Non-contact measurement techniques to study the microsized particle adhesion phenomenon

Airborne microsized particles are generated by artificial activity and natural sources. Due to their characteristics, these airborne particles are highly volatile and can cover thousands of kilometers according to weather conditions (in particular, wind intensity and humidity). As a result, these microsized particles could contaminate engineering systems determining their operation and performance modification. Phenomena such as photovoltaic panel soiling or gas turbine fouling are two of the most detrimental effects generated by the interaction between airborne particles and the relevant surfaces of the system. The present work proposes a set of non-contact measurement methodologies to study the adhesion phenomenon. The non-contact techniques are devoted to studying the deposited layer without altering the deposits. Image and video analysis have been used to show how the deposit can be studied in detail. Stationary and rotating facilities have been employed to show the applicability and the related constraints to the non-contact measurement techniques.


Introduction
The reliability of engineering systems is often related to environmental operating conditions.Beyond the temperature, which is the most controlled variable to ensure the proper operation of each system, the presence of airborne solid particles could determine the modification of system performance.Airborne solid particles are generated by human and natural activities.Usually, these contaminants are microsized and composed of soil and soot particles.Due to their low-density values and diameter, they are highly volatile, and the dispersion action carried out by the wind determines that the interaction could be variable according to the operating site and weather conditions.The deposition of natural particles (soil and dust) on solar panels [1 -5], the deposition of particles on the relevant surfaces of gas turbines (compressor and turbine fouling ) [6 -8] and heat exchangers [9 -11], are the drivers for the losses in terms of efficiency in several applications.The analysis and the capability to predict the effects of this dangerous interaction can effectively increase efficiency and reduce operating costs.The contamination of the surfaces can evolve over time [12].The thickness of the deposited layer varies according to the operating (exposure) time and the adhesion characteristics of the particles [13].
To test and evaluate the interaction between relevant surfaces of energy systems, it is necessary to reproduce the contamination condition and adopt measurement systems that do not alter the deposited layer to reconstruct the root-cause effects of the surface contamination.The adherent layer modifies body shape and surface characteristics, and for these reasons, a non-contact measurement technique helps to detect the contamination results that could be unable to support strength or invasive measurement techniques.The modification of the structure and/or the surface behavior leads to a modification of the result of the deposition process [14 -16].
The challenges related to the analysis of a microparticle deposited layer are due to the accuracy of the detections, which are employed to (i) discover features with a micro-size scale (in the same order of magnitude as the particles), (ii) highlight the differences of the deposition process due to different impact dynamic (velocity or angle) and environmental condition (such as humidity), and (ii) find out the overtime modification of the deposited layer.In light of this statement, in the present work, non-contact measurement techniques have been proposed to detect the effects of soiling and fouling on relevant surfaces.Image detection techniques have been tested on two different test benches.A stationary facility on which a simplified target has been used to discover the capability of the non-contact method to detect essential features during and after the contamination.At the same time, a rotating facility has been considered to show the potential application of the non-contact technique to study the modification introduced by the interaction between a rotating machine and microsized contaminants.

Materials and methods
To evaluate the measurement techniques, two experimental test benches have been used.The common basis of both experimental setups is the dosing system of solid particles.The tests consist of the contamination in a specific manner (with a particular amount of microsized particles) and a fixed airflow that aims to interact with the relevant surface under study.The dosing system is based on the particle feeder TOPAS SAG40U.This system can prepare and inject the solid particle by a Venturi nozzle system.To control the feeding rate, a variable-speed rotating ring is used.The complete description of the system is reported in [17 -19].Finally, to increase the reliability of the comparison between the results, the deposition is carried out with a particle feeding rate of 6 mg/m 3 after the drying process of the powder (12 h at 70 °C).

Stationary facility
The stationary facility was used to investigate the measurement capability of non-contact techniques in the presence of soil contaminants.The Arizona Road Dust (ARD) powder was used with a particle diameter characterization reported in Fig. 1, characterized by a mean diameter of 1.3 μm.This powder comprises silica dioxide and alumina, with a density value of 2717 kg/m 3 .Figure 1 reports an SEM image that shows the grain shape characterized by sharp edges.Details about powder characterization are reported in [15].In [19] are reported the analyses of the high propensity of this powder to interact with smooth and rough surfaces as those found for soot particles.The stationary facility consists of a wind tunnel (square section) that guarantees uniform contamination of the airflow (see Fig. 1).To control the air contamination, the inlet section was equipped with a filtration system.An inverter drove the electric fan to reach and maintain the airflow velocity (40 m/s).The target was held in the center of the section, using two dummy targets to ensure the proper impact and symmetric conditions.As a result, the target surface experienced a thin soil layer adhesion that evolves according to local and global phenomena over time.

Rotating facility
The rotating facility was used to investigate the measurement capability of non-contact techniques in the presence of soot contaminants.The Carbon black powder was used with a particle diameter characterization reported in Fig. 2, characterized by a mean diameter of 1.9 μm.Soot is a paracrystalline carbon derived from the incomplete combustion of a hydrocarbon.As can be seen, the average diameter and the distribution are similar to that reported for the soil powder to make a helpful comparison over the measurement techniques (more details can be found in [15]).The rotating facility comprises an engine test cell equipped with the Allison 250 C18.The compressor units have six axial stages and a centrifugal stage.The pressure ratio is 6.2, and the designed mass flow rate is 1.36 kg/s obtained at the design speed of 51,600 rpm.More details about the rotating facility can be found in [17,18].The compressor unit has been kept at a constant rotating velocity of 20,000 rpm, with a mass flow rate of 0.33 kg/s for 30 min (fixed exposure time).

Experimental investigations
The present section reports the experimental test carried out to explore the capability of non-contact measurement techniques to detect characteristics of the deposited layer.

Image techniques on a simplified specimen
The non-contact technique is based on image analysis.The methods have been proven useful in detecting modification of the surface characteristics and shape [20].Deposited thickness or surface roughness appears challenging to be assessed in a preliminary analysis made with the image techniques.However, starting with the detected images, it is possible to quantify the covered area according to the test condition, and it could be helpful to estimate the reduction of the efficiency for solar panels [1]) or, in turbomachinery, the increment in fluid dynamic losses, which reduce the power and the machine efficiency [21].Therefore, by estimating the fouled (i.e., degraded) area by image detection, it is possible to assess the intensity of the contamination.The procedure compares pictures taken before and after the impact (i.e., deposition) test using a fixed setup of lights, camera, and specimen position.The imageprocessing package FIJI was used to post-process the images [22].As reported in [17,23], a routine was implemented to calculate the difference between the clean and the fouled pictures, obtaining a binary image.These three passages are reported in Figure 3.The greyscale image indicates the differences between the clean picture and the fouled one.After the greyscale conversion, the converted picture can be used to analyze the frequency distribution.A frequency distribution chart can be obtained by assigning a value from 0 to 255 to each pixel according to its color intensity.In Figure 3, the chart shows on the yaxis the number of pixels and on the x-axis the pixel intensity.When the distribution is closer to the white color, the condition is cleaner than when the distribution is closer to the black part of the chart.For a greater number of black pixels, the condition is even far from the clean condition.Starting from this two-dimensional information (detected pictures), it is possible to reconstruct a three-  This technique obtains three-dimensional information [24] starting from two-dimensional data.However, it depends on several mathematical and optical assumptions [25 -27] that generate uncertainty and misinterpret the results.As this strategy recognizes the former shape from the shade, the basic assumption is related to the light intensity and position, which could determine several inaccuracies during the shape reconstruction.Figure 4 shows a three-dimensional shape of the deposited layer obtained by the SFS post-process applied to the result reported in Fig. 3 (greyscale figure).The procedure was carried out with the software package Fiji developed by Sage et al. [28].Starting from the proper region of interest (ROI), the spatial shape of the powder layer can be determined.Scaling the dimension using reference length makes it possible to assign the scale for the third dimension (i.e., the height of the deposited layer) reported as a colorbar in Fig. 4. Using this technique, coupled with the proper light and camera setup, the former twodimensional detection can be interpreted as a three-dimensional data source for shape and morphology estimations (e.g., surface roughness, layer cracks and instabilities [15]).

Video detection on a simplified specimen
Since the image analysis of clean and fouled substrate is useful only for detecting the final fashion of the deposits, the image analysis can be used according to video detection (over time analysis).The overtime detection could increase the difficulties in the test setup, but according to the literature [13], video detection can be used to recognize the sticking and detachment (involving small or wide portions) processes characterized by stochastic but recursive actions.Depending on the camera resolution, the detection could recognize the first or the second type of phenomenon [14].Figure 5 reports the overtime detections.Several frames are reported to show the modification of the deposited layer in terms of fouled and detached regions.Figure 6 shows the detachment effects estimated by considering the footprint area.Using a video frame reported in Fig. 5 and Fig. 6, it is possible to recognize two regions where the layer was detached.Starting from Fig. 5, the picture was projected according to the camera point view using the tool software package Fiji in order to realize the actual footprint view of the specimen.By this technique, the region that experienced detachment are estimated to equal 1 % and 2 % of the substrate area.

Image technique onboard
A similar strategy reported in the previous section has been applied to discover the capability of the noncontact measurement techniques in the onboard investigation of the deposition phenomenon.In the present section, the rotating facility is used to carry out deposition tests using image analysis to discover the deposition process features.The onboard setup imposes constraints and solutions for the camera and light setup, determining some inaccuracies listed in this section.
According to section 2.2 Rotating facility description, the experimental procedure consists of several recursive steps.
To generate a repeatable condition, the cleanliness of the compressor was restored by dissembling the compressor case.The clean procedure involved cleaner, demineralized water, and brushing operation to remove any deposits and particles.Clean pictures were taken for stator vanes and rotor blades at this stage.Therefore, the compressor unit was reassembled, and the start-up phase was performed.At the desired operating point (20,000 rpm), a proper interval was considered to allow the thermal equilibrium of the unit.After that, the contamination injection was activated for the entire test duration (30 min).
During the exposure time (30 min), the particle flow rate injected is kept constant.Finally, the axial part is disassembled at the end of each test, and pictures of the fouled parts are taken.The compressor case (stator vane) was positioned in a tailored holder (the same procedure adopted for taking the clean picture) while the rotor blades were detected directly on the compressor unit installation frame.In Fig. 7, four pictures that describe the position of the light and camera for both detections are reported.The white light (4000 K) setup is adopted for detecting the compressor flow path using a 3264 x 2448 pixels resolution.
Starting from the detection of the clean and fouled images, a similar post-process proposed for the stationary facility (i.e., simplified target) has been proposed.Figure 8 shows the subtraction process results between clean and fouled pictures.The clean and the fouled images differ as a function of the soot deposits.Soot particles are black, determining darker pictures (in contrast with the detections of soil contamination).In the present setup, surface finishing plays an important role.The pictures show that the polished rotor surface determined a light reflection that could mask the deposits, especially in  correspondence with the rotor hub.The light position and the characteristics of the devices represent one of the significant differences between onboard and lab-scale detection.In that case, the non-contact measurement technique has to be matched with the impossibility of modifying the characteristics of the relevant surface.A step forward in this sense can be made by increasing the complexity of the image post-process.As earlier reported, the pictures generated by the subtraction process were first converted into greyscale images and then converted into black-and-white images by assigning a threshold value.This procedure ensures an easier and more direct evaluation of the pixel status but increases inaccuracy due to adopting an arbitrary threshold [29,30].
A more accurate post-process can be done based on the extract the frequency distribution from each rotor and stator surface after the greyscale conversion, similar to that described in the previous analysis.
The reference images are shown in Fig. 9, where the result of the subtraction process is reported in the greyscale strategy.Thus, considering the frequency distribution histogram (0-255 pixel light), it introduces a more accurate method for characterizing the surface condition (not related to the choice of a threshold value) based on a statistical approach.Figure 10 reports the conceptual scheme.The first stage consists of the modelization of the discrete distribution into a Gaussian distribution.Concerning this distribution, the mean frequency and the standard deviation can be calculated to evaluate i) the fouled portion of the surface area of each blade and vane and ii) how much the deposition process was influential in the alteration of surface conditions (directly related to the frequency values).
To post-process the detections, the Kolmogorov-Smirnov (KS) distance is defined.This quantity quantifies the distance between two probability distributions (dKS) and is generally applied to cumulative distribution functions (CDF): where NCDF,1; NCDF,2 are the values of the CDF of the pixel number for a given frequency f obtained, for example, from two contamination processes.The second one is the difference between the average frequencies (Δμ).This parameter is defined as the difference between the average frequencies of the distributions for a specified condition: where μ1 and μ2 are the mean frequencies of the considered distributions.Unlike the KS distance, which is always a positive value (according to its definition), the difference between the mean frequencies can be positive or negative.If Δμ is more significant than zero, the surface conditions after the operation are, on average, lighter and, thus, cleaner than the fouled one.While if Δμ is negative, the analysis results in surface degradation of the surfaces.Concerning this calculation, image detection can generate a qualitative comparison between two fouled conditions, for example, by defining and studying how the deposition process occurs [30].Furthermore, the current approach is a non-contact method, which can be carried out quickly and with low costs.Although, the obtainment of quantitative information (i.e., the mass of deposits on the airfoils) has to be done employing intrusive methods, which lead to the disruption of the deposited layer and require more time [31].
The last consideration about onboard detection is the possible sources of uncertainty due to installation.
Vibrations and thermal effects can misalign the position of the relevant surface between the detection of clean and fouled conditions.In the present investigation, the onboard detection (especially for the rotor blade) implies several difficulties related to correctly detecting the flow path, mainly due to the thermal expansion of the compressor unit.This effect could introduce noise and inaccuracy, especially in the boundaries of the blade, during the onboard detections.
Figure 11 shows the result of the subtraction of the pictures taken after one hour and three hours and the detection of the fouled condition after five hours.Comparing these detections, the effects of the thermal expansion, evaluated by the thickness of the black lines, appear reduced over time.The subtraction between the fouled (after five hours) and the overtime fouled picture of the rotor show a suitable condition to detect the fouled condition after the fouling test.However, from the image analysis, the rotor deposit patterns seem affected by 0.5 mm (at maximum) of shift between the fouled picture and the cleaned one due to the uncertainty source with the onboard detection setup.

Perspectives
The non-contact measurement techniques have been proven to be helpful in estimating the contamination of surfaces.Several non-contact methods are available for detecting the surface characteristics or the internal features of the deposited layer [14 -16], but in the present investigation, the aim is to show the detection in actual experimental installations.The strength of the deposited layer and the need to detect microsized modification of the surface characteristics imply the use of non-contact techniques with the precise control of light and camera setup.The proposed strategies have been applied to two different types of investigation.A lab-scale test with a stationary surface and an actual machine with rotating surfaces.In this sense, constraints, and limits must be included in the measurement process capability and uncertainty due to the installation and device operation.To set up the proper detection strategy, accessibility, geometrical constraints, and repeatability (over time) of the camera and light setup have to be checked and managed carefully.The lab-scale testing can set up a detection and post-process methodology useful for on-field detections.The imaging technique can be used during the machine/system overhaul or inspections to estimate the deposition's intensity.Firstly, the operator can take the picture in the as-is (fouled) condition, and after the cleaning/restoring processes, the image in the restored condition.

Conclusions
In this work, non-contact measurement techniques based on image analysis have been proposed

Figure 1 .
Figure 1.Stationary facility and the soil dust characterization

Figure 2 .
Figure 2. Rotating facility and the soot dust characterization

Figure 3 .
Figure 3. Image analysis on the flat target: clean, fouled, post-processed images dimensional set of data.This approach is an inverse problem [38] called Shape From Shading (SFS).This technique obtains three-dimensional information[24] starting from two-dimensional data.However, it depends on several mathematical and optical assumptions[25 -27]  that generate uncertainty and misinterpret the results.As this strategy recognizes the former shape from the shade, the basic assumption is related to the light intensity and position, which could determine several inaccuracies during the shape reconstruction.Figure4shows a three-dimensional shape of the deposited layer obtained by the SFS post-process applied to the result reported in Fig.3(greyscale figure).The procedure was carried out with the software package Fiji developed by Sage et al.[28].Starting from the proper region of interest (ROI), the spatial shape of the powder layer can be determined.Scaling the dimension using reference length makes it possible to assign the scale for the third dimension (i.e., the height of the deposited layer) reported as a colorbar in Fig.4.Using this technique, coupled with the proper light and camera setup, the former twodimensional detection can be interpreted as a three-dimensional data source for shape and morphology estimations (e.g., surface roughness, layer cracks and instabilities[15]).

Figure 4 .
Figure 4.The three-dimensional reconstructed shape of the soil layer based on the shape from shading image post-process technique

Figure 5 .Figure 6 .
Figure 5.Over time detections for the flat target during the contamination in the wind tunnel facility

Figure 7 .Figure 8 .
Figure 7. Camera and light positions for the detection of rotor blades and stator vanes

Figure 11 .
Figure 11.Estimation of the thermal dilatation of the compressor using the greyscale analysis resulting from the subtraction between the fouled rotor blades overtime