Taking flight
Delta Sigma's engineers use vision-based measurements to help assemble fighter aircraft for Lockheed Martin.
C.G. Masi, Contributing Technical Editor -- Test & Measurement World, 10/1/2004 2:00:00 AM
|
MARIETTA, GA—When Lockheed Martin needed to manufacture the US Air Force's next-generation fighter aircraft, the F/A-22 Raptor, to the highest possible standards while holding down manufacturing costs, the company turned to long-time subcontractor Delta Sigma for assistance. Lockheed Martin asked the Acworth, GA-based company to develop an extremely high-precision alignment machine to fit the aircraft's fuselage together faster and more accurately than could be done manually.
In response, Delta Sigma developed a vision-based sensor that makes three-dimensional position measurements to the required precision, a geometrical algorithm that determines the displacements needed to bring the fuselage sections into alignment, and networked robotic actuators that move the fuselage into position. The resulting system cut the time required for the alignment procedure from hours to minutes.
Precision challenges
|
| Figure 1. The F/A-22 Raptor, the USAF’s next-generation fighter/attack aircraft, is built in three sections. The wings are bolted across the joint between the middle and aft sections. Courtesy of Lockheed Martin. |
The Lockheed Martin F/A-22 Raptor (Figure 1) is the US Air Force's next-generation fighter jet. To achieve its unprecedented performance, it incorporates several advanced-technology features:
-
Stealth makes the aircraft all but invisible to enemy radar and other threat-detection systems;
-
Supercruise allows the aircraft to sustain supersonic speeds for long periods of time; and
-
Thrust vectoring frees the aircraft from the limitations of wing stall, which limits the turning performance of conventional airframes.
Taken together, these features, along with myriad other technical improvements, give the Raptor a decided performance edge over other military aircraft. To achieve the aircraft's unprecedented performance, the major components must fit together within exacting tolerances.
This requirement for tight fitting components is not new to the aerospace industry. In the past, it has been met by aligning the components (a wing, for example, that must align with the fuselage) in a fixture, then match drilling holes to insert fasteners. That procedure results in a strong, stable airframe, but it causes problems when aircraft are damaged. To replace a wing, for example, technicians have to drill out all of the existing fasteners, then match drill all new holes to affix the new wing. It is a long, tedious, and expensive task, and the drilling of new holes weakens the airframe.
To reduce maintenance costs and aircraft downtime, the Air Force wanted the Raptor wings to be field-replaceable units. Maintenance technicians have to be able to easily remove the wings and fit new ones without going through a complex alignment procedure.
Instead of match drilling holes for permanently installed wing fasteners, the Raptor's wings are held on by removable fasteners fitted through permanently mounted wing lugs. The manufacturing task then becomes locating all the wing lugs within 0.004 in. There are 18 wing lugs spread over a wing root that is 2 ft high by 20 ft long! That's a precision of 16 ppm (see " ISO vs. English units," below).
Achieving such precision may be challenging, but it is a challenge that aerospace manufacturers easily meet with existing manufacturing methods. The "head scratcher" for the Raptor project was finding a way to line up the aft and mid-fuselage sections so all of the wing lugs, about half of which are on the aft section while the rest are on the midsection, would align within the required tolerance consistently in a production environment.
For the prototypes and the first few production units, Lockheed Martin developed a manual procedure in which two engineers and several technicians worked for up to four hours to align the fuselage sections. At that point, they could join the aft section to the midsection with the conventional match drilling and fastening process.
While this manual procedure filled the need for building prototypes and the start of production, Lockheed Martin wanted a more efficient system that would run faster with fewer people. The company also wanted an automated system that would be less dependent on highly trained engineers and technicians. Lockheed turned to Delta Sigma to engineer and build the system.
Delta Sigma has a long history with Lockheed Martin. For over a decade, the company has been developing unique manufacturing and test systems to meet the challenges the aerospace giant presented. This nine-person company, located in a rural Georgia community where "fine dining" consists of a trip to Stuckey's Restaurant at the I-75 off ramp, offers a surprising suite of state-of-the-art engineering capabilities in such disciplines as radar, advanced materials, machine vision, and robotics.
"We used a phased approach, because none of this had been done before," recalled Brett Haisty, PhD, who is Delta Sigma's VP of engineering, "We wanted to take a conservative approach."
The company broke the project into three phases: the overall alignment procedure, the subsystem for measuring wing-lug positions, and construction. This way, the engineers isolated the technical risks into the first two phases before a great deal of money had been spent or any irrevocable engineering decisions had been made.
Phase 1: Alignment algorithm
The technical question for the first phase was: How can you take measurements of existing offsets at the wing lugs and use them to compute required displacements at the hard points on the fuselage where the sections are supported?
While the fuselage sections are very rigid, there are only a few places where the structure is strong enough to take the point loads (amounting to tens of thousands of pounds) required to hold them up and manipulate them. It is at these "hard points" (in aerospace lingo) where the displacement actuators have to be located to move the fuselage sections into position so that the wings will fit on.
Unfortunately, the hard points on the fuselage are nowhere near the wing lugs to which the wings attach. It is the wing lugs that all have to line up so field-replaceable wings will fit on. A movement at any one hard point changes the position of all the wing lugs. By the same token, an adjustment of the position of one wing lug requires the coordinated movement of all the actuators at the supporting hard points. Calculating the displacements needed at the hard points to null offsets at the wing lugs becomes an exercise in three-dimensional geometry involving several simultaneous equations.
Because all the wing lugs on a given section are precision aligned with respect to each other during fabrication, not all the wing lugs need to be located. Technicians merely need to locate four lugs (one forward, one aft, one high, and one low) on each side of each section. If those wing lugs are properly lined up with their counterparts on the mating section, then all of the lugs will be properly lined up.
The problem reduces to locating 16 wing lugs (four on each side of two fuselage sections), and aligning them using displacements at four supporting hard points. Simplistically, that would work out to 48 measurements (three vector offset components at each of 16 wing lugs) used to calculate 12 displacements (three vector displacement components at each of four hard points).
The fact that the wing lugs and hard points on each fuselage section are coupled reduces the number of degrees of freedom drastically. In the end, there are 22 independent measurements and seven actuator displacements for each fuselage section.
Haisty developed a geometrical solution to the problem, then wanted to try it out before committing to it. He programmed his algorithm into a Palm Pilot and took it to the Lockheed Martin assembly line where engineers and technicians were lining up a fuselage. The engineers carefully made their 22 measurements. Haisty put those numbers into his Palm-Pilot program, which calculated the 14 displacements needed to align the fuselage sections. Haisty read them out to the technicians, who then manually cranked in the movements.
It worked the first time.
Phase 2: Noncontact measurements
The technical question for the second phase was: How can we make noncontact measurements to the required accuracy?
"[Lockheed Martin's] strong preference was to make noncontact measurements," Roger Richardson, president of Delta Sigma, recalled.
The first feasibility study used contact measurements because that was how the team already did the alignment. But the engineers wanted to change to a noncontact measurement for the automatic system.
If they used contact measurements with mechanical probes touching the aircraft, there would always be the possibility of damaging a probe whenever the fuselage moved. "You would have to have people check each probe to make sure it was out of the way before you moved the fuselage sections," Richardson pointed out, "and there are 22 of them! That would require 22 people, which would defeat the whole purpose of an automated system."
They considered using laser ranging, but that would require 22 separate laser range finders. Haisty reasoned that they could make three orthogonal measurements using one machine-vision camera, and the cameras would cost less than the laser rangers.
|
| Figure 2. The machine-vision-based alignment measuring system consists of a “light pin” and a “vision pin” mounted on a clevis-shaped fixture that straddles the wing lug to be aligned. |
Figure 2 shows Haisty's method for getting positions using a vision camera. The "light pin" is an LED array with a diffuser mounted on one side of a clevis-shaped fixture. The camera mounts in a "vision pin" on the other side, looking at the light pin. In use, the clevis-shaped mount straddles a wing lug.
Haisty chose to use a smart camera manufactured by DVT (Duluth, GA) to simplify the system. The camera incorporates an onboard image-analysis computer with a number of built-in algorithms, such as edge finding and blob analysis. The main algorithms performed edge-finding circle analysis and measured the distance between two points.
What the camera sees is a set of more-or-less concentric circles. The outer circle is the distal edge of the wing lug's cylindrical inside surface (blue dashed lines in Figure 2). The inner circle is the precision circular hole in the mask over the vision pin (red dashed lines). The smart camera's edge-finding routine picks out these two circles and locates their centers. Finally, the computer determines the horizontal and vertical components of the displacement between these two dots.
To get the third dimension (the wing lug's position along the optical axis), the circle-measuring routine calculates the diameter of the inner edge of the outer circle. The further the lug is from the camera, the smaller this diameter appears in the image.
"The only way we can get the subpixel resolution needed to make our tolerance," Haisty said, "is to use a large number of points on the circles that the edge finder locates. The position errors that pixilation introduces cancel each other out."
While Haisty felt that the method should work, he needed to verify it experimentally. He also needed to calibrate the image to get actual x, y, and z coordinates from the image measurements.
Haisty used an experimental test rig to verify the theory and to calibrate the measurements. The fixture provides large aluminum uprights that simulate the arms of the clevis-shaped mounting. A two-axis micrometer stage carries an aluminum block with a hole bored in it to simulate the wing lug. A camera mounted on one clevis leg acts as the vision pin, and an LED array with a diffuser and mask simulates the light pin.
One stage axis moves the "wing lug" parallel to the camera's optical axis. The other stage axis moves it transversally. "We were able to verify the theory and calibrate the system with just one transverse axis," said Haisty. "Rotating the camera by 90 degrees allowed us to calibrate the image along the other transverse axis."
Such a vision pin can locate a wing lug to within 0.0002 in. in the horizontal and vertical image dimensions, and 0.001 in. along the camera axis.
Phase 3: System construction
"This is a 96-axis, 270-foot long, fully automated, robotic system," Richardson said. "There's a lot of money in the production of that. We used the first two phases to eliminate the two technical risk elements. That gave us an off ramp: If either of those phases had failed, we could have stopped the program or modified the design to make it work before we proceeded on to final construction.
"Except for those two risk elements, we knew ahead of time that everything else would work because we had built similar machines before."
|
| Figure 3. A fixture holds eight vision pins in the correct relative locations to align the lugs for the port (left facing forward) wing. Courtesy of Lockheed Martin. |
Figure 3 shows several finished vision pins mounted in a fixture that holds them in the correct relative position. Absolute positions of the wing-lug sets are unimportant. What is important is the positioning of the wing lugs relative to each other. They must match the standard positions of wing lugs on the wings.
Each fixture provides correct relative positioning for a set of vision pins. To align a pair of fuselage sections, the Lockheed Martin assembly team brings them together in very nearly the correct positions with the fixtures pulled well out of the way. "They can easily rough align the sections to well within an inch," Richardson reported.
For this rough alignment, the sections move around on two "skates," or motorized platforms riding on precision steel rails. "The skates are also able to drive freely about on the concrete floor," noted Mike Brewer, mechanical designer on the project for Delta Sigma. The skates have four individually driven "casters" employing fork-lift wheels to contact the floor.
The Lockheed Martin assembly team uses the skates to ferry the aircraft pieces from assembly station to assembly station (there are 11 total assembly stations in the F/A-22 assembly area), until they put the landing gear on. After that, the aircraft roll on their landing gear all the time they are on the ground for the rest of their lives.
To switch from free roaming a skate around the floor to having it locked on the precision rails, the skate operator simply drives it over the rails. A V-shaped guide engages a keel under the skate, forcing it to line up over the rails.
The operator presses a button to engage motors that raise the drive wheels so the skate "squats" down onto the rails. The rails then carry the weight. Spring tension in the castor mechanism keeps the wheels pressed against the concrete floor to provide forward and backward drive along the rails.
With the fuselage sections rough aligned, the team moves the alignment fixtures in place so the vision pins look through their respective wing lugs. Pressing one "button" on the system console's touch screen flashes the LED arrays, recording images through all the vision-pin cameras simultaneously. The system calculates all the displacements and feeds that information to the fine-alignment actuators, which make their adjustments.
It is an iterative process. "If I'm, say, 50 thousandths out," Haisty explained, "The first correction will get within five thousandths. Then, we simply remeasure and make another move. We then repeat the process until we are satisfied."
Each movement takes approximately 10 s. Once the aircraft is ready and all the fixtures are in position, the whole alignment process can take as little as 1 min—although the assembly team usually takes longer "just to make sure."
No related content found.
- 0 rated items found.
Datasheets.com Electronic Parts & Inventory Search
185 million searchable parts
- Part Number
- Description
- Inventory
- Products
- Manufacturers
View All Webcasts
