When American Airlines Flight 309 left LaGuardia Airport in New York bound for Chicago O’Hare the morning of March 11, 2009, no one knew engine number two was doomed to fail just moments after takeoff, forcing an emergency diversion to JFK.
What if aircraft could tell us when a part or a system was about to fail, in time to avert an emergency? Some engineers believe airplanes have always told us this crucial information, but we just haven’t been listening.
Now, recent advances in sensor technology and wireless communications are improving our hearing, making failure prognostication possible. As a result, new ways of indirectly monitoring the health and predicting the longevity of aircraft mechanical and avionics systems are being devised.
One university prognostication system is on the ready ramp. Engineers at Georgia Tech have devised models called adaptive prognostics that use data from real-time sensor measurements to calculate and continuously revise the amount of remaining useful life of various aircraft systems based on their current condition and health status. The predictions are then integrated with aircraft maintenance operations and supply chain policies as part of what the researchers call an "autonomous sense-and-respond logistics paradigm."
"The system can be adapted to constantly monitor any moving part, from a ball bearing to aircraft electronics, sending out electronic alerts well in advance of a breakdown," said Georgia Tech engineer Nagi Z. Gebraeel.
Gebraeel began his research by monitoring the vibration and acoustic emissions signals from rotating machinery, namely bearings. He extracted degradation-based characteristics pertaining to key components on the machinery and used them to develop condition-based signals. Next, he created stochastic models to characterize the evolution of these condition-based signals to predict the remaining life of critical components. (Stochasticism acknowledges that outcomes result from both known and unknown causes.)
"This system will not only say how long a given product or component is going to last, it will also give operations groups insight related to when it makes the best economic sense to replace that component," Gebraeel said.
Experimental results show the technique can potentially reduce total failure costs and costs associated with running an efficient spare parts inventory by about 55 percent.
Encouraged with the results from mechanical systems, Gebraeel turned his attention to developing similar models for electronics. Prognostic systems for avionics present an interesting challenge. Unlike mechanical systems, electronic systems usually don’t provide warnings of impending failure.
Alfred D. Helfrick, an Embry-Riddle Aeronautical University engineer, said developing algorithms for predicting a bearing failure based on acoustic emission makes a lot of sense, but electronic components don’t make sounds like worn bearings.
"Predictive modeling might be possible if the system monitored such items as bit error rates, signal voltages on data buses, power supply voltages and currents — these might provide some indication of an impending failure," Helfrick said.
Gebraeel said all of these inputs can be used. "We’re currently focusing on signal voltages," he said.
Gebraeel recently began working with Rockwell Collins to develop adaptive models designed to estimate the remaining useful life of aircraft electronic components. He said his goal is to embed this prognostic methodology into key avionics systems.
Some critical components in avionics already are equipped with technology that provides diagnostic indicators and/or performance measures that can be easily recorded. Gebraeel’s proposed models have been developed such that they can efficiently piggyback on the existing infrastructure of these systems.
"Efficiently here implies the ability to encode or embed these models in existing microprocessors without necessarily taxing the existing computational power in these systems," Gebraeel said.
Within this construct, he believes his prognostic models can accurately predict the remaining useful life of critical avionic components. "Given this information, it will be possible to make key decisions, such as whether an aircraft is capable of carrying out a specific mission or if it should be assigned to a shorter mission or be grounded," he said.
Ultimately, Gebraeel said the information will be crucial in preventing sudden failures that result from gradual degradation or deterioration in performance.
From a design perspective, most critical systems on aircraft have redundant components, important to ensure safety. In some cases, there can be three or four identical critical components in a given system. If one fails, the others kick in.
"However, given the significantly improved prediction accuracy that can be attained using our prognostic models, it will be possible to reduce, but not necessarily eliminate, the number of redundant components," Gebraeel said.
The result would be a reduction in aircraft weight, which in turn would reduce fuel consumption. Not only will it be safer to fly, it could also be less expensive.
Gebraeel said another aspect of prognostic modeling may be of more interest to the military. Squadron leaders typically must identify a subset of most reliable aircraft to deploy on a specific mission. In many cases, these decisions are spontaneous, making the need for real-time information that much more imperative.
"Within the framework of our research, we have been able to integrate computer science ranking algorithms, such as those typically used in Internet search engines, with our prognostic models to identify the best subset of units within a fleet whose critical avionic components have the largest remaining useful life," Gebraeel said. This information can also be used to plan missions, allocate resources, and guide maintenance activities, he said.
Currently, the knowledge base related to aircraft condition monitoring relies heavily on artificial intelligence techniques and black-box methods, such as the flight operational quality assurance (FOQA) program promoted over the past decade by the Flight Safety Foundation. This effort involves the collection and analysis of data recorded during flight to improve the safety of flight operations, air-traffic control procedures, and airport and aircraft design and maintenance.
Gebraeel said his work differs from FOQA. "FOQA enables airlines to analyze the data for any events or trends that might signal an exceedence of normal, or standard, operating procedures, which yields a binary yes-no, good-bad indicator," he said. "This is good for diagnostic purposes and is not really suitable for planning purposes."
In contrast, Gebraeel said his adaptive prognostic model can be regarded as an enhancement to current capability since it focuses more on predicting remaining life time or the remaining time until a signal will exceed the current normal standards.
"This information can be used to provide an advanced planning horizon to schedule replacement schedule and logistical actions," he said.
Gebraeel said most failures of engineering systems result from gradual damage that occurs during the system’s life cycle, a process known as degradation. "In many applications, it can be extremely difficult to assess and observe physical degradation, especially if real-time observations are required," he said. However, degradation processes are generally associated with some manifestations that are much easier to observe and monitor over time using sensor technology, he said.
Thomas D. Inman, a professor of avionics at the Pennsylvania College of Technology, believes the Georgia Tech system could have a revolutionary affect on the avionics industry. Currently, Inman said, there is no viable way to peer inside a chip or transistor to determine its overall condition. As a result, some transistors apparently work forever, while others fail quickly.
"If Gebraeel’s system works well, not only will safety and maintenance costs be reduced, but also some critical equipment could become life-limited, like those of helicopter components," Inman said.
Wayne Plucker, aerospace industry manager with Frost & Sullivan, said Gebraeel’s approach has several other potential benefits. "First, it localizes the prognostics to the component, which minimizes the I/O between it and a central system, so it is likely to save weight," Plucker said. "Also, it allows component manufacturers to update prognostic routines without needing to be part of a giant aircraft-wide software revision."
Gebraeel also is working with Global Strategic Solutions, of Sterling, Va., which has funding from two U.S. Navy Small Business Innovation Research grants. The focus of one grant is to develop embedded diagnostics and prognostics to predict the remaining life distributions of electrical power generation systems on naval aircraft. The second is to develop advanced health monitoring and remaining useful life models for aircraft communication, navigation and identification systems used on the Joint Strike Fighter.
The Georgia Tech research illustrates how far the state of the aircraft health monitoring art has evolved. Engine systems were first to benefit from health monitoring methods. The reliability of the current jet engine is primarily the result of smarter engines — that is, engines with computer control. Today’s jet engine controllers continually re-trim the engines.
"Engine manufacturers have a vested interest in engine health management systems because they are often responsible for repairs and replacements under various power-by-the-hour programs, like the one started by Rolls Royce over 20 years ago," Plucker said. Power-by-the-hour is the term used to describe performance-based contracts for engines and avionics sold to commercial airlines.
Then, in 2004, electrical systems monitoring and forecasting got a push from the Air Force Research Laboratory, Boeing, and Smiths Aerospace (now GE Aviation), which began a program to manage the health of electrical power systems in the form of diagnostics, prognostics and decision aids. The result is the Aircraft Electrical Power Systems Prognostics and Health Management (AEPHM) program. The first phase of the program, which ended in 2005, addressed electrical systems, including actuation, fuel pumps, valves and arc fault protection. Parts of that effort are now being implemented on several aircraft. The second phase of AEPHM, currently underway, addresses power generation.
Monitoring the health of whole aircraft systems, however, is a difficult challenge. "The parts are often discrete elements that do not talk to each other or to a central system," Plucker said. One of the exceptions is the newer avionics components, which exist on a standard system architecture and could pass health information along with other data.
"At this point, it is hard to sell replacing a box that is working okay when the system says that it is headed in the wrong direction," Plucker said. But as avionics OEMs gain success with their versions of power-by-the-hour, that picture could change, he said.
The F-35 Lightning II is a good current example. BAE Systems is providing "Prognostics Health Management" integration for the F-35 aircraft. BAE says this process includes diagnostic and health management solutions that "identify problems before they become expensive to remedy, enabling guaranteed levels of availability and reducing life-cycle costs."
Pratt & Whitney is providing advanced prognostics and health management systems for the multipurpose fighter’s F135 engine. Similar diagnostic and health monitoring systems are included on the Airbus A380 and Boeing 787.
Most current health monitoring systems rely on acoustic, vibration or heat sensors, and general trend analysis techniques. Some of these for avionics use traditional signal/noise ratios and similar trend data to identify equipment that is potentially beginning to fail.
"One of the greatest current information voids in health monitoring systems is the lack of information that can be reliably used to predict failure modes, Plucker said.
One major distinction between Gebraeel’s sensor-based models and other sensor-based predictive technologies already in use is Gebraeel’s ability to measure the entire history of sensor signals and how they evolved, as opposed to traditional predictive methods that rely solely on current condition monitoring information.