A data-driven health indicator extraction method for aircraft air conditioning system health monitoring

Jinzhong SUN,Choyi LI,Cui LIU,Ziwei GONG,Ronghui WANG

aDepartment of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China

bMaintenance Engineering Department,Xiamen Airlines,Xiamen 361000,China

Received 17 November 2017;revised 22 November 2017;accepted 22 November 2017

Available online 13 April 2018

Abstract Prognostics and Health Management(PHM)has become a very important tool in modern commercial aircraft.Considering limited built-in sensing devices on the legacy aircraft model,one of the challenges for airborne system health monitoring is to find an appropriate health indicator that is highly related to the actual degradation state of the system.This paper proposed a novel health indicator extraction method based on the available sensor parameters for the health monitoring of Air Conditioning System(ACS)of a legacy commercial aircraft model.Firstly,a specific Airplane Condition Monitoring System(ACMS)report for ACS health monitoring is defined.Then a non-parametric modeling technique is adopted to calculate the health indicator based on the raw ACMS report data.The proposed method is validated on a single-aisle commercial aircraft widely used for short and medium-haul routes,using more than 6000 ACMS reports collected from a fleet of aircraft during one year.The case study result shows that the proposed health indicator can effectively characterize the degradation state of the ACS,which can provide valuable information for proactive maintenance plan in advance.

KEYWORDS Air conditioning system;Aircraft health monitoring;Airplane condition monitoring system;Health indicator;Prognostics and health management

1.Introduction

Modern engineering systems,such as aero engines and commercial aircraft,consisting of a very large number of components which closely interact with each other,must run safely and economically for their entire lifetimes.A Prognostics and Health Management(PHM)policy may help achieve this goal by improving reliability,safety,and availability,while reducing operational costs.Such a PHM system typically involves data acquisition and processing,fault detection and diagnostic,failure prognostics and decision support.The main purpose of this system is to detect,diagnose and predict the faults on the system and take appropriate decisions to correct them before they grow into significant problems.1

Modern commercial aircraft are typically equipped with Airplane Condition Monitoring System(ACMS)with a large number of sensors and detectors distributed over the aircraft.The ACMS can collect a wide range of flight data,including the environment,load,status and performance data during the operation of the aircraft system,which can be used for flight quality monitoring and evaluation as well as system and component health monitoring and prognostics.2These ACMS data and diagnostic information are of great significance to ensure the safety,usability,economy and punctuality of aircraft.For airline operators,looking at data trends across a fleet of aircraft can detect deterioration in components and perform proactive maintenance.The primary benefit provided by PHM is the opportunity to help operators identify precursors that are likely to progress to Flight Deck Effect(FDE)faults which will affect airplane dispatch,thus substantially reducing unscheduled in-service interruptions costs.3The vast potential of PHM on modern commercial aircraft is being realized today through the innovative use of available airplane ACMS data.The development of advanced diagnostics and prognostics techniques can enable aircraft health monitoring capability to extend far beyond what is possible using only parameter alerting and trending of raw ACMS data.2-4Therefore,how to effectively explore aircraft ACMS data for aircraft system health monitoring and predictive maintenance to reduce unscheduled aircraft maintenance is currently one of the research focuses.

Complex aircraft airborne systems typically consist of a large number of components closely interacted with each other,which makes it more difficult to develop an effective system health monitoring solutions.The aircraft Air Conditioning System(ACS)is a critical system,which provides the appropriate environmental conditions to ensure the safe transport of air passengers and equipment.With the increasing complexity of the ACS,the complexity of health monitoring solutions also rapidly increases.The limited number of sensors on legacy aircraft ACS provides little information about the health condition,which makes fault detection and isolation a very challenging task in the ACS.This problem is further compounded by the ACS's feedback control loops that can compensate for certain degradation.However,only limited research in the public literatures has been conducted on the PHM of the aircraft air conditioning system.Algarni et al.studied the reliability and quality of the air-conditioning/cooling pack of a particular type of commercial aircraft at component level and system level under actual operating conditions.5The study shows that the ACS of the aircraft operating in harsh atmospheric conditions typically experiences a higher field failure rate than that estimated by the manufacturer.Hare et al.proposed a system-level hierarchical fault detection and isolation method,which is further tested and validated on a simulation data set.6Silva et al.presents a wavelet-based fouling diagnosis approach for the heat exchanger,which is a critical component of the ACS directly determining the efficiency of the ACS.The fouling degradation assessment method is built and tested with the sensor data generated from an experimentally validated aircraft Environmental Control System(ECS)simulation model.7Najjar et al.presented the method for fouling severity diagnosis of the heat exchanger using the principal component analysis and the knearest neighbor classification.8They further studied the optimal sensor selection and fusion methodology to select the most useful sensors that can provide the best diagnosis results.This proposed method is tested on the data generated from an experimentally validated high- fidelity ECS simulation model provided by an industry partner.9Shang and Liu et al.proposed a heat exchanger fouling detection method based on the valve control command of engine bleed air temperature regulation system.The effectiveness of the method is demonstrated on computer simulations and test rig experiments.10Ma et al.proposed a parameter adaptive estimation method based on strong tracking filter and Bayes classification method for fault diagnosis of heat exchanger in aircraft environmental control system,which was demonstrated based on simulation data set.11However,to the best of the authors' knowledge,there is no study to develop an ACS health monitoring solution on the fielded systems or legacy aircraft based on the available ACMS data.The objective of this study is to develop a health indicator extraction method for ACS health monitoring of commercial aircraft that is subjected to an airline's actual use environment.The ACMS data analyzed here were obtained from a popular single-aisle commercial aircraft widely used for short and medium-haul routes.

The remaining sections of this paper are organized as follows.Section 2 presents a brief description of the air conditioning system and the built-in sensors of a legacy aircraft model.Section 3 describes the raw ACMS data preprocessing process.The fourth section presents a data driven health indicator extraction method based on a multivariate state estimation technique and the proposed method is tested and validated on a filed data set from a particular type of commercial aircraft in this section as well.Conclusions are drawn in the final section.

2.Commercial aircraft air conditioning system and built-in sensors

The air conditioning system as a key airborne system in commercial aircraft,directly related to the aircraft cockpit,passenger cabin,and cargo space's normal working and living environment.The study results from 5,12 show that dirt contaminations such as the particulate,dust and sand in atmospheric are identified as the primary causes of the failures for the ACS of the aircraft operating in harsh atmospheric conditions.According to a local airline annual maintenance report,the ACS failures take the first cause for unscheduled maintenance of the particular type of aircraft fleet operating mainly for domestic routes.The ACS'failure not only affects the flight safety of the aircraft but also may affect the dispatch of aircraft,cause flight delays or even cancellations,bringing huge economic losses to the airline.Thus,an effective health monitoring solution to decrease the unscheduled maintenance due to ACS failure is likely to be needed.

2.1.Commercial aircraft air conditioning system

The ACS controls the interior environment of the airplane for flight crew,passengers and equipment.ACS has several sub-systems:distribution,pressurization,equipment cooling,heating,cooling and temperature control.The ACS provides temperature controlled air by processing bleed air from the engines,APU,or a ground air source in air conditioning packs.As a key system in ACS,the main role of the cooling sub-system is to control the quantity of air from the pneumatic system to the pack,remove heat from the air that enters the pack and control the output temperature and moisture of the pack.The cooling system uses these components to cool the bleed air:Flow Control Shut Off Valve(FCSOV),primary heat exchanger,Air Cycle Machine(ACM),secondary heat exchanger and ram air system.The principal configuration of the cooling sub-systems is shown in Fig.1.

The bleed air flow through the FCSOV enters the primary heat exchanger,where the hot bleed air is cooled by the ambient ram air controlled by the ram air system.Then the cooled air flows to the compressor of an ACM where the air is compressed and the temperature increases again.Next,the air circulates through the secondary heat exchanger for additional cooling.The processed cold air is then through water separator,reheater,condenser and then passes through the turbine of the ACM where it is cooled by expansion.The condenser collects and removes moisture from the air before it goes into the distribution system.The processed cold air is then combined with hot air in the mixing manifold,which is then distributed through the left and right sidewall risers to the passenger cabin and the flight deck.5

2.2.ACS built-in sensors

ACS is equipped with many built-in sensing devices such as temperature sensors,position transducers,and pressure sensors mounted at different locations of the ACS for the purpose of feedback and control.Theoretically a lot of operation data can be obtained for ACS health monitoring,however only limited sensor data are acquired and saved to the ACMS for further analysis in real-time or post flight.For the legacy commercial aircraft model studied in this paper,the available ACS sensor parameters recorded in the ACMS are listed in Table 1.

Table 1 List of primary temperature sensors in ACS.

The ACS consists of many feedback,control,and safety mechanisms,therefore simple analysis of the ACS raw sensor parameters may not be sufficient for an effective health monitoring solution,since the redundancy and control mechanisms are able to compensate for a failure.Therefore,other sensor parameters describing the engine and aircraft operating conditions,such as Mach number,altitude and atmospheric conditions were included for further analysis.Domain knowledge and simple signal statistics are used to select an initial subset of sensor parameters recorded in the ACMS as the ACS contextual data,which may more or less cause a certain impact on the operation of ACS(Table 2).

Fig.1 Schematic drawing of the air conditioning systems.

Table 2 Initial subset of ACMS parameters related to operation of ACS.

3.ACMS reports for ACS health monitoring

Fig.2 presents the recorded ACMS data from the temperature sensors,i.e.,the RAMT,the PKT and the MFDT mounted at different locations of the ACS during one flight.It can be observed that the ACS temperature signals show evident fluctuations as the flight mode changes during a flight,since the ACS functions at different modes base on the specific flight phase.Taking the ram air system for an example,the ram air system controls the air flow through the primary and secondary heat exchangers to cool the bleed air.There are at least three modes of control for the ram air system:Ground,Flight( flaps not up),and Flight( flaps up).For the Ground mode(e.g.taxi in/out phase)and Flight( flaps not up)mode(e.g.,take-off phase),the control system makes the ram door full open to allow as much as ram air flow through the heat exchangers to cool the bleed air when the airplane is on the ground or during take-off phase.However in flight( flaps up),the ram air controller controls the ram door to achieve a balanced temperature of the cooled bleed air at the outlet of the ACM compressor.That means the ACS consists of many feedback and control mechanisms,which is able to compensate for an operating contextual condition change or even an ACS abnormal condition,making the ACS health monitoring more complex and difficult.

Fig.2 ACS Temperatures change during a typical flight.

The development of ACS health monitoring solutions begins with a data pre-processing to select a subset of sensor parameters under a specific operating condition or flight phase based on the understanding of system operation and domain knowledge which is typically called ACMS reports.The ACMS report is a customized flight data of particular interest,which is usually sent to ground base station via ACARS(Aircraft Communications Addressing and Reporting System)for aircraft performance evaluation,troubleshooting,and even health monitoring.These reports usually capture aircraft flight data and airborne system operation parameters of particular interest for a given component or fault mode in a specific data format defined by domain experts.The reports can be generated automatically based upon certain triggering criteria,e.g.,a specific operating mode of the system of interest or flight phase of the aircraft,when it is appropriate for system performance characterization or anomalies detection.In this study,based on the understanding of system operation and domain knowledge,the ACMS report for ACS health monitoring is generated when the average engine Exhaust Gas Temperature(EGT)has reached a peak value during takeoff phase.The ACS report includes the airborne system operating parameters such as the ACS temperature sensor data listed in Table 1,as well as the contextual data such as the altitude,Mach number,and the air temperature,etc.,given in Table 2.Then one ACS report is generated during the takeoff phase of each flight.After receiving the ACS report,the reported parameters are parsed and processed for further analysis to extract diagnostic and prognostic information for long term analysis of ACS health.

Fig.3 shows the raw sensor data extracted from the ACS reports generated from about 1000 flights over the course of one year.Only the air temperature,the ACM compressor outlet temperature and the mix manifold temperature are plotted in Fig.3.It should be noted that the system behavior and the sensor data are affected by several factors,such as the operating mode,the contextual conditions as well as the system health state.An obvious varying trend in the ACM compressor outlet temperature can be identified in Fig.3,which is closely related with the contextual conditions,such as the static air temperature and Mach number.The objective of this study is to capture the system health state under varying system operating and contextual conditions.Therefore,advanced analytic methods are necessary to derive enhanced diagnostic and prognostic information from the raw sensor data for ACS health monitoring.

Fig.3 ACS Temperatures data extracted from the ACS reports during one year.

4.Health indicator extraction based on MSET

Many system failure mechanisms can be traced to underlying physical or chemical degradation processes.When it is possible to measure degradation,such measurements often provide valuable information about a system or component's health state.However,only in very few cases is it possible to measure the degradation of a product directly.The measures of system performance(e.g.,temperature,pressure)are available in most situations where the degradation analysis can be carried out on the basis of performance parameters or the features extracted from the monitored performance parameters.Thus,for the complex airborne system health monitoring,the biggest challenge is to find a specific variable,i.e.,the health indicator,which is highly related to the actual degradation state of the system.1A health indicator inferred from a set of raw sensor readings is proposed to characterize the unobserved degradation state of the ACS.A non-parametric modeling technique,Multivariate State Estimation Technique(MSET),is adopted to calculate the health indicator.11,12

4.1.Multivariate state estimation technique

The MSET method is a non-parametric regression modeling technique,which does not need to make any assumptions on the mathematical structure of the relationship among the monitored parameters,but just use the historic data to describe the relationship between the multiple variables.11,12The nonparametric method stores past data samples in memory and processes them when a new query is made.The process of estimating a parameter's value for a new query is made by calculating a weighted average of historical training sample values.13-16

Assume that the state of a system is described byPvariables,denoted by aP×1 vector.Then the system state(or observation)at timetjcan be expressed as:

IfMstate observations are collected from a system,then the training matrix may be represented as follows:

Each row of the matrix is the time series values fromt=1 tot=tMof one variablexi.Each column of the matrix represents an observation of the system at the corresponding time.The training matrix is typically created from a large set of historic reference data covering the full dynamic range of the monitored system.The training matrix D can be used to estimate the values for a query observation Xobs.An estimate of Xobs,defined as Xobs,which is the weighted combination of states in training matrix D,may be calculated with:

where W is a weight vector that decides the contribution of each state in matrix D for the calculation of the estimate.W is derived from the following:

where the symbol⊗stands for a non-linear similarity operator measuring the similarity between each pair of vectors in the DTand D matrices and between each vector in the DTmatrix and the Xobs.17,18The commonly used similarity operator is the Gaussian kernel operator:

wherehis the kernel's bandwidth.

The proposed health indicator for the ACS is the parameter residuals of the temperature of the bleed air at the outlet of the ACM compressor,generated by the difference between the measured value using the RAMT sensor and its estimation.

WhereTRAMT，measuredis the measured ACM compressor outlet temperature, andTRAMT，estimatedis the estimation ofTRAMT，measuredusing the MEST method.

4.2.Case study on a legacy commercial aircraft model

Optimal sensor parameters selection is a key step in nonparametric model development.The correlation coefficients for the initial set of signals identified in Section 2 is computed(Table 3).

To reduce the computation burden,only a subset of signals with strong correlation coefficients(＞~0.7)(i.e.,SAT,TAT,N2,MFDT,RAMT)is selected to construct the state observation vector for further analysis.

The ACS reports and associated maintenance records are collected from an airline over a period of about one year.There are total 6 aircraft,and each one has two identical air conditioning systems,i.e.,the left pack and the right pack.Table 4 shows the summarized data set related to each aircraft.

The normal data samples collected from 4 aircraft(i.e.,Aircraft A,B,C,and D)during a specific period,characterizing the behavior of the aircraft in a health condition,were used to construct the training matrix.The normal data mean no abnormal behavior was observed in the data and no maintenance activity is carried out during that period.Most of the time the ACS is in a healthy state,so an initial set of 4000 training samples characterizing the healthy state of ACS in various operating conditions is selected.To achieve a better estimation performance of the MSET,lots of historic observations are required in the training matrix to cover the full dynamic range of the ACS system,which will easily lead to unacceptable computational loads.Thus to reduce the computation burden,the training matrix has to contain as few historic observations as possible.Several methods,such as minmax selection,vector ordering,fuzzy c-means clustering,are proposed to lessen the computational burden by choosing an optimal subset of the historic observations.11In this paper,the combination of min-max and vector ordering is used.First the minimum and maximum observations for each of the 5 variables are extracted.Then,the remaining observations are chosen through the vector-ordering method,without replacement of the previously chosen observations.19Finally a subset(i.e.,a 405×5 training matrix)is selected using the combination of min-max and vector ordering method:

Table 3 Correlation coefficients for initial set of signals of ACS.

Table 4 Summary of ACS reports and maintenance records.

The training matrix D can be considered as a baseline for the RAMT established on the data collected when the ACS is known to be in a healthy state.The baseline defines the range of RAMT corresponding to acceptable operation conditions;therefore,ACS gradual degradation or abrupt fault will make the measured RAMT deviate from the baseline.

The remaining 3595 data samples from the healthy ACSs are used to test the constructed RAMT baseline model.Fig.4 shows the delta RAMT between the baseline and the measured RAMT,which gives a clearer picture showing that the measured RAMT from the healthy ACSs is close to the baseline with the deviation value around 0.That means the baseline constructed based on the historic performance data of the ACS in a health state can successfully capture the characteristics of the RAMT data under various operating conditions,and further the delta RAMT can be used as a health indicator of the ACS for health monitoring.

Fig.4 Delta between measured RAMT and its estimation.

Fig.5 shows the health indicators computed based on ACS reports data collected from 6 aircraft during about one year.The maintenance actions for each ACS during this period are also indicated in the plots.It can be seen from Fig.5 that evident deviations in the health indicators are observed at the time when a maintenance activity is required,and after the maintenance the deviations restore around zero,meaning a performance improvement due to the maintenance action and the ACS stays in a healthy state.The results from Fig.5 indicate that the proposed health indicator is highly related to the actual degradation state of the ACS,and can provide valuable information about the system's health state.Currently,since there is no effective health monitoring solution for the ACS of the studied legacy aircraft model,the ACS maintenance is mostly triggered due to a carbine or flight deck effect,which may affect airplane dispatch and possibly cause flight schedule interruptions.The main objective of the paper is to extract a health indicator for the ACS,which can help the airliner operators to identify ACS degradation precursors and take proactive maintenance in advance before it progresses to FDE faults.

Fig.5 Health indicators computed based on ACS report data.

Further analysis of the ACS maintenance records showed that most of maintenance is the cleaning or replacement of the ACS heat exchangers due to the fouling problem.The contaminant-prone operating environment in the local area causes a gradual accumulation and build-up of contamination on the heat exchangers,which may lead to a reduction of the system performance over time,and in some case progress to FDE faults affecting airplane dispatch.20For the right ACS of Aircraft A,since there is no evident fouling accumulation on the heat exchangers when it was cleaned,it is hard to observe any performance improvement due to the cleaning from the health indicators shown in Fig.5(a).It can also be seen from Fig.5 most of the ACS faults show a gradual degradation pattern due to the gradual contamination accumulation.Once the health indicator is established,defining the failure threshold in degradation will make it possible to use a prognostic algorithm to predict the failure time,thus providing maintenance personnel the ability to schedule removals and plan maintenance in advance.21This will be the next step of research work.

5.Conclusion

Prognostics and Health Management has become a very important tool in modern commercial aircraft. Derived diagnostic and prognostic information from the Airplane Condition Monitoring System data can be used to enhance aircraft maintenance practice to avoid delays and cancellations.Aircraft system degradation indicators combined with prognostic algorithms can provide maintenance personnel the ability to schedule removals and plan maintenance in advance. For the health monitoring of complex airborne system,considering the limited ACMS data available on the legacy aircraft,one of the challenges is to find a degradation measurement or health indicator that is highly related to the actual degradation state of the system.

This paper presents a health indicator extraction method based on the available ACMS data for the health monitoring of air conditioning system of a legacy commercial aircraft model.Based on the domain knowledge and signal correlation analysis,a specific ACMS report for the ACS health monitoring with associated triggering criteria is defined,which includes the ACS sensor parameters as well as the system operation contextual data.Then a non-parametric regression modeling technique,Multivariate State Estimation Technique,is adopted to calculate the health indicator based on the raw ACMS report data.A case study on a particular type of legacy commercial aircraft is carried out.The proposed health indicator extraction method is validated using more than 6000 ACMS reports collected from a fleet of aircraft over one year. The result shows that the health indicator can effectively characterize the degradation state of the ACS,which can provide valuable information for proactive maintenance plan avoiding service interruptions.Development of prognostic model based on the proposed health indicator for failure prediction will be the future research direction.

Acknowledgments

This work was supported by the National Natural Science Foundation of China(61403198),the Jiangsu Province Natural Science Foundation of China(BK20140827)and China Postdoctoral Science Foundation(2015M581792).