Vision sensors improve pickup truck part handling at DaimlerChrysler

Citation

(2002), "Vision sensors improve pickup truck part handling at DaimlerChrysler", Industrial Robot, Vol. 29 No. 4. https://doi.org/10.1108/ir.2002.04929daf.006

Vision sensors improve pickup truck part handling at DaimlerChrysler

Vision sensors improve pickup truck part handling at DaimlerChrysler

Keywords: Machine vision, Automotive

At the DaimlerChrysler Stamping Plant in Twinsburg, Ohio, an automated process of handling stamped truck-bed parts was needed for the company's full-sized Dodge pickup truck production. Without automation, the 6øft and 8ft box-side inner parts for 2002 Ram trucks would need to be handled by operators. The operators, working side by side in the cell, would have to unload a rack of 18 parts, carry each one over to another station, and place onto a fixture. However, there were a number of problems with this approach.

First, safety issues were a concern. Because the parts are stamped, points are often very sharp and can cut through work gloves. The parts are also large and heavy, making them difficult to carry and leaving the operators susceptible to repetitive motion injuries. Labor costs were also a factor, due in part to higher medical costs associated with injury. Finally, there was also the issue of throughput. The manual loading/unloading process would not guarantee consistent cycle times, and the welding line's throughput was dependent on the rate of load/unload.

To automate the process, DaimlerChrysler called on Cognex, Nachi Robotic Systems and Shafi Inc. to develop a vision-guided robotic system. The main goals of the project were to minimise labor costs and safety issues and speed up cycle time (Plate 6).

"We needed to launch the new line in July 2001, and the idea to automate came up sometime in May", according to Brad Dailey, manufacturing and assembly project and tooling manager at the Twinsburg plant. "With a very strong team effort, we were able to go from initial concept to installation within six weeks."

The integrated system comprises eight Cognex In-Sight 1000 vision sensors, a Nachi SF166 pick and place robot, and Shafi Reliabot PC Vision/Robot Integrated HMI software. In the first stage of the process, stamped parts arrive in a 1,700lb rack, which is placed onto a turntable by a forklift. Before the unloading process begins, the vision sensors are calibrated to the robot using the Reliabot software, which guides operators step by step through a simple calibration procedure. Then, a series of "sanity check" inspections are performed by the vision sensors to ensure the correct parts are present, the parts are spaced properly, the rack is square, and the rack's securing bar is in an open position that will allow the robot access to the parts.

Plate 6 Daimler Chrysler, Twinsberg

A total of six In-Sight 1000 vision sensors, mounted 9ft above the rack, perform these inspections. The In-Sight 1000 is a self-contained general-purpose vision sensor featuring a full library of vision software tools, a vision spreadsheet interface for application set up, and a rugged, lightweight enclosure. The sensors also have built-in network communications, allowing them to be linked together over Ethernet. This enables operators to set up and modify inspections from a central server, perform statistical process control on the combined pass/fail results data from all six sensors. Results are tabulated and logged using the Reliabot software, which enables operators to perform flexible queries about past manufacturing performance.

According to Adil Shafi, president of Shafi Inc., the primary challenge for vision in the sanity check phase was to be able to reliably perform the inspections no matter what the racks looked like. "The plant uses over 200 different racks, none of which are designed with machine vision in mind", explains Shafi. "The racks often get paint and labels on them, and can also be very dirty in appearance. Also, the geometry of the racks is not very repeatable because they can be bent in places. The vision sensor still had to recognise part features in the scene and inspect them amidst a lot of random visual noise. The only other option was to retool each rack and made them perfect for the purposes of the vision sensors, which would be enormously expensive for the company."

To address this challenge, the In-Sight vision sensors use PatFind geometric pattern matching software, which is able to locate part features by geometric shape, despite appearance variations or image background noise. Once PatFind locates the features, blob analysis and edge detection software tools are then used to perform the various inspections.

Once the parts have passed inspection, the turntable rotates the rack of parts 180 degrees into the unloading position. The robot grips the upright parts one at a time using a 4in wide, vacuum suction gripper, turns the part horizontally, and moves it to a point just above the assembly fixture on which it is to be placed. The fixture, which contains two locking pins, requires the robot to place each part within a ±1mm positional tolerance.

To ensure accurate placement guidance, two additional vision sensors, which are mounted to the plant ceiling 12ft above the parts, are used to determine the precise X, Y, and theta coordinates of each part. The coordinates are then sent via serial communications to the robot so it can perform any necessary linear or angular offset prior to placement.

"Image background noise and clutter was an issue at this stage as well", adds Shafi. "There can be sparks flying around from the welding line, and the parts themselves are occluded by the robot gripper cups, leaving only portions of the part visible to the vision sensors. The PatFind tool is able to locate the reference marks we are looking for while ignoring the clutter, resulting in accurate, repeatable placement of the parts."

Using the Reliabot software, operators are able to view a live image of the inspections as they occur, pass/fail data, as well as real-time visual feedback on system status. "When the system is running, our software posts messages on the screen to tell operators and technicians exactly what's going on, such as when the robot is moving towards the rack", explains Shafi. "This feedback empowers the guys on the floor to perform troubleshooting on all aspects of the system without having to drain the plant of technical resources and resultant downtime."

Since the vision-guided de-racking system was installed in July 2001, it has been able to meet all the criteria slated for the project. First, it has reduced labor costs, and has helped make the workplace safer by removing skilled operators from a potentially dangerous material handling environment. Secondly, it has improved line throughput by reducing the cycle time from 20.5 seconds per part down to 17 seconds. The solution has also provided the added benefit of allowing DaimlerChrysler to automate the part handling process without having to alter the existing rack design of more than 200 racks built.

Perhaps most important, the company sees a quick return on investment. "The bottom line is that we've improved throughput issues, and the labor cost savings are significant", says Dailey. "Based on these factors, we're looking at a six-month return on investment."