Author: Sander de Bree, CEO
Having a reliability monitoring program is one of the duties empowered on a Continued Airworthiness Management Organization (CAMO) by means of the aviation regulations (EASA Part-M). The main purpose of such reliability program is to ensure that the aircraft maintenance program tasks are effective and that the periodic tasks are performed adequately (see Acceptable Means of Compliance to EASA Part-M). Generally, this is embodied within a CAMO by means of trend reports displaying the health status of the aircraft per various aircraft system (ATA chapter). Establishing these trend reports allows the organization to identify if corrective actions are required to enhance the performance of a system or component. The other two actions associated to reliability monitoring are determining which corrective action needs to be performed and to monitor the effectiveness of the corrective action.
Sources of data regularly used for reliability monitoring include (but are not limited to): Pilot reports, technical logbooks, on-board maintenance system read-outs, reports on technical delays & cancellations, reports on maintenance findings during checks, reports on findings during shop maintenance, Air safety reports and/or stores reports.
Either on continuous bases or on set intervals this data is gathered and processed into graphical displays that represent the health of aircraft systems and/or components over a certain period. These graphical displays are then used to examine if further investigation is required in certain areas and/or if actions are required to improve the system’s health. Actions in this sense can vary from changes in maintenance procedures and operational procedures. Maintenance changes included frequency change / functional change, changes in maintenance manuals, initiation of modifications, spare provisioning, fleet inspections as well as staff training and / or changes to the manpower / resource planning.
From the previous outline of reliability monitoring, one can conclude that the basic philosophy lies in the fact that a certain event must happen, then we measure how much these events occur and quantify this into the various aircraft systems. Therefore, reliability monitoring is nowadays focused on looking into the history information, identifying trends in this and acting upon what has happened in the past to improve future situations. This focus is also still found in the more recent introduction of on-board Health Monitoring systems in aircraft such as the B787, A350 and Embraer E-Jets. These onboard systems detect certain technical occurrences happening on the aircraft and provide this information to ground stations. Albeit becoming more real-time, the very bases of it is still that a certain occurrence needs to happen to be recorded in the Aircraft Health Monitoring System.
Today, very few models exist in which future behaviour and scenarios are used to predict the reliability of aircraft systems or the areas of an aircraft that will require improvement in the future. The major benefit of applying future scenarios in reliability monitoring is that malfunctions in systems or components can be prevented by acting on these forecasting models and thus preventing unexpected technical ground time of aircraft.
The first step airlines can take to start applying these reliability forecasting models are Early Warning Indicators. (EWI) These are a set of defined indicators that measure the behaviour of aircraft systems (number of pilot reports, maintenance findings, mean time between unit removal (MTBUR etc.) and trigger a response from engineering when set control levels are reached (number of pilot reports, certain level of MTBUR of a component). For EWI’s to be effective a high interval data collection and analysis is required (e.g. daily bases or real-time). This often requires automated solutions to gather and display data.
Advantage of these EWI’s is that deviations are captured earlier and that preventive actions can be taken sooner. However, in a sense it is still the classical model of looking at historical events and acting upon these events. The difference is that we have set the control level stricter and collect data more frequent in comparison with e.g. monthly trend analysis.
To really predict future scenario’s in aircraft system (or component) reliability, forecasting models have to be used. In general Forecasting models depend on two main criteria: 1) Sufficient historical data, 2) cause-and-effect analysis. Once both are available two types of forecasting approaches can be used.
The first option reliability analysts have at hand is the causal model. This model is highly dependent on cause-and-effect data and basically requires identification factors (causes) that have an influence on a certain effect and express these in a linear mathematical formula. Then one focuses on identifying the factors (causes) in the future to predict the future outcome on the effect measured. Limitation of this model is that it requires you to define a set of affects you want to measure and on these effects the forecasting model is based. Hence items not defined in the model are not identified.
The above image illustrates such a causal forecasting model of risk percentage of Aircraft Component failure.
With the increased creation of data by aircraft and associated timely access to this data, it now also start to become possible in aviation to embark on data mining and analysis techniques to search for cause-and-effect relations that were previously unknown. This technique revolves around building decision tree’s on large amounts of data in which percentages are continuously calculated that determine the likeliness of a cause-and-effect relation existing. Once this percentage becomes high enough a cause-and-effect relation potentially exists and can be monitored upon in the applicable related causal forecast model
The second forecasting model available is the time series forecasting method. A time series model heavily depends on historical data and is focused on isolation of patterns in past data to identify trends, seasonal behaviour, cycles, or randomness. Limitation of this model is that if a certain effect has not occurred yet or if the trends leading to a certain effect cannot be identified, this effect will also not be forecasted in the model. Thus, limiting the forecasting power only on that what is known until today. Especially with new aircraft models, this forecasting method is the less favourable one. However, if you have an aircraft type for which much historical data is known, the time series forecasting model prevails above the causal model as data mining techniques (such as decision trees mentioned earlier) can be used to identify trends leading to certain effects. Vice-versa, if you have a new aircraft type with insufficient or no historical data, the causal model would best fit a reliability forecasting model.
First Publication July 2014, latest update October 2017
Follow the link for more information about our aircraft reliability solution Avilytics