8 - UT Aerospace Engineering & Engineering Mechanics



8.0 Carnegie Mellon’s ANDI

8.1 Carnegie Mellon’s Approach

Carnegie Mellon originally considered three ideas when developing the initial design for their automated robot inspection system. JAL (*is it JAL or JPL?**) experimented with the first idea, referred to as the “car wash” approach; it involves the aircraft being “towed through” a machine. Commercial aircraft operators found this approach to be unacceptable for economic and operational reasons. The second approach, referred to as the “cherry picker” approach, utilized a mobile platform that carries an arm and probe manipulator and travels to the aircraft panels that need to be inspected. The US military utilizes a similar approach using neutron beams and x-rays. Commercial operators agree that the “cherry picker” approach is not feasible because the maintenance areas in most commercial aircraft hangars are much too crowded as it is. The third approach, also the approach that CMU pursued, is the “window washer’ approach [23]. In this method, squat mobile robots that are approximately one meter in length and under 15 kg in weight, crawl on the aircraft’s surface and engage in visual inspections [24]. Figures 7 and 8 below and this paper describe CMU’s conceptual approach and design. A marking apparatus and eddy current probe are mounted to platforms on opposite sides of the sensor. ANDI uses these accessories to scan for flaws and mark suspicious areas [16].

[pic]

Figure 9. Operator’s workstation, airplane and mobile inspection robots [24]

[pic]

Figure 10: Sketch of cruciform robot design [24]

[pic]

Figure 11. ANDI undergoing early tests on an airplane panel [25]

8.2 Mechanical System

Carnegie Mellon’s first prototype of the window washer is a “cruciform” robot as shown in Figure 11. ANDI’s “spine” is one long member with a suction cup foot at the forward end and two suction cup feet on a short cross member at the rear of the main member. There are two more short cross members that contain a suction cup foot at each end, which are mounted to a pneumatic actuator [25]. ANDI is supported by either the three feet on the main member or by the four feet that are on the two cross members. The purpose of the cross members, also referred to as “bridges,” are to carry and maneuver the sensors and to allow the robot to walk. While scanning the fuselage, the spine is lowered and the bridge feet are raised so that the bridges are able to scan the fuselage back and forth. Linear stepper motors power the bridges and allow movement The bridges are powered along the spine by linear stepper motors. Compressed air operates the pneumatic cylinders, drives the up and down motion of the feet, operates the linear bearings on the linear motors, and powers the ejectors to create a vacuum effect under the suctions cups. The mechanical system allows ANDI to walk around the circumference of the fuselage [24].

A marking apparatus and eddy current probe are mounted to platforms on opposite sides of the sensor [23]. ANDI uses these accessories to scan for flaws and mark suspicious areas.

8.3 Vision System

The vision system for CMU’s Automated Non-Destructive Inspector, ANDI, employs four cameras. The first set of cameras includes Chinon CX-060 miniature black and white video cameras. These cameras are used for navigation and alignment of probe with respect to the rivet location. Algorithms are developed for finding rows of rivets to provide feedback for automatically aligning the robot. The second set of cameras consist of two Elmo MN401E ½-inch CCD (768H x 494V pixels) color cameras that act as the “proprioception camera” and detailed-view camera. The proprioception camera senses the position, location, orientation and movement of the robot and is capable to pan and tilt. The detailed-view camera provides a close-up view of the rivet being inspected, as well as confirmation of probe location [23].

8.4 Future Development

In February of 1997, CMU completed the third and final design of ANDI. In this last phase, the autonomy, reliability, and ease of operation were optimized. The only improvements made to the final mechanical design of the robot from the preliminary design were the installation of opaque hoods over the two navigation cameras and re-design of the suction cup vacuum system. The opaque hoods were added to improve the quality of the images of the rivets, and the suction cup system was re-designed so that the compressed air requirement was minimal. The final prototype was tested by CMU at US Airways maintenance facility at Pittsburg International Airport [23]. A description of the on-board computer and instrumentation of the robot is described in Figure 12.

[pic]

Figure 12. On-board Computer and Robot Instrumentation of CMU Robot [23]

8.5 Control System

CMU’s control system consists of instrumentation installed on the robot as well as a remote operator workstation. The operator work station serves as the main control point of the robot and contains communication channels that are provided to the remote controllers, on-board computer, and the video-processing computers. The eddy current instrument is also contained in and controlled by the workstation computer. A satellite equipment enclosure houses the support equipment for the linear motors, camera control units, and power supplies [23]. Figure 13 is a picture of the operator workstation.

[pic]

Figure 13. Operator Workstation for CMU Robot [23]

8.6 CIMP

ANDI marks the completion of the first step in a program that CMU is conducting for the FAA’s National Aging Aircraft Research Program. The purpose of the program is to investigate the application of robotic instruments in assisting aircraft inspectors. During the development of AND, the focus was not on producing high quality images for inspection purposes. Instead, the focus of the ANDI project was to develop a mobility and navigation system. Following ANDI’s development, CMU’s focus switched to developing a high quality visual image system for remote visual inspections, This Crown Inspection Mobile Platform (CIMP) is the successor to ANDI and is pictured in Figure 14 [26].

[pic]

Figure 14. Crown Inspection Mobile Platform (CIMP) Robot [26]

CIMP is a wireless remote-controlled robot that inspects the crown areas of aircraft. The robot carries two stereoscopic pairs of cameras (one pair for inspection and the other for navigation), two fixed flood lights, and a rotatable directional light source. Image understanding algorithms have been developed to recognize and classify defects in the surface images produced. Surface crack detection and surface corrosion detection algorithms have both been developed and implemented. A block diagram and a sample output image for the crack detection algorithm are shown in Figure 15 and Figure 16 respectively. As shown in Figure 16, suspected cracks are represented by black marks and non-cracks are represented in grey. “Non-cracks” are edges of rivet holes and scratches that are detected by the image system [26].

[pic]

Figure 15. Surface Crack Detection Algorithm [26]

[pic]

Figure 16. Output of CIMP’s Crack Detection Algorithm [26]

Figures 17 and 18 show the block diagram of the surface corrosion detection algorithm and an example output of the algorithm. As seen in Figure 18, the bright areas represent corroded areas and the dark areas represent the non-corroded areas [site me].

[pic]

Figure 17. Surface Corrosion Detection Algorithm [23]

[pic]

Figure 18. Output of CIMP’s Crack Detection Algorithm [23]

CMU demonstrated CIMP at both Northwest Airline’s Minneapolis maintenance facility and US Airway’s Pittsburg maintenance facility, where aircraft inspector and NDI supervisors alike were willing to accept the method as a replacement method for visual inspections. Aircraft inspections are comprised of 90% visual inspections and 10% non-destructive inspections [16]. It should be noted that CIMP is not a viable alternative for non-destructive inspections. Furthermore, the CIMP system only inspects the crown areas of the fuselage and therefore would not be a suitable system for lap joint inspections. We would also like for our system to be automated, therefore the ANDI robot is more appropriate for our design. The CIMP design is not the final design for CMU’s aircraft inspection robot. The ultimate design will be a marriage of the technologies developed for the ANDI and CIMP inspection systems [26].

In order for the robot to be able to align the eddy current probe to the row of rivets being inspected and to be able to autonomously navigated around the fuselage, algorithms that are able to detect the presence of the rivets and to measure the location of the rivets in the video imagery from the navigation cameras. Such algorithms have been developed at CMU’S robotics Institute. A line-fitting algorithm will create a straight line through the rivet path. Figure 19 illustrates the current line-fitting algorithm used for ANDI.

[pic]

Figure 19. Line-Fitting Algorithm for Navigational Control of Inspection Robot [23]

CMU has developed an open ended approach using network architecture to accommodate for variations in rivets, which are classified by a human trainer. This system is designed to generate a numerical output called “rivetness” that estimated the probability of a rivet being shown in a window produced by the camera output. This method determines and confirms the presence of a rivet in the images outputted by the cameras [26].

8.7 Modifications for Finish Detection

Although Carnegie Mellon’s design embraces the need for crack inspections along the rivet lines of the fuselage, it does not distinguish the difference between anodized and alodined rivets. Because ANDI does not identify the rivet coating, premature cracks may still go undetected.

While CIMP has a superior vision system, ANDI’s mobility and mechanical system is better suited to the needs of this project. ANDI can easily be modified to detect rivet finishes.

ANDI uses the Nortec SPO-1958 eddy current probe that operates within the 100 Hz-50kHz frequency range. Southwest Airlines must replace the probe with one that operates within the 20 to 50 kHz range. The Nortec SPO-2210 falls within this range and can be purchased for $530.

In addition, we propose that an algorithm that detects anodized and alodined rivet coatings be incorporated into the eddy current software for CMU’s design. The algorithm should instruct the marking apparatus to mark the alodined rivets for further inspected by mechanics for cracks, and should record the position of the alodined rivet to the computer.

8.8 Software

For determining the presence of cracks in the lap joint rivets, we recommend implementing a pattern recognition method. Current eddy current inspection procedures do not require quantitative means for classifying the outputted signals. Instead, the results of the outputted signals are determined by comparisons to known eddy current patterns. Pattern recognition methods will aid in the determination of a “good” or “bad” signal during eddy current inspections, therefore providing inspection relief to mechanics and increased accuracy to eddy current inspections. The SmartEDDY 3.0 eddy current software package allows inspectors to set threshold standards, creating a clear partition between signals produced by fasteners with and without cracks [27]. This software would not be appropriate for determining rivet finish, however, because alodined rivets produce a large variation. In other words, a distinct pattern for alodined rivet signals does not exist for the software to be calibrated. Because the SmartEDDY software is a dedicated application, it is not able to exchange information with other parts of the control program. It also requires an entire computer to operate. As a result, Carnegie Mellon developed a custom software interface that integrates the eddy current instrument with the robotic system [28]. Not only does CMU’S eddy current portion of the software manage the instrument inspection mode, the signal level, signal frequency, and sampling rate, it also performs data acquisition and buffering of the inspection data [28]. CMU’s system is able to support a sampling rate up to 1 kHz. Acquired data is able to either be viewed by the inspector at the moment or archived to a file to be analyzed at a later time. Graphical displays for CMU’s software are able to plot the inspection data as in-phase versus quadrature signals or as separate in-phase versus time and quadrature versus time plots [28]. An example graphical display is shown in Figure 20.

[pic]

Figure 20. CMU’s Software – Eddy Current In Phase/Quadrature Versus Time Display [28]

For our project, CMU’s software can be adjusted by engineers for a sliding probe rivet finish detection process as required by Boeing NDT Manual 53-30-21, Part 6. The software may be used to calibrate the probe in accordance with the NDT Manual instructions. In addition, the software settings are modifiable for general crack inspections as described in Boeing’s NDT crack identification inspection process, which is also found in the same NDT manual.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download