by Michael L. Peterson and M.E. Ellis*

 

Advances in the widespread applications of nondestructive testing (NDT) techniques may be achieved through new paradigms of visualization. This paper represent a fusion of advanced visualization technologies with NDT for increased accuracy in interpretation of ultrasonic C-scan data. An example application of visualization and ultrasonic anomaly detection in truck tire casings will be shown. G.P. Singh Associate Contributing Editor

Motivation
This project was motivated by a particular application, the detection of anomalies in truck tire casings. However, the general approach signals significant new opportunities. These opportunities result from the availability of high end graphics capabilities in low cost personal computers. Transferring an existing base of NDT software to more advanced graphical programming interfaces will result in significant progress in the application of NDT. An immediate advantage of transferring software will be a reduction in the rate of errors in the interpretation of NDT data. The improvement in the accuracy of testing is likely both for highly trained operators in a repetitive environment and for less well trained operators who may perform testing in some high volume industrial applications.

 

C-Scan Interpretation
While appropriate use of color mapping in the C-scan facilitates the interpretation of data, the association of particular locations on the image with similar locations on the part may be difficult. The difficulty increases based on the complexity of the part geometry. Without matching geometry to the color map, errors in interpretation are likely. For example, discontinuities often occur in potentially critical areas of a part such as where sharp transitions occur on the physical geometry. Association of the data’s position with respect to these edges is often critical to the evaluation of the part’s integrity. Misinterpretation of C-scans may occur when assumptions not based on calculated positions are made regarding the data/geometry relationship.

---

Viewing internal damage in composites is one application."

---

Once an anomaly has been located by inspection personnel, either visually or through an anomaly detection routine, the significance of the anomaly and the physical location must be determined. When viewing a flat C-scan, variation in data spacing due to object geometry is likely to confuse the interpretation of the image. Mapping of geographical data is the most common example of the challenge of mapping a surface to a flat projection. This fundamental application is still in the process of development of particular projections to reduce this misorientation in geographical mapping (Snyder, 1987). However, these mappings are geometrically dependent and are not likely to be developed for an arbitrary part geometry. Similar to geographical mappings, the distance between locations (anomalies instead of cities) may be critical to the use of the mapping. Distance between anomalies and their stress influence may control the interpretation of the part integrity.

Unlike the flat C-scan, a surface model maps data points in direct proportion to the point of acquisition. This allows better judgments as to the source and implications of the data. The entire dataset may be viewed through combining the surface model with a user interface that enables rotation of the model to an arbitrary orientation. Current technological advances expedite the building of applicable software, as explained below. It has only recently become possible for the reorientation to occur at real time speeds using low cost hardware. Applications of surface models abound; having C-scan information tied directly to surface geometry is applicable to mapping internal discontinuities associated with stress concentrations or mapping potential manufacturing discontinuities to external changes in the coloring scheme.

There are several key ingredients to understanding the data mapped on a surface model. Arbitrary orientation of the component on a 2D display screen assists in forming a mental picture of the 3D object. Allowing the operator to rotate the surface model expedites locating any anomalies and recognizing their position on the actual part relative to a "home" position. As an example, the required Department of Transportation identification number is used to orient the tire casings. Another feature vital to interpretation is interactive zoom in on regions of interest, facilitating close up views of specific data regions. In order to have precise feedback as to the location and dimensions of these data regions, routines are included to give the operator feedback as to the location of the mouse pointer in the part geometry coordinates (for arbitrary views of the model). This allows the operator to make measurements to find the anomaly location and take appropriate measures based on the nature of the anomaly. The location information also provides the potential for feedback to the scanning system for automated marking of discontinuities on the part.

Advanced visualization software developed in a framework such as OpenGL also provides the potential to provide interior views of an object. Transparent objects provide a viewing perspective where internal data from computed tomography or layered ultrasonic C-scans may be displayed. Transparent surfaces provide the possibility of effectively encoding large quantities of 3D data onto a 2D display. Viewing internal damage in composites is an example application. If a volume anomaly in the middle of the data region resulted in an ultrasonic reflection, it could be color coded distinctly. A linear blending of the colors would allow the user to detect the attenuated intensity amidst the other data as the object is rotated. Implementation of these concepts requires advanced graphics capabilities such as those available in engineering workstations.

 

Interface Background
The OpenGL application programming interface (API) has now become available for numerous standard desktop operating systems such as Windows NT/95, OS/2, LINUX and MacOS. OpenGL provides software tools by which full featured, network transparent, 3D graphical applications may be developed for NDT. The OpenGL ARB has linked the use of the OpenGL trademark to passing of a conformance suite which ensures a full, stable feature set across all implementations (Vepstas and Narayanaswami, 1994). Updating software linked with the libraries to future computer hardware/software configurations becomes a straightforward porting of platform/operating system independent code. This avoids building dependence upon a particular operating system and is an effective way of maintaining code investment as technology progresses, or as new needs arise.

The complexity of NDT displays necessitates dependence upon the existence of advanced graphics platforms. Until recently, this required specialized hardware. As recently as 1992, software and hardware for texture mapping (mapping arbitrary images onto a 3D object geometry) were only available on high end graphics workstations such as the SGI machines (Teschner, 1994). While common in a number of other applications (Deffeyes and Spitzer, 1994; Teschner, 1994), texture mapping has previously found limited application to the mapping of indications from NDE scanning systems. The absence of advanced visualization tools in NDE has probably been a result of the cost of the required hardware and the perceived complexity of development.

The OpenGL Performance Characterization Committee (OPC) has developed a benchmark for OpenGL performance on a number of platforms (OpenGL Performance Characterization Committee, 1995). The results for decal texture mapping (pasting the texture map over a surface) are summarized in Table 1. This process corresponds to the mapping of NDE inspection data to a surface model of the object.

Table 1

Performance comparisons of several systems for texture mapping

Company/Product		Memory (MB) RAM/DISK	Frames/s (Test 4)
IBM  Model 730 - P90 S3 864 (OS/2)		24/364	0.59
IBM  Model 730 - P90 S3 864 (NT)		24/364	0.49
SGI  Indigo 2 XZ 133 MHz R4600SC		32/1000	0.77
HP 715/80 Freedom 3150		32/1000	9.61

The unaccelerated rendering speed of a 90 MHz Pentium PC running OS/2 compares favorably with modern workstation performance. A 320 polygon tire image (see mesh in Figure 1) in 24 bit color, texture mapped with a 1024 ´ 256 anomaly map (Figure 2), reorients to new views in near real time.

Figure 1 - Graphical mesh used to map data to tire surface.

 

Figure 2 - The user interface with the 3D texture mapped surface model fully zoomed out.

 

The display rate is 2 frames/s with a texture mapped tire and 2.5 frames/s for a Gouraud shaded tire. Smooth motion ensues when a second memory buffer is used for generation of the image before swapping to the visible buffer (utilizing double buffering). The actual texture mapping of the object occurs within a few seconds, producing an optimized "display list" for the tire in computer memory which is used for redrawing (i.e., when a rotation has taken place).

The configuration of hardware and software used to achieve these results is presented in Figure 3. The large number of hardware and operating system options available for achieving this specific display configuration provides opportunity for meeting the requirements of most NDT analyses. The development of graphical APIs has in the past been a process of playing catch up to the current hardware capability, however, recent progress in open standards has provided the potential for great strides in enhanced NDT visualization using low cost hardware.

Figure 3 - Flow chart which shows the relationship between the OpenGL application program interface and various windowing/GUI systems.

 

Implementation of Tools
C-scan data is applied to the part geometry using texture mapping through OpenGL. Texture mapping facilitates the ability to orient the C-scan dataset as a color map onto the part geometry. With the possibility of hardware acceleration (implementation or optimization of some of the software routines in silicon chips), texture mapping provides a great advance in visualization potential. Texture mapping is done using data defined in texture space. The texture space is a parametric coordinate space (one, two, or three dimensional) which is mapped to the 3D geometry of a surface.

In the example application of visualizing tire inspection data, the angular orientation of the tire (the axis of rotation around the truck axle) and lateral location of the transducer (the linear axes across the width of the tire or along the height of the sidewall) define a 2D texture coordinate system. Texture space is equivalent to the familiar flat C-scan of the object. Each data element in texture space is designated a texel. A mapping function is defined that links each texel to the 3D model geometry. In the standard implementation of OpenGL, this involves mapping corresponding points in texture space to each vertex of the polygons which form the surface model.

In order to handle situations where one texel (data point) does not correspond to one pixel on the display screen, texture mapping provides a method of mapping 3D data to a flat screen in a consistent manner (Catmull, 1975). In the OpenGL API several mapping options are provided to handle magnification and minification; linear, nearest and mipmapping. A "linear" filter performs a weighted linear average of the 2 ´ 2 array of texels that lie nearest to the pixel center (Figure 4a). A "nearest" filter simply assigns the color of the texel nearest to the pixel center as the color of the pixel (Figure 4b). Multiple resolutions of the texture may be specified for improved appearance through the method of mipmapping. To use mipmapping, "provide all sizes of your texture in powers of two between the largest size and a 1 ´ 1 map. For example, if your highest-resolution map is 64 ´ 16, you must also provide maps of size 32 ´ 8, 16 ´ 4, 8 ´ 2, 4 ´ 1, 2 ´ 1 and 1 ´ 1" (Neider et al., 1993). If an indication exists in the original data within the data region represented by a mipmap texel, a filter may be designed which maintains the indication texel (at all resolutions of the data). This mapping of data to a predefined space through filters chosen by the operator make texture mapping an obvious choice for NDT applications. In the example application of ultrasonic C-scan data, a low amplitude region can be mapped to retain its visibility to the operator at all levels of zoom into the part. This approach guarantees that all indicated discontinuities will be visible at all image resolutions.

Figure 4 - Zoomed images of indicated discontinuities showing a "linear" texel map and a "nearest" texel map of data onto surface model.

 

Very small anomalies often need to be detected with high resolution anomaly data projected onto limited resolution display windows. To convey the critical information contained in the original data, it is important to control which data texels are represented at different resolutions of the image. In addition to OpenGL’s capabilities, detection requirements may be addressed by using automatic anomaly recognition routines to amplify discontinuity regions. As scanning resolution increases, the trade off between data overload and resolution must be addressed. For the case of texture mapping, this key consideration in developing an operator display can be optimized with the appropriate use of automatic anomaly detection and minification algorithms.

Texture mapping is applied to a surface model of the part, which is often available from finite element modeling data. OpenGL objects exist in the homogeneous (x,y,z,w) coordinate space of three dimensional projective geometry. The fourth coordinate, w, is a scaling of the radial distance from the point to the origin. The w coordinate is a common addendum to Cartesian coordinates in computer graphics, enabling translation and rotation through a 4 ´ 4 transformation matrix. In the tire inspection system, eight (R,Z) points were chosen to form the cross section of the tire model, which was rotated to form the wire mesh of Figure 1. Polygons in OpenGL are required to be planar to simplify the algorithms which render the polygons (and thereby increase the rendering speed). Triangles were used on the sidewalls to better approximate the curvature. A smooth tire image results by applying lighting calculations to the polygonal surfaces.

 

Results From Tire Inspection System
An application of these tools is demonstrated by mapping anomaly indications onto the image of a truck tire casing. Mapping the data allows association of indications with the crown, sidewall, or other notable aspects of the tire construction. In particular, belt edge separations between the steel belts and the surrounding rubber may evolve at the transition areas on the sides of the crown. Long, thin discontinuities may evolve on the sidewalls (Rogers et al., 1992). These various critical locations are easily identified using a surface model. Most importantly, setup problems for the scanning system may be readily identified in a C-scan image which deteriorates at a location that is associated with an area of transition in the tire casing inspection system. A common setup problem is when a transducer misalignment has caused a change in ultrasonic amplitude, leading to incorrect indications.

Figure 5 provides a second view of the use of high end graphics for anomaly visualization. The figure shows a zoomed view of the texture mapped surface model with different data from that which is shown in Figure 2. The second image shows several circumferential indications mapped to the tire surface. The indications are clearly associated with geometrical transitions on the tire. The sharp edge of the crown discontinuity indication on the shoulder increases the likelihood that the discontinuity indication is actually a result of misalignment of the transducer. The sidewall discontinuity indication, however, appears to be real since the area of discontinuity has acceptable material on either side of the indication. In the flat C-scan of Figure 5, the image is zoomed to a discontinuity which is indicated on the surface model. The zoomed image can be used to facilitate measurement of the discontinuity dimensions which are independent of the texture mapping scale factors associated with mipmapping. Rotation of the surface model allows the operator to view the object from any perspective.

Figure 5 - Zoomed view of texture mapped surface model of tire and zoomed flat C-scan from user interface.

 

Integration of Design and Inspection
Fundamental to the concept of texture mapping is the generation of wire frame models to which the data may be applied. This 3D mesh could easily be constructed through other graphical applications and ported to OpenGL. The Initial Graphics Exchange Specification (IGES) provides a format to describe arcs, lines, text, splines and nodal structure (Smith et al., 1988). An IGES file may be used as input to NDT visualization software. Software packages that allow graphical mesh creation and support IGES format include such industry standards as PATRAN, Pro/Engineer and AutoCAD. By taking advantage of established CAD packages for mesh development, the process of building effective graphical displays tailored to various geometries becomes streamlined and highly automated. The link between NDT visualization and the design process thus becomes transparent.

The flow presented in Figure 6 will facilitate the process by which part design, analysis and testing (with subsequent visualization) takes place. Software specifically tailored toward automatic mesh generation, such as PATRAN and Pro/Engineer, provides cutting edge technology in generating wireframe and surface models of objects. Analysis packages such as ABAQUS, NASTRAN and other software are used to perform finite element analysis related to the given part geometry. Once a prototype part has been constructed and "road tested," nondestructive evaluation tethered with new visualization paradigms facilitate location and display of discontinuities which develop during destructive testing of a design. The geometry from the early stages of the design process is maintained for a consistent analysis throughout the design iteration process.

Beyond meeting the desired display requirements, the approach presented in Figure 6 provides vital information for the NDT analysis. The information available from mesh development usually includes normal vectors to surfaces. The normal vectors determine the angle of refraction for ultrasonic waves passing between materials of dissimilar ultrasonic impedance. Knowing the normal vectors, the optimal scanning trajectory for the transducers can be determined on complex or curved surfaces to most closely approximate normal incidence.

Figure 6 - Integration of design and NDT facilitated by shared data formats can have a significant impact on the future application and component reliability.

 

Conclusions
Potential advances in NDT abound through improved visualization of the NDT data currently being produced. New graphics software and low cost hardware make advanced analysis and display capabilities available even to small commercial applications and within academia. Advanced graphical techniques such as texture mapping are ideal for mapping anomaly data to 3D geometry, where the most information for interpretation may be obtained. New paradigms of graphical display continue to emerge and be integrated into portable software such as OpenGL. Great potential exists for integrating surface modeling, mesh generation, mechanical analysis software and NDT hardware with advanced visualization software for a superb overall design process.

In the case of the used truck tire casing scanning system, the user interface has proven invaluable in debugging the overall system and satisfying the feedback capability desired by the industrial sponsor. Difficulties, such as determining whether air bubbles caught within the tread pattern are affecting the quality of data, are resolved with proper interpretation of the visualization provided. By automating the process of data analysis and display, production testing may be done at numerous sites with valid interpretation of the results for improved product quality.

 

Acknowledgments
All of the scans performed over the course of this research were done using data acquired from the Colorado State University High Throughput Tire Scanning System.

Additional views of the user interface and an animated movie (mpeg) of the user interface can also be seen over the World Wide Web at address http://www.lance.colostate.edu/~mick under the heading "New Life for Old Tires."

 

References
Catmull, Edwin, "Computer Display of Curved Surfaces," Proceedings of IEEE Conference on Computer Graphics, Pattern Recognition and Data Structures, 1975, pp 11-17.

Deffeyes, S., and J. Spitzer, "OpenGL on OS/2," OS/2 Developer (Nov./Dec. 1994, pp 34-41.

Downs, J., III, and M.L. Peterson, "A High-Speed, High-Resolution Ultrasonic Tire Scanning System," (in preparation) Colorado State University, Fort Collins, Colorado.

Neider, J., T. Davis, and M. Woo, OpenGL Programming Guide, 1993. Addison-Wesley Publishing Company, Reading, MA.

OpenGL Performance Characterization Committee, 1995, "CDRS Result Summaries," GPC Quarterly (first quarter, 1995), WWW address: http://net1.uspro.fairfax.va.us/gpc/opc_1_95.html.

Rogers, R., M.L. Peterson, and J.D. Achenbach, "Ultrasonic Testing of Truck Tires, Phase II," Technical Report, 1992, Northwestern University, Evanston, Illinois.

Smith, B., G.R. Rinaudot, K.A. Reed, and T. Wright, "Initial graphics exchange specification (IGES) version 4.0," DOCMIC C13.58:88-3813 Technical Report, 1988. US Department of Commerce, National Bureau of Standards, Gaithersburg, MD.

Snyder, J.P., "Map projections: a working manual," USGS Professional Paper 1395, 1987. US Government Printing Office, Washington, DC.

Teschner, M., "Texture mapping: New Dimensions in Scientific and Technical Visualization," IRIS Universe, 1994, No. 29, pp 8-11.

Vepstas, L., and C. Narayanaswami, "Programming with OpenGL," AIXpert, Feb. 1994.

Williams, T., ed., "Industry group pushing 3-D graphics standard," Computer Design, 1994, pp 50-52.

 

* Mechanical Engineering Dept., Colorado State University, Fort Collins, CO 80523; (970) 491-2813; fax (970) 491-1055; e-mail mick@lamar.colostate.edu.

Copyright © 1997 by the American Society for Nondestructive Testing, Inc. All rights reserved.

[ Materials Evaluation ]

IRRSP, NDT Handbook, The NDT Technician and www.asnt.org are trademarks of the American Society for Nondestructive Testing, Inc. ACCP, ASNT, Level III Study Guide, Materials Evaluation, Nondestructive Testing Handbook, Research in Nondestructive Evaluation and RNDE are registered trademarks of the American Society for Nondestructive Testing, Inc. ASNT exists to create a safer world by promoting the profession and technologies of nondestructive testing.