Competition in the aerospace industry is fierce at all levels, and that includes the factory floor. Achieving production numbers is imperative to meeting both cost and customer expectations, and machine failures and production stops ultimately affect delivery reliability and weaken companies’ competitiveness. Often the problem is only a small defect or normal wear and tear that left undetected could lead to major disruptions and production downtimes.
In an EU-funded project called SelSus, 15 industrial and scientific consortium partners are developing maintenance technology capable of forecasting machine downtimes in production before they occur. This will allow plant managers to rectify faults before the machine breaks down. Based on what’s known as a decision-support system, maintenance personnel can then reach a decision and take targeted action to repair the defect without having to interrupt production.
This is just one of the underlying ideas behind the SelSus project within which the Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) is currently researching.
“The aim is not just to monitor the status of the machines and components. Using intelligent software and sensor networks, the plan is to detect potential weak points or signs of wear and tear early enough for the system to be able to predict potential malfunctions,” said Fraunhofer's Martin Kasperczyk.
The developed diagnostic models also directly provide suggestions or recommendations on how to rectify the problem. As one example, project partner Electrolux, based in Pordenone, Italy, uses such a decision-support system. The system is capable of predicting, with a certain probability, potential failures on a press for washing machine facings and of diagnosing actually occurring malfunctions. The data needed to monitor the machine status is partially provided by sensors. They measure values such as energy consumption, temperature, oil pressure, particles in the oil, and vibrations. Fraunhofer and the consortium worked together to prove that the technology functions reliably in practice. Other consortium partners include Ford Motor Co., Harms & Wende, Xetics, IFE-Werner, the Manufacturing Technology Center, and Gamax.
Self-repairing and -healing systems
The system is capable of sending control impulses to individual machines. A welding control on which a sensor has failed, for instance, can continue to work "almost seamlessly" in a secure mode, without any serious disruptions. The capability for self-repair and sustaining production has also given the project its name. The full project title of SelSus is “Health Monitoring and Life-Long Capability Management for Self-Sustaining Manufacturing Systems.”
However, first a number of technological hurdles had to be overcome.
“One of the biggest challenges was analyzing the flood of data,” said Kasperczyk. “After all, we’re talking here about predicting malfunctions or breakdowns of machines with a high degree of reliability. You don’t get there just by programming a couple of algorithms.”
The Manufacturing Technology Center created a system with self-healing capabilities. In an engine production plant, a dispenser was attached to a robotic arm by means of vacuum. If the dispenser encountered resistance, rather than snapping off, it reacted flexibly. It lost the grip produced under vacuum and dropped a few centimeters until it was stopped by springs. The springs then draw the dispenser back to its original position. Subsequent calibration ensured the tool was in the correct position--and after a brief interruption, the work process continued.
Bayesian networks and sensor data
The experts involved in the research rely heavily on Bayesian networks, which are mathematical models that can be used to compute the probabilities of a certain event or state occurring. Each model represents a set of variables and their conditional dependencies. With the help of the data collected by the sensors, the software computes, for example, the probabilities of a specific high-stress cable breaking in the near future and, where applicable, signals that it should be replaced.
But the SelSus software relies not only on sensors. It also takes the technical characteristics of the machine and its performance parameters into account. This data has to be captured during installation and configuration of the system. An extensive test run tells the system how the machine and its components behave in continuous operation and under load. Only then is it ready for use. The software also registers new data, for instance as a result of machine upgrades or deterioration in performance due to wear, enabling the system to learn.
The software also interacts with operators by analyzing the causes of potential or existing malfunctions and proposing an appropriate course of action.