In this assembly situation, all specified parts must be placed correctly in each product variation. Any errors will reduce product quality or increase costs when faulty assemblies have to be repaired or scrapped. Reducing the number of errors — and incurring some cost to prevent them from occurring — contributes directly to a company’s bottom line. In building a case that justifies spending money to save money, experience has shown that reducing scrap rapidly surpasses all other cost benefits in any process that produces more than a trivial amount of scrap. In addition, any throughput gains will increase the company’s overall revenue.
Polaris Industries can attest to the benefits of investing a relatively small amount of money in automation to reduce a lot of costly scrap. For many years, the company had relied on touch and proximity sensors to verify which parts needed welding, because they feared that a machine vision system would not be able to perform reliably in such a harsh environment.
With touch sensing, a robot uses the end or side of a weld wire or, in some cases, the nozzle to physically touch off on parts. This occurs after a weld cycle has been initiated. Proximity sensors are used on weld tooling to prevent a weld cycle from initiating, or to prevent other operations from occurring unless the sequence is done in a specific order, or a part has been missed.
The time and costs required to sustain this sensor-based inspection system finally drove Polaris to install a machine vision inspection system using cameras supplied by Teledyne Dalsa’s Industrial Products group. Polaris welding engineer Jeff Steiner worked with Tom Wright of Hartfiel Automation to design the system. The new solution has significantly reduced both manufacturing cycle time and product scrap.
Shortcomings of the sensing solution
At the Polaris facility in Spirit Lake, Iowa, an operator loads frames into a fixture at the first of several stations on a welding manufacturing line. These frames contain the necessary parts for a particular model of the product being manufactured. As a first station or loaded fixture is presented to arc welding robots with Ferris wheel-type and turn table-type positioners, a second station is then presented to the robot operator who unloads the previously welded part and loads new parts while the robots are working.
The frame design for a particular product family allows parts to be loaded only in the correct orientation; however, corresponding parts for different models in the same product family often look very much alike. For example, two parts might be slightly different sizes, a discrepancy small enough that a human operator may not notice. An operator also may inadvertently neglect to load one of the parts, or may load it out of position. Therefore, verification that each part is present and correct is critical before the robot welds it to the frame.
In the original process, the robots performed the verification using touch and proximity sensors. This approach created a bottleneck, taking as much as 30 seconds of cycle time per part. If the robot detected a wrong or missing part, the operator had to retract the robot arm, correct the problem, reposition the arm, and run the touch inspection again before allowing assembly to resume. This lengthy process resulted in considerable additional lost production time. And, even with the checks in place, the scrap rate could be as high as five percent per day at any given area. Since the frames cannot be repaired once they emerge from the weld cell, a five percent loss in materials is incurred, as well as the efficiency wasted taking the time to generate scrap.
Vision solution provides reliability and speed
The production environment in any high output welding department does not offer the best platform for incorporating machine vision inspection. Impediments such as poor lighting and smoke from the welding equipment can obscure images being acquired, making a vision-based approach both more complicated and less reliable. Initial assessment of the technique also suggested that programming a camera and keeping it clean would add considerably to cycle time and overall manufacturing costs. Nevertheless, the existing arrangement of utilizing touch-sensing technology and proximity sensor systems encouraged Polaris to look into vision as a viable alternative solution.
Teledyne Dalsa supplied a solution that performs quickly and reliably. One vision controller and two cameras are installed behind the operator, away from the more adverse conditions closer to the welding robot. The 60 frame-per-second, 640 by 480 pixel cameras verify that the frames are loaded correctly and are ready to proceed to welding, reducing inspection time to less than 200 milliseconds per part. A display shows the inspection results, colouring correct parts green and incorrect ones red. An operator can make any necessary corrections before the frame ever reaches the welding robots so that the product flow is not interrupted – another significant time-saver.
Programming touch sense routines for the robot requires downtime from several minutes to hours. More considerable time is required when a toolmaker is required to machine tooling details to house proximity sensors. Depending on the amount of proximity sensors required for a given part, this time could turn into days until completion. By contrast, Polaris engineers programmed the 22 parts in the first frame type to be used in the vision-based process in less than 15 minutes. Programming another 19 scenarios offline took less than an hour. Updating a vision program to incorporate part or tooling changes can be accomplished in only a few minutes.
Looking at the benefits that pay back the cost of the machine vision system requires considering only the cost differences between the new process and the older one. Any costs that remain the same are irrelevant for this purpose.
The investment for cameras, controllers and other components — including engineering time and installation labour — totalled $9,500 US. The new process saved Polaris approximately 4.5 minutes per hour at this robotic workcell. With the current hours of runtime per shift at one shift per day, that savings translated to an additional 1.6 extra frames per hour, or 9.8 frames per day. The increased efficiency allowed Polaris to reduce overall labour for the weldments, or welded joints, produced at this workcell by 5.5 percent.
In addition, the number of frames that had to be scrapped dropped by 23 percent, or eight per day. The cost of scrapping an assembly exactly equals the manufacturing cost of replacing it. With a manufacturing cost of $50 per frame, the vision system increased revenue by $490 per day and reduced scrap costs by $400 per day. That $890 daily benefit means that the vision system paid for itself in only 10.7 work days, or just over two five-day weeks. Other benefits aside, payback from the scrap alone would be 23.8 days, or a little over a month.
Making it better at Polaris
The Polaris company creed, etched in steel at the entrance to each of its locations, is “Understand the riding experience, Live the riding experience, Work to make it better.” Since working to make the manufacturing process of its vehicles better in Spirit Lake, Polaris has installed similar vision solutions at several other robot systems and has benefited from similar cost savings at each of them.
Steve Geraghty, is VP of U.S. Operations, and Director of Teledyne Dalsa’s Industrial Products.