Mining in MRO process optimisation Maurice Pelt Aviation Academy, Amsterdam University of Applied Science m.m.j.m.pelt@hva.nl RAeS Conference, London, 5 September 2017 Increasing Efficiency & Reducing Costs within the Aircraft Maintenance Process using New Technology and Innovative Solutions
Contents Introduction Concept of Mining in MRO Results Understanding Results and preparation Mining in MRO test cases Conclusions and Outlook 2
Need for Mining in MRO process optimization MRO: Unpredictable process times and material requirements Mining promises to improve predictability Focus on SMEs: Limited financial and data resources but important for our economy 2 year applied research project until Q3 2018: already 15 cases Research question: How can SME MRO s use fragmented historical maintenance data to decrease maintenance costs and increase aircraft uptime? 3
Research aim Mining in MRO Generic data mining recommendations for MRO industry mining solutions for specific MRO companies Validation CRISP framework Knowledge development Aircraft uptime: Optimal and accurate MRO planning Toolbox for Mining in MRO Costs: Reduction overprocessing and idle time Demonstration projects Network and sharing Costs: Optimal use remaining life parts 4
CRISP phase preparation Mining models extract information from monitoring data Monitoring data Models to extract information Condition Sensors, data degradation monitoring Load Forces, temperature,.. degradation rate Strong growth in sensors Physical Mathematics, degradation models Knowledge based Domain expert knowledge Usage Hours, cycles, kilometers indication of degradation driven Statistics & learning (Un)supervised Our focus External data Shared data Environmental parameters influences on degradation Strong growth in available data Hybrid Combination of above 5
CRISP phase preparation Maintenance taxonomy Reactive Corrective Failure based Too late Maintenance Preventive Schedule based Usage based Too early Proactive Condition based maintenance Predictive maintenance Model based Physical model Knowledge model driven Right in time Right in time and known in advance 6
CRISP phase preparation First describe and analyse the past, then predict the future and prescribe actions to be taken
CRISP-DM applicable for Mining in MRO? mining: A sequence of steps Cross Industry Standard Process for Mining methodology: CRISP-DM Standard for data mining projects based on practical, real-world experience CRISP-DM is the most used data mining method (Piatetsky, 2014) Source: Chapman, et al. (2000)
CRISP phase preparation Identify the business drivers of a MRO company 1. Identify performance indicators based on these drivers 2. Identify potential DM applications 3. Select relevant data sources Aircraft Uptime break down Total time Aircraft uptime Backlog OEM Aircraft Downtime Corrective maintenance Planned maintenance Interval based maintenance Duration (Turn Around Time) Reliability Engineering/AMP Forecast Accuracy of Mx Checks 9
CRISP phase preparation MRO Costs break down Per unit cost Materials MRO costs Carrying costs Interval of Mx Reliability Engineering/AMP Forecast Accuracy of Mx Checks Inspections Labour costs Manhour per task Repairs Infrastructure and overhead Component replacements (rotables) Variance Manhour estimate Manhour Buffer Nominal Task load Forecast accuracy 10
CRISP phase preparation 3 main categories of data sources: Maintenance data, FDR (AHM) and External data MPD ERP Task Skill Interval Time Since Zone Reference Effectivity Jobcards Vendor P/N S/N Order Qty SB status Removal reason Registration Safety Stock lvl Date stamps Location (on + off a/c) Form 1 P145 Release TSN, TSO P/N S/N Release Registration ATA Discrepancy Corrective Action Manhours Engineer Changed p/n, s/n AMM, IPC reference Date FDR AHM External Fault Codes Actions System parameters Trends Alert messages Diagnostics Date, fh s, fc s OEM databases Wheater data Aircraft position of similar systems Airport / runway data 11
CRISP phase preparation preparation covers activities to construct the final datasets from the initial raw data Intial datasets based on business Deal with imperfect and incomplete data Integrate, format and verify final data set Often tedious, time consuming Cleaning steps Construct data Integrate data Transform data Reduce data Exsyn Remove duplicates; Remove false malfunctions Yes Yes Yes No Jetsupport 1 Remove errors; Fill empty cells; Remove empty cells; Yes Yes Yes Yes Outliner removal; Remove irrelevant data Jetsupport 2 Remove irrelevant data Yes Yes Yes No Jetsupport 3 Correct errors; Fill empty cells; Remove empty cells Yes No Yes No LTLS - Yes No Yes Yes Nayak Correct errors; Fill empty cells; Outliner removal Yes Yes Yes No RNLAF Remove errors; Fill empty cells; Remove irrelevant Yes Yes Yes No data Tec4Jets Remove errors; Fill empty cells; Remove empty cells Yes Yes Yes Yes 12
Case Nayak : Causes of negative performance in high season A/B-checks and line maintenance for KLM Fokker 70 Causes of drop in Fleet Availability during high season CRISP methodology Performance contract: aircraft uptime Correlate ATA (sub)chapter to problems AMOS, weather data, flight data, unscheduled ground time events preparation Cleaned and integrated Descriptive analysis Support Vector Machine to predict problems related to weather Aircraft uptime, part costs Performance drop correlated to ATA subchapter, e.g. tyres, brakes and cabin air quality 13
Case Tec4Jets: Optimal moment to change tyres Line maintenance and A checks, part of operator TUI Increase availability and lower maintenance costs CRISP methodology preparation Issue tree potential applications Selected: Prediction of wheel changes AMOS, FDM cycles, weight, braking action, runway length and temperature Cleaning, integration into single dataset Visualise and calculate correlations Prediction: aircraft uptime, part costs Not statistically significant (yet) 14
Case: Predictive maintenance model of legacy aircraft using external data sources Access tot sensitive flight data is restricted Reduce unplanned maintenance costs excluding sensitive flight data and replace this with other data sources CRISP methodology preparation Predict failures of components (ATA subchapters) Maintenance data, ADS-B data (Flightradar24), weather data (NCEI) Split in different flight phases Averaging of parameters Dimensionality reduction Clustering K-means detected 58 anomalies and DBSCAN 69 Aircraft uptime, part costs Correlated failures and ADS-B data Showed flight anomalies before component (nose wheel) failed 15
Case: Engine Health Monitoring with data that are available for Airlines Inflight data from aircraft engines are sent to the manufacturer only Improve maintenance efficiency using free available data CRISP methodology preparation Economic Replacement Point (ERP), Life Limiting Parts (LLP) and Exhaust Gas Temperature (EGT) define the optimal replacement time of engines Available data: EGT, fuel consumption, oil pressure and oil consumption Select engine type Clean and check data Develop Engine Health Monitoring model Forecast optimal engine replacement point Aircraft uptime, Part costs EGT & LLP limits reached sooner than ERP 16
MAN-HOURS [HR] Case Jetsupport: Predict the duration of planned maintenance checks D E V I A T I O N A C T U A L V E R S U S I N D I C A T E D D U R A T I O N JetSupport is CAMO of two Dornier aircraft of the Dutch Coastguard Increase availability with improved planning of maintenance 48:00:00 36:00:00 24:00:00 Estimated Actual 12:00:00 CRISP methodology 0:00:00 Reduce uncertainty in: Unplanned maintenance Duration planned maintenance (findings) SCHEDULED PACKAGE MRX maintenance system preparation Manual cleaning and integration Automated retrieval Visualisation of planned actual Forecasting algorithms based on actual duration of checkpackages and task cards Aircraft uptime, Maint. efficiency More accurate planning of maintenance 17
Summary of 5 selected cases MRO industry recommendation CRISP DM descriptive, predictive, hypothesis CRISP DM predictive, semi unstructured CRISP DM descriptive, predicive, hypothesis CRISP DM predictive, semi unstructured parameter reduction no sensitive data needed CRISP DM predictive, hypothesis no detailed OEM data needed Solutions for MRO companies Company Solution Contributes to Tec4Jets Predict tyre wear depending on Aircraft uptime destinations and other parameters Part costs Nayak Jetsupport Exsyn Mx Exsyn/ Engines Find components (ATA subchapters) contributing to low performance in high season Predict the duration of planned maintenance checks Predict maintenance needs using external data sources Nose wheel failure as function of landing data Predict optimal engine replacement time (EGT, LLP, ERP) with data that are available for Airlines Aircraft uptime Part costs Aircraft uptime Costs: MRO utilisation rate Aircraft uptime Part costs Aircraft uptime Part costs Costs: MRO utilisation rate 18
Conclusions Source: MRO Air Overall conclusions Case studies proved the value of Mining Aircraft uptime: optimal and accurate planning MRO costs: efficiency, part costs CRISP-DM methodology useful for MRO Understanding preparation Mostly problem (hypotheses) driven approach Supervised data driven approach also applicable Aircraft uptime and MRO costs linked to data sources Distinct DM goals along MRO value chain and business models not aligned for compliance rather than prediction Confidentiality and ownership issues Successful work arounds with own and public data preparation much work Need to improve data structures and capturing Descriptive analyses very useful Promising results with data driven approach Future focus on predictive analyses 19
Thank you for your attention Maurice Pelt m.m.j.m.pelt@hva.nl co-authors: Robert Jan de Boer Jonno Broodbakker 20