Sensors & Actuators: A.

Physical 362 (2023) 114686

Contents lists available at ScienceDirect

Sensors and Actuators: A. Physical

journal homepage: www.journals.elsevier.com/sensors-and-actuators-a-physical

A 3D printed flow sensor for microfluidic applications

Adam Hawke a, Gianmarco Concilia a, Peter Thurgood a, b, Arman Ahnood a, Sara Baratchi b, c,
Khashayar Khoshmanesh a, b, *
a
School of Engineering, RMIT University, Melbourne, Victoria 3000, Australia
b
Baker Heart and Diabetes Institute, Melbourne, Victoria 3004, Australia
c
Baker Department of Cardiometabolic Health, University of Melbourne, Victoria 3052, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: Here, we describe the development and characterisation of a flow sensor for microfluidic applications. The flow
Microfluidics sensor utilises two miniaturised commercial pressure sensors mounted across a 3D printed microfluidic tube to
Flow sensor measure the flow rate based on the viscous pressure drop occurring between the two sensors. The operational
3D Printing
range of the flow sensor can be modulated by varying the diameter of the microfluidic tube. We demonstrate the
Blood viscosity
suitability of the flow sensor for measuring constant and dynamic flow rates driven by syringe, piezoelectric, and
pressure pumps. We characterise the flow sensor against water, water-glycerol solutions, and human blood. We
harness the sensitivity of the sensor for measuring the viscosity of human blood at physiologic and room tem­
peratures. We also show the ability of the flow sensor for monitoring the transitory flow rates generated by a
manual pipette. The flow sensor is compact, low-cost, and highly responsive. It has no moving elements and can
be easily tailored, interfaced, and operated. These features make it appealing for a wide range of applications in
microfluidics.

1. Introduction used as a heating element and a temperature sensor [9]. Thermal-based

flow sensors have no moving elements, and therefore do not induce
Microfluidic systems operate at extremely low flow rates, which is additional mechanical stress (such as shear stress) to the neighbouring
typically smaller than one millilitre per minute (1 ml/min). While this flow; this can be important when working with mechanosensitive sam­
feature significantly reduces the consumption of biological samples and ples such as human blood [10,11]. However, the addition of heat might
reagents and facilitates the formation of highly ordered laminar flow damage thermosensitive proteins within the sample [12,13]. Thermal
profiles within microfluidic systems [1–4], it makes it rather challenging flow sensors enable measuring dynamic flow rates, although their
for conventional flow sensing systems to monitor the flow rate. This response time might be limited due to the high specific heat capacity of
challenge has led to the development of a variety of flow sensors capable water-based solutions [14].
of measuring extremely low flow rates [5,6]. Coriolis flow sensors work on the principles of the Coriolis effect,
Flow sensors can be categorised into two groups, including active which implies that a mass moving in a rotating system experiences a
and passive [5]. An active flow sensor uses external energy to measure force that is perpendicular to its direction of motion and the rotational
the flow rate. The most common commercial active flow sensing tech­ axis of the system. These sensors utilise a U-shaped tube that is oscillated
nologies are thermal, and Coriolis-based flow sensors. Thermal flow at its resonance frequency using magnetic coils. As liquid moves through
sensors are based on the principle of convective heating and utilise a the oscillating tube, it is subjected to Coriolis forces that are propor­
heating element and two temperature sensors [7,8]. The injection of tional to its mass flow rate [15]. The Coriolis force induced in the inlet
heat by the heater leads to a temperature rise in the neighbouring flow, arm of the tube is opposite to the force induced in the outlet arm of the
which is measured by the temperature sensors. The mass flow rate is tube. This causes the tube to twist which is measured using sensitive
inversely proportional to the temperature rise of flow. Single element motion sensors attached to the tube. Coriolis flow sensors are highly
thermal flow sensors utilising the transient thermal offset method have precise and do not rely on the properties of the fluid [16]. However, they
been recently developed, in which a metallic structure is concurrently can be rather expensive due to the implementation of actuators and

* Corresponding author at: School of Engineering, RMIT University, Melbourne, Victoria 3000, Australia.
E-mail address: [email protected] (K. Khoshmanesh).

https://doi.org/10.1016/j.sna.2023.114686
Received 8 June 2023; Received in revised form 9 August 2023; Accepted 22 September 2023
Available online 24 September 2023
0924-4247/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

sensors for tube oscillation. In addition, tube oscillation can exert un­ platform for PCB design. This application allowed a multi-layered PCB
wanted mechanical stress on mechanosensitive cells. using surface-mount technology (SMT) to be conceptualised, designed,
Passive flow sensors do not require external energy to measure the and modeled (Fig. S1a). The design of the system utilised two
flow rate. The most common passive flow sensors are gravimetric and commercially available pressure sensors (MPR Series, Honeywell). Once
cantilever-based sensors. Gravimetric flow sensors determine the flow designed, the PCB was fabricated by PCBWay (China), a large PCB
rate by measuring the fluctuation in the mass of the liquid collected in a manufacturer, while all assembly and soldering of SMT components
reservoir over time [17,18]. These sensors can be highly accurate but were performed in our lab. The use of a custom PCB design allowed for
rely on large scale equipment and provide low acquisition rates. Canti­ the incorporation of two pressure sensors, high-speed universal asyn­
lever flow sensors obtain the flow rate by measuring the deflection of a chronous receiver-transmitter (UART) connectivity, a Bluetooth module
cantilever beam that is embedded in the fluidic structure [19]. The (BM70 BLE, Microchip), and a high-powered microcontroller
deflection is caused by the viscous drag force that is proportional to the (ATmega328, Microchip) to be mounted on a compact platform (50 mm
flow rate [20]. The deflection is often measured using a piezo resistor × 26 mm) (Fig. 1a).
that is attached to the base of the beam. Cantilever flow sensors are
highly responsive but require complex and time-consuming fabrication 2.2. 3D Printing
processes and might be prone to fouling. Strain-based flow sensors have
been recently developed in which flexible polyolefin substrates are A microfluidic tube was designed in CATIA (Dassault Systèmes) and
incorporated into the microfluidic channel. The flow-induced pressure 3D printed to interface directly to the pressure sensors on the PCB. The
leads to the straining of the polyolefin film and changes its electrical design consisted of a single straight tube with a circular cross-sectional
resistance, and this principle is used for measuring the flow rate [21]. area (D × L = 0.5 mm × 40 mm) with two large openings (4.2 mm in
For fluids with a high density of suspended particles or cells, particle diameter) on either side (Fig. S1b). The two large openings allowed for
image velocimetry (PIV) [22] or laser Doppler velocimetry (LDV) [23] the pressure sensors to interface directly with the tube. Before printing,
techniques have been utilised for monitoring the flow rate. PIV measures the 3D model was converted into multiple still images or photomasks
the flow velocity by tracking individual particles in microfluidic struc­ using open-access slicing software (Lychee Slicer, Mango3D). A bench­
tures and is widely used for flow visualisation. The implementation of a top LCD 3D printer (Mono 4 K, Anycubic) along with light-sensitive
high-speed camera and continuous wave laser allows for measuring resin (Eco UV Resin Clear, Anycubic) was used for 3D printing (Fig. 1b).
highly transient flow rates [24]. Despite its high accuracy and ability to Our initial experiments indicated that the drastic expansion at the
map the 3D flow velocity field, it requires a rather expensive optical interface of the microfluidic tube and the sensor mountings cause air
setup involving an inverted microscope fitted with a camera and a bubbles to be trapped within the mountings. The presence of an air
pulsed laser source. The process of analysing the data can be also bubble at the sensor interface decreased the overall sensitivity of the
time-consuming. sensor as the trapped air dampens the force applied to the sensor. To
LDV utilises a coherent, polarised laser beam and a photodetector to overcome this problem, mountings were modified to incorporate a semi-
measure the frequency shift of the light scattered by the moving parti­ circular barrier at the surface of the tube facing the sensor (Fig. S1b).
cles. The frequency shift is due to the Doppler effect and is proportional The semi-circular barrier aimed to mitigate bubble trapping by
to the velocity of the particles [25]. This approach has also been used for decreasing the volume within the mountings. Furthermore, the surface
measuring the flow rate of particle/cell-free fluids by using diffraction
grating generated by moving air bubbles or droplets [26]. LDV allows
for localised and instantaneous measurement of flow velocity. It requires
rather expensive optical equipment, although attempts have been made
to miniaturise this technology [27].
The unprecedented growth and global adoption of 3D printing have
led to rapid advancements in resolution, quality, diversity of inks and
filaments, and a large reduction in cost. These advancements have
enabled the rapid prototyping of customised microfluidic structures. The
high precision and three-dimensional freedom to create complex struc­
tures, enabled by 3D printing techniques, has facilitated the integration
of fabricated or commercially available sensors to provide flow sensing
capabilities [28–32]. We envisage that 3D printing will facilitate the
widespread usage of flow sensors in microfluidics.
In this work, we take advantage of 3D printing technologies to create
a highly responsive, inexpensive, and compact flow sensor. The sensor
utilises two commercial pressure sensors mounted on two sides of a 3D
printed microfluidic structure to measure the flow rate. The flow sensor
is characterised by water-based solutions and whole human blood. We
will study the utility of the flow sensor for measuring both constant and
dynamic flow rates driven by commercial syringe, piezoelectric, and
pressure pumps. We will further demonstrate the ability of the sensor to
measure dynamic flow rates generated by a manual pipette. The versa­
tility of this sensor makes it suitable for monitoring highly transient and
complex flow rate profiles in microfluidics.

2. Materials and methods

Fig. 1. The fabrication process of the flow sensor: (a) A render of the PCB as­
2.1. PCB Design sembly; (b) A render of the 3D case and microfluidic structure; (c) Microgrooves
patterned onto the semi-circular barrier to facilitate the dislodgement of air
The electronic circuit design and the printed circuit board (PCB) bubbles; (d) Exploded view of the 3D printed flow sensor; and (e) The assem­
were designed using Altium Designer (Altium), an industry-recognised bled 3D printed flow sensor.

2
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

of the barrier was patterned with six grooves (W × L = 0.4 mm × 3 mm) We also used our gravimetric platform to calibrate our syringe pump
to reduce the contact area between the bubble and the barrier to facil­ for the flow rates ranging from 25 to 2000 µl/min (Fig. S5).
itate the dislodgement of the bubbles [33] (Fig. 1c).
The next challenge of fabrication was to interface the PCB and the 3D
3.2. Measurement of flow rates generated by a syringe pump
printed microfluidic structure. To ensure a watertight seal between the
PCB and the microfluidic structure, two commercially available O-Rings
First, we tested the accuracy of our flow sensor at constant flow rates
(D × H = 5 mm × 2 mm, Syneco) were added to the system (Fig. 1d). A
driven by a syringe pump (Fig. 2a). To do so, water was infused into the
render of the barrier and the interface between the pressure sensors and
flow sensor at flow rates ranging from 10 to 500 µl/min. Experiments
tube is presented in Fig. S2. Compression of the O-Rings was provided
were performed at room temperature (22 ◦ C) and therefore the viscosity
through eight screw fixtures surrounding the perimeter of the case
of water was set to 0.9544 mPa.s. The diameter of the microfluidic tube
(Fig. 1e). This allowed for easy maintenance and the possibility of
within the sensor was set to 0.5 mm, and accordingly, the ‘k’ correction
different 3D printed microfluidic structures to be interchanged. Addi­
factor was considered as 1.52.
tionally, to ensure compatibility with various microfluidic devices, the
The standard deviation of flow rates obtained over 1600 data points
inlet and outlet of the flow sensor were fitted with traditional tube fit­
measured over 20 s was calculated to assess the ‘noise’ of the flow sensor
tings. The process of the assembly of the flow sensor is rendered in
readings (Fig. S6a). The deviation between the mean flow rates
Video S1.
measured by the flow sensor and the flow rate set by the syringe pump
Supplementary material related to this article can be found online at
was calculated to assess the ‘mean error’ of the flow sensor (Fig. S6b).
doi:10.1016/j.sna.2023.114686.
Furthermore, the standard deviation between the mean flow rates
To facilitate the real-time monitoring of the flow sensor, a multi-
measured during independent readings was calculated to assess the
threaded application was developed to support the simultaneous visu­
‘repeatability’ of the flow sensor (Fig. S6c). Fig. 2b shows the variations
alisation and logging of flow measurements. As pressure measurements
of the flow sensor’s noise, mean error, and repeatability within the flow
are produced by the flow sensor at a sampling frequency of 80 Hz, a
rates ranging from 10 to 500 µl/min. Our measurements indicated that
single thread was dedicated to storing pressure measurements and
our flow sensor’s noise is in the range of 2.20 ± 4.25 % within 0–150 µl/
communicating to the flow sensor via serial communication. A second
min and between 0.54 ± 0.13 % within 200–500 µl/min. The mean
thread was created to calculate the flow rate based on the pressure
error was 2.59 ± 2.5 %, with a maximum error of 5.09 % at 10 µl/min
readings and to facilitate the visual display of information in real-time.
and a minimum error of 0.09 % at 150 µl/min. The repeatability be­
The application was written in Python relying on Matplotlib [34] and
tween independent tests was 99.32 ± 0.47 %.
custom Tkinter libraries, this interface can be found in Fig. S3.
Flow rate measurements were also performed using a calorimetric
flow sensor (LG16–1000D, Sensirion) to assess the performance of our
3. Results and discussions
flow sensor. Fig. 2b shows the variations of the calorimetric flow sen­
sor’s noise, mean error, and repeatability for the flow rates ranging from
3.1. Principles of the flow sensor

The 3D printed flow sensor operates on the principle of pressure loss

through a tube being proportionate to the flow rate. This relationship is
described as the Hagen-Poiseuille equation, which expresses the pres­
sure drop for a laminar flow through a tube with a circular cross-section
[35]. However, due to the complex geometry of the channel caused by
the sudden expansion at the interface of the microfluidic tube and
pressure sensors, a correction factor coined as ‘k’ has been added to the
formula.
128 μ Q L k
ΔP =
π D4

in which, Q is the flow rate through the tube, ΔP is the viscous pressure
drop measured between the two sensors, D and L are the diameter and
length of the tube, respectively, µ is the viscosity of the fluid, and k is the
correction factor.
The correction factor ‘k’ in Equation 1 was determined using gravi­
metric analysis (Fig. S4a-b). Based on our experiments, the correction
factor ‘k’ was obtained as 1.67, 1.58, 1.52, 1.48, and 1.46 when the
diameter of the microfluidic tube interconnecting the two pressure
sensors was set to 0.4, 0.45, 0.5, 0.55, and 0.6 mm, respectively
(Fig. S4c). The slight variation of ‘k’ when with respect to the tube
diameter can be attributed to the change in the geometry of the
microfluidic structure and specifically the sudden contraction and
expansion at the inlet and outlet of the tube, which governs the pressure
drop between the two sensors.
Our further investigation indicated that increasing the tube diameter
increases the operating range of the flow sensor. For example, it was
found that flow rates higher than 2000 µl/min using water and a tube
diameter of 0.4 mm would result in the saturation of the pressure sen­ Fig. 2. A comparison between the constant flow generated by a syringe pump
sors, while a tube diameter of 0.6 mm would only experience 20 % of its and flow sensors; (a) The measured time series flow profile generated by a
measurable pressure range, increasing the maximum range of the flow syringe pump within 10–500 µl/min; and (b) A comparison between the noise,
sensor to 10,000 µl/min. mean error and repeatability of the 3D printed flow sensor and a commercial
calorimetric flow sensor within 10–500 µl/min.

3
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

10–500 µl/min. Measurements revealed that the calorimetric flow sen­ The blood viscosity, as measured by our flow sensor, varied from 5.8
sor’s noise was in the range of 2.5 ± 2.10 % within 0–150 µl/min and to 3.9 mPa.s for the flow rates ranging from 25 to 500 µl/min at 37 ◦ C
between 0.98 ± 0.37 % within 200–500 µl/min. The mean error was (Fig. 3c). Comparably, the blood viscosity at 37 ◦ C provided by the
13.42 ± 13.04 %, with a maximum error of 26.47 % at 10 µl/min and a Carreau-Yasuda model varies from 6.0 to 3.9 mPa.s within the same flow
minimum error of 0.25 % at 400 µl/min. The repeatability between in­ rate range. The measured viscosity from the flow sensor was within ± 3
dependent tests was 99.26 ± 0.48 %. We also demonstrated the % of its theoretical value. Our extended measurements indicated that
compatibility of our flow sensor with various biological buffers (Fig. S7). the human blood viscosity increases by 150–200 % at room tempera­
We further investigated the ability of our flow sensor to measure the ture, varying from 11.5 to 5.7 mPa.s for the same flow rate range.
viscosity of non-Newtonian fluids such as whole human blood. To do so, Further experiments were performed to assess the suitability of our
human blood was applied to the flow sensor using the same syringe flow sensor for measuring dynamic flow rates generated by a syringe
pump at constant flow rates within the range of 25–500 µl/min (Fig. 3a). pump. To do so, we used the same experimental setup presented in
Tests were performed at both room temperature (22 ◦ C) and human Fig. 3a. Dynamic flow rates were generated by programming the syringe
body temperature (37 ◦ C), with experiments indicating the proportion­ pump to turn on for 15 s and then off for 10 s. The maximum flow rate of
ate increase in pressure drop against flow rates for blood samples the pump was set to 250, 375, and 500 µl/min. The flow rate was
(Fig. 3b). The use of a logarithmic scale provided a greater representa­ monitored in real-time using our pressure sensor as well as a commercial
tion of the curves exhibiting a slightly higher gradient at flow rates less calorimetric flow sensor (LG16–1000D, Sensirion) for comparison.
than 200 µl/min, indicating the non-Newtonian nature of human blood Measurements revealed the generation of trapezoidal flow rate pro­
(Fig. S8). files for the case of water (Fig. 4a). Increasing the flow rate shortened the
The Hagen-Poiseuille equation was used to calculate the blood vis­ transition from zero to max on the left side of the trapezoid as well as
cosity at different flow rates. In doing so, the pressure drop was from max to zero on the right side of the trapezoid. The calorimetric flow
measured by our flow sensor while the flow rate was set by the syringe sensor recorded a large overshoot of ~6.2 %, 10.5 %, and 16.1 % at the
pump, and thus the only unknown in the equation was the blood vis­ flow rates of 250, 375, and 500 µl/min, respectively, along the left side
cosity. Measurements proved the higher viscosity of blood samples at of the trapezoid. Following this initial overshoot, the flow rates were
flow rates of less than 200 µl/min (Fig. 3c). The measured blood vis­ recorded as 258 ± 3.3 %, 373 ± 1.3 %, and 490 ± 1.7 % µl/min, indi­
cosities at 37 ◦ C were then compared with the theoretical values ob­ cating a 3.2 %, 0.5 % and 2.0 % error. In comparison, the overshoot
tained by the Carreau-Yasuda model [36], which correlates the viscosity recorded from our flow sensor was ~0.0 %, 3.5 %, and 5.2 % at the flow
to the shear rate: rates of 250, 375, and 500 µl/min, respectively. Following this initial
μ − μ∞ [ ](n− 1)/2
overshoot, the flow rates were measured as 247 ± 2.8 %, 370 ± 1.6 %,
= 1 + (λ ˙ γ)2 and 499 ± 1.4 % µl/min, suggesting a 1.2 %, 1.3 %, and 0.2 % error. Our
μo − μ∞
experiments indicated the superior stability and accuracy of our sensor
in which, µ is the viscosity, µ∞ = 0.0035 Pa.s is the viscosity at infinite compared to the calorimetric flow sensor at transient flow rates.
shear rate, µo = 0.056 Pa.s is the viscosity at zero shear rate, λ = 3.13 is a Further experiments indicated that the size of the syringe would also
time constant,γ̇ is the shear rate within the microfluidic tube, which is affect the response time with a smaller 5 ml syringe, (Inkjet, B|BRAUN)
obtained as γ̇ = 32Q/π D3 , and n = 0.3568. improving responsiveness by around 1–2 s depending on the flow rate
(Fig. S9).
A trapezoidal flow rate profile was also obtained with human blood
(Fig. 4b). However, the flow rate transition from zero to max was
significantly slower compared to water. This slower response can be
attributed to the increased viscosity of blood, which resulted in the

Fig. 3. Characterisation of the flow sensor at constant flow rates driven by a

syringe pump: (a) 3D render of the experimental setup; (b) Variations of
pressure drop at various constant flow rates measured by our flow sensor; and
(c) Variations of viscosity at various flow rates for water and blood derived from Fig. 4. Dynamic flow rates generated by turning on/off a syringe pump ob­
the measured pressure drop. tained by our flow sensor for: (a) Water; and (b) Whole human blood.

4
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

system not reaching its desired flow rate until 10–15 s after turning on (Fig. S12).
the pump. This slow increase in flow mitigated the overshoot at the left Similar sinusoidal flow rate profiles were obtained for the water-
side of the trapezoid compared to water. Following the initial flow rate glycerol solution (Fig. S12a). However, the amplitude of flow rate os­
transition, the calorimetric flow sensor recorded flow rates of 238.5 cillations reduced in response to the higher viscosity of the solution
± 1.8 %, 350 ± 2.8 %, and 487 ± 2 % µl/min at 250, 375, and 500 µl/ compared to water. The reduction of flow rate oscillations led to the
min, respectively, indicating a 4.5 %, 6.7 %, and 2.6 % error. In com­ elimination of backward flows. The amplitude of flow rate oscillations
parison, our flow sensor recorded flow rates of 242.5 ± 2.3 %, 358 ± 2.2 increased linearly while the average flow rate increased exponentially
%, and 478 ± 2.5 % at the flow rates of 250, 375, and 500 µl/min, against input voltage (Fig. S13a).
respectively, indicating a 3.0 %, 4.5 %, and 4.4 % error. In line with the Oscillating flow rate profiles were also obtained for whole human
previous experiment, our flow sensor exhibited a higher accuracy blood (Fig. 5c). The amplitude of flow rate oscillations was further
compared to the calorimetric flow sensor. reduced due to the higher viscosity of blood compared to the water-
glycerol solution. Likewise, the amplitude of flow rate oscillations
increased linearly while the average flow rate increased exponentially
3.3. Measurement of flow rates generated by a piezoelectric pump
against input voltage (Fig. 5c′).
Further investigations indicated that the flow rate oscillations can be
The flow sensor was also used to measure the flow rate profiles
reduced by incorporating a microfluidic structure in the fluidic circuit.
generated by a commercial piezoelectric pump (Micropump MP6, Bar­
For example, the addition of a commercial microfluidic channel (µ-Slide
tels). Piezoelectric pumps are compact, inexpensive, and highly pro­
VI 0.5, ibidi) before the flow sensor significantly reduced the flow rate
gramable, making them appealing for many microfluidic experiments.
oscillations and inhibited the formation of backward flows (Fig. S14).
Tests were performed by coupling the piezoelectric pump to the flow
Likewise, the incorporation of a one-way valve (airline check valve,
sensor. A reservoir (BORO3.3 Beaker, 10 ml) was used to allow for the
Aqua One) or pulsation damper (mp-Damper, Bartels) reduced the flow
recirculation of the liquid. Customised constant flow rate profiles were
rate oscillations (Fig. S15).
provided by a driving circuit (mp-High driver, Bartels) controlled
Further experiments were performed to examine the ability of the
through an external microcontroller (ATmega328, Microchip) (Fig. 5a).
flow sensor to measure complex dynamic flow profiles generated by a
The signal applied to the piezoelectric membrane consisted of two out-
piezoelectric pump (Fig. S16-17).
of-phase oscillating voltages ranging from 2.5 to 76 Vp-p at 78 Hz
(Fig. S10). Tests were performed utilising both water-based solutions
and whole human blood. 3.4. Measurement of flow rates generated by a pressure pump
In the case of water, the flow rate oscillated between positive and
negative values, indicating the existence of transient forward and To demonstrate the versatility of the flow sensor, the next set of
backward flows. For example, at an input voltage of 60 Vp-p, the flow experiments included measuring constant and dynamic flow profiles
rate oscillated between –4000 and + 8000 µl/min (Fig. 5b). The average generated by a pressure-driven pumping system (OB1 MK4, Elveflow).
(net) flow rate (shown as a dashed line) was obtained as + 1257 µl/min. Pressure pumps are often used for their programmability, high precision,
The amplitude of flow rate oscillations and the average flow rate and non-contact pumping capabilities. The pump utilised pressurised air
increased linearly against input voltage (Fig. 5b′). The net flow rates to drive fluids. The pump outlet was coupled to a sealed chamber (Fal­
were verified using a Calorimetric flow sensor (LG16–100D, Sensirion) con tube, 40 ml) containing whole human blood, while the output
(Fig. S11) as well as the gravimetric testing described in Section 3.1 pressure of the pump was controlled and recorded through a USB-B

Fig. 5. Measurement of flow rates generated by a piezoelectric pump subjected to a constant sinusoidal voltage: (a) 3D render of the experimental setup; (b-c)
Measured constant flow profiles of water and whole human blood at 60 Vp-p; (b′-c′) Minimum and maximum flow rates obtained at various voltages applied to the
piezoelectric pump.

5
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

connection. Pressurising this chamber generated pressure at the free (Fig. 6c). The flow rates measured by our flow sensor and the calori­
surface of the blood, which resulted in the smooth flow of liquid from the metric flow sensor had an average difference of only < 1.9 % within 10
outlet of the chamber toward the flow sensor (Fig. 6a). Flow measure­ successive cycles.
ments were also performed using a calorimetric flow sensor (MFS, The second dynamic flow rate profile was generated by setting a
Elveflow) to assess the performance of our flow sensor. square pressure wave varying from 5 to 10 kPa at a frequency of 0.2 Hz.
First, we examined the accuracy of our flow sensor at constant flow This resulted in a smooth trapezoidal flow rate profile to be measured by
rate profiles. Setting the pressure of the pump to 10, 7.5, and 5 kPa led to our flow sensor varying from ~188 to ~405 µl/min (Fig. 6d). The flow
the generation of constant flow rates of 419.3, 299.0, and 178.6 µl/min rates measured by our flow sensor and the calorimetric flow sensor
with the average noise being measured through the system as 0.53 %, varied only < 1.3 % within 10 successive cycles. The flow rate profile
0.34 % and 0.25 %, respectively. In comparison, the calorimetric flow was characterised by the exponential increase of flow rate in response to
sensor measured constant flow rates of 423.6, 306.1, and 180.4 µl/min pressure rise and the exponential decrease of flow rate in response to
with an average noise of 1.31 %, 0.92 %, and 0.30 %, respectively pressure drop. The evolution of exponential profiles is attributed to the
(Fig. 6b). Our experiments indicated the flow rate measured through the delays caused by the viscosity of human blood.
system to be linear with respect to the input pressure, however, a min­ The third dynamic flow rate profile was generated by setting a dual
imum input pressure of 1.3 kPa was required before flow being driven frequency pressure wave, involving two localised peaks varying from
through the system (Fig. S18). 7.5 to 10 kPa (Fig. S19). Our experiments demonstrated the versatility of
To further explore the accuracy of the flow sensor, three dynamic the flow sensor for capturing highly dynamic flow rates.
flow rate profiles were generated. The first dynamic flow rate profile was
generated by setting a sinusoidal pressure wave varying from 5 to 10 kPa
3.5. Measurement of short-term flow rates generated by a manual pipette
at a frequency of 0.2 Hz. This resulted in a smooth sinusoidal flow rate to
be measured by our flow sensor varying from ~203 to ~397 µl/min
Finally, we examined the utility of the flow sensor to measure the
transient flow rates generated by a manual micropipette when using
human blood. To do so, we used a 20 µl Eppendorf pipette, which is
commonly used in biological labs. Three sets of experiments were per­
formed by setting the withdrawal volume of the pipette to 10, 15, and
20 µl. Experiments were conducted by interfacing the pipette tip with
the inlet tube of the flow sensor. In particular, we were interested in
measuring the flow rate changes during the ‘pre-injection’ phase (before
pushing the plunger), ‘injection phase’ (when the plunger is pushed to
its first stop to discharge the majority of the liquid stored in the tip), and
‘extra injection’ phase (when the plunger is further pushed to its second
stop to ensure the liquid stored in the tip is fully discharged) (Fig. 7a).
Our experiments indicated the existence of distinct flow rate profiles
at each phase of the injection (Fig. S20a). The ‘pre-injection’ phase was
associated with a zero-flow rate profile, as expected. The ‘injection’
phase was characterised with an exponential rise in the flow rate until
reaching a peak followed by a slight reduction in the flow rate. The
‘extra injection’ phase was characterised with a small rise in the flow
rate until reaching the secondary peak followed by an exponential drop,

Fig. 7. Dynamic flow rates generated by a manual pipette recorded by our flow
Fig. 6. Dynamic flow profiles generated by a pressure pump using whole sensor: (a) The three stages of flow injection; (b) Dynamic flow rates provided
human blood recorded by the flow sensor: (a) Schematics showing the experi­ by a 20 µl Eppendorf Pipette at different dispense volumes measured for whole
mental set up; (b) Constant flow rate profiles; (c) A sine wave flow rate profile; human blood. (b′) Insets showing the variation of exponents during the ‘extra
and (d) A square wave flow rate profile. injection’ stage as various dispensed volumes of 10, 15 and 20 µl.

6
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

which persisted until the blood was fully discharged (Fig. 7b). MATLAB [DP200101248, Discovery Project], and the National Health and Med­
software was used to fit exponential curves on the experimental flow ical Research Council [2020197, Ideas Grant].
rates obtained during the ‘extra injection’ phase at the withdrawal
volumes of 10, 15, and 20 µl. Characterised by the decaying nature of CRediT authorship contribution statement
the exponential, the exponent (β) consisted of –1.596, –1.520, and
–1.485 for the volumes of 10, 15, and 20 µl, respectively (Fig. 7b′). Adam Hawke: Methodology, Validation, Formal analysis, Investi­
Further experiments were carried out by diluting the whole blood gation, Writing – original draft, Writing – review & editing, Visualiza­
with 10 % of dilution buffer (phosphate-buffer saline) (Fig. S20b). tion. Gianmarco Concilia: Methodology, Investigation. Peter
Thurgood: Methodology, Formal analysis, Investigation. Arman
4. Conclusion Ahnood: Methodology, Supervision. Sara Baratchi: Conceptualization,
Methodology, Formal analysis, Writing – original draft, Visualization,
In summary, we described a pressure-based, 3D printed flow sensor. Supervision, Funding acquisition. Khashayar Khoshmanesh: Concep­
An extensive set of experiments were performed to characterise the flow tualization, Methodology, Formal analysis, Investigation, Writing –
sensor against water and water-glycerol solutions as well as whole and original draft, Writing – review & editing, Supervision, Funding
diluted human blood. acquisition.
First, we measured the constant and dynamic (square wave) flow rate
profiles driven by a syringe pump. Our experiments indicated the ability Declaration of Competing Interest
of the sensor to measure constant flow rates within the range of
10–2000 µl/min with a relative error of ± 1.8 % and to monitor the The authors declare that they have no known competing financial
exponential rise and fall of the flow rate when the pump is turned on/off interests or personal relationships that could have appeared to influence
in repeated cycles. Applying a trapezoidal wave shaped flow profile with the work reported in this paper.
a magnitude of 500 µl/min using a syringe pump, we recorded a 5.2 %
overshoot when rising the flow rate and a 0.2 % error at the peak flow Data Availability
rate using our flow sensor, which were significantly less than the 16.1 %
overshoot and 2.0 % error obtained by a commercial calorimetric flow Data will be made available on request.
sensor.
Additionally, we demonstrated the ability to measure the varying Appendix A. Supporting information
viscosity of non-Newtonian human blood under predetermined flow
rates. The blood viscosity measured by our flow sensor varied from 5.8 Supplementary data associated with this article can be found in the
to 3.9 mPa.s at flow rates ranging from 25 to 500 µl/min at 37 ◦ C, online version at doi:10.1016/j.sna.2023.114686.
showing a relative error of ± 3 % compared to the theoretical values
obtained by the Carreau-Yasuda model. The blood viscosity exhibited a References
150–200 % increase at room temperature, varying from 11.5 to 5.7 mPa.
s at the same flow rates. [1] David J. Beebe, A. Glennys, A. Mensing, G.M. Walker, Physics and applications of
Second, we measured constant and dynamic flow rate profiles microfluidics in biology, Annu. Rev. Biomed. Eng. 4 (2002) 261–286.
[2] S. Baratchi, K. Khoshmanesh, C. Sacristán, D. Depoil, D. Wlodkowic, P. McIntyre,
generated by a piezoelectric pump. Experiments indicated that even A. Mitchell, Immunology on chip: promises and opportunities, Biotechnol. Adv. 32
under constant flow rate conditions, the flow rate oscillated between a (2014) 333–346.
minimum and maximum value. The amplitude of oscillations was [3] T.M. Squires, S.R. Quake, Microfluidics: fluid physics at the nanoliter scale, Rev.
Mod. Phys. 77 (2005) 977–1026.
reduced in response to higher fluid viscosity. We further examined the [4] S. Nahavandi, S.-Y. Tang, S. Baratchi, R. Soffe, S. Nahavandi, K. Kalantar-zadeh,
sensitivity of the flow sensor when generating square wave flow rate A. Mitchell, K. Khoshmanesh, Microfluidic platforms for the investigation of
profiles at 0.1–0.5 Hz. intercellular signalling mechanisms, Small 10 (2014) 4810–4826.
[5] C. Cavaniol, W. Cesar, S. Descroix, J.L. Viovy, Flowmetering for microfluidics, Lab
Third, we measured constant and dynamic flow rates driven by a
a Chip 22 (2022) 3603–3617.
pressure-driven pump. In doing so, we measured dynamic flow rates [6] L. Huang, Microfluidic flow sensing approaches, in: S.M.S. Murshed (Ed.),
generated by sinusoidal, square, and dual-frequency pressure profiles. Advances in Microfluidics and Nanofluids, IntechOpen, 2021.
Experiments indicated the formation of smooth flow rate profiles vary­ [7] A. Rasmussen, C. Mavriplis, M.E. Zaghloul, O. Mikulchenko, K. Mayaram,
Simulation and optimization of a microfluidic flow sensor, Sens. Actuators A: Phys.
ing from 180 to 420 µl/min. 88 (2001) 121–132.
Finally, we measured the dynamic flow rate profiles generated by a [8] J.T.W. Kuo, L. Yu, E. Meng, Micromachined thermal flow sensors—a review,
manual pipette. Experiments provided insight into the three stages of Micromachines 3 (2012) 550–573.
[9] G. Oudebrouckx, D. Nieder, T. Vandenryt, S. Bormans, H. Möbius, R. Thoelen,
flow injection by the pipette. In particular, we studied the exponential Single element thermal sensor for measuring thermal conductivity and flow rate
reduction of flow rate during the ‘extra injection’ stage and showed its inside a microchannel, Sens. Actuators A: Phys. 331 (2021), 112906.
sensitivity to blood viscosity. [10] S.M. Cahalan, V. Lukacs, S.S. Ranade, S. Chien, M. Bandell, A. Patapoutian, Piezo1
links mechanical forces to red blood cell volume, eLife 4 (2015), e07370.
The 3D printed flow sensor is highly sensitive, inexpensive, modular, [11] A.J. Iliff, X.Z.S. Xu, A. Mechanosensitive, GPCR that detects the bloody force, Cell
low-powered, and compact, which makes it amenable for flow moni­ 173 (2018) 542–544.
toring in microfluidics. The flow sensor can be readily coupled to various [12] G. Xu, S.M. Stevens Jr., F. Kobiessy, H. Brown, S. McClung, M.S. Gold, D.
R. Borchelt, Identification of proteins sensitive to thermal stress in human
microfluidic systems for measuring constant or dynamic flow rates neuroblastoma and glioma cell lines, PLoS One 7 (2012), e49021.
generated by various pumping mechanisms [37–40]. These capabilities [13] K.E. Hauschild, R.E. Metzler, H.-D. Arndt, R. Moretti, M. Raffaelle, P.B. Dervan, A.
can be expanded upon through the addition of a manifold system, this Z. Ansari, Temperature-sensitive protein–DNA dimerizers, Proc. Natl. Acad. Sci.
USA 102 (2005) 5008–5013.
would allow for multiple microfluidic devices to be measured using a
[14] T.H. Kim, D.-K. Kim, S.J. Kim, Study of the sensitivity of a thermal flow sensor, Int.
single flow sensor. The pressure drop is proportional to the liquid vis­ J. Heat Mass Transf. 52 (2009) 2140–2144.
cosity, which in turn depends on temperature. As such, the incorpora­ [15] A. Persson, The Coriolis effect, Hist. Meteorol. 2 (2005) 1–24.
tion of a temperature sensor into our flow sensor streamlines the [16] R. Smith, D.R. Sparks, D. Riley, N. Najafi, A. MEMS-Based, Coriolis mass flow
sensor for industrial applications, IEEE Trans. Ind. Electron. 56 (2009) 1066–1071.
calibration process of the sensor. [17] M. Richter, P. Woias, D. Weiβ, Microchannels for applications in liquid dosing and
flow-rate measurement, Sens. Actuators A: Phys. 62 (1997) 480–483.
Funding [18] H. Bissig, H.T. Petter, P. Lucas, E. Batista, E. Filipe, N. Almeida, L.F. Ribeiro,
J. Gala, R. Martins, B. Savanier, F. Ogheard, A.K. Niemann, J. Lötters,
W. Sparreboom, Primary standards for measuring flow rates from 100 nl/min to 1
This work was supported by the Australian Research Council ml/min – gravimetric principle, Biomed. Eng./Biomed. Tech. 60 (2015) 301–316.

7
A. Hawke et al. Sensors and Actuators: A. Physical 362 (2023) 114686

[19] N. Noeth, S.S. Keller, A. Boisen, Integrated cantilever-based flow sensors with Gianmarco Concilia received his Bachelor’s degree (with
tunable sensitivity for in-line monitoring of flow fluctuations in microfluidic Honours) in Biomedical Engineering in 2019 from RMIT Uni­
systems, Sensors 14 (2014) 229–244. versity, Australia. He is currently a Ph.D. student at the School
[20] H.P. Lang, M. Hegner, C. Gerber, Nanomechanical cantilever array sensors, Springe of Engineering, RMIT University. His research interests include
Handb. Nanotechnol. (2017) 457–485. microfluidics, image processing, and organ-on-a-chip.
[21] E. Skotadis, E. Aslanidis, G. Kokkoris, E.A. Vargkas Kousoulas, A. Tserepi,
D. Tsoukalas, Flow determination via nanoparticle strain sensors for easy Lab on
Chip integration, Sens. Actuators A: Phys. 344 (2022), 113765.
[22] R. Lindken, M. Rossi, S. Große, J. Westerweel, Micro-Particle Image Velocimetry
(µPIV): recent developments, applications, and guidelines, Lab a Chip 9 (2009)
2551–2567.
[23] F. Bürkle, J. Czarske, L. Büttner, Simultaneous velocity profile and temperature
profile measurements in microfluidics, Flow Meas. Instrum. 83 (2022), 102106.
[24] F.G. Ergin, B.B. Watz, N.F. Gade-Nielsen, A review of planar PIV systems and image
processing tools for lab-on-chip microfluidics, Sensors 18 (2018).
[25] J. Czarske, L. Büttner, Micro laser doppler velocimetry (μ-LDV), in: D. Li (Ed.),
Peter Thurgood received his Ph.D. in Electrical Engineering
Encyclopedia of Microfluidics and Nanofluidics, Springer New York, New York,
from RMIT University, Australia in 2018. He is currently a
NY, 2015, pp. 1825–1829.
Research Fellow and a key member of the Mechanobiology and
[26] L. Stern, A. Bakal, M. Tzur, M. Veinguer, N. Mazurski, N. Cohen, U. Levy, Doppler-
Microfluidics Research Group at RMIT University. He has
based flow rate sensing in microfluidic channels, Sensors 14 (2014) 16799–16807.
published more than 30 peer-reviewed journal papers. His
[27] N. Morita, W. Iwasaki, H. Nogami, F. Nakashima, E. Higurashi, R. Sawada,
research interests include microfluidics, lab-on-a-chip, and
Hematocrit-insensitive absolute blood flow rate measurement in 0.5-mm-diameter
microfabrication.
flow channel using MEMS-based Laser doppler velocimeter with signal
modification for detecting beat frequency from broad power spectrum, Sens.
Mater. 30 (2018) 3009.
[28] K. Adamski, B. Kawa, R. Walczak, 3D printed flowmeter based on venturi effect
with integrated pressure sensors, Proceedings 2 (2018) 1509.
[29] G. Wolterink, A. Umrani, M. Schouten, R. Sanders, G. Krijnen, 3D-printed
calorimetric flow sensor, IEEE SENSORS2020 (2020) 1–4.
[30] T. Koutsis, P. Pikasis, A. Psyrris, G. Kaltsas, A thermal flow sensor with a 3D printed
housing for spirometry applications, Microelectron. Eng. 226 (2020), 111286.
[31] F. Daniel, J. Peyrefitte, A.D. Radadia, Towards a completely 3D printed hot wire Arman Ahnood received his Master’s degree in Engineering
anemometer, Sens. Actuators A: Phys. 309 (2020), 111963. from University of Cambridge in 2006 and his Ph.D. in Elec­
[32] M.R. Khosravani, T. Reinicke, 3D-printed sensors: current progress and future tronics and Electrical Engineering from University College
challenges, Sens. Actuators A: Phys. 305 (2020), 111916. London in 2011. He is a Senior Lecturer in the School of En­
[33] I. Pereiro, A. Fomitcheva Khartchenko, L. Petrini, G.V. Kaigala, Nip the bubble in gineering, RMIT University. He is an expert in biomedical
the bud: a guide to avoid gas nucleation in microfluidics, Lab a Chip 19 (2019) signals and systems and is passionate about applying in­
2296–2314. novations in electronics and optoelectronics for better health
[34] J.D. Hunter, Matplotlib: a 2D graphics environment, Comput. Sci. Eng. 9 (2007) outcomes.
90–95.
[35] A. Ostadfar, Chapter 1 - fluid mechanics and biofluids principles, in: A. Ostadfar
(Ed.), Biofluid Mechanics, Academic Press, 2016, pp. 1–60.
[36] S. Baratchi, K. Khoshmanesh, O.L. Woodman, S. Potocnik, K. Peter, P. McIntyre,
Molecular sensors of blood flow in endothelial cells, Trends Mol. Med. 23 (2017)
850–868.
[37] J. Han, M. Mayyas, J. Tang, M. Mousavi, S.A. Idrus-Saidi, S. Cai, Z. Cao, Y. Wang,
J. Tang, R. Jalili, A.P. O’Mullane, R.B. Kaner, K. Khoshmanesh, K. Kalantar-Zadeh,
Liquid metal enabled continuous flow reactor: A proof-of-concept, Matter 4 (2021) Sara Baratchi received her Ph.D. in Cellular Biology from
4022–4041. Deakin University, Australia in 2011. She is currently the head
[38] M. Mohammed, P. Thurgood, C. Gilliam, N. Nguyen, E. Pirogova, K. Peter, of the Mechanobiology and Microfluidics laboratory at the
K. Khoshmanesh, S. Baratchi, Studying the response of aortic endothelial cells Baker Heart and Diabetes Institute, Melbourne, Australia. She
under pulsatile flow using a compact microfluidic system, Anal. Chem. 91 (2019) has published more than 65 peer-reviewed journal papers and
12077–12084. is the recipient of several nationally competitive grants from
[39] N. Nguyen, P. Thurgood, A. Arash, E. Pirogova, S. Baratchi, K. Khoshmanesh, the Australian Research Council and the National Health and
Inertial microfluidics with integrated vortex generators using liquid metal droplets Medical Research Council. Her research interests include
as fugitive ink, Adv. Funct. Mater. 29 (2019), 1901998. vascular biology, mechanotransduction, and bio-microfluidics.
[40] P. Thurgood, J.Y. Zhu, N. Nguyen, S. Nahavandi, A.R. Jex, E. Pirogova, S. Baratchi,
K. Khoshmanesh, A self-sufficient pressure pump using latex balloons for
microfluidic applications, Lab a Chip 18 (2018) 2730–2740.

Adam Hawke received his Bachelor’s degree (with Honours)

in Electrical and Electronics Engineering in 2020 and Master’s Khashayar Khoshmanesh received his Ph.D. in Biomechan­
degree in Electronic Engineering in 2021 from RMIT Univer­ ical Engineering from Deakin University, Australia in 2011. He
sity, Australia. He is currently a Ph.D. student at the School of is currently an Associate Professor and the co-leader of the
Engineering, RMIT University. His research interests include Mechanobiology and Microfluidics Research Group at the
microfluidics, 3D printing, and lab-on-a-chip. School of Engineering, RMIT University. He has published
more than 135 peer-reviewed journal papers. He is the recip­
ient of multiple grants, awards and fellowships from the
Australian Research Council and the National Health and
Medical Research Council. His research interests include
microfluidics, organ-on-a-chip, hemodynamics, and vascular
mechanobiology.