Engineering researchers at the University of Michigan are working to increase factory productivity by spotting glitches on the factory floor in real time and reconfiguring around them.
Advanced manufacturing plants today hold hundreds of software and hardware components. Robots, conveyor belts, sensors, control systems and communication networks work closely together.
The rise of automation, which while increasing efficiency and raising quality, also brings with it vulnerabilities. For example, while networked robots can be remotely controlled or repaired for increased efficiency of operations, such capabilities can also be weak links in the cybersecurity of the system.
Machine failures, operators' mistakes and cyberattacks can halt production, and lead to expensive unscheduled downtime or potentially dangerous situations. Unscheduled factory downtime is one of the most prevalent causes of inefficiency in manufacturing. When a cyberattack or a broken machine stops a factory in its tracks, the cost can run tens of thousands of dollars a minute.
The researchers are working on a new methodology for controlling manufacturing systems, called "software-defined control". The work will combine techniques from multiple disciplines, spanning control theory, modeling of physical properties, machine learning, and cybersecurity.
Central to the new approach is a continuous and full simulation of the manufacturing plant. Special software will compare the plant's actual operation with the simulation.
"The idea is you have the physical manufacturing plant and the simulated model of the plant so if there's a difference between the two, you can detect a fault or a cyber-intrusion," said project principal investigator Dawn Tilbury, associate dean for research and professor of mechanical engineering at the U-M College of Engineering.
But the system is not just for diagnostics or fault detection.
"The goal is to develop control systems for manufacturing systems that are secure and reconfigurable automatically," explained Tilbury.
The solution could reprogram how parts flow through the plant to avoid a faulty piece of equipment, for example.
If successful, the researchers could transform manufacturing systems from their current paradigm of low efficiency and high susceptibility to system disruptions to a new era of system-level anomaly detection, classification and action.
This will lead to less downtime, faster responses to disruptions and a more efficient manufacturing system.