in cooling the device faster during the off-state of the input power pulse, which in turn will move it to the inactive temperature zone earlier.This phenomenon is attributed solely to convection, with higher flow rates providing faster cooling, which in turn modulates the optical duty cycle.Thermal conduction plays a secondary role because the device is mounted in a free-standing manner, and it has no supporting substrate in direct contact.We can then establish a relationship between the optical duty cycle and the applied flow.The period of the input power pulse is 5 s and corresponding duty cycle is fixed at 26.5%.Under zero flow, the resulting device optical duty cycle is 64%, meaning that the device is optically activated for 3.2 s and inactivated for the rest 1.8 s (figure 8(a), No Flow).By progressively increasing the flow, the duty cycle decreases as described above (figure 8(a), rest of the pulses).We can then extract the relationship that correlates the stimulation of our device (flow rate) with the sensor response (optical signal duty cycle), as presented in figure 8(b).By fitting the experimental data points with an exponential decay function, we can define the transfer function for the proposed sensor for the given input (pulses with T = 5 s and 26.5% duty cycle) (equation ( 2)).

Duty cycle
where Q is the input flow in SLPM.
The main drawback of the proposed, all optical output device is its relatively limited accuracy; the applications this device should be used in are those that demand fast and remote flow sensing, without complex readout systems and software.
The group is actively working on evaluating different geometries and positioning of the heater elements, as well as arrays of different thermochromic materials for a wider range of temperature-flow measurement.Additionally, a complementary model is being trained with images, for enhancing the automatic extraction of flow using a smartphone camera.

Conclusions
In this work, a fully printed device for flow monitoring using optical means was presented.The device was fabricated by screen printing on a flexible PET substrate in three layers: the electrodes were printed using an Ag ink, the heater using a carbon-black ink, and the TC layer using a commercial TC ink having a standard activation temperature.To comprehensively assess the device's performance, electrical, optical and thermal measurement were conducted, by simultaneously monitoring with a smartphone and an IR camera.Subsequently, we evaluated the device's capabilities as an optical flow sensor using a dedicated experimental setup.
The device integrates the carbon-black heater for allowing periodic activation of a TC layer over the critical temperature; once flow is applied, the time interval in which the device is activated decreases, leading to a distinguishable change in the on-off pattern of the TC layer.That way, the flow magnitude can be easily observed remotely, allowing for sensing without an electrical readout system; the algorithm for flow extraction based on the optical duty cycle can be transferred to a smartphone, with which the user can convert the optical information to flow rate.
The preliminary experimental results of the sensor are promising, showcasing a clear correlation between flow and optical duty cycle.This innovative device offers a contactless, cost-effective, and userfriendly solution for flow detection, making it an appealing choice for applications that require efficient and affordable flow monitoring methods.

Figure 1 .
Figure 1.Device cross-section (a); fabrication process steps (b); photograph of the fabricated device; the carbon black layer is hidden below the thermochromic layer, scale bar is 1 cm (c).Dimensions in (a) and (b) in mm, not to scale.

Figure 3 .
Figure 3. Setup for the device characterization (a).The smartphone stand (1) and the IR camera (2) can be seen, alongside the mass flow meter (3) and the sample under test; view from the smartphone screen (b); view from the IR camera (c).Scale bar is 5 mm.

Figure 4 .
Figure 4. RGB (a), IR (b) and grayscale (c) images of the device under different supplied power (0, 24, 48, 79 and 139 mW); max temperature extracted from IR image (d) and grayscale (e) with respect to supplied power.

Figure 5 .
Figure 5. Normalized temperature response for rising (a) and falling (b) input power pulse phases under no flow; RGB and IR images of the sample biased, under 3.3 V and 11 V, while being subjected to either no flow or 25 SLPM of dry air (c); temperature-flow relationship for a bias voltage of 3.3 V (d) and 11 V (e); normalized temperature-flow relationship for both bias voltages (f).

Figure 6 .
Figure 6.The driving pulse applied to the heater (a); the resulting temperature profile (under no flow) (b); the corresponding optical output of the system (c).

Figure 7 .
Figure 7. Different flow rates result in different temperature profiles, with both maximum temperature and time-duration over critical temperature being affected.The temperature profiles correspond to flows in the range [0-25 SLPM] with a step of 5 SLPM.

Figure 8 .
Figure 8. Resulting optical output for all the flows tested (a); the relationship between flow and optical duty cycle (b).