Today, as airlines face stiff competition and travelers become more cost conscious, increased operational challenges are placed on airlines to constantly seek ways of cutting costs. One area that offers great potential for doing so is maintenance. Even with recent improvements in efficiencies, it is estimated that more than 25 percent of maintenance spending is due to unplanned maintenance, which also drives five percent of additional wasted fuel consumption. A combination of predictive maintenance and data analytics promises to yield significant benefits for addressing these problems. However, there are hurdles that must be overcome to implement these solutions.
Predictive maintenance uses data that is generated by each aircraft, in combination with operational data, to determine the health of the systems onboard the aircraft. Sensors on the aircraft are used to monitor key parameters, such as air pressure, temperature, airspeed and fuel flow. These sensors can provide useful data to show if the system is performing optimally. Conversely, if the data shows that an avionics system has a problem, the appropriate maintenance can be scheduled at a suitable time. Ideally, predictive maintenance data should indicate how much time the airline has before there the avionics system will experience a significant decrease in performance or, in worst case, a complete failure.
The sensors used to monitor each aircraft system are connected to electronic units, commonly referred to as Flight Data Acquisition Systems (FDAU), that are dedicated to collecting the data for analysis. An example of an FDAU is shown in Figure 1. Some of the desired system data is already monitored by the aircraft avionics and can be transmitted to the FDAU via a data bus, such as ARINC-429.
After being sent to the FDAU, the predictive maintenance data can be stored on removable media such as compact flash drives, or can be transmit over a network such as ACARS or WiFi. FDAUs can also host integrated processing modules to support onboard computation, reducing the amount of data that needs to be transmitted or stored.
There are a range of options for accessing the data. An advantage of transmitting predictive maintenance data over ACARS is that it becomes available for analysis virtually in real-time. Unfortunately, this approach can be prohibitively expensive if there is a lot of data to transmit. Similarly, data can be transmitted over a cellular network, which, in addition to also being quite expensive, requires a strong and reliable signal to be effective. One alternative is to store the captured sensor data on removable compact flash modules. While these storage devices can support large quantities of data, they need to be physically removed from the aircraft prior to analysis. Lastly, the data can be transmitted off the aircraft via WiFi after it arrives at the gate.
After the sensor data from the FDAU is made available, it needs to be combined with a variety of airline operational and maintenance data derived from other sources and formats, including paper and PDF documents. All of this data needs to be collected and consolidated so that it can be analyzed. But first, it must be checked for erroneous or missing data in a process called “data cleaning.” This time consuming and painstaking process is essential for checking the data and correcting any errors before the data is analyzed.
In recent years, the aircraft industry has seen a growing trend to use Artificial Intelligence (AI) and Machine Learning (ML) to analyze the cleaned data in order to identify any anomalies that show whether a component or system is not performing correctly. The resulting information can be used to plan the suspect component’s removal, so that it can either be tested or replaced. At this point in the predictive maintenance process testing, a component, even if known to exhibit anomalous behavior, might be determined to have “No Fault Found” (NFF), since it didn’t fail. This situation causes a quandary for airlines as they learn to optimize the potential of predictive maintenance, since data might show that the suspect component is component is reaching the end of its life but still passes bench or shop testing.
Modern aircraft must monitor a huge amount of parameters, as many as 24,000 parameters. Combine that with the aircraft’s operational data and a single flight can generate Terabytes of data. The problem is that aircraft today are data rich and information poor. Collecting data is typically not difficult, but data itself is useless. The challenge is to establish a "normal" condition from which exceedances and trends that would otherwise remain unidentified can be algorithmically calculated. Identifying these trends enables maintenance and operation practices to become proactive.
The purpose of the aircraft’s sensors is to show that the onboard systems are performing correctly and to keep them performing as designed. As a system ages, and a failing component starts to degrade it, there may not be sufficient data to pinpoint which component is causing the problem and why.
The opportunity for improving aircraft maintenance and reducing costs is to collect the right type of data to determine why some of the systems onboard an aircraft are unreliable. It’s not enough for the maintenance team to replace the problem component. They also need to eliminate the root cause if possible so that a reoccurrence doesn’t happen. While the sea of data being collected today may and often does help, it is also likely that additional and precise data may need to be collected.
There will always be several systems on an aircraft that are more unreliable than the others, which lead to most of the unscheduled maintenance costs. If sufficient amounts of the right type of data are being collected from these potentially problematic systems, AI and ML can be used to make better decisions about their testing and replacement. If the existing data is not sufficient for pinpointing the source of the problem it may be beneficial to install more sensors and a custom FDAU in order to collect and transmit or store the desired data. A thorough cost/benefit analysis will need to be done in order to show the cost and time savings that can be achieved. One successful approach for quantifying the cost/benefit results is to use degraded components, ones that have been previously removed due to problems, with an instrumented ground test rig in order to prove that the root cause of the problem can be determined. If successful, this instrumentation can be replicated fleet-wide.
Modular and scalable aircraft data acquisition systems can be used to monitor sensors on an unreliable system that doesn’t have sufficient data. The acquisition hardware enables data to be collected from parts of the aircraft that today have little or no instrumentation. This modular and scalable hardware approach allows a great deal of hardware commonality where data can be collected. Examples include aircraft passenger air conditioning systems, landing gear systems, APUs, etc.
Because the FDAU can be situated close to the system, installation costs are significantly reduced. Each FDAU can be custom built to meet the precise needs of the customer. If, in future, a need for more data is identified, the FDAU can be upgraded to include the new requirements. The use of size, weight and power optimized off-the-shelf data acquisition solutions enables customers to quickly address predictive maintenance requirements.
Michael Doherty, product line manager for Curtiss-Wright Defense Solutions in Dublin, Ireland, has wide a wide range of experience in aircraft systems testing, structural testing and ground vibration testing. He has a BSc in Electronics and an MSc in Engineering Management.
