**The Design and Experimental Development of Air Scanning Using a Sni**ff**er Quadcopter**

#### **Endrowednes Kuantama 1, Radu Tarca 2,\*, Simona Dzitac 3, Ioan Dzitac 4,5, Tiberiu Vesselenyi <sup>2</sup> and Ioan Tarca <sup>6</sup>**


Received: 1 August 2019; Accepted: 2 September 2019; Published: 6 September 2019

**Abstract:** This study presents a detailed analysis of an air monitoring development system using quadcopters. The data collecting method is based on gas dispersion investigation to pinpoint the gas source location and determine the gas concentration level. Due to its flexibility and low cost, a quadcopter was integrated with air monitoring sensors to collect the required data. The analysis started with the sensor placement on the quadcopter and their correlation with the generated vortex. The reliability and response time of the sensor used determine the duration of the data collection process. The dynamic nature of the environment makes the technique of air monitoring of topmost concern. The pattern method has been adapted to the data collection process in which area scanning was marked using a point of interest or grid point. The experiments were done by manipulating a carbon monoxide (CO) source, with data readings being made in two ways: point source with eight sampling points arranged in a square pattern, and non-point source with 24 sampling points in a grid pattern. The quadcopter collected data while in a hover state with 10 s sampling times at each point. The analysis of variance method (ANOVA) was also used as the statistical algorithm to analyze the vector of gas dispersion. In order to tackle the uncertainty of wind, a bivariate Gaussian kernel analysis was used to get an estimation of the gas source area. The result showed that the grid pattern measurement was useful in obtaining more accurate data of the gas source location and the gas concentration. The vortex field generated by the propeller was used to speed up the accumulation of the gas particles to the sensor. The dynamic nature of the wind caused the gas flow vector to change constantly. Thus, more sampling points were preferred, to improve the accuracy of the gas source location prediction.

**Keywords:** quadcopter; drone; pollutant; carbon; air monitoring; kernel

#### **1. Introduction**

Several types of air, water, and soil pollutants are impossible to avoid, being encountered in almost all countries. Some of them represent real threats, rising risks to human health and environment degradation, such as (for the case of big cities) air pollution generated by stationary sources (e.g., factories, power plants), by mobile sources (e.g., cars, buses), and also natural sources (e.g., windblown dust, wildfires). Pollutants can be classified into two types: pollutants having known sources, and

pollutants having unknown sources. Each of them have specific approaches related to measurement methods and source identification. Most developed countries have developed law regulations to organize regular air quality supervision for primary urban pollutants [1–3]. The regulations could be put into practice only by monitoring regularly and accurately the concentration of the pollutant. Usually, stationary monitoring stations with network systems located in critical areas for data collection and processing are used [4–6]. This type of monitoring is limited by the sampling location and data accessibility [7]. With the advance in technology, this old paradigm gets surpassed by the utilization of sensor technology which is superior in respect to the low-cost material, simplicity and affordability, and also the portability of the air pollution monitoring systems [8,9]. The technology above can be integrated into an air scanning system by using an unmanned aerial vehicle (UAV) characterized by good maneuverability, controllable altitude, and location. [10,11]. Although the recent developments in UAV technology are promising, there are studies that show that in particular areas of applications (for example environmental restoration monitoring) there are still many issues to be addressed [12]. A comprehensive review of using autonomous vehicles in environmental monitoring for pollution source localization is presented in [13]. A design and manufacturing protocol for the low cost and low weight quadcopter platform prototype for the purpose of environmental monitoring and research in order to assess ecological devastation of the natural environment is presented in [14].

Different recently published studies show the interest of researchers in the applications of quadcopters used in different fields of environmental monitoring like sampling of different substances, monitoring soil contamination, and response to natural disasters. Lally, H.T. et. al. [15] addresses the development of water sampling by using drones specially equipped with water-sampling devices. The issues of biological and physico-chemical sampling are described as well as some solutions for these issues. The authors show that it is envisaged that drone-assisted water sampling will act as a pivotal supporting tool if the cost benefit analysis of the application gives positive results. In [16] the author's goal was to introduce and test a method able to predict copper accumulation points, using aerial photos taken by drones and micro-rill network modeling. In this case the drone collected photogrammetric data, which was compared with the results obtained by computer modeling. According to the results of the study, the authors were able to predict zones of copper accumulation at a plot scale. Other important applications of drones are presented in [17] where is stated that UAV's can have a crucial role in the case of natural disaster response and humanitarian relief aid. The key areas of intervention in this case are: aerial monitoring of post-natural disaster damage, natural disaster logistics and cargo delivery, and post-natural disaster aerial assessment. An application which is close to our paper's subject is presented in [18] where gas concentrations resulting from an underground coal fire (carbon dioxide emissions) were measured using aerial monitoring with drones. The authors state that it is estimated that these fires generate as much as 3% of the world's annual carbon dioxide emissions and that drone collected gas concentration data provides a safe alternative for evaluating the rank of burning coal deposits.

A portable air scanning system was developed using a quadcopter equipped with an air scanning sensor to perform air quality measurement, thus called a 'sniffer' quadcopter. The development began with analysis of the correlation between propeller's air trajectories and the sensors placement, determination of an appropriate flying pattern to optimize the air measurement, and investigation of ways to minimize wind effects in the measurement process. A computational method was used to ascertain the sensor's placement on the quadcopter, and the result was proven in the field test. The low-cost and portable gas sensors MQ-2 and MQ-135 were used to measure carbon concentration in the air. To develop this system, the sensor placement in the quadcopter is critical. Using computational fluid dynamic (CFD) simulation, the vortex field generated by the propeller was analyzed to determine the best place for sensor mounting. With an appropriate mounting place, the response time and the accuracy of data collected by the sensor can be increased [19]. Even with low-cost instruments, the data accuracy and detection range can be as good as conventional monitoring [20]. Two types of flight patterns for air measurement were used to detect the direction of gas dispersion and discover

the gas source location and gas concentration level. The present system was designed to perform flight pattern measurement methods which consist of point source measurement and non-point source measurement. A path planning with eight sampling points around the gas source was used to obtain the gas concentration at a known point source, for example, an industrial emission or chimney [21]. The sample points formed a square with the gas source in the middle. Thus, more accurate data can be obtained despite the dynamic environmental change. On the other hand, for a non-point or unknown sources such as forest fires or pipe gas leaks, a grid pattern with 24 sampling points was used to detect the gas source location based on the gas dispersion measurement and analysis. To optimize the measurement results, wind effect was considered. Gas source location and gas concentration level can only be estimated with statistical methods. In the case of fires, an extensive study had been made to detect sources with UAVs equipped with thermal detection capabilities [22]. Gas concentration detection (for example CO concentration) can give additional information besides the thermal data. The method described in this paper is much more cost effective and can be used by smaller communities (at the level of small cities and towns) to check for fire sources that can be a threat for the population or economy. The results from both methods were evaluated using analysis of variance (ANOVA) to obtain the gas concentration at the source, and using a Gaussian dispersion model to analyze the gas dispersion. In the Gaussian dispersion model, parameters such as wind speed and direction, source term, etc., were obtained by monitoring data to acquire positioned trajectories with bivariate input environmental data [23,24]. In a small scale of gas measurement (<100 m), a sparse Gaussian kernel method was used as a statistical evaluation on a two-dimensional spatial model of a grid pattern to deal with the specific properties of gas dispersion, including the turbulent features of the wind [25,26]. Spatial integration is made by convolving sensor readings and modeling the information data of the point measurements with a Gaussian kernel method. The grid pattern size and data retrieval time depend on the sensors' sensitivity. This research is limited with the usage of low-cost sensors and a narrow pattern, but these don't affect our goal which is to prove that the Gaussian kernel method is suitable to analyze gas dispersion vectors and detect gas source locations.

The sniffer quadcopter was designed to work automatically according to the command input in the pre-flight setting, one of which is a GPS coordinate. In automatic mode, the quadcopter will fly to the target point to perform the measurement with the pre-programmed flight pattern. In this research, the measurement target is carbon monoxide (CO) because it was easily found and/or made and was measured with low-cost and portable gas sensors MQ-2 and MQ-135. The tests were performed on an aerial zone with a maximum of 24 sampling points, each measuring 1 m2. On each point, the sampling time was 10 s and the data was collected whilst maintaining the quadcopter in hover mode. The aerial zone was intentionally small so the research could be focused on the algorithm for gas source determination and gas concentration level, besides optimizing the sensor placement on quadcopter. The main purpose of our paper is to study the possibility of using low-cost aerial monitoring system and pollution source detection, which can be available to smaller communities (towns, NGOs, environmental protection associations) in order to detect and prevent threats to the environment.

#### **2. Sni**ff**er Quadcopter Design and Analysis**

Prior to a quadcopter design, it is imperative to know the application of the drone itself, thus one can estimate the required lifting force in accordance to the total weight of the quadcopter. The physical factors of quadcopter design related to lifting force are propeller diameter, propeller's angle of attack, quadcopter size, and rotor angular velocity. Greater lifting force and high-speed flight are not synonymous with larger propellers or high-speed rotors because they can cancel out the advantages of using a quadcopter. For example, an oversized propeller produces substantial air flows which can cause turbulence and flight instability.

To optimize the sniffer quadcopter design, a detailed analysis of the vortex field generated by the propeller's angular velocity and its effect on sensor placement on the quadcopter was performed. This research used 13-inch propellers, a 380 rpm/v rotor, and 18.5-volt lithium polymer (LiPo) batteries. With this specification, it can be deduced that the maximum no-load rotor speed is 7030 rpm. To get a more accurate calculation of the lifting force, the computational fluid dynamic (CFD) module of the SolidWorks software was used. Table 1 shows detailed specifications of the quadcopter. The frame size was 460 mm, the quadcopter weight was 1800 g, and the sensor weight was 100 g. Quadcopter flight time depends on the capacity of the LiPo batteries used and the maximum carriable load depends on the total lifting force. The magnitude of the force analyzed with the CFD method and the correlation between thrust, frame size, and level of stability will be explained in detail in the propeller vortex field section.


**Table 1.** Detail specifications of sniffer quadcopter.

Due to short flight duration (<40 min), several settings must be initialized to permit the quadcopter to perform optimally, such as the target point, in the form of GPS coordinate, and quadcopter flight behavior, in the form of altitude, speed, and measurement pattern. The settings can also be made on the ground station system through the designed mission planner system which can communicate with the sniffer quadcopter using 433 MHz telemetry [27,28]. The architecture of the aerial platform system is presented in Figure 1. The ground station using an open source web application platform was also used to monitor all measurement processes. With the autonomous flight pattern, the quadcopter will run the sequential process according to the pre-flight command list. For intentional interruption of the process, an emergency system was designed so the user's command can be delivered via the ground station control which will force the quadcopter either to make a landing or to fly back to the home coordinates. On board the sniffer quadcopter system one can find two controllers: the flight controller and sniffer microcontroller. The flight controller serves to maintain the stability of the maneuvers, together with an orientation sensor and a tracking pattern based on the input coordinate [29]. The sniffer system serves to perform air scanning and save the data on a memory card. Both controllers are always communicating with each other to determine when to carry out the data retrieval process.

#### *2.1. Vortex Field Analysis*

Data retrieval was performed while keeping the quadcopter in the hover state at each analyzed point. It was noticed that data were profoundly affected by the air trajectories of the propeller's vortex field and also by the wind. The vortex method was applied [30,31] to analyze the aerodynamic behavior of the aircraft. The authors' previous studies [32,33] explain the quadcopter's propeller design and flight stability analysis. As mentioned previously, the maximum no-load rotor speed was 7030 rpm. When connected with a 13-inch propeller, the maximum rotor speed was 4080 rpm. To obtain the total thrust, a CFD analysis was performed using the SolidWorks software. The computationally analyzed parameter was the propeller rotation with a set up rotation area, with 0.1% turbulence intensity with 0.002 m turbulence length, at a thermodynamic pressure of 101,325 Pa and a temperature of 293.2 K. Each propeller had a rotation region, in which a diagonal pair of propellers rotated clockwise (CW) and the other diagonal pair rotated counterclockwise (CCW) with a velocity of 0–4080 rpm. The simulation generated the values of the vertical force and the total air velocity in all propeller rotation areas. With a velocity of 4080 rpm, each rotor can produce a total thrust of 24 N. Besides that, the vortex field was

also modeled using the same software with the same parameters. Figure 2 below shows the design of the sniffer quadcopter.

**Figure 1.** Sniffer quadcopter system architecture.

**Figure 2.** The sniffer quadcopter design.

In this study, the generated vortex field and its correlation with sensor position and the effect of the environmental change have been analyzed. The vortex field was modeled using computational fluid dynamics (CFD). The results showed that the air trajectories generated by each propeller rotation with maximum velocity didn't affect each other because the air velocities produced were the same. Figure 3 presents the analysis of the vortex field using CFD. The propellers having a maximum angular velocity (Ωmax) of 4080 rpm generate an air velocity of 6 ms−<sup>1</sup> (υair) oriented downwards (along the *z*-axis); the air velocity measured between propellers (−30 ≤ x ≤ 30) mm and (−30 ≤ y ≤ 30) mm was 0.5 ms−1. For this study, we used a sniffer quadcopter with a total load of 1900 g keeping it in hover state at 80% of the maximum speed, which corresponds to 3264 rpm. Consequently, the angular velocity generated an air velocity of 4.8 ms−<sup>1</sup> on the propeller, but the vortex fields generated by the propellers did not affect each other.

**Figure 3.** Air velocity dispersion.

This means that the vortices generated during the hover state with no wind effects didn't lead to turbulence on the quadcopter frame, which proved that the selected propeller diameter, quadcopter size, and rotor speed were appropriate. In the presence of wind effects, the quadcopter control system will work to maintain stability.

#### *2.2. Correlation Between Vortex Field and Sensor Position*

The transport of the monitored gas towards the gas sensor is a critical process to obtain accurate data which can be easily affected by the disturbances generated by the propeller's vortex. Sensor position, gas distribution, and wind resistance are factors which influence the measurement results [34,35]. The downside of using a low-cost gas sensor is the low response time. To determine the sensors' placement, one must consider the sensors' response time whilst still maintaining the quadcopter stability. The use of an extended pole to place the sensors outside of propeller vortex field is not advisable because it can affect quadcopter stability. Even though it might not matter much with a small-scale sensor, an extended pole attached with a heavier sensor will surely be effect by the wind. Thus, the design and computational analysis for the gas scanning sensors' placement on the quadcopter's frame were only tested for two points, i.e., point A and point B.

Point A corresponds to the placement of the sensors at the bottom of the main frame; thus, the gas flow is not influenced by the propeller's vortex. Point B corresponds to the situation in which the sensors are mounted on the front side of the frame so that the propeller's vortex blows the gas directly on the sensors. Both positions were analyzed using CFD prior to the field tests. The results of the simulation process for the case of the propeller's maximum angular velocity (4080 rpm) are presented in Figure 4. It can be noticed that the maximum air velocity occurs in each propeller rotation area while the minimum corresponds to the center of the frame. The placement of the air scanning sensor relies heavily on its type and response time. The following coordinates describe the point A's position: (−<sup>50</sup> <sup>≤</sup> *<sup>y</sup>* <sup>≤</sup> <sup>50</sup>) mm; (−<sup>130</sup> <sup>≤</sup> *<sup>x</sup>* <sup>≤</sup> <sup>130</sup>) mm, and (*<sup>z</sup>* <sup>&</sup>gt; <sup>−</sup>50) mm, where air speed is <sup>υ</sup>air <sup>≈</sup> 0 ms−<sup>1</sup> as seen in Figure 4c.

**Figure 4.** Air velocity on quadcopter in various perspectives: (**a**) Point A—top view; (**b**) Point B—front view; (**c**) Point A—bottom view.

It is evident that at point A, each sensor is surrounded by the propeller's vortices, with air trajectories going downward to produce thrust. For the analyzed gas to reach the sensor while the quadcopter is in the hover state, it is essential that the gas velocity be higher than the air velocity generated by the propeller. The vector of gas velocity towards the sensor, and the magnitude of total resultant velocity are very crucial to determine whether the gas can reach the sensor or not. Another way is to place the sensor on the top of the frame (*z* > 70) mm or higher than the propeller as seen in Figure 4a. Point B corresponds to the placement of the sensor on the front of the frame (*x* > 100) mm and takes advantage of the air trajectories generated by the propeller, as shown in Figure 4b. In this position, the gas around propeller is suctioned out by the propeller and passed through the sensor before going downward. This placement facilitates the gas to reach the sensor and is suitable for sensors with low response time.

Figure 5 shows the comparison between three sensor placements in a field test with durations of 160 s and a stable gas source position. The aim of this test was not to compare the reading of gas concentration, but rather to see the sensors' response time in different mounting positions. Analysis of point A showed that the gas concentration decreases as the speed of the propeller increases, and the gas concentration increases as the speed of the propeller decreases. This result was linear with the CFD analysis results. This means the sensors were obstructed by the propeller's air trajectories and thus unable to properly measure the gas concentration. On the other hand, analysis of point B showed that the sensors' reading was a bit affected by propeller speed change, but the sensor could work well when the propeller speed was stable. This was due to the propeller's air flow 'directing' the gas towards the sensor. Lastly, the analysis was conducted on sensors only. The result showed the most stable measurement because it didn't get affected by the propeller's air trajectory. In conclusion, taking advantage of the propeller's air trajectory to 'direct' the gas toward the sensors resulted in valid data reading, on the condition that the measurement was done while the quadcopter was in a hover state to ensure stable propeller speed in every measurement.

**Figure 5.** Comparison of sensor positions.

#### **3. Grid Pattern Analysis**

The grid pattern and wind algorithm were integrated into the gas measurement process which was dynamically distributed [36]. Measurements can be performed for the case of either point source or non-point source. The point-source measurement is used for gas emissions with known locations, for example, chimney exhaust gases in industrial districts. However, as the gas dispersion depends on the wind direction an eight-point (P1–P8) square pattern is used to cover all wind blowing directions, as illustrated in Figure 6a. On the other hand, non-point source measurement is used to locate the gas source based on the particle density and wind direction. Data acquisition for this method allows the user to observe the gas dispersion gradient toward the closest point to the source (which has the highest particle concentration), exemplified in Figure 6b. In the field test, the grid pattern with 24 sample points (S1–S24) placed in a 4 × 6 matrix was able to be extrapolated using the Gaussian kernel method. The size of the cell depends on the sensitivity of the gas sensor.

**Figure 6.** Pattern model with (**a**) point source, (**b**) non-point source.

The quadcopter's flying sequence and sample points took place according to the grid numbers shown in Figure 6. The methodology consisted of collecting sample data of gas concentration *<sup>G</sup>*(*i*) *s* on each cell k, at the location *x*(*i*) where the symbol *i* represents the number of the measurement sample.

As we mentioned above, sometimes position error occurred while collecting data. Thus, position adjustments (Δ*G*) were needed to ascertain that the measurements were made in the center of each grid cell (*Gc*). The post-processing data errors relative to positioning errors on each cell can be minimized by data recovering process using Equation (1). The symbol *k* represents the number of the grid cell.

$$
\Delta G\_i^{(k)} = \left| G\_c^{(k)} - G\_s^{(i)} \right| \tag{1}
$$

where, for the *<sup>k</sup>* cell, in which a number of *<sup>m</sup>*(*k*) measurements were made, *<sup>G</sup>*(*k*) *<sup>c</sup>* value can be estimated with

$$\mathbf{G}\_{\mathbf{c}}^{(k)} = \frac{\sum\_{i=0}^{m^{(k)}} \mathbf{G}\_{\mathbf{s}}^{(k)}(i)}{m^{(k)}} \tag{2}$$

The differentiation of gas concentration between the center of the grid cell and the value acquired by the sensor was used to locate the gas source. If the relationship Δ*G*(*k*) > Δ*G*(*k*−1) is true, it means that the flight pattern moves toward the gas source, and conversely Δ*G*(*k*) < Δ*G*(*k*−1) means that the flight pattern moves away from the gas source. This post-processing of gas concentration variation in each cell helps in understanding the correlation between flight position and gas dispersion. In order to analyze the gas dispersion behavior, both for point source and non-point source models, an adaptive threshold with binary sample was used, as seen in Equation (3) [37].

$$\overline{P}^{(k)} = \overline{S}^{(k)} = \begin{cases} 1 & \to \left(\Delta G\_t^{(k)} > \Delta G\_t^{(k-1)}\right) \\ 0 & \to \left(\Delta G\_t^{(k)} \le \Delta G\_t^{(k-1)}\right) \end{cases} \tag{3}$$

For both methods, sample acquisition was done with the same iteration (*t*) The first value of gas concentration acquired has been used as the reference for measurements. The *P* (*k*) <sup>=</sup> *<sup>S</sup>* (*k*) = 1 indicates an increase in gas concentration, whereas *P* (*k*) = *S* (*k*) = 0 indicates a decrease.

#### *3.1. Gas Dispersion*

In order to analyze the gas dispersion measured by the quadcopter, a statistic method has been used to generate a two-dimensional gas distribution map, using the DM + V kernel algorithm presented in [38,39]. This algorithm treats the gas distribution model as a density estimation problem which can be solved using convolution with a two-dimensional Gaussian kernel. The kernel's shape regulates the amount of extrapolation. When the wind is not blowing, the kernel's shape is a circle, thereby: σ*<sup>x</sup>* = σ*<sup>y</sup>* = σ0.

The vector of gas dispersion with spatial extrapolation was analyzed using the Gaussian weighting function (N) which represented the importance of the gas reading value obtained for each cell. The first step of the algorithm is the weight calculus *fi* (*k*)(σ0), which, intuitively represents the information content of a single measurement, *i,* of the sensor inside a net's cell. The weight is calculated by the mean of a Gaussian kernel (N) evaluation applied to the distance between measurement location *x*(*i*) and the center point *x*(*k*) of the *k* cell.

$$w\_{\mathbf{i}}^{(k)}(\sigma\_0) = \mathcal{N}(\left|\mathbf{x}^{(k)} - \mathbf{x}^{(\bar{i})}\right|, \sigma\_0) \tag{4}$$

*Sensors* **2019**, *19*, 3849

Starting from equation (4), the following values are integrated and placed in a temporary grid map: weights *W*(*k*)(σ0), weighted sensor readings *G*(*k*)(σ0), and weighted variance contributions *V*(*k*)(σ0), as follows:

$$\mathcal{W}^{(k)}(\sigma\_0) = \sum\_{i=1}^{n} w\_i^{(k)}(\sigma\_0) \tag{5}$$

$$G^{(k)}(\\\sigma\_0) = \sum\_{i=1}^{n} w\_i^{(k)}(\\\sigma\_0) G\_s^{(i)} \tag{6}$$

$$V^{(k)}(\\\sigma\_0) = \sum\_{i=1}^n w\_i^{(k)}(\sigma\_0)\tau^{(i)}\tag{7}$$

where τ(*i*) = <sup>Δ</sup>*G*(*k*) *i* 2 is the variance contribution of reading *i*.

From integrated weight map *<sup>W</sup>*(*k*)(σ0) a confidence map *<sup>x</sup>*(*k*) (σ0) can be obtained showing the degree of trust with which for one cell the readings is considered to be in the vicinity of the respective grid cell's center and is expressed as shown in the Equation (8).

$$w\_{l}^{(k)}(\\\sigma\_{0}) = N(\left|\mathbf{x}^{(k)} - \mathbf{x}^{(i)}\right|, \sigma\_{0}) \tag{8}$$

In normal dispersion, the confidence value is within the interval (0–1) which can be affected by the trajectory of the quadcopter, the size of the grid cell, the width of the kernel (σ0), and the scaling parameter (σ*r*).

Normalizing the integrated weighted sensor readings *G*(*k*)(σ0) with the integrated weights *W*(*k*)(σ0), then applying the confidence value and adding with the best guess for the cells with a low confidence (i.e., for cells for which we do not have sufficient information from nearby readings, indicated by a low value of *x*(*k*) ) results in the map estimation of the mean distribution *g*(*k*)(σ0):

$$\mathbf{g}^{(k)}(\sigma\_0) = \overline{\mathbf{x}}^{(k)} \frac{\mathbf{G}^{(k)}(\sigma\_0)}{\mathcal{W}^{(k)}(\sigma\_0)} + \left\{ 1 - \overline{\mathbf{x}}^{(k)} \right\} \mathbf{G}\_0 \tag{9}$$

As the best guess of the mean concentration *G*<sup>0</sup> we use the average over all sensor readings.

In the same way, the corresponding variance map *v*(*k*)(σ0) results from normalizing the weighted variance contributions *V*(*k*)(σ0) with the integrated weights *W*(*k*)(σ0) then multiplying with the confidence value and adding with a best estimate for the cells with a low confidence:

$$
\boldsymbol{\sigma}^{(k)}(\boldsymbol{\sigma}\_{0}) = \overline{\mathbf{x}}^{(k)} \frac{\boldsymbol{V}^{(k)}(\boldsymbol{\sigma}\_{0})}{\mathcal{W}^{(k)}(\boldsymbol{\sigma}\_{0})} + \left\{ 1 - \overline{\mathbf{x}}^{(k)} \right\} \boldsymbol{v}\_{tot} \tag{10}
$$

The estimate *vtot* of the distribution variance in regions far from measurement points is computed as the average over all variance contributions.

The mean value of gas concentration from each cell was used to make a predictive model in ANOVA. The spatial structure of the dispersion variance provided information on the gas dispersion vector and on the highest gas concentration which surely is located near the source.

#### *3.2. Correlation Between Wind and Gas Dispersion*

The gas dispersion is in linear correlation with the wind dynamic movement vector. Knowledge about the wind vector is helpful in locating the gas source. The extrapolation of gas measurement using bivariate Gaussian kernel provides information about the wind vector [10,40]. Two possible models were considered, i.e., an idle state in which wind velocity is zero, and a windy state, correlated with wind direction. An example of a detailed model is presented in Figure 7.

**Figure 7.** Wind direction model: (**a**) Idle state and (**b**) model considering wind velocity and direction.

The idle state with zero wind velocity is obtained from a normal dispersion having symmetrical kernel width (σ0) along the *x*-, *y*-, and *z*-axis on grid cell; a diagonal matrix with variance data (Σ) represents this state as seen in Equation (11).

$$
\Sigma\_{\rm{iilk}\varepsilon} = \begin{bmatrix}
\sigma\_0^2 & 0 \\
0 & \sigma\_0^2
\end{bmatrix} \tag{11}
$$

Wind velocity creates gas dispersion in the form of an ellipse with linear dependency. The wind vector changes the ellipse position according to the amount of change in the rotation matrix *R*(α). Rotation matrix is an orthogonal matrix in which *R*(α) <sup>−</sup><sup>1</sup> = *R*(α) *<sup>T</sup>* rotate bivariate Gaussian kernels around the *x*- and *y*-axis. Angle alteration (α) in the horizontal position (*x*- and *y*-axis) determines a two-dimensional wind vector which can be calculated using Equation (12). Data was enough to determine the pollutant source using the grid cell.

$$\sum\_{\overline{R}} = \begin{bmatrix} \sigma\_x^2 & a\sigma\_x\sigma\_y \\ a\sigma\_x\sigma\_y & \sigma\_y^2 \end{bmatrix} = R(a)\Sigma\_w R(a)^T \tag{12}$$

$$R(\alpha) = \begin{bmatrix} \cos \alpha & (-\sin \alpha) \\ \sin \alpha & \cos \alpha \end{bmatrix} \tag{13}$$

$$
\sum\_{\mathbf{w}} = \begin{bmatrix} a^2 & 0 \\ 0 & b^2 \end{bmatrix} = \begin{bmatrix} \left(\sigma\_0 + \gamma \left| \over \mathbf{v} \right| \right)^2 & 0 \\ 0 & \frac{\sigma\_0 - 2}{\sqrt{1 + \frac{\gamma \left| \over \mathbf{v} \right|}{\sigma\_0}}} \\ \end{bmatrix} \tag{14}
$$

The gas dispersion along with the wind velocity and vector γ → ν were obtained from the wind sensor's measurements located on the gas dispersion contour; *x*-axis values were proportional to the wind velocity, while *y*-axis values decreased with the wind velocity. The variable γ is the stretching parameter which depends on many environmental variables. The bivariate Gaussian kernel was rotated according to the wind vector.

#### **4. Environmental Monitoring**

The experiment for measuring carbon concentration on each sample point was done in an open field for both the point source and the non-point source case. The pollutant source consisted of burning coals placed in a burner with a height of 0.5 m. The 24 grid cells covered an area of 6 m (width) × 4 m (length). Each cell was 1 m2 in size, and the distance between measurement points was also 1 m. This distance depends on the sensor sensitivity; the more sensitive the sensor is, the larger the grid cell size may be, and a wider aerial zone requires a bigger gas source. Gas concentration measurement was done with the quadcopter in the hover state at the center of each cell during the 10 s sampling time.

#### *Sensors* **2019**, *19*, 3849

Each measurement consisted of 50 readings, used to calculate the mean value for that measurement. The setup of the experiment is shown in Figure 8.

**Figure 8.** Sniffer quadcopter field test.

Data analysis of the gas dispersion and the point source was done manually by moving the quadcopter in a pattern using a remote control (RC), at a fixed altitude of 1 m from the ground. When the quadcopter reached the center of the cell, it remained in hover mode, and an interrupt control system was sent through the RC to collect data. Continuous flight patterns were completed in the order of the data acquisition sequence shown in Figure 6. This driving method has been used for the aim of manual correction during the field test and also to minimize errors; thus, a more accurate gas dispersion post-processing algorithm has been achieved.

The test for sensor positioning was done based on the CFD result, as presented in Figure 4. With the gas sensor mounted on the middle-bottom frame, unstable data reading and sometimes even zero value readings resulted. On the other hand, by having the gas sensor placed on the front of the quadcopter, the airflow generated by the propeller always passed the gas beyond the sensor. In hover mode, the quadcopter always adjusts its position to a stable state and produces equal air velocity on each of the rotors. Each quadcopter's propeller rotation generates a vortex that draws the air from above and directs it downwards, to the CO sensors. When sensors were placed in the front of the quadcopter, the gas sensor responded well. The recorded data was analyzed for two flight patterns, as follows:

#### *4.1. Point Source*

In the first stage of the experiment, the sniffer quadcopter was used to determine the level of CO concentration in the surrounding atmosphere. The quadcopter flew in a square or circle pattern to read the CO concentration in the center of each of the eight sample points. As much as 8 sample points were measured during the test, and the total flight time of quadcopter was 130 s without landing and take-off time. Every time a position error occurred, an adjustment was made using Equations (4) and (8) under the condition that the point sample for one grid cell must be greater than one sample; thus, the variant and average value near each cell's center can be obtained. Figure 9 shows the value of CO concentration measured for each cell. All data were distributed normally and the hypothesis from Equation (4) was applied to determine the vector of gas dispersion.

**Figure 9.** Point source experiment results: (**a**) CO concentrations; (**b**) CO vectors; (**c**) Gas dispersion.

The analysis result for the gas dispersion vector with respect to the gas source position ) (*X*,*Y*) = 2 \* showed that the vector was ) (*X*,*Y*) > 2 \* . The carbon reading was modeled using a 2D contour in order to see the vector more clearly. The CO reading correlated with location was formulated in a matrix form shown in Equation (15).

$$G\_i^k = \begin{bmatrix} 19 & 6 & 27 \\ 5 & (>160) & 160 \\ 80 & 106 & 145 \end{bmatrix} \tag{15}$$

The gas dispersion contour along the *x* and *y* axes showed that the dispersion started from the highest to the lowest concentration, more precisely from the cell's center to the P1 point. In the point source method, the wind model of gas dispersion could not be seen clearly.

ANOVA analysis was used to compare cells two by two in all possible combinations, to get the estimation of CO concentration in the source. The analysis yielded the mean value (Δ*G* = 6.89) , the standard deviation (σ<sup>0</sup> = 0.36) . After that, based on the gas dispersion vector, the irrelevant values were eliminated. To calculate the gas source concentration, the values from the first column (*X* = 1) and the first row (*Y* = 1) were eliminated. The sample points (P6, P7, P8) were used in the calculation since they were in accordance with the vector. The gas concentrations in each cell were summed up with the standard deviation value, and the results were averaged to get the CO source value, found to be 200.29 ppm. In comparison with real measurements of the gas source with sensors only, as much as 8.85% error was detected for CO concentration.

#### *4.2. Non-Point Source*

The non-point source experiment was done by collecting 24 sample points of CO concentration during 360 s of flight time. Data acquisition was done in the same manner used for the point source method. Two types of MOX sensors were used, i.e., MQ-135 and MQ-2 which were placed together on the quadcopter's front frame. Both sensors were used simultaneously to ascertain the validity of the data and to get better analysis over the gas dispersion. In the field test, some burning coals were used as the CO source, placed on the position (X, Y) = (5, 2.5). Data of CO concentration were saved in the matrix form as presented in Equations (16) and (17). The CO concentration of the source was predicted, and the location of the source was analyzed through gas dispersion using the variant data of each cell.

$$S\_{MQ135} = \begin{bmatrix} 19 & 19 & 19 & 35 & 82 & 40 \\ 24 & 44 & 44 & 67 & 110 & 28 \\ 49 & 18 & 16 & 18 & 33 & 28 \\ 16 & 17 & 25 & 19 & 20 & 23 \end{bmatrix} \tag{16}$$

$$S\_{MQ2} = \begin{bmatrix} 37 & 30 & 27 & 39 & 65 & 45 \\ 34 & 67 & 71 & 70 & 98 & 34 \\ 54 & 33 & 29 & 42 & 38 & 18 \\ 29 & 25 & 28 & 13 & 24 & 34 \end{bmatrix} \tag{17}$$

All 24 sample points in the matrix were computed, yielding a 2D gas concentration contour for each of the sensors, presented in Figure 10.

**Figure 10.** Non-point source experiment results in a 2D contour: (**a**) for MQ135 sensor; (**b**) for MQ2 sensor.

Based on the highest value of CO, the gas tended to drift toward the S13 cell or (*X* < 5) in the horizontal direction and toward the S20 cell or (*Y* > 3) in the vertical direction. Tentatively, it can be concluded that the gas source was around the red zone on the 2D contour. Further analysis using Equation (10) was done to discover the connection between gas dispersion and wind effect. A gentle breeze of5ms−<sup>1</sup> blew around the field test area. Variant data of each sample was distributed normally to obtain standard deviation in form of kernel width for wind speed. Uncertainty of wind was conditioned as non-constant wind flow and represented by the (γ) parameter. Its value was estimated at 0.4 m based on the highest concentration stretch point and affected the stretch kernel shape. The calculation resulted in σ*o*(*MQ*135) = 0.247 and σ*o*(*MQ*2) = 0.236 with rotation vector *R*1(α) = 0 and *R*2(α) = <sup>π</sup> <sup>2</sup> , thus a wind vector with range value a = 2.24–2.25 m and b = 0.08 m was obtained. From the analysis of wind direction and gas dispersion, two possible positions for CO source resulted, which were at the coordinates (5 ≤ *x* ≤ 6.12), (2.92 ≤ *y* ≤ 3.08) and (4.92 ≤ *x* ≤ 5.0),(1.88 ≤ *y* ≤ 3) . Compared with the real position of the gas sensor (*X*,*Y*) = (5, 2.5) , it can be concluded that only one out of the two possibilities is the source of pollutant area; bivariate Gaussian kernel analysis was used to assess the source location and minimize the error of CO reading.

#### **5. Conclusions and Future Work**

Air monitoring using a sniffer quadcopter with a flight pattern was designed to measure CO concentration in each cell. The post-processing analysis was used to determine the source location and its CO concentration. The results from CFD and the field test showed that the sensor placed in front of the frame (*x* > 100) mm of the quadcopter was able to utilize the air trajectories generated by the propeller to direct the gas straight to the sensor. The best result was achieved when the data acquisition was made in hover mode to ensure constant airflow. Data acquisition was made using two methods, i.e., point source and non-point source. The point source method with a known location of the gas source was done using eight sample points forming a square pattern, having the source in the middle, which also is useful in facing the unpredictable wind effect.

The concentration of CO in the source was quantified using post-analysis by the means of the ANOVA method which was ran on eight samples during 130 s of flight time. Compared to the real data, the analysis showed as much as 8.85% error.

The other method was the non-point source, used to pinpoint the location of the gas source and also its concentration. This method adapts a grid pattern with 24 cells to collect data of CO with two types of gas sensors used simultaneously to ascertain data validity. The gas dispersion analysis results showed that the gas dispersion vector had changed twice, thus indicating two possible positions for gas source location. The gas dispersion vector has been analyzed using both the measurement position and CO concentration matrices. The readings of both sensors showed the same gas dispersion pattern, indicating the highest value of CO was in the S17 cell. The differences in the accuracy of data reading were affected only by the sensitivity of the MOX sensors. The correlation between gas dispersion and wind behavior must be known to overcome the possibility of result misinterpretation due to wind influence. Logically, the gas source must be in the vicinity of the cell having the highest CO concentration. A bivariate Gaussian kernel has been used to locate this cell's position. The gas source was calculated with the same method used in the case of the point source; thus, the weight cell in the form of standard deviation was obtained with the value of CO between 118.06–133.24 ppm while the actual value was 125 ppm. Overall, the field tests were done by manipulating the gas source; the quadcopter's altitude maintained at 1 m from the ground to collect data which then were calculated accordingly to acquire the gas source location. For the case of a small amount of burning coal, the experiment was possible only at low altitude with manual control of the quadcopter flight. The sensitivity of the sensors must pass the reliability test before being placed on the quadcopter because it affects the size of the cells. Finally, the analyses using normal dispersion and ANOVA were essential to obtain the gas concentration and gas source position.

This study using sniffer quadcopter has been limited to carbon monoxide measurements. The measurement method and gas source location detection method still have room for improvement. In the future, other pollutant compounds will be investigated and different gas sensors such as optical sensors will be used in comparison with the tested sensors. There are various application of mapping and measurement using a sniffer quadcopter, such as gas pipe leakage measurement, early warning systems for volcanic-prone areas, water pollution mapping with the pattern method, etc. From the perspective of flight pattern, with improvements in sensitivity and accuracy of reading, a larger scale grid pattern can be designed to save time on the data collection process. Quadcopter capability to withstand wind effects or heatwaves from the gas source can also be more developed. The higher goals are to utilize sniffer quadcopter as unmanned aerial security patrols to cope with environmental issues and monitor dangerous zones. In future experiments, we intend to test the data collection from the sensor at larger scales in quasi-real situations and also to use multiple drones which can communicate with each other and better map the field of interest.

**Author Contributions:** Conceptualization, R.T. and E.K.; methodology, R.T.; vortex field analysis, E.K. and T.V.; correlation between vortex field and sensor position, E.K. and I.T.; grid pattern analysis, E.K., R.T., S.D., I.D., I.T., and T.V.; environmental monitoring, E.K., R.T., and T.V.; writing—original draft preparation, E.K. and I.T.; writing—review and editing, V.T., S.D., and I.D.; supervision, R.T.; project administration, R.T.; funding acquisition, I.D. and R.T.

**Funding:** This research was funded under the LEADERS—Erasmus Mundus Grant (agreement number 2014-0855/001-001) by the European Commission, through the Education, Audio-visual, and Culture Executive Agency, in the Action Plan 2 for the years 2014–2018, supported by PNCDI III Programme P2—Transfer of knowledge to the economic operator (Bridge Grant PN-III-P2 2.1 BG-2016-0296) funded by UEFISCDI, Romania and also by PNCDI III Programme P2—Experimental demonstration project (PN-III P2-2.1-198PED⁄2017) funded by UEFISCDI, Romania.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Novel PDMS-Based Sensor System for MPWM Measurements of Picoliter Volumes in Microfluidic Devices**

**Mihăi¸tă Nicolae Ardeleanu 1,2, Ileana Nicoleta Popescu 1,\*, Iulian Nicolae Udroiu 3, Emil Mihai Diaconu 3, Simona Mihai 4, Emil Lungu 5, Badriyah Alhalaili <sup>6</sup> and Ruxandra Vidu 7,8,\***


Received: 30 September 2019; Accepted: 4 November 2019; Published: 8 November 2019

**Abstract:** In order for automatic microinjection to serve biomedical and genetic research, we have designed and manufactured a PDMS-based sensor with a circular section channel using the microwire molding technique. For the very precise control of microfluidic transport, we developed a microfluidic pulse width modulation system (MPWM) for automatic microinjections at a picoliter level. By adding a computer-aided detection and tracking of fluid-specific elements in the microfluidic circuit, the PDMS microchannel sensor became the basic element in the automatic control of the microinjection sensor. With the PDMS microinjection sensor, we precise measured microfluidic volumes under visual detection, assisted by very precise computer equipment (with precision below 1 μm) based on image processing. The calibration of the MPWM system was performed to increase the reproducibility of the results and to detect and measure microfluidic volumes. The novel PDMS-based sensor system for MPWM measurements of microfluidic volumes contributes to the advancement of intelligent control methods and techniques, which could lead to new developments in the design, control, and in applications of real-time intelligent sensor system control.

**Keywords:** microinjection sensor; PDMS capillary; micro-wire; picolitre volume measurement; MPWM microfluidic transport

#### **1. Introduction**

The increasing demands for a complete and real-time diagnosis of a larger population and the improvement of the quality of life in general has led scientists from all over the world to study and develop multidisciplinary science and technologies [1–6]. Among them, microfluidics are based on life sciences and medical technologies of the future, which will revolutionize biology and medical diagnostics [4,5] through manipulation of very small volumes of fluids (from nano to pico and femtoliters) [1,6] and miniaturized devices [7].

To integrate one or more laboratory analyses and synthesis on a single chip of just millimeters in size, researchers and specialists have used a miniaturized device called lab-on-a-chip (LOC), which allows a quick response and diagnosis at low energy consumption and cost, using a very small volume of samples [3,8–10]. LOC devices, including microfluidic ones, have many advantages such as reducing the use of chemicals (reduced sample and reagent usage) that can be rare, expensive and polluting, reducing waste production, miniaturization, integration, portability, and automation [4,7,8,11]. In addition, the LOC offers a precise analysis of a large numbers of individual molecules and cells, flexibility of device design, high experimental control, and reduces measurement times [4,5,12]. Another advantage of the microfluidic cell culture device is the ability to incorporate analytical biosensors into the culture platform, thus combining, in a non-invasive way, living cells and sensors for detecting cellular physiological parameters and analyzing external stimuli in situ [12,13].

Nowadays, microfluidic devices can be used for biological and medical analysis, detection, control and manipulation of biological samples and cell biology research such as: analysis of the unpurified blood samples, analysis of complex mixtures and molecules (especial DNA and proteins), DNA sequencing, single cell manipulation, electrophoretic separations, drug screening, screens for protein crystallization conditions, cell culture studies and reproductive cell selection [2,9,12,14–18]. Also, with the help of microdevices, we can carry out environmental analysis, food analysis and control, detection and screening of residues and explosives [19,20]. Microfluidic chips provide unrivaled control over droplets and jets, which have advanced all natural sciences [21]. The development of new types of bioassay for monitoring patient response to therapy is another application [7]. LOC is a complex microsystem that includes a network of mechanical, electronic and fluid functions microchannels. These microchannels serve as pipes, valves, sensors, electronics and other structures [7,8]. LOC, bio-microelectromechanical systems (bioMEMS) and micro-total-analysis-systems (μTAS) are technologies that include both microfluidics and detection capabilities [10,22]. In microfluidic devices, the chemical and physical microenvironment can be easily controlled by using on-chip valves that allow the release of fluids containing target molecules and substances with precise timing [23]. The droplet-based microfluidic LOC platform has significant advantages for high-throughput, continuous flow and ultra-low volume studies of biological and chemical experiments [10].

For microfluidic design and fabrication it is very important to select the appropriate material that will be used and also the size and the geometry (i.e., cross-section) of the channel, and the surface characteristics of the microchannel to be made in accordance with the specific applications of the chip must be taken into account and they are briefly reviewed in the following section.

#### *1.1. Microfluidic Device Materials*

Microfluidic devices can be fabricated using several materials: (a) inorganic materials such as: silicon and glass; (b) polymeric materials such as polymethylmethacrylate (PMMA), polydimethylsiloxane (PDMS) and a relative novel one, the hydrogel; and (c) Paper-based for microfluidic chips [22,24,25]. Glass is transparent, but because it is amorphous, vertical side are more difficult to etch than Si. The production of microfluidic devices made of glass or silicon is generally expensive and time-consuming [2]. Paper is a very inexpensive material and easy to work with. It is compatible with biological samples and can be chemically treated to bind to molecules or protein, but the main disadvantage when using paper–based microfluidic devices is the difficulty in detecting channels on the chip [22]. The hydrogel is a colloid made of polymeric chains of molecules spread in water. Hydrogel is very malleable, and various feature designs and sizes can be molded onto it. Furthermore, the hydrogel is commercially available, non-toxic to the cells and cost effective. One common polymer used to make hydrogel is sodium polyacrylate. Their application for microfluidic devices depends on their suitability for biological experiments. For instance, hydrogel could serve as a matrix material of

new classes of solar cells and photoreactors with embedded microfluidic networks and also it is used for tissue engineering due to its high permeability and biocompatibility [25]. The most used material (both for academic and industrial applications) for prototyping and fabrication LOC device is PDMS, a mineral-organic polymer in the siloxane family. PDMS is an excellent material for the fabrication of microchannel systems/master mold because of its specific characteristics that will be discussed in detail in the next section, selection of materials for experiments [12,16]. The PDMS has some drawbacks, such as diffusivity and swelling in organic solvents [26]. The chemical structure of PDMS consist in repeating units of -OSi(CH3)2- groups, which leads to a hydrophobic surface (θ*H*2*<sup>O</sup> <sup>a</sup>* <sup>=</sup> <sup>108</sup>◦) [27], which is another major disadvantage of this material. There are some solutions to solve this problem, e.g., the hydrophobicity can be solved by changing the surface properties of PDMS using specific surface treatments or modification [26].

#### *1.2. Techniques for Manufacturing Microfluidic Devices*

Generally, microfluidic devices consist of different components, such as reservoirs, chambers, and microchannels [22]. Microchannels with varied geometries and surface properties were produced and studied over time by many researchers. There are many techniques for making microfluidic systems, all of which are consistent with their use in different fields of application [2,11,28–37]. The methods for fabrication of microfluidic devices include prototyping techniques, such as: hot embossing [29,38,39], injection molding [40] soft lithography (photolithography followed by etching and bonding) [2] rapid prototyping [2] and replica molding [16,41]. Other methods used are direct fabrication techniques like laser photoablation or laser micromachining [42] photolithography/ optical lithography [43] or photolithography followed by etching and bonding) [44] and x-ray lithography [37, 45]. These techniques have advantages and disadvantages also that may or may not be used effectively in the manufacture of microchip devices. For instance, according with Qi et al. [39], the hot embossing technique has the following advantages: cost-effective, accurate and fast replication of microstructures and mass production. This technique has some disadvantages such as: (i) the restriction of applying only to thermoplastics material and (ii) the difficulty to fabricate complex 3D structures. With the micro-injection molding method [40], it is easy to manufacture complex polymer geometry, fine features and 3D geometries. This technique allows mass production, is highly automated and has a short cycle time. But has also disadvantages, i.e., the difficulty of forming large undercut geometries, high cost mold, and is applied only to thermoplastic materials, as in the case of the hot embossing technique. Another type of prototyping technique capable of fabricating 3D geometries is soft lithography, which is a cost-effective method and high resolution features (down to a few nm) can be obtained. One of the major disadvantages is the vulnerability to defects [2]. Rapid prototyping is capable of producing prototypes of almost any geometrical complexity in relatively short time [41]. The advantages of this technique compared to the methods that used the chromed mask in the photolithographic step are the reduction of time (i.e., hours instead of weeks) and reduction of costs (20–100 times more expensive than the chromatic mask that replaces the transparency. The disadvantage is the lower resolution of transparency (>20 μm) compared to the chromed mask (approx. 500 nm). The low transparency resolution leads to two walls with rough edges [2].

Replica molding has the role to form the microchannels in PDMS by generating a negative replica of a master (made by metal/hard material) in PDMS through casting of pre-polymer against the master [2]. Despite rapid and large format production, laser photo-ablation has multiple treatment sessions, and is applied for a limited number of materials [42].

Ideal for microscale features is the conventional photolithography/optical lithography but usually it requires a flat surface to start with and chemical post-treatment steps [43]. Finally, x-ray lithography allows high resolution for nano-patterns fabrication, absorption without artifacts and is capable of producing straight, smooth, walls, but imposes difficulties in the manufacturing process, consuming time and high costs [45].

#### 1.2.1. Geometry and Quality of Microchannels Surface

From the point of view of the geometry of the channels, the soft lithography technique can only make rectangular channels, which is a disadvantage in replicating the circular cross-sections of blood vessels, as well as for replicating cardiovascular flow conditions [30]. Rectangular microchannels limit cell growth on the bottom of the channel, exposing the cells to a non-physiological geometry [44]. Cells growing on the bottom of the channel have different shear stresses, which directly influence the alignment, elongation, differentiation and expression of the genes [44]. Microfabrication by photolithography method [32] is manufactured in the form of rectangular sections of channels. The rectangular channel has several disadvantages to generate drops [44,46], the fluid phase wets the upper and lower walls of the rectangular channel at a low flow rate of fluid in a continuous phase, which causes the attachments of cells to the channel walls, or even destroyed due to them: (i) surface shear force [36] and also due to (ii) capillary instability [10,47]. By increasing the flow of a continuous phase, the wetting surface of the disperse phase might be eliminated [47].

Using channels of circular section (or cylindrical geometry) instead of rectangular ones, the velocity profile can be generated evenly in the direction of the cross section, formation of stagnation areas at the corners is eliminated and a stable equilibrium position is established in the center of the channel, thus generating uniform, well controllable and monodisperse drops at a wide range of flow rates [44,47–49]. Another advantage of circular microchannels is that it can increase the efficiency of light transmission inside on-chip waveguides for light-sensing and light actuation methods implemented in lab-on-a-chip devices [44,50].

The surface chemistry of the microchannels is another important aspect to consider when making microchips. For example, due to its high hydrophobicity, PDMS absorbs certain organic solvents and hydrophobic analytes, causing contamination (fouling) of the inner surface channels [26]. Hydrophobic materials are difficult to fill (wet) with aqueous solutions and easily nucleate air bubbles (which makes it difficult to remove air bubbles from the channels), and consequently affect the quality of the material [2]. For this reason, to remove hydrophobicity it is necessary for PDMS to be treated with oxygen plasma to create hydrophilic PDMS surfaces by oxidation. After oxygen plasma treatment, it is necessary to maintain surfaces in contact with water or polar organic solvents [26] so as to avoid contact with the air. The best solution for surface modification is the treatment of surfaces with silanes or polyelectrolyte multilayers [27]. However, the direct contact with air results in surface rearrangements and the PDMS surface becomes hydrophobic in about 30 min [26,27].

#### 1.2.2. The Microwire-Molding Technique

Recognizing the benefits of circular microchannels, several research groups have been motivated to search for new techniques for making circular channels [44]. In this context, to improve the manufacturing techniques and to solve some of the problems or disadvantages that arise in the fabrication of microchannels, especially in the circular channels, researches developed novel, simple and fast fabrication methods. Based on pouring of liquid PDMS pre-polymer on microwires, this method is called suggestive microwire-molding [51–55]. The advantages of the microwire molding process are: rapidly and simply fabricate the circular-channels with perfect circular cross-sections in bulk PDMS and also the possibility to make different and/or complex topological shapes from straight channels to helical or curving micro-channel fabrication [51–55].

These benefits result from two main properties of PDMS: low surface energy and reversible swelling by solvents [27,52–55]. Thus, Verma et al. [55] developed for the first time a simple and cost effectiveness methods for generating straight (1D), cross-connectors (2D) and more complex (3D orientation structure) microchannels inside cross-linked PDMS blocks. They used nylon wires of varying diameters (50–250 μm) as a template to generate channels and to ease the nylon removal. Chloroform and triethylamine solvents were used for swelling the cross-linked PDMS Sylgard 184 Type, (Dow Corning, Wiesbaden, Germany) [51–55]. For PDMS mixtures, pre-polymer were used: curing agent for 10:1 (w/w) [51–53,55] and 9:1 (v/v) for [54] (Table 1).



In the same way, by embedding the microwires through molding process of PDMS mixture and then removal of microwires by swelling PDMS in different types of solvents, Jia et.al. [53], also made microchannels with different topological shapes (1D, 2D and 3D structures), in economic, flexible and convenient way in comparison with photolithography and conventional soft lithography [53]. To realize these channels, besides the casting and treatment parameters, of major importance are the preparations of microwires, substrates or supports (Table S1, Supplementary Materials).

For instance, at fabrication of straight channels [51–55], it is a very good stretching out the nylon/stainless steel/copper etc. microwires between rigid supports, also cleaning, drying of glass substrate/wires or degassing of PDMS mixture are required. For 2D micro-channel fabrication [53,55] it is necessary templates fabrication (2D mesh channels), formed by mechanical (hot) pressing/micro contacting and microspotting.

Thus more complex 3D structures [52–55], can be made by different techniques, such as (i) forming of a helix around a rigid cylindrical rod (Φ = 100–500 μm), heated (100 ◦C) to fix the helical form permanently [55], casting the PDMS mixture on 2-D nylon mesh template and curing mixture at 90 ◦C for 1h, or (ii) rolling up by winding around a rigid rod, casting PDMS mixture at 100 ◦C for 35 min [52], followed by microwire removal in different solvents mixture [52,53,55], or removal the microwire by heating and suction with vacuum pump [54]. By the technique with heating and suction removing the wires pump [54], a roughness inner surface it is obtained. Due to possible remaining of solder material after the wire removal process, additional coating treatments is necessary (i.e., wall coating of channels using PDMS solutions diluted with curing agent, in different proportions) and also corona discharge treatment to increase the hydrophilicity (for improving the fluidic flow and cell adhesion); The final step, the sterilization device by filling with 70% ethanol and washing with deionized water, make the device suitable for biological application [54].

#### *1.3. The Purpose Statements*

For the general purpose of generating different experiments focused on the biological environment, we present in this paper the design and fabrication by microwire molding a PDMS-based sensor with circular cross-section microchannel with a good surface quality, for the measurement of microfluidic (nanoliter to picoliter) volumes. To control the microfluidic transport at the nano-, pico- and femto-liter level, we used Microfluidic Pulse Width Modulation (MPWM) [56] system. We designed and fabricated the novel sensor especially to calibrate a microinjection MPWM system that will be used for very high precision intra-cellular insertions. The calibration of MPWM system it is necessary for reproducing with high precision and the detection and measurements of microfluidic volumes.

#### **2. Materials and Methods for PDMS –Based Sensor Fabrication**

#### *2.1. Selection of Materials*

Based on literature data [2,12,26,27,38,44,57] we selected a polydimethylsiloxane (PDMS) polymer (Sylgard 184 type) for the fabrication of a circular-cross section channel microfluidic device due to its remarkable characteristics such as: (i) excellent optical transparency, down to 280 nm [57] (ii) flexibility (i.e., it is a ductile material); (iii) elasticity (its elasticity can be "tuned" using cross-linking agents); (iv) biocompatibility (v) capacity to seals materials like glass, polystyrene and PMMA [57]; (vi) high thermal stability up to T = 300 ◦C [38]; (vii) permeability to gases (is more permeable to CO2 than to O2 or N2) [12] and also to water vapor; (viii) cost-effective production at microscale and (ix) does not require clean room environment [57]. To get a circular cross section channel we selected metallic microwires (enameled copper and steel) due the smooth surface and high flexibility of them. For creating different size of channels, we selected the following wire diameters: 63 μm and 120 μm for the enameled copper and 190 μm for the steel wire. The low surface energy and reversible swelling by solvents of the PDMS [26,27] allow the microwire removal with proper solvents. In this case, of our experiments, chloroform (with 1.39 swelling ratio) [27] were used for swelling the crosslinked PDMS. We made

ultrasonic cleaning of the mold and wires in isopropyl alcohol (commercially used for degreasing and cleaning electronic components). The ultrasonic cleaning was performed with a USC-TH type ultrasonic cleaner (VWR International, LLC) at 27 ◦C for 30 min.

#### *2.2. Fabrication Methods of PDMS Sensor*

For fabrication of PDMS based sensor it is necessary to generate straight microchannel inside PDMS block. The main steps include: (i) Preparation of mold support and microwires, (ii) Microwire fixing and alignment, (iii) PDMS Casting, Degassing and Curing, (iv) Demolding and Microwire removal, (v) Needles insertion and sealing by needles connection to the polyethylene tubes. Preparation of molds and devices before casting: The cylindrical polymer molds were punched diametral opus, with needles (300 μm in diameter). For fixing and aligning the wires there were custom made a custom stretching device with many working positions with screw (Figure 1) respectively with magnets for attachment of the wires were made. Every position is composed of two diametric disposed elements. The stretching was performed using a microscopic system by which a linear landmark was aligned with the stretched wire. There are two stretching devices: the first one was equipped with screws and locking nuts and the second one used permanent magnets for wires fixation. At the M4 × 1 screw, we used the nut & lock-nut system to lock the convenient position to ensure the linearity of the wire. The linearity was tested with a standard edge by parallelization with the stretched wire and microscopic observations. The permanent magnets have been chosen so that by application they will secure the firm lock of the pre-tensioned wire manually. The linearity was proved as in the case with a screw.

**Figure 1.** Customized stretching device with three working positions and the casted PDMS mixture inside of polymeric molds.

The degassed of PDMS mixture (removing of air bubbles) was made using a low level vacuum. After casting of mixture, the casted chip made of PDMS has a height of 2.5 mm, which allows a quick natural degassing, no more than 3 h. However, if a bubble remains stuck in the PDMS table, it is mechanically removed with a needle tip. All molded chips did not show any bubbles left after casting. The degassed PDMS mixture (components A to B = 10/1 w/w) was poured in the polymer molds (Figure 1). The PDMS mixture was cured at 80 ◦C for 2 h and then it was cooled slowly in a Venticell type heating oven (MMM Group, Planegg, Germany), with natural air circulation,. After degassing and curing, the demoulding step consist into gentle pulled out from the PDMS block of the microwires by swelling PDMS in chloroform, leading to the formation of microchannels in the PDMS block.

The characterization of obtained microchannels using microwire-molding technique were done from microstructural point of view using different types of microscopes: (i) 1000 × 8 LED Zoom Digital USB Handheld Microscope PC Endoscope Camera Test TE306 (AmScope, Irvine, CA USA) and (ii) 40× – 800× Inverted Tissue Culture Trinocular Microscope with different video camera (custom and AmScope FMA type, Irvine, CA USA).

#### **3. Volumes Detection Method**

#### *3.1. The Principle of Volume Detection Using PDMS-Based Sensor*

By applying a constant pressure in microchannel, the flow of the liquid is defined by the microchannel geometry due to the common laws of the laminar flow [58]. The general characteristics of the liquids (incompressibility, fill and copy the shape of the vessel) [1,2], allow that the volumetric measurement of a fluid flow along the channel (with constant circular section) to be reduced to the measurement of the distance traveled between two given impulses. In this experimental work, impulses are given by a system called microfluidic pulse width modulation (MPWM). Based on a concept from the electrical engineering field, the MPWM system controls the transport of fluids through the microfluidic channels. Ardeleanu et al. [56] used MPWM to measure volumes/flow rates for future application in cell/particle trapping and release. Ainla et al. [59] used pulse width modulation (PWM) for microfluidic diluter and Unger et al. used PWM for dynamic duty cycle over time for liquid flows [60]. The experiments are performed with very small volumes of liquid using microchannels obtained by microwire molding [51–55] using wires of 63, 120 and 190 μm in diameter.

For precise flow control, many methods including optical volumes and flow rates measurements were developed [56,58]. An optical volumetric measurement technique has been developed to enhance the measurement in the picoliters region. Usually, a microscope with 500x magnification is used when working with picoliter volumes. The visualization of the capillary requires a very good clarity to determine the distinctive front of the transported liquid. For this reason, we used a microchannel with 120 μm in diameter, and a good magnification to have a clear image of the boundary between two immiscible fluids like air-liquid and/or liquid1-liquid2. The computerized generator of absolute coordinates needs to have a clear edge for internal accurate measurements [56,58]. Thus, in this volume determination method, the image of this boundary is very important. This generator is in fact a complex microscopic system that calculates automatically the current absolute coordinates of the mouse cursor (pointer) within the microscopic field of view. Summarizing all the above considerations, the principle of volumes detection and measurement mainly involve three elements: 1) a cylindrical capillary; 2) a fluid boundary as detectable tracking element; 3) an absolute coordinate's generator in microscopic field of view. The capillary must be fabricated based on PDMS molding techniques [51–55] and it must have finally the imposed cylindrical geometry with the precise dimensions determined using microscopic tools [56,58].

The PDMS chip element (Figure 2a) sustains the spatiality of the microchannel (capillary) creating the mechanical premise for the microinjector connector and inverse microscopic visualization (Figure 2d). The tracked fluid boundaries are air-liquid (Figure 2b) and the visual volume detection is obtained by computing two successive boundary positions (Figure 2c). The liquid is a commercial red color ink.

The PDMS-based sensor chip is connected to a microinjector system (Figure 3a). This microinjector system is a MPWM equipment type that will be used to circulate the working fluid using the pulse modulation principle (Figure 3b). The generation of fluid volumes is quantified by controlling the closing and opening times of the valve. When the valve is opened (ON position), the pressure difference between the two chambers connected by the valve will allow a precise volume of fluid to flow. When the valve is closed (OFF position), the fluid circulation is blocked. The time when a sequence of closed/open states occurs is called a period. The ratio between ON time and period is called duty cycle. The duty cycle is the basic parameter of the volume generation from the experiment, keeping the pressures and the MPWM signal frequency constant. Figure 3c shows the schematic Digital Tracking Coordinates System (DTCS) of Celteh Company (Targoviste, 130092, Romania), which allows the determination of absolute coordinates in the videomicroscope active screen. The DTCS or also called bi-dimensional micrometric terminal blocks (BMS) [61], is presented in detail in the paper of by Ardeleanu et al. [61].

**Figure 2.** PDMS-based sensor: (**a**) Sensor-chip and (**b**) Boundary air-liquid; (**c**) Illustration of two distinct positions of successive boundaries accounted in the measurement of the microfluidic volume; (**d**) Inverted microscope field of view for PDMS-based sensor.

**Figure 3.** Experimental set up and measurements: (**a**) Schematic representation of the experimental set up, (**b**) MPWM principle, (**c**) The measurement performed by taking into account the advancement of the liquid boundary with the length (L) (the cross-sectional area of the channel is constant).

The mouse pointer is the object to which DTCS interactively assigns absolute coordinates. With the "double-click" command, the current position of the mouse pointer is registered by the DTCS in a database with a certain order number. The recorded coordinates are also visible on the interface, the operator being able to use them for the purpose of determinations and calculations.

The minimum transported volume by microinjector is in the femtoliters range. In the field of view of the inverted microscope, the air-liquid boundary will be detected using DTCS (Figure 3a). Figure 3b shows the fluid advance from the initial moment at different times, t, represented by closing/opening (on/off) of the chamber with valves, controlled by the MPWM signal. The fluid advance after each MPWM signal is thus controlled and precisely determined by the advancement of very small volumes of liquid from the microfluidic level. A volume determination involves two distinct positions of the fluid boundary at two distinct moments (Figure 3b). The absolute coordinate generator has the ability to determine with high precision of less than 1 μm the current mouse cursor position. A simple click allows the position data to be recorded, in absolute coordinates of x and y. A single (pL) volume measurement involves a succession of two distinct boundary positions and two distinct position data records. The researcher uses these data sets to determine the length L of injected fluid volume by one MPWM impulse that generates the fluid movement, as shown in Figure 3b. The liquid volume determination consists in using the Equation (1):

$$V = L \cdot \frac{\pi \cdot D^2}{4} \tag{1}$$

For every injected volume in capillary, the database will be populated with distinct coordinates of the position of the fluid boundary. The capillary is aligned with one of the two coordinates, which is *x*-axis in this case. The *x* becomes the main coordinate for calculating the length L of an injected volume, calculated by subtracting two successive values from the database, i.e., *x*k-1 and *x*k.

#### *3.2. Method of Experimental Measurements of Picoliter Volumes*

The measurement of the movement of the boundaries of liquid during fluid movements was done by two methods: the first one is the current coordinate mouse registration "mouse head tracking method" (MHT) and the second one is the "pointing method" (P). In the MHT method, the current coordinates of the mouse pointer are displayed on the interface at the associated positions on the screen. The tip of the mouse is a visible landmark that offers precision due to its small size relative to the tracked object. Figure 4 illustrates the MHT method, which was also used for MPWM calibration.

In Figure 4a,b, the mouse pointer coordinates are shown as pointed on the interface with xm and ym. These xm and ym values change interactively with the change of the position of the mouse, which allows the tracking of elements of interest on the screen of the video-microscope.

After the calibration was performed with the MHT method, the second method, P, was used to test the fluid volume generation. For a series of successive steps of injected volume with picoliter amounts, we use the P method presented in Figure 5. We dimensioned the Pointer Width (diameter) (PW) bigger than the determined Boundary Width (BW) (Figure 5a).

In this way, the operator will position the point symmetric related to the liquid boundary in the same horizontal line (Figure 5a). The precision of measurements is determined by the operator's scoring precision (human factor) (Figure 5b).

For calibration, we used a magnification of 800× for visualization, because the volumes of fluid passing through microchannels were of the order of 25 - 50 pL, i.e., 1 MPWM step, and this condition assumes that the distance between two successive boundaries was very small. The MHT method is appropriate in this case, because the mouse size on the working screen is quite large in relation to the thickness of the boundary and allows a good positioning of the mouse tip relative to the center of the boundary, which is visually approximated by the human operator.

**Figure 4.** Representation of the MHT method: (**a**) microscope image capture of the first position of the mouse pointer, (**b**) microscope image capture of the second position of the mouse pointer, (**c**) schematic representation of the MHT method, applied to the liquid boundary measurement on a PDMS-based sensor

**Figure 5.** Microscope image captures of (**a**) method P for visual volumes detection (the insert shows the magnified pointer point and boundary); (**b**) Successive pointing positions.

For measurements, the amounts of fluid passing through microchannels are large, usually tens of MPWM steps, which means that the volumes are of the order of hundreds and thousands of pL. To include two successive boundaries on the screen, a smaller magnification (i.e., >500×) is required. In this case, the thickness of boundary is defined with fewer pixels, which leads to an increase in the uncertainty of the position of the mouse tip in the center of boundary. Therefore, a point in the shape of a red circle (marker), with a diameter slightly larger than the boundary thickness has been proposed. This marker can be identified very well if it has been positioned by the operator in the center of boundary. The tip of the mouse becomes inaccurate as a reference to boundaries, therefore the

geometry of the reference object is increased, i.e., a circular point with a diameter slightly larger than the thickness of boundaries. In this way, the thickness of the boundary is positioned relative to the point, becoming very visible the symmetry of crossing the boundary through the middle of the point (red circle).

To measure, the operator uses the resulted coordinates from the interface and processes these data using Equations (2) and (3). The Equation (2) is the theoretical base for Equation (3), when we took in consideration the calibration factor *ucalibration* = 44, which results from Figure 6.

**Figure 6.** Validation of measurement precision of the distance between two successive gradation lines on the calibration scale representing 10 μm, taking into account the value of the calibration factor of 44.

After each MPWM executed step (impulse), a new boundary position is formed in the microscope field of view. The operator must point every new position boundary that generates automatically the *X* and *Y* absolute coordinates. The microchannel is aligned with the x-axis and so it will be used to calculate only the x-coordinate from the DTCS interface table. The volume depends on the *xk* and *xk*–1 coordinates, which are the only variables in Equation (2) and, taking into account the following constants: *D*- microchannel diameter and *ucalibration* – calibration DTCS factor. The volume measured at "Point *k"* is given by Equation (2):

$$V\_k = \frac{\pi \cdot D^2}{4} \frac{x\_k - x\_{k-1}}{u\_{\text{calibration}}} 10^{-3} \text{ [pL]} \tag{2}$$

where: *Vk* is the volume measured at "Point k" [pL], *xk*–1 is the absolute coordinate at "Point (k-1)" [a.u.]; *xk* is the absolute coordinate at "Point k" [a.u.]; *D* is the microchannel diameter [μm]; *ucalibration* is the calibration. Equation (2) is obtained from the volume calculation of a cylinder with the surface of a circle of diameter D and the length deduced from the absolute coordinates xk–1 and *xk*.

In our case, for a channel with *D* = 120 μm and *ucalibration* = 44, the Equation (2) becomes:

$$V\_k = 0.257039 \cdot (\mathbf{x}\_k - \mathbf{x}\_{k-1}) \text{ [pL]} \tag{3}$$

The displacement of the liquid through the channel appears when a pressure difference is applied between the two chambers bounded by the liquid-air front. The channel has a very small section compared to the rest of the liquid path (the internal tubes of the MPWM system), which will produce a hydraulic resistance behavior. The pressures used must be appropriate to the geometry of the microchannel. The applied pressures used in these experiments were in the range of hundreds of millibars.

The volume determination will be performed automatically in the near future, through the image processing boundary recognition. At the moment, the measurement process is manual, the human factor being "the trigger" of coordinate recording, which leads to the determination of the length (L) and volume calculation. In order to generate picoliter volumes, the MPWM system must be calibrated at this working level.

#### **4. Calibration of MPWM for Picolitre Volumes**

The MPWM device is a complex technical system [56] that was built specifically for microfluidic transportation and measurements. An on/off valve is actuated to allow an energetic transfer of fluid between a pressurized and a depressurized space. For calibration, the pressures, frequency of the command signal and duty cycle are combined to quantify the fluidic volumes that pass through the on/off valve. Each volume generated by the MPWM system represents the quantity and the force that has an impact on the fluidic environment. In bioprocess microfluidics, extracting an adherent cell from a Petri dish requires a force applied for a short period of time, which means for MPWM a great negative pressure, high frequency and medium to high duty cycle. To transport a cell after extraction to a remote microfluidic room, speed and precision is required, this means for MPWM high pressure, high frequency and small to medium duty cycle. In this way, the MPWM signal can be tailored on specific application. For this experimental work, the final goal is to obtain a very high resolution for the 120 μm channel and a very high precision of the transported fluidic volumes. To detect the smallest volumes as possible, we used the PDMS-sensor along with the MPWM system.

On this occasion, we also tested the working capacity of the MPWM device, dimensioning the volumes generated by the MPWM system. For this purpose, we used small pressure levels, medium frequencies, and small duty cycles factor. The MPWM system presents a high repeatability of the generated volumes. The PDMS-based sensor allows to precise measuring the volumes generated by the MPWM device and storing the information along with the associated values of the input parameters. The correlations between causes (pressures, frequencies and duty cycles) and effects (generated volumes) will be the calibration process and will be stored as a map. We used in the calibration process a frequency of 100 Hz, which is one previously used for cell extraction and membrane cell elasticity testing [62].

Tables S2 and S3 (Supplementary Materials) present experimental data for the calibration in the following conditions: a constant pressure of 256 mbar, frequency of 100 Hz and duty cycle range [13.5%, 16.8%]. These tables include the *x*-coordinates measured with the DTCS system and the calculated using Equation (2) and corresponding to the determined volumes. These data allow the MPWM user to determine the calibration function. This function allows us to calculate the appropriate duty cycle parameter for the transport of certain required volumes (imposed).

For data processing, the function of approximating the correlation between the generated volume size and the duty cycle was sought. For this, the average value of the volume generated by the MPWM system was calculated for each duty cycle tested. For the average values resulting from each filling factor tested, the calibration MPWM function was obtained as shown in Figure 7.

**Figure 7.** Chart of the MPWM experimental data calibration.

The use of this function must also take into account the deviation of measurements corresponding to each average value, deviation of measurements determined by reporting the average value at fitted values (those obtained by applying linear regression). By processing the MPWM calibration data, for each duty cycle, a series of average values of the volumes generated were obtained, which are presented in Table 2. In addition, the deviations corresponding to each average value are shown. Considering the average value as a reference, the relative deviation was calculated as the maximum fluctuation relative to the average value. The calibration function (Figure 7) is a mathematical tool for the human operator to choose the optimum MPWM parameters into a given application (transportation small/tiny volumes).


**Table 2.** A series of average values of the volumes generated by processing the MPWM calibration data, for each duty cycle.

#### *4.1. Experimental Results*

#### 4.1.1. Microscopic Analyses of Microchannels

Figure 8 presents the images of the circular cross-sectional area of channels obtained after 120 μm copper wires removal. A smooth surface of microchannel was observed, which allows an uniform and well controllable flow of the fluid.

The microwire-molding techniques allow fabricating rounded channel shapes in PDMS. These fabricated channels have a circular cross section (Figure 8), that allow a very good visualization in the analyzed field, due to the efficiency of light transmission inside on-chip waveguides for light-sensing and light actuation methods.

**Figure 8.** The circular cross-sectional area of microchannels produced by microwire molding technique, after 120 μm copper removal, at different magnifications (80 ×, 200 ×, 500 ×, 800 ×).

By extracting the wire under the conditions of swelling of PDMS, the longitudinal micro-irregularities have been created that do not affect the laminar flow of the fluid, but for the precise determination of the maximum error affecting the calculation of the fluid volume, the variation of the channel section due to the roughness can be taken into account. The surface irregularities affect in a small proportion (maximum 3.8%) only the area of the channel section at the diameter level (Figure 8).

The experiments were performed by researchers from Valahia University of Targoviste in collaboration with Celteh Mezotronic. An automatic microinjection system was adapted to existing MPWM equipment, by integrating a sensor based on image processing of fluid flow through the microchannel realized with the technique known as microwire-molding. The image processing software was developed by Ardeleanu in collaboration with Company Celteh and is an innovation in the field of microscopic mechanical displacement determination with submicrometric precision.

The validation of the idea that the sensor allows the measurement of microfluidic volumes is the subject of this work.

The precision of the equipment for determining the absolute dimension is below 1 μm. The display in the interface (Figure 9) means the absolute dimensions of the mouse pointer in the field of view of the microscope, expressed in virtual numerical units.

**Figure 9.** *Cont*.

**Figure 9.** Calibration method of the DTCS system: (**a**) screen-print showing 11 red points (virtual units) representing 10 μm in length, (**b**) screen-print showing the measurement of the cannel diameter of 120 μm by DTCS.

The measurements required for the experiment represent specific distances within the field of vision, i.e., the distance between two liquid fronts generated by the MPWM signal. Figure 9 shows, for example, the calibration mode of the DTCS system. After the calibration it was found that: 11 vital points (units) represent 10 μm (Figure 9a) and thus, with this calibrated system we were able to determine the diameter of the microchannel (Figure 9b). The DTCS system can associate a number of arbitrary units to a real calibration distance (μm) so that distances of less than 1 μm can be measured with very precise approximations.

In Figure 9a calibration of 11 virtual units at a real distance of 10 μm is rendered by a specially graded ruler for this purpose. The meaning of the two red dots (1 and 2) from Figure 9b represents the positions of reaching the extremities of the visualized channel for measurement. The clarity of the image allows a precise reproduction of these edges, which leads to a precise diameter calculation. In order to determine the diameter of the microfluidic channel, a calibration of the DTCS with 44 virtual numerical units for a real distance of 10 μm was used. As can be seen in the image of the DTCS interface in Figure 9, the two reference points recorded by the human operator with the mouse on the screen, have the absolute *x* coordinates of x1 = 56.684 and x2 = 51.404. The difference between the two coordinates is the diameter of the channel expressed in virtual numerical units, i.e., 5280.

#### *4.2. Results of Experimental Tests for Microfluidic Transportation*

The microfluidic transportation was tested to determine the deviation of measurements obtained when small volumes were generated by the MPWM system, using the calibration function obtained previously. Using a frequency of 100 Hz, a pressure of 256 mbar and a duty cycle of 14.5%, a fixed number of MPWM steps were generated in order to obtain a given total volume. This volume is obtained by multiplying the volume corresponding to a single MPWM step by the number of steps required.

To determine the deviation of measurements due to execution, we repeated 10 times each test. Thus, the repeatability of the volumes generated by the MPWM system was obtained for a given parametric conditions. According to the calibration function, the volume for a single step under the given conditions is 67.5 pL. The four experimental tests corresponding to 5, 10, 15 and 20 MPWM steps have been generated. Tables 3 and 4 shows the values of volumes measured in these experiments and the associated deviation of measurements (dev. of meas.) with respect to the reference volume calculated with the calibration function.

**Table 3.** MPWM repeatability experimental tests performed for 5 and 10 steps in the following conditions: frequency of 100 Hz, pressure of 256 mbar and duty cycle of 14.5%.


**Table 4.** MPWM repeatability experimental tests performed for 15 and 20 steps in the following conditions: frequency of 100 Hz, pressure of 256 mbar and duty cycle of 14.5%.


The next parameter of interest is the standard deviation. In Figure 10, the values of the standard deviation for each of the four volumes tested under the same conditions 10 times are shown.

**Figure 10.** The values of the standard deviation for the four expected volumes: (**a**) for 5 and 10 steps cases; (**b**) for 15 and 20 steps cases.

#### **5. Conclusions**

A PDMS-based sensor with circular cross section microchannel was obtained experimentally through the microwire molding technique. PDMS-based sensor with microchannels of 63, 120 and 190 μm were obtained and analyzed to check the quality and the micrometric sizes of the channels. The PDMS-based sensor was integrated with the MPWM system and the methodology to measure picoliter volumes was developed. We have shown that the set-up proposed in this paper was able to verify the microinjection-sensor conceptualization. The measuring process was based on absolute coordinates in the microscopic field of view with high magnification factors. The DTCS system works based on image processing, mainly detecting the mouse pointer coordinates. The human operator had the possibility to pointing with the mouse pointer, into a very precise mode, different positions of fluid boundary as main trackable element from the micro-channel, watching it on the video-microscope screen during experiments. For microfluidic volumes circulation a very high precise generator was used.

The sensor was specially designed to allow the calibration of the MPWM-injector. MPWM is based on a PWM signal that is, in fact, a "train" of identical pulses. One single pulse is a passed volume through the on/off valve. Certain applications request a given precise fluid quantity to be transported, i.e., an MPWM fluidic signal with a given number of pulses. Once the unique relation between a volume (value) and a pulse (pressure, frequency, duty cycle) is established, it only remains to generate a precise PWM signal with a precise number of pulses. In addition to the calibration of the device, the sensor serves the MPWM system to close the control loop in the microinjection process. This is why this novel PDMS-based sensor system for MPWM measurements of microfluidic (nanoliter to picoliter) volumes contributes to the advancement of intelligent control methods and techniques, and could lead to new developments in the design, control, and in applications of real-time intelligent sensor system control.

Following the experiments, the MPWM calibration function was determined, using the relationship between the microfluidic (nanoliter to picoliter) volumes and the duty cycle of the system, thus obtaining the MPWM experimental calibration diagram. Following the calibration, at 100 Hz, 14.5 duty factor and 256 mbar the microfluidic transport and the resulting picoliter volumes were tested using the pointing method. The results of the experiments confirm the quality and precision of the measurement at picoliter volumes, making it possible, in future system developments, to be applied in high-end research fields such as the automatic microinjection of biological cells.

#### **Supplementary Materials:** The following are available online at http://www.mdpi.com/1424-8220/19/22/4886/s1.

**Author Contributions:** Conceptualization, M.N.A., R.V. and I.N.P.; Methodology, M.N.A., R.V., I.N.P. and S.M.; Software, M.N.A. and E.L.; Hardware, I.N.U. and E.M.D.; Validation, M.N.A.; Formal Analysis, M.N.A., B.A. and I.N.P.; Investigation, M.N.A. and I.N.P.; Resources, M.N.A.; Data Curation, M.N.A.; Writing—Original Draft Preparation, I.N.P.; Writing—Review and Editing, R.V., B.A. and I.N.P.; Visualization, M.N.A.; Supervision, R.V.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank all members of CELTEH MEZOTRONIC for providing the MPWM system. We also thank the management and staff of the Scientific and Technological Multidisciplinary Research Institute (ICSTM-UVT), Valahia University of Targoviste, for their collaboration and access to their laboratories.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Review*
