The same often happens with machines. For instance, many turbines used to generate electricity inside power plants have an official life of 30 years or so. But some of the machines that GE maintains have been operating for twice as long — and one is still running after 82 years, says Colin Parris, vice president of software research at GE’s Global Research Center (GRC). What happens is that parts get replaced and service contracts extended long beyond the product life span. “In some of these steam turbines we’ve changed every part,” says Parris. “Maybe just the frame of the casing is the only thing that’s left. It makes you think, wow, this is the beginning of an immortal machine.”
Making machines last forever is actually the goal of one of Parris’ projects at GRC. The idea is to take the kind of life extension that often occurs by happenstance in the field and turn it into a formal engineering practice. That would not only increase the lifetime of machines, but also ensure that they’re running more efficiently. The project involves combining three important technologies — computer simulation, artificial intelligence and 3D printing — into a new way of looking at engineering and maintenance.
Computer simulation is a common tool in the design of new machines. It also comes in handy later in a machine’s life cycle, when it breaks down and engineers are trying to figure out what went wrong. Parris’ idea is to link the two into one continuous simulation that shadows each machine through its working life. He calls it a “digital twin” — a computer-simulated version of a machine that is a realistic, up-to-date model of the machine in the field. The advantage of having such a virtual model is that engineers can then “run” it inside a computer to figure out precisely what maintenance must be done at what specific time. “It’s a way of keeping the mind of the machine alive,” Parris says.
For instance, if a part fails and needs to be replaced, humans can use 3D printing to make the replacement part. But more than that, we could the use the opportunity to make a better, more highly customized part that not only fixes the machine but also it enables it to perform better. Artificial intelligence, and machine learning in particular, is an essential part of the digital twin project. Parris and his colleagues are now running tests on a gas turbine in the lab at GRC in part to gather data about how the machine wears down over time. By running the turbine under high heat and pressure, they can simulate the passage of five or six years of operations in the space of a few weeks. They take ultrasound and X-ray images of the machines to look for signs of wear — small cracks and other kinds of damage. All that data is fed into the machine-learning algorithm, and it’s ultimately used to predict the right time when maintenance is the least disruptive. “Normally a model will describe how a turbine works, but every machine degrades in different ways depending on a variety factors,” says Parris. “We use AI machine learning to figure out how to perfectly match the asset to your simulation.”