Methods providing corrosion detection using thermal images and related systems

Abstract

According to some embodiments of the present disclosure, corrosion on may be detected. A test area of a surface is heated using a heater, and the test area includes a plurality of test regions. After moving the heater away from the test area, a respective plurality of sequential temperature measurements are recorded for each of the plurality of test regions, and the respective plurality of sequential temperature measurements for each of the test regions represents a thermal decay of the respective test region. For each of the test regions, a cumulative distribution transform is performed on the plurality of sequential temperature measurements for the respective test region to provide a cumulative distribution transform result for each of test regions. Whether a test region is a corroded region or a non-corroded region is determined for each test region based on the respective cumulative distribution transform result for the test region.

Description

METHODS PROVIDING CORROSION DETECTION USING THERMAL IMAGES AND RELATED SYSTEMSCROSS-REFERENCE

[0001] This Application is a Nonprovisional Utility Patent Application and claims the benefit of priority under 35 U.S.C. Sec. 119 based on U.S. Provisional Patent Application No. 63 / 698,759 filed on September 25, 2024. The disclosures of Provisional Application No. 63 / 698,759 and all references cited herein are hereby incorporated in their entirety by reference into the present disclosure.FEDERALLY- SPONSORED RESEARCH AND DEVELOPMENT

[0002] The United States Government has ownership rights in this invention. Licensing inquiries may be directed to Office of Technology Transfer, US Naval Research Laboratory, Code 1004, Washington, D.C. 20375, USA; +1.202.767.7230; nrltechtran@us.navy.mil, referencing Navy Case # 212247.TECHNICAL FIELD

[0003] The present disclosure related to methods of detecting corrosion and related detection systems.BACKGROUND

[0004] From subterranean oil applications (see, Reference [1]) to outer space exploration (see, Reference [2]), corrosion and the detection thereof presents significant challenges across industries throughout the world and beyond. The corrosion problem itself tallies up to an estimated 3-4% of GDP cost for the US and other nations (see, Reference [3]). To mitigate the impact of corrosion, techniques have been developed to monitor, search for, and detect corrosion defects as early as possible to inform maintenance operations.

[0005] Among the various monitoring methods, non-destructive techniques (NDT) are important in industries because they can enable monitoring without affecting the lifespan of thecomponent and / or disrupting operations (see, Reference [4]). Among traditional methods, visual testing is the most common corrosion monitoring technique. However, it may involve exhaustive and / or labor-intensive inspection, and is susceptible to human error. Acoustic emissions (AE) techniques rely on AE signals from highly sensitive microphones placed on the surface of the structure to detect stress corrosion cracking (see, Reference [5]). Although AE techniques are cost- effective, they may be prone to false alarms caused by scattered wave motions (see, Reference [6]). In recent years, several ultrasonic inspection (UI) based corrosion detection methods have been proposed (see, References [7], [8], and [9]). However, performance of these methods may degrade with increased surface roughness. Radiographic NDT methods (see, References

[0010] ,

[0011] , and

[0012] ] study variations in electromagnetic wave absorption to detect defects in a component. Another group of structural health monitoring techniques employs infrared thermography (IRT) (see, References

[0013] ,

[0014] ,

[0015] , and

[0016] ). IRT methods analyze variations in thermal radiation of a component to identify defective areas. Among numerous techniques used to detect corrosion in materials, IRT stands out for its versatility and applicability (see, Reference

[0017] ). IRT techniques, in particular, may be especially effective to identify invisible subsurface and surface corrosion.

[0006] Methods of corrosion detection from thermal images have been extensively studied in the recent past (see, References

[0018] ,

[0019] , and

[0020] ). Sakagami, et al. (Reference

[0021] ) proposed a moving average image binarization process to automatically detect corrosion under a coating. Kopf and Tighe (Reference

[0022] ) presented an adaptive thresholding-based method to detect and accurately quantify corrosion extents from IRT images. Lim, et al. (Reference

[0023] ) proposed an accelerated region-based convolutional neural network (CNN) that integrates visual and thermo-graphic images to automate the detection and classification of surface and subsurface corrosion in steel bridges. While these methods have shown promising results, they generally rely on pre-established classification techniques applied to IRT pixel measurements, without a clear connection to the physical processes governing the emission of heat by corrosive materials. Consequently, the performance of these techniques may be adversely affected by environmental factors.SUMMARY

[0007] This summary is intended to introduce, in simplified form, a selection of concepts that are further described in the Detailed Description. This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter. Instead, it is merely presented as a brief overview of the subject matter described and claimed herein.

[0008] According to some embodiments of inventive concepts, methods of detecting corrosion on a surface are provided. A test area of the surface is heated using a heater, and the test area includes a plurality of test regions. After heating the test area, the heater is moved away from the test area. After moving the heater away from the test area, a respective plurality of sequential temperature measurements are recorded for each of the plurality of test regions, and the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region. For each of the plurality of test regions, a cumulative distribution transform is performed on the plurality of sequential temperature measurements for the respective test region to provide a cumulative distribution transform result for each of the respective test regions. For each of the test regions, whether the test region is a corroded region or a non-corroded region is determined based on the respective cumulative distribution transform result for the test region.

[0009] Performing may further include performing a signed cumulative distribution transform on the plurality of sequential temperature measurements to provide a signed cumulative distribution transform result for each of the respective test regions. Determining may further include determining whether the test region is a corroded region or a non-corroded region based on the respective signed cumulative distribution transform result for the test region.

[0010] Recording may include recording the respective plurality of sequential temperature measurements for each of the plurality of test regions based on output from an infrared camera system defining a pixel array, and each of the temperature measurements may be based on output from an image area of the pixel array.

[0011] The IR camera system may move relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of testregions, and for each of the test regions, at least two of the plurality of sequential temperature measurements may be based on output from different image areas of the pixel array. In addition, the pixel array may include a first pixel subarray provided using a first semiconductor detector chip and a second pixel subarray provided using a second semiconductor detector chip, and the different image areas may be from different ones of the first and second pixel subarrays.

[0012] The IR camera system may be stationary relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of test regions, and the sequential temperature measurements for a respective test region may be based on output from a respective image area of the pixel array.

[0013] The plurality of test regions may include first and second test regions. Recording may include recording a first plurality of sequential temperature measurements representing a thermal decay of the first test region and recording a second plurality of sequential temperature measurements representing a thermal decay of the second test region. Performing the cumulative distribution transform may include performing the cumulative distribution transform on the first plurality of sequential temperature measurements to provide a first cumulative distribution transform result for the first test region and performing the cumulative distribution transform on the second plurality of sequential temperature measurements to provide a second cumulative distribution result for the second test region. Determining may include determining whether the first test region is a corroded region or a non-corroded region based on the first cumulative distribution result and determining whether the second test region is a corroded region or a non-corroded region based on the second cumulative distribution result.

[0014] Responsive to determining that the test region is a corroded region, the location of the first test region on the surface may be identified. Identifying the location may include identifying the location of the first test region on the surface based on information from a global positioning receiver, and / or marking the location of the first test region on the surface.

[0015] A background temperature of the surface may be recorded. For each of the test regions, determining may include determining whether the test region is a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region and based on the background temperature of the surface.

[0016] Recording the background temperature of the surface may include recording the background temperature for each of the plurality of test regions before heating the test area, and determining may include determining whether the test region is a corroded region or a noncorroded region based on the respective cumulative distribution transform result for the test region and based on the respective background temperature of the test region.

[0017] Recording the background temperature of the surface may include recording a representative background temperature for the surface, and determining may include determining for each of the test regions whether the test region is a corroded region or a noncorroded region based on the respective cumulative distribution transform result for the test region and the representative background temperature for the surface.

[0018] According to some other embodiments of inventive concepts, methods of detecting corrosion on a surface are provided. A test area of the surface is heated using a heater, and the test area includes a plurality of test regions. After heating the test area, the heater is moved away from the test area. After moving the heater away from the test area, a respective plurality of sequential temperature measurements are recorded for each of the plurality of test regions based on output from an infrared camera system defining a pixel array. At least two of the plurality of sequential temperature measurements for each of the test regions are based on output from different image areas the pixel array, and the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region. For each of the test regions, whether the test region is a corroded region or a non-corroded region is determined based on the respective plurality of sequential temperature measurements for the test region.

[0019] The IR camera system may move relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of test regions.

[0020] The pixel array may include a first pixel subarray provided using a first semiconductor detector chip and a second pixel subarray provided using a second semiconductor detector chip, and the different image areas of the pixel array may be from different ones of the first and second pixel subarrays.

[0021] For each of the plurality of test regions, a cumulative distribution transform may be performed on the plurality of sequential temperature measurements to provide a cumulative distribution transform result for each of the respective test regions, and determining may include determining for each of the test regions whether the test region is a corroded region or a non-corroded region based on the cumulative distribution transform result for the respective test region.

[0022] Performing the cumulative distribution transform may include performing a signed cumulative distribution transform on the plurality of sequential temperature measurements for each of the respective test regions to provide a signed cumulative distribution transform result for each of the respective test regions, and determining may include determining for each of the test regions whether the test region is a corroded region or a non-corroded region based on the respective signed cumulative distribution transform result for the test region.

[0023] The plurality of test regions may include first and second test regions. Recording may include recording a first plurality of sequential temperature measurements representing a thermal decay of the first test region and recording a second plurality of sequential temperature measurements representing a thermal decay of the second test region. In addition, determining may include determining whether the first test region is a corroded region or a noncorroded region based on the first plurality of sequential temperature measurements and determining whether the second test region is a corroded region or a non-corroded region based on the second plurality of sequential temperature measurements.

[0024] Responsive to determining that the first test region is a corroded region, the location of the first test region on the surface may be identified. Identifying the location may include identifying the location of the first test region on the surface based on information from a global positioning receiver, and / or marking the location of the first test region on the surface.

[0025] A background temperature of the surface may be recorded, and for each of the test regions, determining may include determining whether the test region is a corroded region or a non-corroded region based on the plurality of sequential temperature measurements for the test region and based on the background temperature of the surface.

[0026] Recording the background temperature of the surface may include recording a respective background temperature for each of the plurality of test regions before heating the test area, and determining may include determining whether the test region is a corroded region or a non-corroded region based on the respective plurality of sequential temperature measurements for the test region and based on the respective background temperature of the test region.

[0027] Recording the background temperature of the surface may include recording a representative background temperature for the surface, and determining may include determining for each of the test regions whether the test region is a corroded region or a noncorroded region based on the respective plurality of sequential temperature measurements for the test region and the representative background temperature for the surface.

[0028] According to still other embodiments of inventive concepts, detection systems are provided to detect corrosion on a surface. These detection systems include a heater, an infrared camera system, a drive system, and a controller. The heater is configured to heat a test area of the surface, and the test area includes a plurality of test regions. The infrared camera system includes an array of pixels defining a plurality of image areas configured to receive thermal energy emitted from the surface within a field of view. The drive system is configured to propel the detection system so that the heater moves away from the test area and so that the test area is within the field of view of the IR camera system, and the controller is coupled with the IR camera system. The controller is configured to record a respective plurality of sequential temperature measurements for each of the plurality of test regions from the IR camera system after the test area is within the field of view, and the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region. The controller is also configured, for each of the plurality of test regions, to perform a cumulative distribution transform on the plurality of sequential temperature measurements for the respective test region to provide a cumulative distribution transform result for each of the respective test regions. In addition, the controller is configured, for each of the test regions, to determine whether the test region is a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region.

[0029] According to yet other embodiments of inventive concepts, detection systems are provided to detect corrosion on a surface. These detection systems include a heater, an infrared camera system, a drive system, and a controller. The heater is configured to heat a test area of the surface, and the test area includes a plurality of test regions. The infrared camera system includes an array of pixels defining a plurality of image areas configured to receive thermal energy emitted from the surface within a field of view. The drive system is configured to propel the detection system so that the heater moves away from the test area and so that the test area is within the field of view of the IR camera system, and the controller is coupled with the IR camera system. The controller is configured, after moving the heater away from the test area, to record a respective plurality of sequential temperature measurements for each of the plurality of test regions based on output from the infrared camera system. At least two of the plurality of sequential temperature measurements for each of the test regions are based on output from different image areas the pixel array, and the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region. The controller is also configured, for each of the test regions, to determine whether the test region is a corroded region or a non-corroded region based on the respective plurality of sequential temperature measurements for the test region.BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Examples of embodiments of inventive concepts may be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings in which:

[0031] FIGs. lA-1, 1A-2, 1A-3, IB-1, IB-2, 1C, ID, and IE are diagrams illustrating proposed methods to detect corrosion using IRT according to some embodiments of inventive concepts, where each pixel from the thermal image sequence is labeled as non-corroded or corroded by classifying the ID thermal signal extracted at the corresponding pixel location;

[0032] FIG. 2 is a schematic illustration of an IRT system used to detect corrosion according to some embodiments of inventive concepts, where a sample is heated temporarily using a moving heat source, after which acquisition of the spatiotemporal data (x = 0, y, z, t) begins,with each pixel location (x = 0, y, z) in the domain of the imaging setup Q being determined as being consistent with a "normal" or a corroded material region;

[0033] FIG. 3 illustrates signals across regions in a test coupon including a corrosion defect according to some embodiments of inventive concepts, where the coupon contains a corroded region (left most small square), and time decay signals from corroded and non-corroded regions differ from one another;

[0034] FIG. 4A illustrates a manually labeled ground truth test coupon with corroded regions, and FIGs. 4B, 4C, 4D, 4E, 4F, and 4G illustrate detection of corroded regions using different techniques;

[0035] FIGs. 5A and 5B are graphs respectively illustrating average pixel accuracy and similarity index for different detection methods as a function of a number of training coupons;

[0036] FIG. 6A is a schematic diagram illustrating a corrosion detection system used to detect corrosion on a metal surface according to some embodiments of inventive concepts;

[0037] FIG. 6B is a schematic diagram illustrating test areas including test regions on the metal surface of FIG. 6A according to some embodiments of inventive concepts;

[0038] FIGs. 6C-1, 6C-2, 6C-3, 6C-4, 6C-5, and 6C-6 are schematic diagrams illustrating mappings of test regions TR on the surface of FIG. 6A to image areas IA of a pixel array of an IR camera system to record sequential temperature measurements of test regions according to some embodiments of inventive concepts; and

[0039] FIG 7 is a flow chart illustrating operations providing corrosion detection using the system of FIG. 6A according to some embodiments of inventive concepts.DETAILED DESCRIPTION

[0040] Inventive concepts will now be described more fully hereinafter with reference to the accompanying drawings, in which examples of embodiments of inventive concepts are shown. Inventive concepts may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of present inventive concepts to those skilled in the art. It should also be noted that theseembodiments are not mutually exclusive. Components from one embodiment may be tacitly assumed to be present / used in another embodiment. Moreover, certain details of the described embodiments may be modified, omitted, or expanded upon without departing from the scope of the described subject matter. Moreover, in the drawings, like reference numerals refer to like elements throughout, and the sizes of each of the elements may be exaggerated for clarity and / or conveniences of explanation.

[0041] The present disclosure introduces novel approaches used to detect corrosion defects on metal substrates using infrared (IR) thermal images. Among numerous non-destructive techniques, infrared thermography (IRT) is notable for its effectiveness in identifying invisible surface and subsurface corrosion in materials. Existing methods for corrosion detection from IRT images may lack the connection to the physical processes governing the emission of heat by defective areas. The present disclosure proposes a transport-based mathematical model to describe the difference in heat flow characteristics between non-corroded and corroded material regions. Novel detection techniques are also disclosed, utilizing the signed cumulative distribution transform (SCDT) and a subspace classifier to classify one dimensional (ID) thermal signals derived from IRT image sequences. Experiments demonstrate that the approach of the present disclosure may be capable of detecting corrosive regions on metal substrates with high accuracy while being data efficient with respect to a number of machine learning-based detection methods.

[0042] In the present disclosure, methods are disclosed to detect corrosive defects using IRT in an aluminum coating on a steel substrate, representative of deck material found on naval vessels (see, Reference

[0024] ). Starting from a partial differential equation (PDE) physical model for the experimental setup, a recently developed signal transform (the signed cumulative distribution transform, see Reference

[0025] ) is used to represent PDE solutions (IRT measurements) in a convenient form to provide modeling and classification using a supervised learning process. This method may be tailored to overcome superfluous variations in temporal displacements which may interfere with the ability to discern signals corresponding to corrosion spots from those corresponding to non-corroded spots. A method of the present disclosure is compared with numerous other pattern recognition and classification methods, and its accuracy and robustness are demonstrated.

[0043] FIGs. 1A-1, 1A-2, 1A-3, IB-1, IB-2, 1C, ID, and IE are diagrams providing an overview of a method to detect corrosion using IRT according to some embodiments of inventive concepts. In this method, each image area (i.e., each pixel or group of pixels) of IR camera 105 corresponds to a respective test region TR from the surface of test sample 103, and thermal information from each image area from the thermal image sequence is used to label the respective test region TR as non-corroded or corroded by classifying a ID thermal signal extracted for each corresponding image area.

[0044] As shown in FIG. IB-1, a test area TA from a surface of test sample 103 may include a plurality of test regions TRi.i to TRy,z, and each test region TR corresponds to a respective image area of a pixel array of IR camera 105. Each image area of the pixel array may include a single pixel, or each image area may include a group of pixels. Each image area of IR camera 105 can thus capture a plurality of sequential temperature measurements for the respective test region (TR) representing a thermal decay of the respective test region (TR). If an image area includes a single pixel, the temperature measurements from that image area are based on information from the single pixel. If an image area includes a group of pixels, the temperature measurements from that image area are based on an aggregation of information from that group of pixels (e.g., based on an averaging). FIG. IB-2 illustrates a sequence of thermal images from which the sequential temperature measurements for the test measurements can be obtained.

[0045] First, a ID thermal signal is extracted from the sequence of IR thermal images at each image area (i.e., each pixel or each group of pixels) of IR camera 105 corresponding to the respective test regions TR. The extracted signal for each test region TR is then classified by the proposed detector as either corroded or non-corroded. All the extracted thermal signals from the IRT sequence for the respective test regions can thus be classified to identify corrosion defects on the surface of test sample 103.

[0046] In FIG. 1A-1, heater 101 (also referred to as a heat source, heating element, etc.) is provided adjacent the surface of test sample 103 to heat test sample 103. In FIG. 1A-2, heater 101 is moved away from test sample 103, and in FIG. 1A-3, IR camera 105 generates thermal images of test sample 103 overtime as test sample 103 cools, and examples of the resulting thermal images 107a, 107b, 107c, ... 107n are shown in FIG. IB-2. Each image area of camera 105 and thermalimages 107a-n corresponds to a respective test region TR on the surface of test sample 103 (also referred to as a pixel location), and each image area may include a single pixel or a group of pixels. By using a same image area of thermal images 107a-n to determine a temperature of a corresponding test region TR on the surface of test sample 103 as that test region TR of the test sample 103 cools, a plot of the thermal decay at that test region TR over time can be determined as shown in the graph of FIG. 1C. This process can be replicated for each image area of IR camera 105 to determine a respective thermal decay at each corresponding test region TR of the test sample corresponding to a respective image area of images 107a-n. In the embodiment of FIGs. 1A-1, 1A-2, 1A-3, IB-1, IB-2, and 1C, IR camera 105 is static with respect to test sample 103, and each image area of IR camera 105 is thus associated with a same respective test region TR of test sample 103 in each of the thermal images 107a-n. According to some other embodiments, the IR camera may move across a test surface, and the temperature measurements for a same test region TR on the test surface may be represented by a different image area in each of thermal images 107a-n. In such embodiments, the IR camera and heater may be provided on a moving system (e g., a rover) that moves across a large test surface such as the deck of a ship.

[0047] By categorizing each test region TRi,i to TRx,yof test sample 103 corresponding to a respective image area as either non-corroded or corroded, thermal decal characteristics corresponding to these test regions and / or image areas can be characterized in the training data of FIG. ID as either non-corroded §(1) or corroded §(2). As shown in FIG. IE, a signed cumulative distribution transform nearest local subspace (SCDT-NLS) classifier can be used to predict whether a test region on the test sample (corresponding to an image area of images 107a-n) is noncorroded or corroded based on the measured thermal decay of FIG. 1C and the training data of FIG. ID.

[0048] The remainder of this disclosure is organized as follows. A discussion of the physics underlying heat dissipation and a statement of the corrosion detection problem will be provided. Next, a transport-based model for signal classes will be defined and the proposed method of corrosion detection will be discussed. Collection of IRT images, data preprocessing, and results will then be discussed. Finally, concluding remarks will be provided.

[0049] FIG. 2 illustrates a schematic overview of an infrared thermography (IRT) system used to provide corrosion detection according to some embodiments of inventive concepts. Infrared (IR) camera 105 is affixed to aluminum frame 123 and placed over a surface area (“coupon") of test sample 103 to be analyzed. Linear actuator arm 121 is coupled to heater 101 to cause heater 101 to pass directly over the coupon of test sample 103 resulting in a transient / temporary heating of the test sample material. Once linear actuator arm 121 removes heating element 101 from the coupon of test sample 103, respective image areas of IR camera 105 record the spatio-temporal thermal decay for each of the test regions TR.Here Q refers to the spatial domain (field of view in the x-y plane) of the image, and CO and T correspond to initial and end acquisition times, respectively. An objective of some embodiments of inventive concepts is to determine whether each image area (x = 0, y, z) in the domain of the imaging setup Q is consistent with a "normal" or corroded material test region TR.

[0050] While static IRT images may include important geometric clues regarding interesting object features, including possible corrosion regions, the dynamic behavior of IRT measurements may include even more salient features capable of discriminating between corrosion vs. non-corrosion features. FIG. 3 shows time dependent IRT measurements in four adjacent locations / regions from a coupon of test sample 103 with a corrosive spot. As the test regions (also referred to as locations, Regions of Interest, or ROIs) transition from the center of the corrosive spot into regions without corrosion, a noticeable difference between the thermal time traces (showing thermal decay at the respective locations) is visible. Some embodiments of inventive concepts may provide a signal processing system capable of differentiating normal, non-corrosion test regions TRs (including, bolts, welds, and other geometric features) from locations / regions known to contain corrosion spots. As with all detectors, false positive and false negative errors should be reduced / minimized.

[0051] This portion of the disclosure outlines practical and / or effective solutions to provide corrosion detection using a transport-based modeling approach. A physical description of heat dissipation and corrosion detection (classification) are discussed below, and then experimental procedures are disclosed to obtain training data, based on which a detection model is disclosed.

[0052] A dynamic system describing heat dissipation in a metal coating on a steel surface (illustrated in FIG. 3) is considered. From the 1st law of thermodynamics (statement of energy of a system) and a control volume analysis, a general heat conduction equation may be defined as: Equation (1)where u := it(x , t) is the temperature field, K is the thermal conductivity with the 3D spatial coordinate x E R3, p is the material density, c is a specific heat, and q is the rate of internal energy generation (source). Assuming a homogeneous isotropic substance, Equation (1) reduces to a typical heat conduction equation, namely the Fourier-Biot equationthe thermal diffusivity. In practice, however, materials to be inspected are not typically homogeneous, and neither is the conductivity, so Equation (1) may be more realistic (see, Reference

[0026] ).

[0053] As shown in FIG. 3, time decay signals 311, 312, 313, and 314 may be generated for respective locations / regions 111 (e.g., TRj,k), 112 (e.g., TRj+i,k), 113 (e.g., TRj+2.k), and 114 (e g., TRj+3.k) of a coupon of a test sample 103 with corrosion defect 321. In FIG. 3, leftmost locations / regions 111 and 112 are on the corrosion defect 321, location / region 113 crosses an edge of corrosion defect 321, and rightmost location / region 114 is outside of corrosion defect 321. Time decay signals 311 and 312 from corrosion regions 111 and 112 thus differ from time decay signal 314 from non-corrosion region 114. Moreover, time decay signal 313 (for location / region 113 which crosses an edge of corrosion defect 321) differs from time decay signals 311 and 312 (for locations / regions 111 and 112 on corrosion defect 321) and from time decay signal 314 (for location / region 114 on non-corrosion region).

[0054] At time t = to (after the heat source is withdrawn from the test region) the temperature field (x, t) at each spatial location (including locations / regions 111, 112, 113, and 114) is then measured as a function of time for a given time interval, to < t < T. The ID thermal signal (t) is the measured temperature field (heat dissipation) at each location / region x / , i.e., s7(t) = H(u(Xj, t)), Xj E R2. Here H(-) is a linear shift invariant operator that includes intensity modulation and diffraction, modeling the image acquisition process. Given that this operator is independent of the actual diffusion process being observed, and that the goal is simply to classifymeasurements as “non-corroded" or “corroded", for the remainder of this disclosure, blurring, diffraction, etc., may be neglected from the analysis. The task is then to determine whether the measurement time series (t), t E [to, T] originates from an image area corresponding to a location / region TR without corrosion (Class 1) or with corrosion (Class 2).

[0055] While heat flow following impulse surface heating in materials with possible defects may require complex numerical modeling techniques (see, Reference

[0027] ), simplified models for measured data s(t) exist. One such model is provided by Almond et al. (Reference

[0026] ) as: Equation (2)

[0056] where p, c, k, a are density, heat capacity, thermal conductivity and thermal diffusivity respectively, and Jo is the optical energy intensity (J / m2) from the excitation source absorbed at the surface. Here d refers to the thickness of a layer describing a delamination-like defect, D refers to the diameter of the defect, and R refers to a thermal reflection coefficient. While such a model is a simplification from the actual PDE shown in Equation (1), they suggest that corroded regions would have different thermal time decay relative to non-corroded regions given that they will have different parameters. As there are different parameters, it is postulated herein that there exist sets §(1)and §(2)that denote the ‘non-corroded’ and ‘corroded’ classes, respectively. FIG. 3 shows different time series (taken immediately after the heater is removed from each region) from a region 111 in a material showing a corroded spot (shown as time series graph 311) transitioning to a region 114 without corrosion (shown as time series graph 314). It is postulated herein that the classes are distinct and don’t overlap: §(1)A§(2)=0. Note that this assumption may neglect partial volume effects, related to a region of pixel measurements (for example, near the border of corroded and non-corroded regions) containing contributions from both corroded and non-corroded materials that can contribute, partially, to the same measured signal s (as shown by time series graph 313 corresponding to region 113). The present disclosure describes methods to estimate a mathematical model for both signal classes from sample training data and an algorithm to ascertain the membership of a measured signal sj(t), t E [to, T] to one of the classes.

[0057] As discussed above, basic considerations regarding heat flow properties in normal materials and materials with defects (i.e., corrosion) enable the assumption that IRT measurements Sj(t) belonging to non-corroded regions versus corroded regions have different characteristics. Therefore, it is postulated that the existence of two sets(1), and §(2)denoting two (non-corroded vs. corroded) signal classes. A heat transport-type model is thus proposed where signals belonging to class c, c E { 1, 2} can be modeled as a rearrangement of a template time decay function <p: Equation (3)

[0058] Thus signal class c is represented by a mathematical set(c)where <p(c)refers to a typical (template) time decay function, and g E (G refers to a time warping function that acts to warp the template, and a is a positive multiplicative calibration constant. Any signal measured for class c will correspond to a version of the template(c)that has been possibly distorted according to s(t) = g'(t)(p^c\g(t}). The distortion time warp g will belong to a particular subset C of increasing time warp functions. For example, they can be translated versions of the template, i.e., s(t)= (p(c\t~g) for some random translation factor g, which could be due to slight imprecisions in the heater timing mechanism. The time warp could also represent a scaling plus translation, in which case g(t)=at+g, with a, g parameters pertaining to the physical process, etc. In this case the scaling constant (a), related to the speed of decay, can vary slightly from different regions due to slight changes in material composition (unrelated to corrosion), for example. Other, higher order warping functions could also be used to account for normal (i.e., non-corrosion-related) variation in material properties, air temperature, humidity, etc.

[0059] While the model expressed in Equation (3) may seem intuitive given the discussion above, it also has close ties to actual solutions of partial differential equations (PDEs), including specifically the diffusion equation. It has recently been shown in Reference

[0028] that the model above can actually be used to represent the solution of PDEs related to transport (i.e., convection, diffusion, wave propagation, etc ). Samples (PDE solutions) Sj(t)=u(xp t) obtained at different times can be represented with Equation (3) exactly, and the functions g obtained from these representations ‘encode’ information about the system (PDE) that yielded the solution (signal) measured and can be used to ‘recover’ the system parameters (see, Reference

[0028] ). In our case,we are not interested in recovering system parameters (e.g., diffusion constant, thermal conductivity, etc.), but rather only recovering information about the system related to whether that location has signs of corrosion or not. This is described in greater detail below.

[0060] Given that a particular class r(i.e., no defect) may contain numerous features such as paint blots, joints, and other material characteristics that do not correspond to corrosion, the assumption that there exists only one template(r)that models this class can be restrictive. Following Reference

[0029] , a given class may contain Mctemplate functions instead and the following extension to the class model proposed in Equation (3) is proposed below:

[0061] where^ refers to the mthtemplate function for the cthclass, and where the notation

[0062] Given a measured IRT signal s(t), the task now becomes deciding whether Sj belongs to class §(1)or §(2). A solution described in References

[0030] and

[0029] may be adopted. Training signals belonging to both classes are used to estimate the class model described in Equation (4b). Utilizing a recently proposed transform, denoted the signed cumulative distribution transform (see, Reference

[0025] ), the data model is first transformed using the signed cumulative distribution transform (SCDT), and the problem is solved in transform domain. Because of certain linearization properties of the transform being used (described below), the class model in transform space can be thought of as a union of linear subspaces, facilitating the solution of the classification problem which is implemented using a nearest subspace technique.

[0063] The signed cumulative distribution transform (SCDT) presented in Reference

[0025] extends the cumulative distribution transform (CDT) introduced in Reference

[0031] for a class of normalized positive smooth functions. The CDT is an invertible nonlinear ID signal transform from the space of smooth positive probability densities to the space of diffeomorphisms. It can be described as discussed below.

[0064] represent a given signal and a known reference signal, respectively, with , and both s(l) and so(y) being positivewithin their respective domains Qs and flSo. The CDT of (t), denoted as s*(y), is the function that satisfies the following equation:An alternative expression for s*(y) is given by, s*(y) = S’1(So(y)), Equation (6) For a uniform reference signal (i.e., can be written and therefore s'(y)=5' '(y).

[0065] The SCDT was proposed in Reference

[0025] as an extension of the CDT to general finite signed signals. For a non-negative signal (t) with arbitrary mass, the SCDT is given by:where | |s| |Lis the Li norm of signal s, and s* is simply the CDT (Equation (6)) of the normalized signa For a signed signal (t), the signal is decomposed as s(t)=s (t)-s (t), where s+(t) ands"(t) are the absolute values of the positive and negative parts of the signal s(t). The SCDT is given by: Equation (8)where s+(y) and s_(y) are the transforms, as defined in Equation (7), for the positive signals s+(t) and s~(t), respectively. The SCDT has a number of properties which may help simplify signal processing problems, including the composition property and the convexity property.

[0066] The Composition property is discussed as follows. The SCDT of the signal sg=g's°g is given by: Equation (9)where, the SCDT of the signed signal s is given by Equation (8), g(t) is an invertible and differentiable increasing function, s°g = s(g(t)), and g'it') = dg(tydt (see, Reference

[0025] ).

[0067] The Convexity property is discussed as follows. Given, a set of signals § is defined as follows: where (p is a given signal and represents a set ofID spatial deformations of a specific type (such as translation, scaling, etc.). According to theconvexity property, the set is convex for anyif and only ifis convex (see, Reference

[0025] ).

[0068] The set S can be interpreted as a mathematical model for a signal class while (p being the template signal corresponding to that class. In the next section, a mathematical modelbased problem formulation for corrosion detection using ID thermal signals is discussed. Subsequently, it is shown how the composition and convexity properties of the SCDT help render certain nonlinear signal classes into convex clusters, thereby simplifying the corrosion detection problem.

[0069] The data models defined in Equations (3) and (4b) may generally yield nonconvex signal classes, which may cause the above detection problem to be difficult to solve. As shown in References

[0032] and

[0029] , under certain assumptions, the geometry of the signal class can be simplified in the SCDT domain. Hence, proposed corrosion detection methods may begin by applying the SCDT on the ID signals extracted from the thermal images. The signal class models defined in Equations (3), (4a), and (4b) may then given by: andrespectively, in the transform domain. If the data follow the model defined in Equation (10), the SCDT-NS classifier proposed in Reference

[0032] can provide the solution to the detection problem as:where s is the SCDT of an unknown test signal s,is the subspace corresponding to class c, and d(., .) is the Euclidean distance between s and the nearest point in V(c). Thermal signals extracted from training coupons are used to model the subspace V(c)during the training phase of the experiment for both corroded and non-corroded classes. As shown in References

[0029] and

[0028] ,such signal classes can be described using the model defined in Equations (I la) and (11b). Given this model, the unknown class label c of a test signal s can be determined by solving, Equation (13)In Reference

[0029] , the nearest local subspacesearch algorithm was exploited in SCDT domain to solve this problem.

[0070] In the present disclosure, a modified approach is adopted to solve the problem defined in Equation (13), diverging from the methodology proposed in Reference

[0029] , Given the high computational complexity of applying the nearest local subspace technique proposed in Reference

[0029] , the procedure may be simplified by constructing subspacesfrom the training data by using time decay signals s where i is a small neighborhood inthe training data. Once transform domain local subspaces are defined, Equation (13) may beused to determine the class of IRT sample Sj collected at pixel location x / . Subsequently, all the IRT signals extracted from the test substrate are classified to identify corrosive defects present on the substrate.

[0071] As discussed above, IR thermography is a versatile non-destructive remote sensing technique that can be used to monitor for the potential presence of corrosion with corroded and non-corroded regions differentiated by distinct temperature decay characteristics. In order to experimentally examine the thermal decay characteristics, under controlled conditions, of materials used for various (critical) navy infrastructure on platforms exposed to harsh environmental conditions, surrogate samples termed “coupons" were used. The coupon substrate is a square 8 inch x 8 inch area and 0.5 inch thick, steel material, coated with an aluminum “nonskid" material. In order to precipitate occurrence of a corroded spot in the coating, a blemish is added to the surface of the substrate either in a score mark of fixed depth or shallow drill hole. Next, to accelerate the corrosion process, the coupon is fully immersed in a salt bath. The coupons are removed from the immersion, and tested periodically via sounding, to determine if corrosion has occurred. Upon confirmation of corrosion defect development, the coupon surface is thencovered with a sealant. Baseline coupons containing no corrosion are prepared simply by applying the coating to an unblemished substrate and applying sealant.

[0072] The experimental setup is shown in Fig. 2. A 1.5 kW linear heater 101 (also referred to as a heating element, heat source, etc.) is attached to a pair of extruded aluminum rails, and a 12 V linear actuator arm (121) is used to drive the heater back and forth over coupon 103. IR camera 105 is provided over coupon 103, and a data acquisition system is coupled with IR camera 105. This double pass scheme is used to account for the limited range of motion of actuator arm 121 in the experimental setup, and the necessity to remove heater 101 out of the field of view of IR camera 105. Procedurally, coupon 103 is placed at one end of the rails, opposite to actuator arm 121, far enough away to reduce / prevent any spurious heating from heater 101. Actuator arm 121 is activated to drive heater 101 to the far end past the location of coupon 103, IR camera 105 is activated, and the direction of heater 101 is reversed sweeping heater 101 over coupon 103 in ~ 3.5 seconds. Heater 101 may be a single element electric radiant heater, and IR camera 105 may be a Forward Looking Infrared (FLIR) model SC660 un-cooled long-wave infrared (LWIR) system with 640x480 pixels and a thermal resolution (NDET) of 45 mK. For each run, IR camera 105 was set to acquire data at a frame rate of 10 Hz and image sequences were taken for a total of 90 seconds, resulting in 900 thermal images. The experimental procedure was performed using several non-corroded baseline coupons and a collection of corroded coupons.

[0073] After data acquisition, data is preprocessed to discard the portion of the time series that is acquired while heater 101 is on and between coupon 103 and camera 105. This operation was performed ‘manually,’ though it can be automated with knowledge of the time interval that heater 101 is on. The average IR intensity prior to heater 101 being turned on (also referred to as a background temperature) is used as a background measurement. The time series s(t) = u(x7, t) measurements are divided by their respective background measurements in order to reduce / remove variable baselines due to differences in room / material temperature and related effects. The normalized time series Sj(t) is then classified using the proposed method to detect whether the corresponding pixel at x7belongs to a corroded area.

[0074] Table 1 - Cross validation results: accuracy and similarity index performances of the comparing methods.

[0075] Table 1 provides results of a comparative analysis of the proposed corrosion detection method performed relative to several time series classification -based detection techniques, including: linear Support Vector Machine (SVM) (see, Reference

[0033] ) classifier; ID Residual Network (ResNet) (see, References

[0034] and

[0035] ); Long Short-Term Memory Fully Convolutional Network (LSTM-FCN) (see, References

[0035] and

[0036] ); and SCDT-nearest subspace (SCDT-NS) (see, Reference

[0032] ) classifier. Furthermore, comparison was performed against a Fourier transform (FT)-based traditional approach, which employs a linear-SVM classifier to classify the thermal time series in the Fourier domain. The detection performance was studied in terms of pixel accuracy, similarity index (Dice score), and data efficiency. FIG. 4A is the thermal image of a manually labeled ground truth test coupon.

[0076] Experiments were conducted on IR thermal images collected from five coupons, where four contained corroded areas, and one was in sound condition (i.e., without corroded areas). First, a leave-one-out cross-validation experiment was conducted on the five coupons, meaning that the detection methods were trained using four coupons and tested on the remaining one. This process was repeated for each coupon. Pixel accuracy and similarity index (SI) were measured as: andwhere TP and FP represent true positive and false positive, respectively.

[0077] Table 1 presents the average accuracy and similarity index, along with the corresponding standard deviations, for all methods. The results demonstrate that the proposed method outperformed the comparative methods, achieving higher pixel accuracy and similarity scores. FIG. 4A illustrates a manually labeled corroded region. FIGs. 4B, 4C, 4D, 4E, 4F, and4G illustrate the corrosion detection results on a test coupon for all methods. It shows that the proposed method (FIG. 4B) demonstrated superior accuracy in identifying the corroded region (SI = 93.6%), whereas the other methods (FIG. 4C for FT-SVM, FIG. 4D for SVM, FIG. 4E for ID ResNet, FIG. 4F for LSTM-FCN, and FIG. 4G for SCDT-NS) experienced higher rates of false positive detection. These results show that the proposed method has the potential to effectively detect corrosion spots in a metal coating.

[0078] In addition to the effectiveness, the data efficiency of the detection methods was also evaluated. To showcase the data efficiency of the proposed method, the cross-validation experiment was redesigned. All methods were tested on each of the five coupons while the number of training coupons was varied from 1 to 4. Pixel accuracy and similarity index were measured for each training split. FIG. 5 presents the average accuracy and similarity scores as a function of the number of training coupons for the various detection methods. The error bars represent the standard deviation for each split. These plots demonstrate that the proposed method outperforms the comparative methods in terms of accuracy and similarity scores, even with fewer training samples.

[0079] Corrosion and detection thereof represent a significant challenge that, if overcome, could bring significant benefits to oil, aerospace, maritime, and other industries. Techniques that employ infrared thermography (IRT) measurement systems could yield practical and effective methods to provide early identification of invisible subsurface and surface corrosion defects in a variety of materials. To date, multiple pattern recognition and artificial intelligence (Al) methods have been employed to analyze IRT spatiotemporal data in order to differentiate measurements pertaining to corroded regions from measurements pertaining to non-corroded regions. Building robust enough detectors such that the detection rate is increased / maximized while the false detection rate is reduced / minimized using standard Al methods, is challenging due to the volume of training data necessary for these models to not be confused by the variety in each signal class (corroded vs. non-corroded). One way to build such technology is to coherently combine physics and mathematical modeling methods with available training data to calibrate model parameters. Using physics-based mathematical models, this need for large training sets may bereduced / overcome, and detectors may be provided that are robust to unseen situations not present in the available training data.

[0080] In the present disclosure, one such approach has been presented. In particular, an IRT system has been described (see, FIGs. 1A-1, 1A-2, 1A-3, IB, and 1C for an overview) that uses heater 101 to irradiate / heat the surface of the material to be inspected, and then measures the ensuing heat decay (FIG. 1C) using IR camera 105. Based on a mathematical model for solutions of the differential equation dictating heat in solid materials, the existence of two classes of measurements is hypothesized, with one set encompassing measurements belonging to a corroded class and another set encompassing measurements belonging to a non-corroded class. The elements of each set (corroded vs. non-corroded) are expected not to be identical to one another due to temporal variations (time warps) induced by imprecisions in calibration (e.g., measurement timing), different material properties (e.g., composition), ambient temperature, humidity, etc., all of which may affect the measured signals. A mathematical model is proposed that can describe such time variations in each set and utilize a new SCDT signal transform (see, Reference

[0025] ) to facilitate estimation of model parameters from available training data.

[0081] FIG. 5A is a graph illustrating average pixel accuracy as a function of the number of training coupons for each of the techniques discussed above with respect to Table 1 and FIGs. 4B (Proposed), 4C (FT-SVM), 4D (SVM), 4E (ID-ResNet), 4F (LSTM-FCN), and 4G (SCDT- NS). FIG. 5B is a graph illustrating similarity index as a function of the number of training coupons for each of the training techniques discussed above with respect to Table 1 and FIGs. 4B (Proposed), 4C (FT-SVM), 4D (SVM), 4E (ID-ResNet), 4F (LSTM-FCN), and 4G (SCDT-NS).

[0082] Performance of the proposed system was evaluated experimentally, using real data measured from coupons with known corroded and non-corroded regions. An approach of the present disclosure was compared to multiple existing pattern recognition methods using a cross validation technique, where a portion of the data is used for training or system calibration, and the remaining coupons are used for testing. The performance of the system is improved in terms of correct detection vs. false alarm overall. The system is also computationally in-expensive to compute, and results also show that it performs well even when small amounts of training data are used for training.

[0083] While these results are encouraging, they should be considered preliminary. The experiments demonstrated here were performed under controlled lab settings, where variations in temperature, humidity, air flow, etc. are small. Moreover, the material property differences presented in the coupons investigated are also small. As such, it is important to note that it is not possible to conclude that as is, the proposed system, could perform as well in more realistic scenarios where variations in experimental conditions could be much larger. Future work will include expanding experimental capabilities to encompass variations that are closer to what can be expected in the ‘real world’ (as opposed to controlled lab settings) and designing calibration procedures to enable the method to handle such variations, should such calibration procedures be necessary.

[0084] FIG. 6A is a schematic diagram illustrating a corrosion detection system according to some embodiments of inventive concepts that may be implemented, for example, as an autonomous rover using drive system 607 to propel the corrosion detection system over a large metal surface 600 such as the deck of a ship. Drive system 607, for example, may be coupled with traction elements 621 (e.g., wheels, tracks, etc.) to propel the corrosion detection system. The corrosion detection also includes heater 601, controller 603, IR camera system 605, GPS receiver 609, and IR camera system 615. In addition, the corrosion detection system may include marking system 617 and / or output port 610.

[0085] When the corrosion detection system is propelled across surface 600 in direction 650, heater 601 heats a portion of surface 600 before that portion of surface 600 passes through the field of view FOV of IR camera system 605. IR camera system 605 can thus capture a plurality of sequential temperature measurements for each of the plurality of test regions TR as discussed in greater detail with respect to FIGs. 6B, 6C-1, 6C-2, 6C-3, 6C-4, 6C-5, and 6C-6.

[0086] FIG. 6B illustrates surface 600 with test areas TAX(each including respective test regions TRx,i to TRx,y) in a path of the corrosion detection system of FIG. 6A, and FIGs. 6C-1 to 6C-6 illustrate mappings of test regions TR to respective images areas IAI.I to IAm.y of IR camera system 605 as the field of view FOV of IR camera system 605 moves across test area TAi. For example, IR camera system 605 may include a semiconductor detector including an array of pixels, with each image area IA defined by one or a plurality / group of the pixels, and the field of viewFOV from surface 600 is focused onto the image areas IA. Accordingly, each row of image areas IA matches a respective row of test regions TR, but the number m of columns of image areas IA may be less that the number y of columns of test regions so that the system can provide corrosion detection over an essentially infinite length of surface 600 (e.g., the deck of a ship). Moreover, the number of columns of image areas IA can be increased by using multiple IR cameras in IR camera system 605, so that each IR camera includes a separate semiconductor detector chip with some image areas lAk.i to IAk,yprovided by one semiconductor detector chip and some other image areas IAk+i,i to IAm,yprovided by another semiconductor detector chip, with k<m. Each of FIGs. 6C-1 to 6C-6 thus illustrates different test regions TR mapped to the same image areas IA of IR camera system 605 as the field of view FOV advances across surface 600.

[0087] As previously noted, heater 601 heats portions surface 600 before those portions of surface 600 enter into the field of view FOV of IR camera system 605 as the system moves across surface 600 in direction 650. In FIG. 6C-1, image areas IAi,i to IAi,yprovide respective first temperature measurements for test regions TRi,i to TRi,y, and these measurements are recorded by controller 603. In FIG. 6C-2, image areas IA2.1 to IA2,yprovide respective second temperature measurements for test regions TRi,i to TRi,y, image areas IAi,i to IAi,yprovide respective first temperature measurements for test regions TR2,I to TR2.y, and these temperature measurements are recorded by controller 603.

[0088] In FIG. 6C-3, image areas IA3.1 to IA3,yprovide respective third temperature measurements for test regions TR1.1 to TRi,y, image areas IA2.1 to IA2,yprovide respective second temperature measurements for test regions TR2.1 to TR2,y, and image areas IAi,i to IAi.yprovide respective first temperature measurements for test regions TR34 to TR3,y, and these temperature measurements are recorded by controller 603.

[0089] In FIG. 6C-4, image areas IA4,I to IA4,yprovide respective fourth temperature measurements for test regions TRi,i to TRi,y, image areas LAM to IA3.V provide respective third temperature measurements for test regions TR2.1 to TRzy, image areas IA2.1 to IA2,yprovide respective second temperature measurements for test regions TRa,i to TRa,y, and image areas IAi,i to IAi,yprovide respective first temperature measurements for test regions TR4,I to TR4.y, and these temperature measurements are recorded by controller 603.

[0090] In FIG. 6C-5, image areas IAs,i to IAS.V provide respective fifth temperature measurements for test regions TRi.i to TRi,y, image areas IA44 to IA4,yprovide respective fourth temperature measurements for test regions TR2.1 to TR2,y, image areas I A3.1 to IA3.V provide respective third temperature measurements for test regions TR?,i to TRa.y, image areas IA2.1 to IA2.V provide respective second temperature measurements for test regions TR4,I to TR4.y, and image areas IAi,i to IAi,yprovide respective first temperature measurements for test regions TR5 1 to TRs.y, and these temperature measurements are recorded by controller 603.

[0091] In FIG. 6C-6, image areas IAe,i to IAe,y provide respective sixth temperature measurements for test regions TRi,i to TRi,y, image areas IAS.I to IAs,yprovide respective fifth temperature measurements for test regions TR.21 to TR2.y, image areas IA4.1 to IA4,yprovide respective fourth temperature measurements for test regions TR3.1 to TRs.y, image areas IA.3,1 to lAs.y provide respective third temperature measurements for test regions TR4,I to TR4,y, image areas IA24 to IA2.V provide respective second temperature measurements for test regions TR5 1 to TRs.y, and image areas IA1.1 to IAi,yprovide respective first temperature measurements for test regions TRe.i to TRe.y, and these temperature measurements are recorded by controller 603.

[0092] Operations of FIGs. 6C-1 to 6C-6 may be continued until each test region TR has passed through field of view FOV resulting in m temperature measurements being recorded by controller 603 for each test region. The m temperature measurements for a test region TR can thus provide a temperature decay for that test region, and controller 603 may thus use these temperature measurements for each test region TR to determine whether that test region is corroded or non-corroded using a cumulative distribution transform (e.g., a signed cumulative distribution transform) as discussed above. Regarding test region TR14, for example, m sequential temperature measurements are provided respectively by image areas IAi,i to IAm.i and recorded by controller 603 to provide a plurality of m sequential temperature measurements representing a thermal decay of test region TRi,i. Controller 603 can then perform a cumulative distribution transform (e.g., a signed cumulative distribution transform) on the plurality of m sequential temperature measurements to provide a cumulative distribution transform result for test region TRi,i, and controller 603 can use the cumulative distribution transform result to determine whether test region TR1.1 is corroded or non-corroded. Similar operations are performed using m sequentialtemperature measurements provided by image areas IA1.2 to IAm,2 for test region TR1.2, and using m sequential temperature measurements provided by image areas IAi,yto IAm.yfor test region TRl,y.

[0093] FIG. 7 is a flow chart illustrating operations of the system of FIG. 6A as the system of FIG. 6A moves across surface 600 including test regions TRi,i to TRx.y. Operations of FIG. 7 may be performed, for example, as drive system 607 powers traction elements 621 to move the system autonomously in response to controller 603 and / or GPS Receiver 609. According to some other embodiments, the system of FIG. 7 may be pushed and / or steered manually.

[0094] At block 703, IR camera system 615 provides a background temperature of surface 600, and the background temperature is recorded by controller 603. According to some embodiments, the background temperature may include separate background temperatures for respective test regions TRi,i to TRi,yof test area TAi. According to some other embodiments, the background temperature may be a representative background temperature for multiple test regions of one or more test areas.

[0095] At block 705, test area TAi of surface 600 is heated using heater 601 as heater 601 moves over test area TAi including a plurality of test regions TRi,i to TRi,y. At block 709, heater 601 is moved away from test area TAi. At block 715, after moving heater 601 away from test area TAi, IR camera system 605 provides and controller 603 records a respective plurality of sequential temperature measurements for each of the plurality of test regions TRi,i to TRi,yas discussed above with respect to FIGs. 6C-1 to 6C-6, where the respective plurality of sequential temperature measurements for each of test regions TRi,i to TRi,yrepresents a thermal decay of the respective test region.

[0096] At block 719, controller 603 performs a cumulative distribution transform (e.g., a signed cumulative distribution transform) on the plurality of sequential temperature measurements for each of test regions TRi,i to TRi,yto provide a cumulative distribution transform result for each of test regions TR1.1 to TRi.y. The cumulative distribution transform may be a signed cumulative distribution transform, in which case, each cumulative distribution transform result may be a signed cumulative distribution transform result.

[0097] At block 725, controller 603 determines whether each of test regions TRi,i to TRi,yis a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region and based on the background temperature of the surface.

[0098] If the background temperature includes separate background temperatures for respective test regions TRi,i to TRi,y, whether each test region is a corroded region or a noncorroded region is determined based on the respective cumulative distribution transform result for the test region and based on the respective separate background temperature of the test region.

[0099] If the background temperature is a representative background temperature for all of test regions TRi.i to TRi,y, whether each test region (TR) is a corroded region or a non-corroded region is determined based on the respective cumulative distribution transform result for the test region and based on the representative background temperature.

[0100] At block 729, responsive to determining that test regions TRi.i is corroded, controller 603 identifies the location of the corroded test region on surface 600. According to some embodiments, controller 603 may identify locations of all corroded test regions based on information from a global positioning receiver (609). For example, controller 603 may save GPS information for locations of all corroded test regions that are detected, and a list of such GPS information may be transferred via wired / wireless output 610 to an output device (e.g., to a computer, printer, etc.). According to some other embodiment, controller 603 may identify locations of all corroded test regions by controlling marking system 617 to mark locations of all corroded test regions that are detected on surface 600. Marking system 617 may mark surface 600, for example, with paint.

[0101] Operations discussed above with respect to FIG. 7 may be repeated for each of test regions TRi.i to TRx.yin a path of the corrosion detection system.

[0102] According to some other embodiments of inventive concepts, the corrosion detection system may stop to test an area of surface 600 within the field of view FOV of IR camera system 602, so that each of the plurality of sequential temperature measurements for a test region TR is recorded using information from a same image area IA. Once the pluralities of sequential temperature measurements have been recorded for all of the test areas within the field of viewFOV, the corrosion detection system can move so that the field of view FOV covers a new test area.

[0103] A listing of References cited in the foregoing disclosure is provided below, and the disclosures of these References are hereby incorporated herein in their entireties by reference.

[0104] Reference [1]. FAYOMI, O.S.I., et al., “Economic impact of corrosion in oil sectors and prevention: An overview,” Journal of Physics: Conference Series, Vol. 1378, Issue 2, 9 pages (2019) 022037.

[0105] Reference [2], CALLE, L. M., “Corrosion control in the aerospace industry,” in: Webinar: Control de la Corrosion en la Industria Aerospacial, 74 pages, September 22, 2020.

[0106] Reference [3], NASH, W., et al., “Deep learning corrosion detection with confidence,” npj Materials Degradation, Vol. 6, Article number 26, 13 pages, March 31, 2022.

[0107] Reference [4], VASAGAR, V., et al., “Non-destructive techniques for corrosion detection: A review,” Corrosion Engineering, Science and Technology, Vol. 59, Issue 1, pages 56-85, February 2024.

[0108] Reference [5], OHTSU, M., et al., “Principles of the Acoustic Emission (AE) Method and Signal Processing,” in: Practical acoustic emission Testing, Springer, pages 5- 34, 2016.

[0109] Reference [6], ZAKI, A., et al., “Non-destructive evaluation for corrosion monitoring in concrete: A review and capability of acoustic emission technique,” Sensors, Vol. 15, pages 19069-19101, August 5, 2015.

[0110] Reference [7], HONARVAR, F., et al., “Ultrasonic monitoring of erosion / corrosion thinning rates in industrial piping systems,” Ultrasonics, Vol. 53, Issue 7, pages 1251-1258, September 2013.

[0111] Reference [8], SHAH, J. K., et al., “Ultrasonic monitoring of corroding bolted joints,” Engineering Failure Analysis, Vol. 102, pages 7-19, April 12, 2019.

[0112] Reference [9], MARCANTONIO, V., et al., “Ultrasonic waves for materials evaluation in fatigue, thermal and corrosion damage: A review,” Mechanical Systems and Signal Processing, Vol. 120, pages 32-42, April 1, 2019.

[0113] Reference

[0010] , PRIYADA, P., et al., “Intercomparison of gamma scattering, gammatography, and radiography techniques for mild steel nonuniform corrosion detection,” Review of scientific instruments, Vol. 82, pages 035115-1 to 035115-8, March 18, 2011.

[0114] Reference

[0011] , JAMSHIDI, et al., “Simulation of corrosion detection inside wellbore by x-ray backscatter radiography,” Applied Radiation and Isotopes, Vol. 145, pages 116-119, March 2019.

[0115] Reference

[0012] , EDALATI, K., et al., “The use of radiography for thickness measurement and corrosion monitoring in pipes,” International journal of pressure vessels and piping, Vol. 83, Issue 10, pages 736-741, October 2006.

[0116] Reference

[0013] , CHUNG, L, et al., “Infrared thermographic technique to measure corrosion in reinforcing bar,” Key engineering materials, Vols. 321-323, pages 821-824, October 2006.

[0117] Reference

[0014] , GRINZATO, E., et al., “Hidden corrosion detection in thick metallic components by transient IR thermography,” Infrared physics & technology, Vol. 49, Issue 3, pages 234-238, January 2007.

[0118] Reference

[0015] , MEOLA, C., et al., “Recent advances in the use of infrared thermography,” Measurement science and technology, Vol. 15, No. 9, pages R27-R58, July 23, 2004.

[0119] Reference

[0016] , OMAR, T., et al., “Infrared thermography model for automated detection of delamination in rc bridge decks,” Construction and Building Materials, Vol. 168, pages 313-327 April 20, 2018.

[0120] Reference

[0017] , BAGAVATHIAPPAN, S., et al., “Infrared thermography for condition monitoring-a review,” Infrared Physics & Technology, Vol. 60, pages 35-55, September 2013.

[0121] Reference

[0018] , GOFFIN, B., et al., “Use of infrared thermal imaging to detect corrosion of epoxy coated and uncoated rebar in concrete,” Construction and Building Materials, Vol. 263, 12 pages, July 23, 2020, 120162.

[0122] Reference

[0019] , DOSHVARPASSAND, S., et al., “An overview of corrosion defect characterization using active infrared thermography,” Infrared physics & technology, Vol. 96, pages 366-389, January 2019.

[0123] Reference

[0020] , WICKER, M., et al., “Detection of hidden corrosion in metal roofing shingles utilizing infrared thermography,” Journal of Building Engineering, Vol. 20, pages 201-207, July 2018.

[0124] Reference

[0021] , SAKAGAMI, T., et al., “Nondestructive detection of corrosion damage under corrosion protection coating using infrared thermography and terahertz imaging,” in: Proceedings of 13th International Workshop on Advanced Infrared Technology & Applications, pages 229-233, September 2015.

[0125] Reference

[0022] , KOPF, L., et al., “Thermographic identification of hidden corrosion,” in: 2021 36th International Conference on Image and Vision Computing New Zealand (IVCNZ), IEEE, 6 pages, 2021.

[0126] Reference

[0023] , LIM, H. J., et al., “Steel bridge corrosion inspection with combined vision and thermographic images,” Structural Health Monitoring, Vol. 20, Issue 6, pages 3424-3435, February 2021.

[0127] Reference

[0024] , WEGAND, J., et al., “Status of high temperature resistant thermal spray nonskid coating within the U.S. Navy,” in: Proceedings of NACE International, Washington, DC, USA, Paper No. 2015-6478, 16 pages, November 2015.

[0128] Reference

[0025] , ALDROUBI, A., et al., “The signed cumulative distribution transform for 1-D signal analysis and classification,” Foundations of Data Science, Vol. 4, Issue 1, 29 pages, March 2022.

[0129] Reference

[0026] , ALMOND, D. P., et al., “An analytical study of the pulsed thermography defect detection limit,” Journal of applied physics, Vol. I l l, pages 093510-1 to 093510-9, May 3, 2012.

[0130] Reference

[0027] . SAINTEY, M., et al., “Defect sizing by transient thermography. II. A numerical treatment,” Journal of Physics D: Applied Physics, Vol. 28, pages 2539-2546, September 1995.

[0131] Reference

[0028] , RUBAIYAT, A. H. M., et al., “Data-driven identification of parametric governing equations of dynamical systems using the signed, cumulative distribution transform,” Computer Methods in Applied Mechanics and Engineering, Vol. 422, 23 pages, March 2024, 116822.

[0132] Reference

[0029] , RUBAIYAT, A.H.M., et al., “End-to-end signal classification in signed cumulative distribution transform space,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 46, Issue 9, pages 5937-5950, March 1, 2024.

[0133] Reference

[0030] , SHIFAT-E-RABBI, M., et al., “Radon cumulative distribution transform subspace modeling for image classification,” Journal of mathematical imaging and vision, Vol. 63, No. 9, pages 1185-1203, 2021.

[0134] Reference

[0031] , PARK, S.R., et al., “The cumulative distribution transform and linear pattern recognition,” Applied and computational harmonic analysis, Vol. 45, Issue 3, pages 616-641, November 2018.

[0135] Reference

[0032] , RUBAIYAT, A. H. M., et al., “Nearest subspace search in the signed cumulative distribution transform space for Id signal classification,” in: ICASSP 2022- 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, pages 3508-3512, May 2022.

[0136] Reference

[0033] , HEARST, M.A., et al., “Support vector machines,” IEEE Intelligent Systems and their applications, Vol. 13, pages 18-28, 1998.

[0137] Reference

[0034] , FAW AZ, H.I., et al., “Data augmentation using synthetic data for time series classification with deep residual networks,” arXiv preprint arXiv: 1808.02455 (2018).

[0138] Reference

[0035] , IWANA, B.K., et al., “An empirical survey of data augmentation for time series classification with neural networks,” Pios One, Vol. 16, No. 7, 32 pages, July 15, 2021, e0254841.

[0139] Reference

[0036] , KARIM, F., et al., “LSTM fully convolutional networks for time series classification,” IEEE access, Vol. 6, pages 1662-1669, December 4, 2017.

[0140] Reference

[0037] , WEGAND, J., et al., “Status of High Temperature Resistant Thermal Spray Nonskid Coating Within the U.S. Navy,” Paper No. 2015-6478, 2015 DoD-Allied Nations Technical Corrosion Conference, Pittsburg, PA, pages 16 pages, November 2019.

[0141] Reference

[0038] , MORGAN, J.S., et al., “Corrosion detection from IR thermal images in signed cumulative distribution transform domain,” NDT&E International, Vol. 154, 8 pages, 2025, 103390.

[0142] Reference

[0039] , Adams, M. I., et al., “Method of and Apparatus for Detecting Corrosion Utilizing Infrared Analysis,” U.S. Patent No. 4,647,220, Issued March 3, 1987.

[0143] Reference

[0040] , Hutchinson, M. N., et al., “Methods, Apparatuses and Systems for Time Delay Estimation,” U.S. Patent No. 11,290,212, Issued March 29, 2022.

[0144] Additional disclosure is provided below.

[0145] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a,” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and / or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. The term "and / or" includes any and all combinations of one or more of the associated listed items.

[0146] Spatially relative terms, such as “beneath,” “below,” “lower,” “above,” “upper,” and the like, may be used herein for ease of description to describe one element's or feature's relationship to another element(s) or feature(s) as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if thedevice in the figures is turned over, elements described as “below” or “beneath” other elements or features would then be oriented “above” the other elements or features. Thus, the term “below” can encompass both an orientation of above and below. The device may be otherwise oriented (rotated 90 degrees or at other orientations), and the spatially relative descriptors used herein may be interpreted accordingly.

[0147] It will be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. Thus, a first element discussed herein could be termed a second element without departing from the scope of the present inventive concepts.

[0148] It will also be understood that when an element is referred to as being “on”, “connected” to / with, or “coupled” to / with another element, it can be directly on, connected to / with, or coupled to / with the other element, or intervening elements may be present. In contrast, when an element is referred to as being “directly on”, “directly connected” to / with, or “directly coupled” to / with another element, there are no intervening elements present. Similarly, when an operation / element is referred to as being “responsive to” or “in response to” another event / operation / element, it can be directly responsive to or directly in response to the other operation / element or intervening events / operations / elements may be present. In contrast, when an operation / element is referred to as being “directly responsive to” or “directly in response to” another event / operation / element, there are no intervening events / operations / elements present. Moreover, if an element is referred to as being “on” another element, no spatial orientation is implied such that the element can be over the other element, under the other element, on a side of the other element, etc.

[0149] The operations of any methods disclosed herein do not have to be performed in the exact order disclosed, unless an operation is explicitly described as following or preceding another operation and / or where it is implicit that an operation must follow or precede another operation. Any feature of any of the embodiments disclosed herein may be applied to any other embodiment, wherever appropriate. Likewise, any advantage of any of the embodiments mayapply to any other embodiments, and vice versa. Other objectives, features and advantages of the enclosed embodiments will be apparent from the description herein.

[0150] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which inventive concepts herein belong. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0151] While inventive concepts have been particularly shown and described with reference to examples of embodiments thereof, it will be understood that various changes in form and details may be made therein without departing from the spirit of the following claims.

Claims

CLAIMS:

1. A method of detecting corrosion on a surface, the method comprising: heating a test area of the surface using a heater, wherein the test area includes a plurality of test regions; after heating the test area, moving the heater away from the test area; after moving the heater away from the test area, recording a respective plurality of sequential temperature measurements for each of the plurality of test regions, wherein the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region; for each of the plurality of test regions, performing a cumulative distribution transform on the plurality of sequential temperature measurements for the respective test region to provide a cumulative distribution transform result for each of the respective test regions; and for each of the test regions, determining whether the test region is a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region.

2. The method according to Claim 1, wherein performing further comprises performing a signed cumulative distribution transform on the plurality of sequential temperature measurements to provide a signed cumulative distribution transform result for each of the respective test regions, and wherein determining further comprises determining whether the test region is a corroded region or a non-corroded region based on the respective signed cumulative distribution transform result for the test region.

3. The method according to Claim 1, wherein recording comprises recording the respective plurality of sequential temperature measurements for each of the plurality of test regions based on output from an infrared camera system defining a pixel array, and wherein each of the temperature measurements is based on output from an image area of the pixel array.

4. The method according to Claim 3, wherein the IR camera system moves relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of test regions, and wherein for each of the test regions, at least two of the plurality of sequential temperature measurements are based on output from different image areas of the pixel array.

5. The method according to Claim 4, wherein the pixel array comprises a first pixel subarray provided using a first semiconductor detector chip and a second pixel subarray provided using a second semiconductor detector chip, and wherein the different image areas are from different ones of the first and second pixel subarrays.

6. The method according to Claim 3, wherein the IR camera system is stationary relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of test regions, and wherein the sequential temperature measurements for a respective test region are based on output from a respective image area of the pixel array.

7. The method according to Claim 1, wherein the plurality of test regions includes first and second test regions, wherein recording comprises recording a first plurality of sequential temperature measurements representing a thermal decay of the first test region and recording a second plurality of sequential temperature measurements representing a thermal decay of the second test region, wherein performing the cumulative distribution transform comprises performing the cumulative distribution transform on the first plurality of sequential temperature measurements to provide a first cumulative distribution transform result for the first test region and performingthe cumulative distribution transform on the second plurality of sequential temperature measurements to provide a second cumulative distribution result for the second test region, and wherein determining comprises determining whether the first test region is a corroded region or a non-corroded region based on the first cumulative distribution result and determining whether the second test region is a corroded region or a non-corroded region based on the second cumulative distribution result.

8. The method according to Claim 7 further comprising: responsive to determining that the test region is a corroded region, identifying the location of the first test region on the surface.

9. The method according to Claim 1 further comprising: recording a background temperature of the surface; wherein for each of the test regions, determining comprises determining whether the test region is a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region and based on the background temperature of the surface.

10. A method of detecting corrosion on a surface, the method comprising: heating a test area of the surface using a heater, wherein the test area includes a plurality of test regions; after heating the test area, moving the heater away from the test area; after moving the heater away from the test area, recording a respective plurality of sequential temperature measurements for each of the plurality of test regions based on output from an infrared camera system defining a pixel array, wherein at least two of the plurality of sequential temperature measurements for each of the test regions are based on output from different image areas the pixel array, and wherein the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region; andfor each of the test regions, determining whether the test region is a corroded region or a non-corroded region based on the respective plurality of sequential temperature measurements for the test region.

11. The method according to Claim 10, wherein the IR camera system moves relative to the test area while recording the respective plurality of sequential temperature measurements for each of the plurality of test regions.

12. The method according to Claim 10, wherein the pixel array comprises a first pixel subarray provided using a first semiconductor detector chip and a second pixel subarray provided using a second semiconductor detector chip, and wherein the different image areas of the pixel array are from different ones of the first and second pixel subarrays.

13. The method according to Claim 10 further comprising: for each of the plurality of test regions, performing a cumulative distribution transform on the plurality of sequential temperature measurements to provide a cumulative distribution transform result for each of the respective test regions; wherein determining comprises determining for each of the test regions whether the test region is a corroded region or a non-corroded region based on the cumulative distribution transform result for the respective test region.

14. The method according to Claim 10, wherein the plurality of test regions includes first and second test regions, wherein recording comprises recording a first plurality of sequential temperature measurements representing a thermal decay of the first test region and recording a second plurality of sequential temperature measurements representing a thermal decay of the second test region, andwherein determining comprises determining whether the first test region is a corroded region or a non-corroded region based on the first plurality of sequential temperature measurements and determining whether the second test region is a corroded region or a noncorroded region based on the second plurality of sequential temperature measurements.

15. The method according to Claim 14 further comprising: responsive to determining that the first test region is a corroded region, identifying the location of the first test region on the surface.

16. The method according to Claim 15 wherein identifying the location comprises identifying the location of the first test region on the surface based on information from a global positioning receiver.

17. The method according to Claim 15 wherein identifying the location comprises marking the location of the first test region on the surface.

18. The method according to Claim 10 further comprising: recording a background temperature of the surface; wherein for each of the test regions, determining comprises determining whether the test region is a corroded region or a non-corroded region based on the plurality of sequential temperature measurements for the test region and based on the background temperature of the surface.

19. A detection system to detect corrosion on a surface, the detection system comprising: a heater configured to heat a test area of the surface, wherein the test area includes a plurality of test regions; an infrared camera system comprising an array of pixels defining a plurality of image areas configured to receive thermal energy emitted from the surface within a field of view;a drive system configured to propel the detection system so that the heater moves away from the test area and so that the test area is within the field of view of the IR camera system; and a controller coupled with the IR camera system, wherein the controller is configured to, record a respective plurality of sequential temperature measurements for each of the plurality of test regions from the IR camera system after the test area is within the field of view, wherein the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region, for each of the plurality of test regions, perform a cumulative distribution transform on the plurality of sequential temperature measurements for the respective test region to provide a cumulative distribution transform result for each of the respective test regions, and for each of the test regions, determine whether the test region is a corroded region or a non-corroded region based on the respective cumulative distribution transform result for the test region.

20. A detection system to detect corrosion on a surface, the detection system comprising: a heater configured to heat a test area of the surface, wherein the test area includes a plurality of test regions; an infrared camera system comprising an array of pixels defining a plurality of image areas configured to receive thermal energy emitted from the surface within a field of view; a drive system configured to propel the detection system so that the heater moves away from the test area and so that the test area is within the field of view of the IR camera system; and a controller coupled with the IR camera system, wherein the controller is configured to, after moving the heater away from the test area, record a respective plurality of sequential temperature measurements for each of the plurality of test regions based on output from the infrared camera system, wherein at least two of the plurality of sequential temperature measurements for each of the test regions are based on output from different image areas thepixel array, and wherein the respective plurality of sequential temperature measurements for each of the plurality of test regions represents a thermal decay of the respective test region, and for each of the test regions, determining whether the test region is a corroded region or a non-corroded region based on the respective plurality of sequential temperature measurements for the test region.

Images