Predictive maintenance is a technique that uses historical and real-time data to detect possible defects and anticipate problems before they happen. Mechanical and Biomedical Engineering Professor Abdel Bayoumi’s work with predictive maintenance dates to 1998 when he combined conventional engineering methods and data to help maintain U.S. Army helicopters. But as technologies have evolved over the last 24 years, Bayoumi is now working on implementing a digital transformation to create a sustainable predictive maintenance program for the next generation of mechanical, electrical and structural systems.
Bayoumi’s current U.S. Army research includes a three-year program funded through $1.1 million from Carnegie Mellon University, the U.S. Army and Department of Defense. The project started this past March and builds on previous predictive maintenance research. Other current programs are funded by the U.S. Army Future Vertical Lift and the U.S. Air Force.
“We focus on the operation and battalion testing using data and models into artificial intelligence (AI). This is where we now stand after 24 years working on predictive maintenance. We are breaking some components to understand what is causing a part to fail,” Bayoumi says.
Bayoumi’s team plans to implement futuristic systems that include increased levels of integration, complexity and adaptability. The digital transformation process will utilize technologies and techniques such as AI, digital twins and Internet of Things, which aim to decrease costs and time while increasing overall quality.
We’re trying to make the life of the crew chief, pilots, maintainers and engineers a lot easier.
- Professor Abdel Bayoumi
In the United States, sub-optimal maintenance methods exceed $85 billion annually. The high costs are attributed to a lack of understanding in implementing the best maintenance practices to prolong the life and safety of a component or system. Bayoumi is director of the Center for Predictive Maintenance at the University of South Carolina. The center’s research focuses on improving maintenance practices based on need, approach, benefits per costs and competition.
“The framework is built on integration and user interfaces where all types of data such as design, manufacturing and operation are utilized. Adding AI helps translate between steps to provide additional insights, which are independently learned by a machine and presented for decision making, feedback and analysis,” Bayoumi says.
Since each aircraft component affects its performance, physics-based digital twins of components are being developed and tailored to specific needs. The methods developed intend to support systems for higher levels of electrification since aging and failure modes are dramatically different from traditional mechanical systems.
“We're focused on showing how a digital twin and digital thread can be applied to maintenance and how they connect the process, from design to use and even decommissioning,” says research assistant Rhea Matthews. “Our goal is to share this data and the lessons learned since it can be applied to different platforms for many applications.”
Bayoumi’s team will also utilize models for use cases in developing sensor-based test stands for various mechanical components to help facilitate data collection. Each Blackhawk or Apache helicopter that undergoes a phase inspection will be examined on a test stand, and the data will assess current health and predict future performance.
“We have to decide on the specific use cases by focusing on fundamental research and field data,” says research assistant Evan Barnett. “An Army helicopter has a lot of mechanical and electrical components with installed sensors. We want to generate that data and then leverage it to a digital model. AI could then be used to determine which specific components are about to fail and might need to be replaced.”
The final phase of Bayoumi’s research will implement a digital thread by building on the use cases developed for creating a digital twin. This includes developing and using digital tools such as dashboards, AI, virtual and augmented reality, and simulators to assist with decision making. The combination of users and digital tools creates a digital thread where data and knowledge can be exchanged throughout a component’s life cycle.
“The digital twin allows for workflow and analysis, which will orientate models to make decisions for the physical system. These complimentary systems act on decisions and are updated together in real time,” Bayoumi says. “The objectives for creating digital twin framework for maintenance is for the physical system and digital twin to learn and advance based on feedback from each other. This will simulate and provide preventative maintenance capabilities for end users.”
The digital twin is the main component for any decision making. Instead of having a part and measuring it as a configuration, the digital twin provides the data to see how the product is the same physically and digitally.
“This decision making comes with collecting data to provide a model of prediction. For example, if you need three hours to reach a destination, you can determine whether it’s safe to go or risk your safety because a part can fail. The AI component also collect this data online while flying,” Bayoumi says.
Bayoumi states that the digital twin that was developed using only historical data is usually not 100% accurate, so it may need to “talk” with the physical system to correct each other until they are identical. The digital twin has a decision-making feature in a dashboard that is a connected system between the digital and physical that could identify different areas on the aircraft itself.
“You can solve engineering components after creating the digital twin. This digital tool is instantaneous and deep in terms of condition,” Bayoumi says. “For example, a digital twin can reflect the condition of takeoff or actual conditions when flying. If you are flying in the Middle East compared to Alaska, the digital twin should reflect the environment to adapt to the exact conditions.”
According to Bayoumi, his team’s digital twin of the entire system is like a composite material, which is comprised of two or more sub-materials. These sub-digital twins help provide features that make the overall system digital twin useful to everyone involved in the product life cycle. His team is also working on replicating additional applications such as manufacturing, assembly, supply and chain.
“We have applied different processes and applications with all of the digital tools,” Bayoumi says. “We’re trying to make the life of the crew chief, pilots, maintainers and engineers a lot easier because you can see it happening based on the real data.”