A Software-Based AI Approach to Robot Control
The rapidly evolving and increasingly complex hardware industry demands updated and more flexible automation in manufacturing, but such updates are often costly and impractical considering the scale of production and pace of changes. Artificial Intelligence (AI) can help bridge this “flexibility gap.”
ZF, one of the world’s largest automotive suppliers, recently looked to automate the workpiece feed of a large-volume milling station that produces gears.
In the work process, metal rings are removed from a crate and placed on a conveyor belt, where they later flow into the gear production. Several factors made this process difficult to automate:
- The production step is highly variable, as the rings shift in the delivered mesh box, so they are randomly arranged.
- The placement and shape of the box often varied as did light conditions.
- The surface of the rings could be shiny and clean, oil-smeared, or even corroded, making classic automation impossible.
Today, ZF uses an artificial intelligence (AI)-based control system with a collaborative robot (cobot) in an automated workpiece fixture. ZF’s own controller first positions the cobot over the rings in the crate. The AI-based control system then takes over, moving the robot independently to the next ring to bring the gripper into the correct three-dimensional gripping position. The cobot then resumes control, picking up the ring and moving it to the conveyor belt to be placed. The complete setup of the cobot took only a few days, and it solved a long-standing problem.
This is just one example of how robots and other forms of industrial automation enable manufacturers to improve their agility, adapt production processes to address rapidly changing markets, increase innovation, produce more cost-effectively, and shrink product delivery times. Automation also relieves human workers from performing repetitious, monotonous and sometimes dangerous tasks.
Traditionally, automation has been primarily a hardware-based process, with purpose-built machines that can provide substantial production savings for high-volume applications with long product life cycles. However, companies like ZF are beginning to recognize that manually programming, reconfiguring and maintaining hardware-based control systems is time-consuming, and that it can be unprofitable to deploy industrial robots, especially in high-mix, low-volume manufacturing. This is especially true for markets increasingly demanding individual, high quality products.
THE FLEXIBILITY GAP
Prior to deploying this AI-based control system and cobot, ZF faced a problem that has hampered modern automation for some time—the “flexibility gap,” the range between viable small batch manufacturing and high-volume automation. In short, the more that flexibility is required in a manufacturing process, the more complex the automation must be to accommodate it. When machines are built to handle specialized, complex tasks, the cost can only be justified by high volumes and long product life cycles. Further, this type of automation is not easy to repurpose as products continue to evolve in faster and faster cycles.
When volumes are not high enough or a product has too many variances, people can take on the production tasks that require more flexibility and lower costs than conventional robots can perform. Unfortunately, the ongoing labor shortage has made finding employees willing to perform these types of “dull, dirty and dangerous” tasks difficult, ultimately halting production lines.
With more manufacturing falling into this flexibility gap, manufacturers need a new approach to reliably meet demand for their products.
SOFTWARE-BASED AI CONTROL FOR ROBOTS
As ZF and others have experienced, AI is the key to overcoming the flexibility gap. Instead of relying on specific, purpose-built hardware, manufacturers need to be able to bring the flexibility of a human into robot-based processes. With AI, they can enhance existing robots to cost-effectively deal with variances in real-time. In this way, processes can be planned, optimized and automated with maximum agility and efficiency.
Instead of investing in increasingly complex and expensive hardware with a shorter viable lifecycle, manufacturers can use smart software to extend the capabilities and lifespan of their robots. Advances in AI, especially in machine learning (ML), offer companies efficient ways to plan and improve their processes. In fact, production facilities are already using ML to substantially improve monitoring and maintenance.
ROBOTS + AI = ROBUST HANDLING OF VARIANCES
Software closes the flexibility gap by combining robots with AI to automate manual workstations. With the help of AI-driven controls connected to a camera, robots gain hand-eye coordination, one example of human-like flexibility. These enhanced robotic systems can be easily trained by a human to understand their general tasks. Using AI, the system then generalizes the training across new situations and comparable variances in the manufacturing process, including differently shaped or positioned workpieces. The robot is now able to accommodate variances by independently adjusting movements in real time as needed.
A robot in combination with an AI controller such as MIRAI, the AI control system used by ZF (Figure 1), can be trained in just a few hours through human demonstration. Numerous activities, including picking individual parts, infeed movements, joining, and tracking, can be implemented for the first time with a single small camera mounted on the robot‘s wrist, as well as a special robot controller. No AI skills or programming skills are required. Instead, with the camera in place, the robot is taught what to do by demonstrating the required activity in typically occurring variances (Figure 2). During training, MIRAI aggregates all of the data necessary to build out a neural network in the cloud that can efficiently and accurately perform the task and its variances. To assure reliability and safety of operations, all cloud activities meet the highest cloud safety standards.
This software enhances the capabilities of an existing tool by extending the robot’s ability to accommodate variance by building positioning or motion skills the software allows. In this simple and straightforward way, the AI can be quickly trained, making it possible to automate high-mix, low-volume production with intelligent robot systems—even under a time crunch. This approach combines the best of robotic automation with the flexibility of manual processes while minimizing cost and time-to-market.
RELIABLE QUALITY CONTROL MANAGEMENT: AI CHECKS FOR COOLANT LEAKS AT BSH SPAIN
Another premise for success in manufacturing is a consistently high level in product quality, which is why accurate quality management is so important. Equally important, however, is the overriding goal of reducing the workload for humans.
BSH Hausgeräte GmbH, a major European white goods manufacturer, proved that it is possible to find an economically feasible and intuitive solution to both requirements. BSH Spain relied on automation to check coolant leaks on refrigerators. So-called “fridge sniffing” is a monotonous, error-prone activity, where refrigerator manufacturers use hand-held probes to search for coolant leaks in the solder joints of compressors and copper piping. When leaks go undetected, harmful substances can escape, so this is an important safety test. To ensure the pipes are leak-proof, a probe is brought to within a millimeter of a solder joint, which can vary in position. Now, BSH Spain uses an AI control system to guide the robot to the joint to detect possible leaks. In this way, the robot can perform a tedious task with repeatable precision and consistent quality.
SOFTWARE-BASED AI—THE FUTURE OF MANUFACTURING
Enhancing a robot’s ability to accommodate variances is just one example of how AI bridges the flexibility gap. With AI, production efficiency, reliability, and quality can be improved and maintained. However, the benefits of AI can extend far beyond automating these manufacturing processes.
Through AI and ML, for example, decision-makers can have faster access to real-time information about production lines, supply chains, and product operation. This enables them to better evaluate future product development, new business models, and overall strategic decisions.
This information can also be used by robots to improve their own operation. Predictive maintenance, for example, is a technology where AI can track robotic operations across an entire factory floor. Over time, the AI can predict when various equipment will need maintenance. Rather than shut down production when equipment fails, such maintenance can be proactively scheduled so that the impact on production is minimized.
AI in automation is the key factor for future success for manufacturers from all industries. Instead of relying solely on complex and expensive hardware solutions or manual workstations, AI software can extend the necessary agility for robotic equipment to overcome the flexibility gap. AI not only accelerates automation and the training of robots, it also changes how developers and designers can plan production to be more economical as products become more complex.
As the ZF and BSH Spain use cases demonstrate, companies that adopt AI-based sofware to augment their automated processes can quickly adapt to changing market conditions, shifts in customer requirements, and shorter product lifecycles. As a result, they can beat their competition to market, putting them on the right path for continued success.
Micropsi Industries | www.micropsi-industries.com
Micropisi Industries provides AI-based software for industrial and collaborative robots (cobots). MIRAI, the company’s flagship product, allows these robotic arms to be controlled in real time, in direct response to sensor information. This is made possible by AI that enables the robots to learn from humans and deal with variances such as changing lighting conditions or workplaces whose position or composition deviates so they can more easily and cost-effectively operate in dynamic environments.
PUBLISHED IN CIRCUIT CELLAR MAGAZINE • NOVEMBER 2022 #388 – Get a PDF of the issueSponsor this Article