La caméra associée au robot accroît la qualité et réduit l'investissement
The conventional approach to replacing the slides with machine vision would have required as many as 30 different fixed cameras, each with special lighting requirements. The Piston Group developed a much less expensive and more flexible solution by mounting a single Cognex In-Sight® 5603 vision system on a Fanuc® robot. The robot moves the vision system into position to capture the 30+ images in less than 45 seconds, completely inspecting the module. The In-Sight vision system can be modified to inspect for future design changes easily with a few hours of programming time. The new inspection system has substantially improved quality by inspecting more points at a higher level of accuracy while reducing initial investment by 40% and retool cost by 80%.
The Piston Group provides sequenced and non-sequenced sub-assembled components for complex modular assemblies. The company assembles modules ranging from front end cooling systems, suspension and chassis systems, interior systems, and power train systems. The cooling module produced in this application consists of essentially everything between the motor and front bumper: the core support, radiator, electric fan, AC condenser, power steering and transmission coolers, reservoirs, hoses, wiring harnesses, and many other small components. The modules are built in many different configurations. For example, most lines have over 20 different wiring harnesses are used depending in the model and options selected by the customer. The customer sends The Piston Group a daily release that indicates the required front module configurations and their build sequence.
Demanding quality requirementsThe quality requirements for the modules are demanding. First, each module must be correctly configured for the vehicle it will be installed on, with the proper wiring harness and other components. Second, many components need to be installed within tight dimensional tolerances. Several hoses and clamps must be installed within 1 mm of a specified location. All electrical connectors must be fully seated and engaged.
In the past, many of these inspections were performed in a check station by mechanical probes mounted on slides. This approach required that a custom slide be designed and built for each dimension that was checked. Whenever the dimension was changed, the slide had to be modified or replaced with a new design. The cost of retooling for a all new vehicle model was typically about $150,000 and required approximately two weeks of downtime to retrofit the entire check station. Prior to model changeover, the company would build small quantities of pre-production new model parts and these parts had to be inspected manually due to the station not being changed over. Many inspections, such as determining whether parts were identified and installed properly, were performed by 200% visual inspection by Quality Engineers. The flexibility of the new system allows for accommodation of both current and new model production, reducing this reliance on visual inspection.
“Every time the customer made a single engineering change, the cost was a minimum of $15,000,” said Kevin Miller, Director of Manufacturing Engineering for The Piston Group. “We wanted to implement a flexible vision system to reduce cost and turnaround time on changes. We also wanted to reduce the amount of required manual inspection to improve quality. The normal practice is to use one camera per inspection point. This approach would have taken up too much space on the existing equipment and the cost was too high.”
Using a single vision system for multiple inspectionsSeparate cameras have traditionally been used for each inspection point because each point generally requires very specific lighting and camera focal distance in order to achieve the required level of accuracy. Camera speed has also been a concern when considering the idea of using a single camera to take multiple images within a single cycle. But the evolution of Cognex vision system technology makes it possible for a single robot-mounted camera to inspect a large number of points with a high level of accuracy.
“We worked with the Vision & Traceability Group of McNaughton McKay Electric to identify the right camera for this application,” said Patrick O’Dell, Control Engineer for The Piston Group. “We selected Cognex because its vision systems provide the wide range of tools needed to inspect the many different points involved in this application. Also, Cognex’s PatMax® geometric pattern matching tool provides substantial improvements in accuracy by accurately determining the part location. A single Cognex In-Sight 5603 met the requirements of this application by accurately inspecting 30+ very different features in many different locations in less than 45 seconds. Plant wide we are using this camera for more than 90 different inspections”
Programming the vision systemO’Dell programmed the inspection application with Cognex’s In-Sight software which uses a spreadsheet programming interface. “The spreadsheet interface provides the ultimate in programming flexibility,” O’Dell said. “It provides access to every imaginable vision tool and lets me create a new inspection operation simply by copying a similar one and making a few tweaks.” The first operation of the inspection station consists of reading the parts RFID tag to determine the module type. The type of module determines both the robot program / path and vision inspection program.
O’Dell used the PatMax tool to determine the position of the fixed location of the cooling module and then based all subsequent inspection operations on this position. He has used histogram tools to determine the presence or absence of components in checking for the right module content. He determined the position of the hose clamps by using PatMax tools to locate the hose clamp and a feature on the matting component, such as a form rolled bead stop. Then he used a distance measurement tool to determine the distance between the hose clamp and this feature. For each measurement, O’Dell entered a high and low value in the spreadsheet. The values in the spreadsheet represent brightness for the histogram and distance for the measurements. Bar codes are also read on several components to be sure the correct part number is installed.
O’Dell tried several different lighting arrangements and found one, a ring light integrated into the camera, that worked for each application. All images from each inspection are serialized and collected on a network server for traceability back to each assembled module. O’Dell tried several different methods for storing images and found the fastest was to store them locally during the cycle and then transfer them to the network while the part is transferred. A ControlLogix programmable logic controller (PLC) controls both the robot and the vision system. Only small sections of the spreadsheet are enabled when processing each image to save processing time. Cognex Connect™ includes support for the most commonly used open-standard Industrial Ethernet and Fieldbus communications protocols for trouble-free connection to PLCs and a wide range of automation devices from Mitsubishi, Rockwell, Siemens and other manufacturers.
O’Dell performed a Gage Repeatability and Reliability (R&R) study on the vision system as part of the Production Part Approval Process (PPAP) submission for the cooling module. He set the values in the spreadsheet so that no bad parts would escape. This setting requires that a small number, less than .4%, of good parts may fail the inspection. These false rejects require inspection by a Quality Technician and can be passed with a password, which is recorded for traceability. There are a few items that cannot be inspected by the vision system because they have too much variation, such as the wire harness push pin locations. He provided feedback to the customer’s design organization to push for changes that will make these items easier to inspect with a vision system.
When a part fails the inspection, the program captures and processes another image. If the part fails again, the PLC sounds an alarm and the RSViewME human machine interface (HMI) presents the inspection image that failed on the screen along with a verbal description of the failure. Either the operator or the team leader reacts to the failure. In some cases, such as if the hose clamp is out of position or a push pin is missing, the operator can unlock the gate and fix the part. All inspection operations are then re-run to make sure that fixing the issue did not cause another failure. If the part passes the test, then the line continues from the point at which it stopped.
Storing images avoids penaltiesThe ability to connect to a server and store each image is valuable. In one case, a customer said they had received a brake corner with a missing brake pad. The stored image, however, showed that the brake pad was present when the module was built at Piston. The customer investigated and discovered that the vehicle was repaired for another reason and during the repair the brake pad was knocked off. “In the past, we would have had to pay a penalty and our quality score would have been reduced,” O’Dell said. “Now we can use hard data to prove we are not at fault.”
The vision system is able to inspect to a much higher level of accuracy than previous inspection methods. For example, mechanical probes were only able to determine hose clamp position to an accuracy of +/- 4 mm, due to the amount of stack up tolerances in the components and how that was tracked back to our fixture datum points. While the vision system now provides +/- 1 mm accuracy, because we are able to locate specific points within the view of the camera utilizing PatMax and take a dimensional reading from there instead of trying to gage against the fixture datum points. The vision system also inspects twice the number of points that the company was able to inspect in the past with mechanical probes. “The bottom is line that we have improved quality by inspecting more points with a higher level of accuracy with 40% less upfront investment and 80% lower cost for changes than would have been required by a traditional check station,” Miller concluded.