Thermal barrier coatings (TBCs) are a key technology enabler in many industrial applications, such as spanning aerospace, automotive, and energy. These coatings are typically deposited on metal components to provide protection from harsh operational environments; specifically, they prevent part failure by selectively increasing resistance to corrosion, elevated temperatures, and abrasion.1,2 Due to the important role that these coatings play for increasing a part’s reliability, maintaining the coating’s quality during deposition is critical. The most fundamental parameter to control during the coating’s application is the deposited material’s thickness (50–250 μm thick) and uniformity. Yet, this has historically been difficult to measure on actual parts.3 Rather, approximate thickness is inferred from previous runs; e.g., the number of required torch passes equals the desired thickness divided by the empirically measured thickness per torch pass. Validation involves measuring the thickness using eddy current sensors, calipers, or micrometers after the coating has already been deposited. This inferred procedure leaves the process open to changes in the coating’s deposition parameters, leading to reduced yield and reliability.
A successful application of thermally sprayed coatings relies on many interrelated variables. Small deviations in these variables (e.g., torch power, arc strike, powder size distributions, booth ventilation, gas flow rates, etc.) can create a substantially different coating in terms of both geometry and reliability.45.–6 To compensate for these errors, process variables are currently maintained by implementing process controls on the impact plume’s properties, including particle velocity, diameter, temperature, and dispersion.78.–9 Yet, even with this feedback, coating failures and yield issues still persist, especially on critical components that operate near the temperature, thermal cycling, and high-velocity impact limits of the material and coating. Ultimately, since many of these procedures do not test the deposited coating, increased time and resources are required to assess pass or fail criteria. Thus, an in situ thickness measurement of the coating would help to understand the real-time variations of the process, enabling more consistent results. Furthermore, an active mechanism to understand the coating thickness and uniformity would help both designers and fabricators to optimize the numerous variables based on the direct real-time feedback.
With the aim of increasing reliability and coating yield, several noncontact thickness gauge approaches have been demonstrated, such as photothermal (PT) or laser-ultrasonic (LU) techniques.10 However, utilization of these technologies is limited by the inherent complexity and costs, and usually requires a priori knowledge of the coating properties or geometry, which adds additional complexity and uncertainty. Furthermore, the spray process imparts a significant temperature gradient to the part’s substrate. This causes the part to expand; a situation PT and LU techniques are inadequate to compensate against. This type of error means that the measured coating thickness could have a smaller magnitude than the net substrate’s thermal expansion. Finally, optical techniques such as line projection have been successfully attempted; however, their applicability is limited to simpler part geometries11
In this paper, a viable approach to in situ coating thickness measurement is described. Leveraging the sensitivity of optical interferometric methods while ruggedizing it for use in the harsh environment of thermal spray production, we demonstrate a fringe projection sensor that enables real-time per-pass thickness measurements. This new approach allows for the direct measurement of the coating layer’s thickness, independent of material properties, and thermal expansion. To this end, Sec. 2 describes the basic theory and operational concept for the fringe projector and Sec. 3 discusses the laboratory and production spray booth devices. This is, then, followed by Sec. 4, which provides our results and discussions. Finally, a summary of our results is provided in Sec. 5.
Background: Fringe Projection
In fringe projection, the target under test is illuminated with an off-axis sinusoidal intensity pattern. After including a triangulation factor, this projected pattern modulates the three-dimensional spatial information of the target onto its phase. Comparison of the measured phase to that of a reference measurement taken of a flat planar target enables one to relate the measured phase to the sample’s physical depth. An illustration of the fringe projector’s geometry is provided in Fig. 1.
The geometry of Fig. 1 shows that the projected sinusoid’s illumination is incident onto the reference plane at an angle with an observed fringe period .12 The detected reference plane intensity can be represented at point by1) establishes a reference phase . When the reference is removed and the sample is measured at point , the detected sample plane intensity can be expressed as
Assuming is small,12 it means that the physical distance between the reference and sample surfaces can be calculated as
Thus, measuring the phase difference can be accomplished at each spatial location within the scene. Such a phase extraction can be achieved within a single snapshot using Fourier transform techniques.13184.108.40.206.–18 Expressing Eqs. (1) and (2) in terms of the general spatial coordinates (, ) yields5) and (6) yields
Inverse Fourier transformation of and produces two exponential functions within the spatial domain
Differential Measurement Approach
Accuracy in the aforementioned measurement methodology relies on critically measuring the phase difference between the sample and reference planes. Typically, the designs of such systems are configured to avoid any modulo phase unwrapping errors.19 This leads to the assumption that the target geometry is smoothly varying within the region of a single fringe. For thermal spray this is often the case, as most surface coatings have continuous derivatives (smoothly varying geometry as imparted by the plasma process). However, other sources of phase ambiguity and error can arise due to part motion. Herein, we refer to such motion errors as “jitter,” or “jitter error.” For instance, a nominal part undergoing thermal spray may be mounted on a spindle chuck, rotating in excess of 300 rotations per minute (RPM), with a run-out greater than 12.5 mm. This translates into a variable imager-to-object distance versus time. Unaccounted for, this error is substantially larger than the micrometer level changes in coating thickness that we are attempting to measure.
As part of our system development, we selected a fringe period to maintain the peak-to-peak phase error caused by jitter to within . Additionally the jitter error is compensated by incorporating a single object-based reference frame in combination with a differential technique between two separate regions on the target’s surface. While differential measurements to compensate jitter have been implemented by others in past research using line projection,11 its application to fringe projection within the context of thermal spray, to the best of our knowledge, has not been studied.
By incorporating a differential approach, we can reduce the measurement’s sensitivity to target jitter. However, there are also differential effects that occur due to the geometry of the target or part. Part geometry can be accounted for with a measuring pass prior to the coating process’s start, thus acquiring a part-dependent reference image. The differential operation is described by expanding () in Eq. (13) to include both the part’s geometry and the coating’s thickness for two unique spatial locations in the scene. This geometry is depicted in Fig. 2 for a cylindrical target of radius . In this case, a reference frame is taken of the uncoated but sandblasted (e.g., diffusely reflecting) part. At a later time, the coating is deposited on the surface at the previously measured location. However, due to jitter, it is separated by a distance away from the original reference plane. To quantify and to isolate it from the coating thickness , two points are compared: one with () and the other without () the coating.
Assuming that and are close to 90° enables us to use Eq. (13) as before to calculate the phase () at both locations and , yielding and , respectively. These phase terms can be represented as a combination of the coating’s thickness and the jitter separation as
The difference of the arguments results in the differential phase term, with both the geometry and coating effects present. Subtracting from isolates the coating’s thickness from the measurement such that
An example of this differential technique is illustrated in Fig. 3, which depicts an image of a stainless steel witness coupon as viewed under fringe illumination. This coupon has a layer of yttria-stabilized zirconia (YSZ) on the right half (denoted by box labeled “A”) and sandblasted metal on the left half (box “B”). Following any fringe (black or white line), there is a noticeable phase shift in the transition from the bare metal to the YSZ deposited layer at this magnification. This relates to a change in thickness between the two regions due to the coating. In the differential measurement approach, two regions are identified on separate sides of the image that are aligned with the torch’s motion. By treating one as the reference, a snapshot differential phase measurement can be produced.
Assuming the plasma torch deposits the coating material over the surface moving from the right to the left in Fig. 3(a), then the increase in thickness would be detected by box (a) before box (b). This change can be directly related to an increase in the coating’s thickness. Applied more generally, as depicted in Fig. 3(b), this differential (single frame) measurement approach can occur over many points in the image at once. For instance, Fig. 3(b) demonstrates that if the image was broken up into zones, then the phase difference across the image could be measured and calibrated to the thickness across a 2-D space. This approach assumes that the plasma torch’s spot size is small; less than half of the camera’s field of view (FOV), in addition to a frame rate that is high enough to capture the torch at several positions across the FOV. This allows a significant increase in sampling points and measuring area for any given snapshot and for each individual coating pass. Thus, by measuring these individual passes and summing the total phase change during the transitions, a measurement of the overall coating thickness can be determined across the part’s surface.
A laboratory prototype was developed as a proof of concept for the fringe projector’s differential measurement technique. In this approach, the fringes are generated and projected onto the target of interest, forming a non-normal angle relative to the plasma stream’s direction and the imaging camera path. This approach is captured in Fig. 4, showing relative geometry and orientation.
The fringe projector assembly is shown schematically in Fig. 5(a), while an image of the completed (but unsealed) prototype is depicted in Fig. 5(b). A 650-nm laser diode (average output power around one (1) Watt, continuous wave) was chosen. This wavelength allows for a nearly maximized detector response while simultaneously avoiding peak spectral emission lines from the plasma, as is shown from plasma gas mixture measurements per Fig. 6.
A lens was used to nearly collimate the laser diode and present a slightly diverging optical wavefront through the linear polarizers ( and ) and Savart plate (SP). Note that in Fig. 5(a), this diverging beam is exaggerated to illustrate the operational concept. The first linear polarizer () was orientated parallel to the laser’s linear polarization state to maximize the laser’s output power. The SP was then set at 45° (extraordinary and ordinary index vectors) relative to in order to split equal power into both of the SP’s eigenstates. For an SP, the net shear distance is20 The final polarizer, , sets the fringe coherence by combining the two polarization states, enabling interference fringes to be localized at the sample. Note that the small divergence of the beam results in a projected fringe frequency that depends on the distance between the projector and object. While error is expected from varying object distances, this error is within approximately 0.1% given a 1.0% error in projector-to-object distance induced by jitter. While this approach is prone to some small error, it was chosen because it provides an effective means to vary the fringe frequency by adjusting the beam’s divergence. This allowed us to select a fringe frequency that was large enough to maintain the jitter within modulo phase.
The fringe illuminator assembly was then mounted in a sealed aluminum tube to prevent degradation of the optics from particulates and powder debris in the coating chamber. The imaging side of this setup was an industrial CMOS detector, capable of integrating down to a minimum of three (3) microseconds, to minimize rotation-induced motion blur. A notional view of the setup is depicted in Fig. 7, where the full sensor geometry is shown relative to a rotation target in Fig. 7(b). It is also noted, that in this approach, a remote trigger is provided in order to sync measurements to a repeatable position. Here, we have accomplished triggering through an optical pickup and reflective tape per Fig. 7(b) at one location. In this orientation, the robot is moved vertically, which is parallel to the long dimension and the cylinder’s rotation axis.
In Situ Testing Results
In order to validate the approach and fringe projection method, the sensor was tested in conventional thermal spray booth configurations. As shown in Fig. 8(a), the sensor was mounted in a production booth to measure deposited metal coatings (nickel substrate). During this trial, a 100-mm diameter cylindrical stainless steel substrate, approximately 300 mm long, was mounted in a rotation chuck that was operated at 300 RPM. The fringe projection sensor was mounted to a tripod that was positioned perpendicular to the plasma stream [Fig. 8(b)], while the camera was triggered by an optical pickup and reflective tape to measure a repeatable area. Utilizing this approach, the robot moved up and down the rotating part, depositing a thin coating layer on each pass. The sensor took an image on each rotation and created a database of measurements over the course of approximately 6 min. Indexing each measurement according to time and displaying the full time sequential database, Fig. 9 shows the unaltered measurement phase (green and blue lines), and the extracted actual deposition measurements (red line) utilizing the differential approach. In fact, without the differential technique, the induced jitter from this setup masks any single-pass measurements that are clearly depicted in the peaks and nulls of the jitter compensated dataset, although the general increase in phase from the full coating is visible in the trend of both the uncorrected measurements.
Processed Results and Analysis
Processing and calibrating the jitter-compensated results of Fig. 9 yield the calibrated data depicted in Fig. 10. Starting at time , the plasma torch is first moved over the target to provide uniform heating (no powder), which registers a spike in the dataset. The torch then returns to the “home” position as it awaits a stable feed of powder stock (denoted as no activity from approximately 40–130 s). After this point, the plasma operation proceeds with the coating passes (14 individual passes), which are captured for the next approximately 230 s. These are denoted by an individual pass number in Fig. 10. Each single-pass measurement is then accumulated to yield the total deposited coating thickness. This operation calculated a coating thickness of , where the traditional measurement, taken with a digital caliper, estimated ; this is less than 5% deviation and well within a significant value for both. The instrumental accuracy was calculated by taking the standard deviation during the “no pass” image set from the time block of 40 to 130 s, while the micrometer measurement was averaged over four measurements across both the inner and outer diameters, to mitigate heating and localized point effects.
The fringe projection approach was demonstrated as a viable in situ coating thickness measurement technique, accurate to the micron level, by accumulating thickness changes on individual plasma coating deposition passes. Results were shown on metallic plasma coatings in an in situ measurement environment. Measurements from the system agreed with conventional data taken with a caliper. Overall, this novel measurement capability will allow improved control for a key thermal barrier coating deposition, and demonstrates a practical manner to implement in an industrial setting.
In the future, this technique could allow the operator to measure numerous points at a significantly faster rate than currently allowed through the use of automation and 2-D data acquisition over the target surface. This would provide real-time feedback on the coating results, decrease operator dependence for physical measurements and downtime, increase yield, and enable more optimal control of coating tolerances.
Nicholas D. Trail received his BS and MS degrees in optical science and engineering in 2007 and 2010, respectively, and anticipates completing his PhD degree in optical sciences in 2015 from the University of Arizona, Tucson, Arizona. He also works for a major defense company as a senior optical engineer, where he designs and develops visible to far-infrared, passive and laser-based electro-optical systems.
Michael W. Kudenov received his BS degree in electrical and computer engineering from University of Alaska Fairbanks, Fairbanks, Alaska, in 2005 and his PhD degree in optical sciences from the University of Arizona, Tucson, Arizona, in 2009. He is an assistant professor of ECE at North Carolina State University in Raleigh, North Carolina. His lab researches compact high-speed hyperspectral, polarimetric, and interferometric sensors and sensing systems within multidisciplinary applications spanning remote sensing, defense, process monitoring, and biological imaging. He is a member of SPIE.
Eustace L. Dereniak is a professor of optical sciences and electrical and computer engineering at the University of Arizona, Tucson, Arizona. He is a co-author of several textbooks, including Optical Radiation Detectors, Infrared Detectors and Systems, published by Wiley-Interscience, and Geometrical and Trigonometrical Optics, published by Cambridge Press. He has written chapters in Imaging in Medicine, edited by S. Nudelman and D. Patton, related to research and development using thermograph instrumentation for the early detection of breast cancer. His publications also include over 100 authored or co-authored refereed articles. He is a fellow of SPIE and OSA, and a president of SPIE in 2012.