Vision System Keeps Engine Production Moving
What started out as a simple pass-fail test system
ended up performing over thirty automated inspections.
The following is a manuscript for an article published in Vision
Systems Design magazine. Vision Systems Design magazine holds the
copyright for the finished article.
by C.G Masi, Contributing Editor
When Milwaukee, Wisc. motorcycle manufacturer Harley Davidson approached
Scott Woida, President of Midwest Engineering Systems in West Allis,
Wisc. to design a system that would verify the correct timing-chain
installation in their new Twin-Cam 88 engines, they only envisioned
making the one pass-fail inspection.
"As soon as Harley learned that we were considering a vision
system to do this," Scott recalls, "they started to say:
'Well, maybe we could also ensure that the washers are put on correctly,
and the sprockets are tightened down all the way. Then, we could
see that the guides are pressed down completely. As long as we are
looking at these things that span the entire engine, let's look
at all this stuff!'
"So, we ended up performing 30 some-odd inspections on the
engine."
To make it run more smoothly than their previous engines, Harley
had designed the new motor with two balance shafts installed parallel
to the crankshaft. One balance shaft is ahead of the crankshaft
to drive the intake and exhaust valves for the front cylinder. The
other is behind the crankshaft to drive the rear-cylinder valves
(See Fig. 1). A timing chain running over sprockets on the three
shafts keeps them turning in synchrony.
|
Fig. 1: Harley Davidson's Twin-Cam 88 engine has three
main shafts: the crank shaft and two balance shafts. A timing
chain keeps all three shafts running together. Courtesy Midwest
Engineering Services, West Allis, Wisc. |
The original timing-chain inspection was needed to ensure that
the timing chain was installed correctly so that the intake and
exhaust valves would open and close at the proper time with respect
to the crankshaft's rotation. When the timing chain is installed
correctly, marks cast into the sprockets line up with discolored
links in the timing chain, as Fig. 2 shows.
|
Fig. 2: The original inspection consisted of ensuring that
marks cast into the timing-chain sprockets lined up with discolored
links in the timing chain. Courtesy Midwest Engineering Services,
West Allis, Wisc. |
This is quite an easy task for a machine-vision-based inspection
system to perform. Additional inspection tasks that involve verifying
the presence of all the components that should have been installed
by this point in the assembly process are also fairly simple for
an automated inspection system. When Harley asked that the vision
system be able to verify that certain components were properly tightened,
however, things became more difficult.
For example, there are two spring-loaded chain tensioners visible
in Fig. 1. To ensure proper chain tension, the
tensioners have to be rammed all the way down in their sockets.
The vision system can verify that they were rammed all the way home
by measuring the distance from the casting surface to the top of
the chain-tensioner body. Making that measurement, however, requires
precise positioning of the camera with respect to the target.
Precisely positioning the camera is complicated by the fact that
the engines are carried through the assembly area on J-hooks, which
are approximately 12 feet (4 m) long, suspended from an overhead
conveyor. These J-hooks can only be relied on to position the engine
in the inspection cell to within +/- 2 inches (5 cm) in x, y, and
z, and several degrees rotation around the vertical (z) axis.
Later, Midwest engineers realized that there was also a large planarity
error as well. That is, engines do not always sit flat on the J-hook,
so the plane in which the timing chain runs can not be relied on
to line up with the x, y (horizontal) plane.
System Architecture
The engineers' first idea was to use a mechanical fixture to position
the engine under inspection with respect to several cameras. They
soon realized, however, that they could make a more robust, flexible
system that was faster and easier to use by mounting a single camera
on a five-axis robot arm as shown in Fig. 3.
|
Fig. 3: The final design consisted of a single camera mounted
on a five-axis robot arm. Courtesy Midwest Engineering Services,
West Allis, Wisc. |
Fig. 4 shows the system they came up with. The camera is a Series
600 SmartImage Sensor from DVT in Atlanta, Ga., which incorporates
a sophisticated image-processing system in the camera body. The
robot is a Model 1400 robot manufactured by ABB in New Berlin, Wisc.
The HMI (human-machine interface) terminal is a desktop PC running
software the engineers wrote to communicate with the other system
components. The system components communicate with each other by
being nodes on an Ethernet network, which also communicates with
the factorywide intranet.
|
Fig. 4: The inspection system has three components (camera,
robot and HMI interface terminal) communicating via Ethernet.
Courtesy C.G. Masi, Golden Valley, Ariz. |
The robot's internal computer controls the inspection process.
It knows which inspections should be done in what order.
The SmartSensor camera, on the other hand, knows how to do the
inspection. Its internal computer is configured with all of the
image-processing routines needed, and programmed with the particular
image analysis procedure needed for each inspection.
Woida divides the procedures into two classes: orientations and
inspections. Orientations are visual-feedback-directed robot motions
used to move the camera to the precise location and orientation
to capture the image needed. An inspection is the capture and analysis
of that image.
Inspections are generally done in groups of up to six. Thus, the
system might perform an orientation to capture the image in Fig.
5, where the camera is looking the timing chain edge on. The camera
captures the image and performs its group of inspections. These
may include determining whether the sprockets have been tightened
down (shown) and whether the chain is aligned with the horizontal
plane. The camera's internal computer then sends an Ethernet packet
through the Ethernet hub to the HMI interface to inform the operator
whether the unit passes. The packet is formatted to carry results
of up to six tests.
|
Fig. 5: Navigating the camera to capture an in-plane image
allows the system to make sure several bolts have been tightened
down. Courtesy Midwest Engineering Services, West Allis, Wisc. |
If any of the tests fail, the camera also sends a failed-image
file, which is stored in a database on the PC. Then, the system
goes on to complete the rest of the inspection sequence to find
any other failures.
After finishing the entire inspection sequence, the PC tells the
operator whether the unit passes or, if there has been one or more
failures, what corrective action to take. There are 20 seconds built
into the 80 seconds each engine spends at this inspection station
to allow repairs.
Using Ethernet to tie the system components together has a number
of advantages. Installation and cabling is much simpler than it
would be using other interconnection protocols. Each component (smart
camera, PC and robot) has its own IP address, so intra-system communication
can use standard network protocols and software. Finally, Ethernet
allows seamless integration into the factory wide network, so the
inspection system does not become an island of automation apart
from the factory management system.
"I can sit anywhere in the plant and monitor that camera,"
says Woida. "I can see what it's seeing. If I have the correct
access and security, I can actually program that camera from anywhere
on the network."
Making It Work
Lighting, of course, is one of the most critical elements of any
inspection system development. The inspection cell itself is curtained
off to control lighting variations as well as to prevent human intrusion
into the space while the robot is operating.
Three lighting sources are used. First, area floodlights provide
bright, even illumination throughout the cell. The SmartSensor camera
comes with a high-intensity LED light source mounted on its body,
which the inspection system uses at the start of the inspection
sequence, when it is first trying to locate the engine in the cell.
Finally, when the camera gets in so close that its body may block
the area lighting, a ring light mounted around the lens floods the
visual field with even illumination.
The inspection sequence starts with the robot well away from the
J-hook carrying the engine to be inspected. Supposedly, the J-hook
is locked into a certain orientation, carries the engine laying
flat on its side (with the crankshaft pointed vertically), and stops
in a certain spot.
Any of those assumptions, however, could be wrong. The inspection
system, therefore, has to start by checking to make sure it isn't
going to damage itself by bumping into the engine or the hook. From
that rough evaluation of the scene, it can determine where it should
start maneuvering.
All of the orientation procedures are designed as feedback loops.
To locate itself with respect to the engine in the x, y plane, for
example, the camera takes an overhead image of the engine. The camera's
image processor locates the crankshaft end and calculates how far
it appears from the image center. The processor then sends a two-dimensional
displacement vector over the Ethernet to the robot, which moves
the camera. The camera then takes another image and sends a correction
back to the robot, which moves the camera again. The feedback loop
keeps going until the displacement error falls within a preset tolerance.
Once the camera is centered over the crankshaft end, the system
uses a second feedback loop to rotate itself into q-alignment using
a second feature to determine the required rotation angle at each
step.
To locate itself in the z direction, the system locates a certain
machined hole in the engine case that has a known diameter. The
image processor measures the hole's diameter (in pixels) at the
image plane, compares that with a target apparent diameter, and
estimates how far to move the camera toward or away from the engine
to achieve that target. That estimate then goes to the robot, which
moves the camera, and so forth until the apparent diameter comes
within its tolerance.
Finding the planarity error is the most complicated orientation
procedure. To estimate the planarity error, the image processor
measures the apparent diameters of three holes and uses solid trigonometry
to calculate a correction, which will comprise a three-dimensional
displacement and a j-angle correction. Again, the feedback loop
continues until the correction magnitude falls below a tolerance.
It took about a year from the time Harley Davidson first approached
Midwest Engineering Services with the inspection problem to the
time the first system was fully operational. The engineers spent
about six months to analyzing the problem, developing a solution
and building the system. Then, it took about four months to get
it installed on the production floor-mostly because the installation
had to be coordinated with Harley's annual Christmas shutdown. "Then,"
Woida recalls, "we sat on their floor for another two months
and made all of those changes.... Such as where you watch 100 parts
go by and make sure the system can handle all the different variations
between the parts."
There were two main difficulties the engineers had to overcome
in this project: planarity errors and part variations. "Once
we licked [the planarity] problem," Woida recalls, "most
of our problems went away."
Part variations arise from the fact that Harley uses multiple suppliers
for many engine components. Harley's designers define the significant
engineering specifications, but many properties are left to the
suppliers' discretion.
For example, the dimensions of a washer inside of the engine were
defined, but the finish (which makes no difference to the engine's
operation) was not. So, washers arrived with anything from a polished
surface to a matte anodized finish. While that variation didn't
make any difference to the engine, it gave the inspection system
fits!
For some variations, it was possible to open up the tolerances
on what the vision system would accept. In other instances, such
as the washer situation, Harley had to add a finish specification
to their part requirements.
"This was our initial [inspection system] project," Woida
points out. "It was a learning experience. Since then, we have
developed a piece of software that almost automates the creation
of the inspection routine. There has been a lot less custom code
that had to be written for subsequent versions."
|