·Table of Contents
·Methods and Instrumentation
Path Controlled Contour Following for Ultrasonic Imaging of Complex Shaped Composite ComponentsDaniel J. Cotter, Jennifer E. Michaels, Thomas E. Michaels,
Igor V. Ivakhnenko, and Daniel Kass
Panametrics Incorporated, Waltham, MA 02453-3497
Utilization of CAD (Computer Aided Design) or CATIA (Computer Augmented Three-Dimensional Interactive Application) data for teaching complex contour paths has been successful for several reasons that are detailed. Surface points and normal vectors directly accommodate contour following with transducer perpendicularity and off normal refracted longitudinal and shear wave inspection angles readily being incorporated. Regions where the radius of curvature does not support ultrasonic imaging can be smoothed to maximize inspection coverage. Computer models can be adjusted to follow interior plies, which are neither parallel to the front nor back surfaces to maximize ultrasonic sensitivity to de-lamination. Developments are exhibited with display of improved ultrasound data from difficult contours and actual production component scans.
Fig 1: Gantry Ultrasonic System. The complex contoured component is automatically imaged with a single scan plan in less than 1 hour.
The ultrasonic system shown here has two robotic manipulators employing a water squirter system to couple the ultrasound to the component under test. The approximate XYZ scan envelope is 7 m x 3 m x 3 m (approximately, 24 ft x 10 ft x 10 ft). The through transmission inspection exhibited scans the complex contoured component automatically with a single scan plan in under 1 hour. Generally, the transducer must be kept normal (perpendicular) to the component surface for pulse-echo and through-transmission longitudinal wave inspection, or at a prescribed angle for refracted longitudinal or shear wave inspection. The water path distance from the transducer to the component surface also affects sensitivity in ultrasonic imaging; therefore, it must be precisely controlled.
The focus of the present effort is on improved understanding of path controlled pulse-echo scanning of difficult geometries encountered in composite inspection applications. Influence of the front surface geometry has been modeled in the past employing ray tracing (3). Herein production inspection data are augmented with results from a simulated component where better control of taught surfaces and reduction in the complexity of the microstructure aid in interpretation. The discussion is related to the pulse-echo inspection of composites, generally performed at 5 MHz and below. The emphasis is placed on input of surface geometry from computer data because it has been the most successful approach.
2.1 Referencing the Component and Teaching of the Contoured Surface
Key requirements for following the surface contour of an object in the scan envelope include knowledge of the component location, orientation, and surface topology with respect to the scanner coordinates. Scanners typically operate from a predefined absolute coordinate system. The component orientation and location in the scan envelope can be defined uniquely by three reference points selected on the component surface. These reference points can be used to re-reference a component that has been taught and moved or replaced. A pointer device or the transducer used for the inspection can be employed to re-reference the component placed in the scan envelope to the robotic positioning system. This three-point correction procedure can be adapted or transferred to component fixtures for production inspection. Of course, component-to-component dimensional repeatability must support scanning with the same three-point correction and surface topology.
Teaching methods also rely on referencing each surface point acquired to the robotic positioning system. The robotic positioning system software must have knowledge (measured or apriori data) of the offset in the water path and ultrasonic transducer location or the mechanical offset of the teach pointer or touch probe tip from the scanner. The robotic positioning system is taught the coordinates needed to direct these teaching implements to the surface points. In actuality, the surface point is attained and the reference information is used to calculate the coordinates of the point on the surface from actual robotic positioning information. Teaching methods include ultrasonic surface ranging and normalization, mechanical fixed pointer probe, pneumatic touch probe, and conversion and referencing of CATIA data.
2.1.1 Ultrasonic Teaching Methods
Ultrasonic methods can be manual, semi-automated, and automated. A 5 MHz focused transducer is typically employed for teaching, even if the inspection requirement is at lower frequencies. In the manual mode, the operator manipulates the transducer to optimize the signal from the component in pulse echo. Typically, the amplitude of the front surface echo is peaked. In semi-automated mode, the manipulator moves to a grid of points (points obtained with other teach methods or calculated from coarse grids) and the operator again manually optimizes the signal. Automated ultrasonic teaching and normalization based on a coarse grid of points is possible in some cases using auto-normalization with virtual swivel and gimbal motions. The coordinates and normal vectors are interpolated based on algorithms providing localized motion and monitoring of feedback on the peak signal response. In essence, the manipulator is automatically moved to both sides of the peak signal position with virtual swivel and gimbal moves, and then to the maximum signal position for acquisition of the coordinates.
A powerful capability of robotic-based positioning software is semi-automated or automatic ultrasonic normalization using virtual motion axes. For a curved component, it can be very difficult to manually peak the signal from a front surface echo. During the manual peaking process, the operator typically changes the swivel and gimbal angles until the amplitude of the front surface echo is maximized. However, as these angles are changed, the ultrasonic signal is not necessarily being reflected from the original point on the surface, but from wherever the beam wanders on the surface during the motion. With virtual axis peaking, the initial front surface echo defines a point on the surface, and the coordinated motion of the axes is controlled such that the beam is maintained at that point during the peaking process.
All the ultrasonic ranging and normalization methods calculate the surface point given the scanner coordinates, offset to the transducer, and the measured time-of-flight to the front surface echo. The normal vector for the surface point is obtained from the angular position of the manipulator.
2.1.2 Mechanical Teaching Methods
With the pointer probe method, a fixed pointer is attached to the transducer or squirter nozzle. The operator touches the probe to the surface of the component to obtain the surface point. An alternative is a touch probe that has an air actuated linear probe whose position is monitored with an encoder. For each teach point, the probe is extended until it makes contact with the component surface. With the touch probe, it is possible to automatically teach the entire component after the operator manually teaches a coarse grid. The touch probe is very efficient for teaching components for through transmission imaging that have a contour that can be characterized with the probe. The mechanical methods again calculate the surface point given the scanner coordinates and mechanical offsets. The normal vectors for the surface points are not obtained, therefore they must be calculated.
2.1.3 CATIA Based Contour Following
The utilization of CATIA (Computer Augmented Three-Dimensional Interactive Application) data for teaching complex contour paths has been successful for several reasons. The surface points and normal vectors directly accommodate contour following with transducer perpendicularity. Off normal refracted longitudinal wave or shear wave inspection angles can be readily incorporated in the CATIA data. Regions where the radius of curvature does not support ultrasonic imaging can be smoothed to maximize inspection coverage. There are cases where following the outside surface is not optimal for imaging. For example, in scanning multi-layer composites, the interior plies, which need to be interrogated for delaminations, are not necessarily parallel to the outside surface. In this circumstance, following a CATIA mid-plane perpendicular to the plies can maximize acoustic sensitivity to delaminations. In comparison to ultrasonically or mechanically taught surface points, the CATIA data are ideally smooth and speeds for robotic motions can be greatly increased with less locally tortuous paths.
Points selected from the CATIA data provide the basis for the description of the component's surface contour. CATIA data files are often used by the aerospace industry in machining processes and coordinate measurement system analysis of components, and the surface points (X, Y, and Z) and normal vectors (i, j, and k) can be readily converted to the robotic positioning system employed for ultrasonic scanning. The points can typically be output as ASCII or text files. A 25.4 mm by 25.4 mm (1 in by 1 in) point spacing in the grid is generally recommended for gradual contours; however, spacing may be as fine as required locally to better represent contour detail. It is important to note that the index and scan increments ultimately specified for acquiring ultrasonic data are not limited directly by the grid. These values can be much finer, e.g., 1 mm (0.040 in) square increments or larger are common for composites, and 0.025 mm to 0.25 mm (0.001 in to 0.010 in) are more typical of metals.
|Fig 2: Teach Grids. The component presented a challenge due to compound radii and the small radius of curvature at the their transition. Complete coverage was attained.|
Several planar views of teach points are shown in Figure 2 for the component previously exhibited in Figure 1. The component presented a challenge due to compound radii and the small radius of curvature at their transition. The component was oriented to be scanned with the Z-axis (vertical axis) to overcome the extensive swivel movements that would have been required at the transition with horizontal scanning. Complete coverage is attained in through transmission with little difficulty considering the geometry, and the single scan is acquired in less than 1 hour. Previously, the component was manually scanned with long through transmission yokes.
Typical spatial increments for inspection of aerospace composite components range from 1 to 3 mm (0.040 to 0.120 in) and the frequency of inspection is generally 5 MHz and below. To ensure complete coverage at the lowest spatial increment in through transmission scanning, the overall system positional accuracy error must be less than +/- 0.2 mm (+/- 0.008 in). The motion requirements for pulse echo inspection are more rigorous because the transducer must maintain normal incidence to the component within approximately +/- 0.5 degrees. The contour must be characterized fine enough to maintain inspection requirements, such as a percentage in loss of signal in through transmission or loss in back wall echo. The component teaching methods, numerical processing techniques, mechanics, and motion controls must all meet these requirements in order to perform a valid inspection.
2.2 Generation of the Mathematical Description
A mathematical description of the component to be inspected must be determined from the geometrical information, which is input as an unordered set of teach points in three dimensions. A mathematical description that can be used is a grid on the surface of the component called a "control grid" (2). The control grid is defined by a set of points organized as N rows by M columns with surface points and normal vectors defined at each point. A generalized outline of the steps for determining this control grid is provided below.
The final result of this procedure is the control grid, which is the grid of points and normal vectors in three dimensions that are directly used to generate the specific scan plan. The overshoot parameters allow the scan to extend beyond the original component surface to ensure inspection coverage. The implementation also includes a feature to add left or right hand boundaries to the control grid to skip areas not of interest, such as cut outs within the component or clamps on its edge.
2.3 Scan Plan Path Calculation and Execution
The scan plan path is calculated based upon scan and index increments specified by the operator. There is no limitation on the resolution of the scan; it can be either coarser or finer than the control grid resolution. The coordinates for the scan are determined from the control grid by interpolation of the points and normal vectors. A cubic-spline fit is performed along the scan line and a linear interpolation in the index direction. The operator specifies the speed and acceleration for the path following the surface, and the actual travel time along each scan line is calculated at all of the control grid intersection points. The resultant scan plan is a list of coordinates and times along the path for each scan line. A separate list of times along the path where the pulser fires is also part of the overall scan plan.
The scan plan is executed on a line-by-line basis. At the beginning of each scan line, the computer loads the motion path for each axis to each respective motion controller and the pulser firing times to a sync generator. A path controller synchronizes all scanner axes by sending timing information from the common clock to all of the motion controllers. Each controller monitors the actual time and position and controls the axis using a PID (proportional integral derivative) control algorithm to maintain its position along the path. The sync generator fires the pulser at the proper times along the scan line. The digitizer electronics acquire data at each sync pulse, and the computer displays and stores the data in real time.
The accuracy of the mathematical description can generally be improved as much as needed by adding teach points and using a finer control grid. The accuracy of the scan plan execution depends upon the control algorithm and parameters, and generally degrades as the speed increases. The best absolute accuracy is typically five to ten times the encoder resolution, which generally enables servo control to 0.0254 mm (0.001 in). Ultimately, it is the speed and acceleration capabilities of the individual axes that limit positional accuracy along a complex path. If the variance in actual scanner coordinates from those required by the path at a given time exceed a threshold determined by the user, a contour error results and scanning is stopped. If any data are missed due to exceeding data throughput rates, the scan is also halted.
2.4 Adaptive Techniques in Path Controlled Contour Following
Operator imposed software limitations on the axes speeds and accelerations can be used to determine a new kinematic solution to the path to accommodate the complexity of the contour while remaining within contour error tolerances. The scan plan path and timing is calculated in view of the adaptive limits. In practice, a menu driven algorithm allows the operator to set individual axis limits on speed and acceleration below the real motion capabilities, execute the scan, and view the impact on numerical readout of errors in positional accuracy. This technique is much more powerful than simply limiting speed and acceleration for the entire scan to stay within contour error tolerances. Consider, for example, a 4.88 m (16 ft) long gradually contoured component that can be scanned at about 0.38 m/s (15 in/s) until a steep contour 0.3 m (1 ft) from its end is reached that can only be scanned at 0.05 m/s (2 in/s). In production inspection, the adaptive technique would allow the scan time to be more than 5 times faster.
2.5 Verification of Scan Plans
Surface scans can be employed to verify the performance of robotic-based transducer positioning. The amplitude response of a pulse-echo signal is uniquely sensitive to normality to the front surface, and water path variations can be monitored with time-based measurements. A 5 MHz focused transducer is typically employed to scan the surface, while a gate is used to monitor the peak reflection. The resulting amplitude and time-of-flight to the peak data indicate the variation in response to the followed contour.
The sensitivity of the inspection method is typically monitored employing reference standards containing known flaws. Inserts or flat bottom holes are often employed at angles of interest and prescribed depths. Often, regions of interest result from experience in component failure or known manufacturing difficulty such as delamination at composite ply drops, porosity in high stress locations, etc. Feedback on the response of the inspection method can be used interactively with CATIA operations to improve sensitivity. Of course, the inspection method and structural detail to be evaluated must be reconciled.
3.1 Considerations on Ultrasonics
Clearly, the ultrasonics employed are application dependent; nonetheless, key considerations with regard to contour following include selecting a transducer that will create a small beam diameter which can be focused in the water stream or near-surface in the material. The smaller the beam diameter in relationship to the local radius of curvature of the component, the more flat the entry surface appears. Manufacturers of composites often require that the material is imaged in the far field to accommodate flaw sizing, and this is favorable to the contour following. Of course, decrease in sensitivity at depth must be considered. Decrease in the nozzle and water stream diameter in squirter-based systems is often favorable, as long as the amplitude reduction is not too great and nozzle noise (spurious signals from the nozzle) does not occur. Typically, steep, high gain TVG (time varied gain) curves are employed with composites because of the high attenuation of the laminar material, and a spurious signal or multiple, albeit initially insignificant, is grossly amplified under the TVG curve. Through transmission composite inspection ultrasonics are addressed in detail in earlier published work (4). As previously discussed, higher frequency and more focused transducers are typically employed for ultrasonic teaching or optimization even if the inspection requirements are less demanding.
3.2 Contour Following a Cylindrical Transition in Wall Thickness
The teach grid shown in Figure 1A represents the outside surface of one quadrant (about 90o ) of a cylindrical composite component with a right circular cylinder internal diameter; i.e., like a straight pipe. The cross-section is symmetric, so only the upper half is shown in Figure 1B. The parallel walls change in thickness from about 25 mm to 60 mm (1 in to 2.4 in). The transition between the two parallel walls presents a difficult geometry for ultrasonic inspection due to the varying cross-sectional thickness with non-parallel front and backwall surfaces, and the ply drops illustrated in Figure 1C.
|Fig 3: Cylindrical Transition in Wall Thickness. The teach grid (A), cross-section (B), and transition (C) of a cylindrical composite specimen. The transition presents a difficult geometry due to varying cross-sectional thickness, non-parallel surfaces, and ply drops.|
A simple (often over simplified) approach generally employed is to ensure coverage and perform the inspection by monitoring the loss in back wall echo. As illustrated in Figure 1B, the longitudinal wave impinging normal to the front surface (left) maximizes pressure entering the material but little or no back wall echo is returned. Normalizing to the backwall (right) greatly reduces the pressure entering the material, and refraction again results in little back echo being returned. Experience has shown that it is favorable in contour following to error on the side of getting sound into the material. The compromise (middle) propagates a refracted longitudinal wave for reflection from the back surface, which tends to maximize the back wall echo. It is important to note that sensitivity to inter-ply delaminations is not necessarily maximized at all levels because the plies are not perpendicular to the propagating longitudinal wave. An improved approach sometimes utilized is to use apriori knowledge of the problematic regions in fabrication (intermediate cures, etc.) and define a mid-plane CATIA model that provides normalization of the refracted longitudinal wave to the ply level of interest, as exhibited in Figure 1C. The necessary surface points and normal vectors can be selected for conversion to teach points. It is often difficult to adequately model the influence of ply drops on ultrasound because each ply and intermediate cure presents an interface of some significance.
In practice, the compromise in the reflected pressure, or amplitude, from the front surface echo and back surface echo can be readily monitored as a function of position along the scan line. The software and virtual axis swivel and gimbal angular motion previously described can be employed to adjust the response in situ. Simulated disbonds (inserts that promote delamination) are typically used in critical locations to aid in setup and monitoring of the inspection technique. The response from the known delaminations can also be monitored while performing virtual angle motions. Individual teach points, or in this case, entire lines of teach points, can be adjusted to the angle needed to maximize sensitivity using menu driven algorithms. The teach points can be employed directly or the information on the offset angle can be used to permanently adjust the CATIA data.
3.2.1 Ultrasonic Imaging of the Transition
A C-scan slice of full waveform data is shown in Figure 4A, along with selected B-scans through simulated disbonds in the straight wall, Figure 4B, and in the transition, Figure 4C. Imaging was performed using a 3.5 MHz 19 mm (0.75 in) diameter element 127 mm (5 in) focal length broadband transducer with the focus established near surface. The angle of incidence was adjusted to increase sensitivity to the ply orientation and inter-ply disbonds in the transition region. Some loss-of-back wall echo amplitude was noticed. Evaluation of the transition is often done by viewing B-scans from stored full waveform data in a movie like presentation because of the morass of spurious signals caused by ply drops. Also, a fixed flaw gate width cannot be employed in the region of varying thickness for C-scan presentation. All the critical simulated defects in the transition could be located.
Fig 4: Scans of Transition. A C-scan slice is shown (A) along with selected B-scans through simulated disbonds in the straight wall (B) and in the transition (C). Angle of incidence adjustment in the scan plan increased sensitivity to the ply orientation and inter-ply disbonds.
||Fig 5: Scanning Along a Corner or Shoulder. The teach grid shown (A) represents the outside surface of 1 quadrant (about 90°) of the component. An expanded view of the teach points is provided (B) along with a representative illustration of the cross-section (C).
3.3 Contour Following Along a Corner or Shoulder
Another difficult geometry for pulse echo inspection with contour following involves scanning along a corner or shoulder, as is illustrated in Figure 5. The overall length of the multi-layered composite component is about 3 m (10 ft) with greatly varying wall thickness. The primary scan path would be along the corner or shoulder in the X-axis or longitudinal direction. The teach grid shown in Figure 5A represents the outside surface of 1 quadrant (about 90o ) of the component. An expanded view of the teach points is provided in Figure 5B, and a representative illustration of the cross-section is shown in Figure 5C.
Pertinent scans along the shoulder region of a production component are provided in Figure 6. The original front surface echo scan is displayed in Figure 6A with the associated loss in back echo scan shown below it in Figure 6B. The gain levels are established such that ideal scans would result in an 80% of full-scale height signal and appear completely uniform and nearly white. It was apparent that some of the loss in back echo could be attributed to loss of normality to the front surface during contour following. The component surface points were adjusted employing automated ultrasonic teaching with automatic virtual swivel and gimbal correction as describe above. With the difficult geometry and some localized machining roughness, about 2 in 20 points required operator intervention because the automated peaking algorithm could not find the peak signal position; nonetheless, the teaching process was greatly aided. The resulting front surface scan is displayed in Figure 6C and the associated loss in back echo scan in Figure 6D. Only the problematic central region along the corner or shoulder and the lower 0.9 m (35 in) of path length was re-taught and scanned, so it should be noted that the image as viewed in the scan is magnified. Coverage was improved for the single scan plan.
|Fig 6: Pertinent Scans of the Corner or Shoulder. The original front surface echo scan is displayed (6A) with the associated loss in back echo scan (6B). The component surface points were re-taught and the resulting front surface (6C) and associated loss in back echo (6D) showed improved coverage.|
A possible solution to provide complete coverage with a loss in back echo monitor, if required, is to scan with an additional channel of higher gain, or in the case of single channel ultrasonics, re-scan only the region of the corner or shoulder with higher gain. The later approach was employed for demonstration purposes on the production component with complete coverage being attained. An alternative to these approaches is to consider relaxing the demand for complete coverage with the loss in back echo monitor in the corner or shoulder region because it is recognized, and will be shown, that some loss of back echo is a consequence of the imaging geometry. C-scan images from flaw gates in the region or B-scan presentation of stored full waveform data, as discussed for the transition region, could be employed. A concern is that fully teaching the back wall echo for complete coverage in a single scan with one gain setting could adversely impact imaging of the bulk material.
3.4 Experiments with a Simulated Component
Production composite component inspection data were augmented with results from simulation of the geometry where better control of taught surfaces and reduction in the complexity of the microstructure aided in interpretation. The specimen of a corner or shoulder segment is shown as scanned by the squirter in Figure 7A. It was fabricated employing a solid polymer material of similar sound speed to graphite-epoxy composite. Three flat bottom holes were machined parallel to the front surface of the specimen with one centered in the corner and one on either side. The front surface was ultrasonically taught such that a monitoring gate showed an amplitude variation of less than 5 % anywhere across specimen as the right angle corner or shoulder was scanned. The front surface scan was uniform and nearly white as imaged in Figure 7B. A flaw gate C-scan is shown in Figure 7C and loss in back echo scan in Figure 7D with the transducer ideally normal. The images are symmetric and the sensitivity to the flat bottom holes is excellent. There is loss of back echo due the geometry alone, and this can be interpreted as shown in Figure 7E. On the flat regions, most of the signal is directly reflected back to the receiver (upper left illustration). At the apex of the corner or shoulder, the divergent beam in the material is necessarily reflected at angles that will not return the signal to the receiver (lower left illustration). If angle is introduced through errant teaching, the image becomes asymmetric and again reflected signal is lost (upper and lower right illustrations). Flaw gate and loss of back echo scans following the same plan with a normal vector offset of +5o , Figure 7F and 7G, and -5o , Figure 7H and 7I are provided. It is important to note that the back wall echo amplitude becomes somewhat more uniform with the off angle approaches and the gain could be increased to promote coverage, but the loss in sensitivity to the holes in the flat section becomes severe. The central flat bottom hole has the benefit of increased sensitivity due to the lens effect of the geometry of the front surface (5). The simulated geometry aids in interpretation, and perhaps provides a basis for selecting the best methodology for imaging in view of inspection requirements.
|Fig 7: Simulation of Corner or Shoulder. Segment of corner or shoulder is shown (7A), along with a front surface scan (7B), flaw scan (7C), and loss in back echo scan (7D) with the transducer ideally normal. There is loss of back echo due to the geometry alone, and this can be interpreted with the aid of the signal path illustration (7E). Flaw gate and loss of back echo scans following the same plan with a normal vector offset of +5° (7F and 7G) and -5° (7H and 7I) are provided.|
|© AIPnD , created by NDT.net|||Home| |Top||