Ultra High-Tech Monitoring
Software that learns
By Greg Napert
March 1998
One of the biggest problems the maintenance community has been faced with related to turbine engine trend monitoring has been having to interpret the data they collect. Coming up with something meaningful from the sometimes overwhelming amounts of data can be quite challenging.
A company, called Sensa Technologies, recently set out to try to solve this problem by developing software called JENA that would do some of the analysis legwork for maintenance personnel.
According to John Howell, president of Sensa Technologies, most of today's engine monitoring systems monitor a multitude of parameters that must be monitored by an individual. That individual is not only expected to find parameters that are out of limits, but is expected to interpret the data and determine what is wrong with the engine. Howell says that with so many parameters to monitor, that's just not possible.
It is possible, however, to develop profiles for good and bad engines and have computers monitor whether the engines in the fleet match any of those profiles. It then alerts maintenance personnel when an engine needs attention.
Howell explains, "To understand how this concept might work, you first have to understand how current systems work."
The current situation
The typical airline already collects operating data every day. Maintenance personnel at the airlines receive a long list of information that they
have to compare against trending baseline data. They then observe the data and identify engines that are tracking out of limits and select them for removal. They have no idea what the problems might be with these engines — only that they aren't tracking properly.
"JENA," says Howell, "takes the information that the airline already collects and takes it one step further. It gives them probabilities on what the engine problems actually are and lets maintenance personnel make the decision if they want to do something about it.
"Another difference with our system is that when it finds a bad engine, it generates an exception report on that engine. The maintenance people are only getting reports on engines that are not meeting the parameters. With their current system, they are getting reports every day on every single engine and it's up to them to determine the bad engines. Some airlines are getting reports for thousands of engines. I've found that many facilities don't even bother using the report. Essentially, their current system is worthless. There is too much information and nothing to interpret it for the operators," Howell explains.
The philosophy JENA subscribes to is that any problem in the engine will actually change a number of parameters, and these parameters will take on certain characteristics based on what the problem is. Howell says, "If you look at each of the parameters individually, they are oftentimes not significant enough to tell you anything. But if you look at them all together, they form a pattern that can tell you what the problem is. In order to tell what these patterns mean, we rely on input from the operator and from past engine data that we have obtained from actual engine inspections and test cell runs. Data obtained at a particular time when an engine was taken out of service is saved as a pattern for the particular problem that was found."
In practice, the customer continuously collects data on engine problems, download operating information, and updates and improves the database.
Another major difference between existing technology and JENA is that it allows the user to improve it's data by incorporating actual findings into the database.
The user interface that allows the customer to program these findings into the system is called JADE. JADE allows the user to build structures, collect data on test cell runs, and classify flight data. JADE allows the customer to classify data based on engine tear-down.
This is a really critical difference between JENA and other systems out there. The data is continuously imperially classified. That is, verified against engine condition. The operator has a problem, tears into the engine, and classifies the data according to actual findings. The software actually gets "smarter" as time goes on. In other words, it learns!
Howell says, "Some trend companies provide data which attempts to locate the basic area of the engine where a problem exists and give the operator a go or no-go indication, but these programs are not based upon data that is continuously collected. It is based on a specific set of data which was determined at some point in the past. These systems don't learn."
Although most of what JENA does is much more sophisticated than trend monitoring, trend monitoring is a part of the overall system. Howell says "Trending is also used with the JENA system, but it uses it as a tool to classify data. The way we trend is rather unique. We look at the data from the last test cell run and use that as the base line. We standard-day correct the flight data and run that against the base line. Then we match up pattern to pattern.
"We don't check one parameter against another. Instead, we say that the patterns exist in multi-dimensional space. We then take these patterns and compare good engine patterns against flight data patterns. This is similar, in a sense, to what other trending methods do, except the other systems plot individual parameters against one another. We look at patterns that are based on many parameters," he says.
"The reason we do that is to identify change," says Howell. "We don't use it to identify problems. Our charts identify change with a graph that tells the maintenance personnel where the deviation began so they can classify all the data from that point."
Howell continues, "One other point about JENA is that many of today's trend monitoring companies require you to send your data to them so they can analyze the data and tell you if there's a problem. Ours provides instant analysis of your problems in real time."
