This invention generally relates to diagnostic systems, and more specifically relates to the detection of surface anomalies such as corrosion and other types of physical defects.
There are many applications where it is of critical importance to determine the amount of degradation that is or has occurred on a device. For example, in modern manufacturing techniques it can be critical for efficient manufacturing to know whether physical degradation, such as corrosion, is occurring on critical surfaces. As another example, in aircraft engines, the surface integrity of turbine blades is of extreme importance to safe and reliable operation. In all of these applications there is a strong need for effective techniques to determine if surface anomalies have formed, or are being formed on a device.
One particularly serious type of surface anomaly is pitting corrosion. Generally, pitting is a localized form of corrosion that is characterized by the formation of holes or pits on the surface. Pits often have a small surface area, and can thus be difficult to accurately detect or characterize. However, pits can penetrate to a relatively large depth and can thus cause severe failures in some applications.
Additionally, in some applications there is also a strong need to determine the size and rate of the growth of surface anomalies. For example, determining the rate of pitting corrosion growth can be used to calculate the structural integrity and hence the remaining component lifetime, a process generally referred to as component lifing. Without an accurate determination of the rate of anomaly growth may not be possible to determine the remaining component life.
Unfortunately, previous techniques for detecting surface anomalies in general, and pitting corrosion in particular, have had limited usefulness because they typically require an expert evaluation to interpret the data. This prevents, for example, the system from being used in an automatic feedback control loop. Thus, what is needed is a robust automated system and method for detecting and characterizing pitting corrosion and other surface anomalies.
The embodiments of the present invention provide a surface anomaly detection system and method. The detection system and method provides the ability to automatically detect and characterize anomalies on a surface. For example, the detection system and method may be particularly applicable to detecting and characterizing pitting corrosion on metal surfaces.
The surface anomaly detection system and method uses an image of the surface under evaluation. A variety of different types of sensors and methods can be used to obtain the image. As one example technique, the image data comprises data created using a synthetic aperture focusing technique (SAFT). This technique applies guided waves to the surface under evaluation. For example, the waves are guided using a piezoelectric transducer array. In this embodiment, the guided waves are reflected back to the transducer by the physical defects in the material. The reflected waves received by each of the transducers in the array are delayed and summed on an individual frequency basis to create an image of the scanned region. The image indicates regions of high intensity that correspond to surface anomalies, and regions of low intensity that correspond to normal surfaces.
The image data is passed to a noise filter. The noise filter uses an estimated characterization of the background noise to remove the background noise from the image data. In one embodiment, the background noise is removed by using a stochastic model to estimate a smoothed value for the intensity at each pixel. In one specific implementation, the estimate from the stochastic model is implemented using a modified Kalman filter.
The smoothed pixel intensity is then compared to a threshold to identify significant departures. In one embodiment, the threshold detector uses a threshold calculated using a constant false positive rate. In this embodiment, a desired constant false positive rate is selected and used to determine the threshold value for each of the pixels in the image, and the threshold can then be used to establish the distribution of anomalies in the image data.
The detected anomalies are then passed to a contour detector, which determines the contours of anomalies in the surface to provide more precise characterization of the surface defects. In one embodiment, the detected anomalies are joined together by drawing closed contour boundaries around the anomalies. This groups anomalies together that may all be part of the same surface defect. In one specific implementation, the contour boundaries are identified using a principle curves technique that removes islands of noise from the data. This results in closed contours that encompass the detected anomalies.
The detected closed contours of anomalies may then be passed to a surface defect severity estimator. The defect severity estimator provides a metric referred to herein as an anomaly factor. Specifically, the defect estimator can estimate the severity of the surface anomaly by calculating a parameter of the closed contour. For example, the anomaly factor can be expressed as the area of the closed contour together with a center. In another embodiment, the defect severity estimator can aggregate the area of closed contours at the same location calculated over a length of time to determine the rate of change in the detected anomalies. These different characterizations of the anomalies can then be used to determine the extent of the structural damage and its likely impact on the associated device.
The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of a preferred embodiment of the invention, as illustrated in the accompanying drawings.
The preferred exemplary embodiment of the present invention will hereinafter be described in conjunction with the appended drawings, where like designations denote like elements, and:
The embodiments of the present invention provide a surface anomaly detection system and method. The detection system and method provides the ability to automatically detect and characterize surface anomalies on a surface under test. While the surface anomaly detection system and method can be used for a variety of anomalies, the system and method may be particularly applicable to detecting and characterizing pitting corrosion and material cracks.
Turning now to
The image data 110 is passed to the noise filter 102. The noise filter 102 suitably uses an estimated characterization of the background noise to remove the background noise from the image data. In one embodiment, the background noise is removed by using a stochastic model to estimate a smoothed value for the intensity at each pixel. In one specific implementation, the stochastic model is implemented in the noise filter 102 using a modified Kalman filter. The removal of the noise results in smoothed image data that can then be used to locate surface anomalies.
The smoothed image data is then passed to the threshold detector 104. The threshold detector compares each pixel in the image data to a threshold value(s) to identify significant intensity departures. In one embodiment, the threshold detector 104 uses a threshold calculated using a constant false positive rate. In this embodiment, a desired constant false positive rate is selected and used to determine the threshold value for each of the pixels in the image which can then be used establish the distribution of anomalies in the image data.
The detected anomalies are then passed to the contour detector 106, which determines the contours of anomalies in the surface to provide more precise characterization of the detected anomalies. In one embodiment, the detected anomalies are joined together by drawing closed contour boundaries around the anomalies. This technique combines anomalies together that may be all part of the same surface anomaly. In one specific implementation, the contour detector 106 defines boundaries using a principle curves technique that removes islands of noise from the data. This results in closed contours that encompass the detected anomalies.
The detected closed contours of anomalies are then passed to a defect severity estimator 108. The defect severity estimator 108 generates an anomaly factor 112. The anomaly factor 112 is a metric that can be used to characterize and quantify the severity of a defect. In one embodiment, the defect estimator 108 can generate the anomaly factor 112 by a parameter of the closed contour(s) located in the image data. For example, by calculating an area along with a center from the closed contour. In another embodiment, the defect severity estimator 108 can aggregate the area of multiple closed contours corresponding to the same surface anomaly taken at different times to determine the rate of change in the detected anomalies, and calculate the anomaly factor 112 based on the rate of change. In both cases, the characterizations of the anomalies provided by the anomaly factor 112 can then be used to determine the extent of the structural damage and its likely impact on the associated device.
As described above, the surface anomaly detection system 100 receives image data of the surface under test. In one embodiment, the image data is created using a synthetic aperture focusing technique (SAFT). This technique applies guided waves to the surface under evaluation. For example, the waves are guided using a piezoelectric transducer array. In this embodiment, the guided waves are reflected back to the transducer by the physical defects in the material. The reflected waves received by each of the transducers in the array are delayed and summed on an individual frequency basis to create an image of the scanned region. The image indicates regions of high intensity that correspond to surface anomalies, and regions of low intensity that correspond to normal surfaces. There are several other devices and approaches that can be used to induce the guided waves into the material and to receive the reflected waves. These include Electromagnetic Acoustic Transducers (EMAT), magneto-restrictive transducers and laser based systems. More information about how SAFT can be used to create the image of the surface under test can be found at Sicard, R., J. Goyette, D. Zellouf (2002) “A SAFT Algorithm for Lamb Wave Imaging of Isotropic Plate-Like Structures”. Ultrasonics, 39, 7: 487-494.
The image data can be created and stored in a variety of formats. In one suitable format, the image data is in the form of a red-green-blue (RGB) image of the surface. For example, regions of high intensity reflected waves, which are indicative of surface anomalies, can be indicated by colors with significant red components. Conversely, regions of low intensity can be indicated by colors with significant blue components. Of course, this is just one type of exemplary format in which the image data can be stored.
Turning now to
One issue with image data created using a SAFT technique is that the image data is often very noisy. In this case the high intensity regions of the image may indicate noise caused by manufacturing and sensing irregularities rather than surface anomalies. The noise filter 102 provides an automated mechanism for eliminating noise from the image data. The noise filter 102 determines the mean intensity of the image data, where the mean intensity, also referred to herein as the smoothed intensity, is defined as the intensity after the noise has been at least substantially removed. In this application, the subscript i,j is used to denote the pixel element at location i,j. The intensity of each image element (e.g., a pixel or collection of pixels) can then be modeled as:
xi,j=
where xi,j is the observed (noisy) intensity of the element at the i,j location,
Several different techniques can be used for determining and modeling the background noise component εi,j. The background noise component εi,j can be characterized using experimentation. For example, by taking multiple images of clean samples with no surface anomalies, the noise in typical images can be calculated and characterized. This characterization of the noise may then be used to implement the filter, which is then used in operation to filter image data from surfaces under examination. Additionally, several different techniques can be used to characterize the noise εi,j using multiple images.
Several specific examples of noise characterization techniques will now be discussed. In one example, no noise is assumed to exist and εi,j is thus equal to a 0, ∀i,j. In another example, a constant noise is assumed to exist and εi,j is a constant, ∀i,j. In this embodiment, 0 is valid value for the constant, implying a noise free background. In this embodiment the noise subtracted from the observed intensity to calculate the corresponding mean intensity value at location i,j. In another example, the statistical characteristics of noise is assumed to be uniform such that εi,j=N(0,σ), ∀i,j. In this notation, N stands for a Gaussian distribution with mean 0 and standard deviation σ. In this model the background noise is assumed to belong to a single distribution with the same mean and variance. Thus, the background noise is assumed to vary spatially, although at each location the distribution is well characterized. The standard deviation σi,j can be calculated using xi,j observations from multiple image data.
All of these techniques provide a mechanism for modeling the noise that can be used to filter the image data and calculate the mean intensity value
In another embodiment, the mean intensity
where qi,j(t) is a random input variable with zero mean and known variance Q, a is a time constant of the process which characterizes how fast the intensity may change with respect to time. Multiple sets of image data may be collected over a series of time interval. For example, the pixel intensity at location i,j from image collected at time t may provide the noisy observation of the hidden state at time t, with the state itself is evolving according to equation 2.
A noise filter based on a stochastic model is implemented using a modified Kalman filter. In general, a Kalman filter is a statistical estimator used to separate random noise from an underlying signal. In this approach the underlying signal (i.e., the pixel intensity
In one specific embodiment of the invention the basic Kalman filter is modified to account for the non-Gaussian nature of the random variable qi,j(t). For example, the Kalman filter is modified to account for a qi,j(t) that follows a Weibull distribution. In this embodiment the Kalman filter is modified to follow the random variable qi,j(t) up to its first two moments. For example, the Kalman filter is modified to replace qi,j(t) with another random variable pi,j(t)˜N(m, v) such that the mean (i.e., the 1st moment) of qi,j(t) is equal to the mean of pi,j(t), and the variance (i.e., the 2nd moment) of qi,j(t) is equal to the mean of pi,j(t). In this embodiment, the original Kalman filter equations are re-derived following this approximation. Thus, the Kalman filter is modified such that the underlying dynamic system is excited by non-Gaussian white noise measure, allowing the Kalman filter to obtain the mean intensity value
Implementing such a modified Kalman filter generally involves two steps. Step one is a prediction step. Given a linear dynamic model, such as the model described by equation 2, one can predict the value of the state at t+1 given its value at t. In other words, it can predict
A variety of different techniques can be used to implement a Kalman filter. For example, a series of image data from clean samples is used to identify the “a” and the covariance of the random variable in equation 2. This step can be referred to as system identification. In one example, the parameters of equation 2 are identified using a least squares minimization. In this example, the innovation error (difference between what equation 2 predicts and actual intensity observation at location i,j) is minimized subject to optimal values of “a” and “Q”.
The modified Kalman filter can thus be implemented to work as a Stochastic filter and calculate the mean value for each pixel in the data. This provides the ability to characterize a noise distribution without assuming that it has a particular type of distribution (e.g, Gaussian.) Thus, the modified Kalman filter is one mechanism that can be used to calculate the mean value for the intensity of each pixel
The smoothed image data is then passed to the threshold detector 104 to identify significant variations in the intensity of the image. In general, threshold detector compares the mean or smoothed intensity at each pixel
As one example, the baseline threshold value is set to a pre-determined constant δ. In this example, if the mean intensity
rij≡
In this case the residual ri,j is defined for each pixel as the difference between the mean intensity value
In a second embodiment, the threshold detector uses a constant false positive probability technique. In this example, experimentation using samples, commonly referred to as coupons, is used determine the distribution of the mean intensity
In a third embodiment, the threshold detector uses a constant false positive rate technique to determine the threshold value. In general, a false positive rate is defined as the number of false positives per successful detection of an anomaly. To achieve a desired false positive rate, calibration is performed using samples with varying degrees of anomalies. The calibration samples are analyzed to determine a value for the threshold that achieves the desired rate.
In this embodiment, the
As stated above, a constant false positive rate technique uses calibration results from good and corroded samples to select a threshold δ that will result in the desired rate of false positives. Stated another way, the threshold δ will be set such that the corroded residual r*i,j area to the left of the threshold δ will have a desired ratio to the area of the good residual r*i,j area to the right of threshold δ. When so calculated using sufficient numbers of known good and samples with confirmed anomalies, the threshold δ can then be used for detecting pixels with excessive intensity levels. This provides a threshold δ that minimizes the probability of false classification while maintaining a pre-defined detection probability. Thus determined, the threshold δ can be used in the threshold detector to determine when intensity values cross the threshold δ such that pixels with excessive intensity are detected.
The pixels with excessive intensity that were detected by the threshold detector are then passed to the contour detector, which determines the contours of anomalies in the surface to provide a more precise characterization of the detected anomalies. The main objective of the contour detector is to establish the boundaries of the excessive intensity pixel cluster and thus provide a less ambiguous output for the detection of anomalies. In one embodiment, the detected anomalies are joined together by drawing closed contour boundaries around the identified pixels. This combines pixels together that are all part of the same cluster or surface defect, e.g., all part of one corrosion pit. In one specific implementation, the contour detector defines boundaries using a principle curves technique that removes islands of noise from the data. This results in closed contours that encompass the detected cluster of pixels.
As one example, the input to the contour detector comprises a matrix with rows representing discrete positions along the length of the surface under investigation, while the columns representing discrete positions along the width of the surface. The contour detector takes this matrix input and outputs a smooth continuous curve along the outer rim of the excessive intensity pixel cluster.
In one specific embodiment the contour is modeled as a principal curve. A principal curve is a flexible family of one-dimensional non-parametric curves which pass through the middle of a d-dimensional probability distribution of 0 and 1 pixel intensity values generated during the threshold step. Several different techniques can be used to calculate the principal curves from a given data set of pixel intensity values. For example, one technique for calculating a principal curve is Polygon Line Algorithm. This technique starts with a straight line segment (a part of the first principal component line or the least squares regression line), and in each iteration it increases the number of line segments by adding a vertex to the polygonal line produced by the previous iteration. After adding the vertex, the position of each vertex is updated in an inner loop. Typically the technique will continue until the number of vertices exceeds a predetermined limit.
The inner loop includes of a projection step and an optimization step. In the projection step the data points are partitioned into “nearest neighbor regions” according to which segment or vertex they project. In the optimization step the new position of each vertex is determined in a line search to minimize an objective function that consists an average squared distance term and a curvature penalty. These two steps are iterated until convergence is achieved. Thus, the contour detector can use an iterative technique to draw a polygon that passes through the mean of the pixels that exceed the threshold.
Thus, the principal curve technique identifies edges clustered around the principal curve. Each identified cluster then corresponds to a possible anomaly and the corresponding principal curve is the estimated outline or contour of the anomaly. Turning now to
In this technique it will generally be desirable to define a minimum size of the anomaly that will be identified for further evaluation. Typically, the minimum size of an anomaly that will be identified needs to be translated into pixel count. For example, in one application a user may not care about anomalies that are less than 1 mm×1 mm. Based on the resolution of the image, this may correspond to n pixels. This number can thus be used to determine the minimum number of points that will be used to draw the polygon edges. This also prevents the principal curves from merely joining every pixel and generating a very noisy contour, such as a worst case polygon.
The detected closed contours of anomalies are then passed to a defect severity estimator that provides a metric referred to herein as an anomaly factor. The anomaly factor can then be used to calculate a magnitude for a surface defect under test. In one example, the anomaly factor can be used to characterize the growth rate of the anomalies by calculating the rate of change over time. As another example, the defect severity estimator can estimate the severity of the defect by calculating a parameter, such as the area, from the closed contours. These different characterizations of the anomalies can then be used to determine the extent of the surface anomaly and its likely impact on the associated device.
In one embodiment, the anomaly factor is calculated as the area enclosed by the principal curve defining the closed contour. The area can be calculated from principal curve using any suitable technique. In one example, if the principal curve is described by a n-polygon, then one can calculate the area of the circumscribing circle, where the circumscribing circle connects every vertex. In a second example, if the principal curve is described by a n-polygon, then one can calculate the area of the circum-inscribing circle, where the circum-inscribing circle is inside the polygon and is a tangent to every edge of the polygon.
In another embodiment, the principal curve is mapped as a regular geometric shape and the anomaly factor is selected to be a parameter of that shape. For example, the principal curve can be mapped to a circle and the diameter used as the anomaly factor.
In all these cases, the anomaly factor is selected to characterize the anomaly and provide a metric that can be used to evaluate the detected anomaly. Furthermore, multiple examinations of a surface under test can be used to trend the anomaly factor as a function of time. Such a time-series plot can be used to provide an end user with a quick summary of how the anomaly is progressing and thus can assist in determining the potential effects of future corrosion.
The surface anomaly detection system and method can be implemented using a variety of different platforms. Turning now to
The processor 110 performs the computation and control functions of the system 50. The processor 110 may comprise any type of processor, including single integrated circuits such as a microprocessor, or may comprise any suitable number of integrated circuit devices and/or circuit boards working in cooperation to accomplish the functions of a processing unit. In addition, processor 110 may comprise multiple processors implemented on separate systems. In addition, the processor 110 may be part of an overall vehicle control, navigation, avionics, communication or diagnostic system. During operation, the processor 110 executes the programs contained within memory 180 and as such, controls the general operation of the computer system 50.
Memory 180 can be any type of suitable memory. This would include the various types of dynamic random access memory (DRAM) such as SDRAM, the various types of static RAM (SRAM), and the various types of non-volatile memory (PROM, EPROM, and flash). It should be understood that memory 180 may be a single type of memory component, or it may be composed of many different types of memory components. In addition, the memory 180 and the processor 110 may be distributed across several different computers that collectively comprise system 50.
The bus 170 serves to transmit programs, data, status and other information or signals between the various components of system 100. The bus 170 can be any suitable physical or logical means of connecting computer systems and components. This includes, but is not limited to, direct hard-wired connections, fiber optics, infrared and wireless bus technologies.
The interface 130 allows communication to the system 50, and can be implemented using any suitable method and apparatus. It can include a network interfaces to communicate to other systems, terminal interfaces to communicate with technicians, and storage interfaces to connect to storage apparatuses such as storage device 190. Storage device 190 can be any suitable type of storage apparatus, including direct access storage devices such as hard disk drives, flash systems, floppy disk drives and optical disk drives. As shown in
In accordance with the preferred embodiments of the invention, the computer system 50 includes the surface anomaly detection program. Specifically during operation, the surface anomaly detection program is stored in memory 180 and executed by processor 110. When being executed by the processor 110, the surface anomaly detection program receives image data of surfaces under test and generates an anomaly factor that is indicative the surface anomalies on the surface under test.
It should be understood that while the present invention is described here in the context of a fully functioning computer system, those skilled in the art will recognize that the mechanisms of the present invention are capable of being distributed as a program product in a variety of forms, and that the present invention applies equally regardless of the particular type of computer-readable medium used to carry out the distribution. Examples of computer-readable medium include: recordable media such as floppy disks, hard drives, memory cards and optical disks (e.g., disk 195).
The embodiments of the present invention thus provide a surface anomaly detection system and method. The surface anomaly detection system and method provides the ability to automatically detect and characterize corrosion and other anomalies on a surface. The embodiments and examples set forth herein were presented in order to best explain the present invention and its particular application and to thereby enable those skilled in the art to make and use the invention. However, those skilled in the art will recognize that the foregoing description and examples have been presented for the purposes of illustration and example only. The description as set forth is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching without departing from the spirit of the forthcoming claims.
Number | Name | Date | Kind |
---|---|---|---|
4484081 | Cornyn, Jr. et al. | Nov 1984 | A |
4603584 | Bartle et al. | Aug 1986 | A |
5442462 | Guissin | Aug 1995 | A |
5549002 | Chiao et al. | Aug 1996 | A |
5571966 | Tsuboi | Nov 1996 | A |
5648613 | Kiefer | Jul 1997 | A |
5698787 | Parzuchowski et al. | Dec 1997 | A |
5760904 | Lorraine et al. | Jun 1998 | A |
6038335 | Yokoyama et al. | Mar 2000 | A |
6128092 | Levesque et al. | Oct 2000 | A |
6182512 | Lorraine | Feb 2001 | B1 |
6245016 | Daft et al. | Jun 2001 | B1 |
6285449 | Ellingson et al. | Sep 2001 | B1 |
6487909 | Harrold et al. | Dec 2002 | B2 |
6490927 | Braunling et al. | Dec 2002 | B2 |
6592523 | Avinash et al. | Jul 2003 | B2 |
6606909 | Dubois et al. | Aug 2003 | B2 |
6771807 | Coulombe et al. | Aug 2004 | B2 |
6775400 | Zhao et al. | Aug 2004 | B1 |
6792357 | Menon et al. | Sep 2004 | B2 |
6831469 | Foreman et al. | Dec 2004 | B2 |
6870950 | Houge et al. | Mar 2005 | B2 |
6874365 | Deveney et al. | Apr 2005 | B2 |
7286964 | Kim | Oct 2007 | B2 |
Number | Date | Country | |
---|---|---|---|
20070098245 A1 | May 2007 | US |