Presentation

ASNE Fleet Maintenance & Modernization
Symposium 2014
Intelligent Asset Management on
Submarines and Ships through
Embedded Intelligence
Presented by Mr David Lofting MEng MIET
OUTLINE
•  Embedded Intelligence •  PHM – Prognos5c Health Management •  Benefits of PHM •  Case studies v CMAPSS dataset v FMMEA of a power relay WHAT IS EMBEDDED INTELLIGENCE?
The ability of a product, process or service to reflect on its own opera5onal performance, usage load, or environment to enhance the product performance and life5me, to increase quality or to ensure customer sa5sfac5on.” “
EMBEDDED INTELLIGENCE
•  Applications to dictate the technology solution sets
•  Determine solution spaces for: sensor modalities, communications, power, processors
•  Physical partitioning of
functionality
•  Packaging for cost, size and
compatibility
•  Assembly methodologies
Design for EI
Packaging &
Interconnect
Intelligent
Software
•  Diagnostic and prognostics
•  Layered ontologies, mapping
data
•  Compatible source-target
semantics
•  Semi-automated linking
Applications
engineering
Manufacturing
Solutions
•  Process consolidation
•  Cradle-to-cradle: remanufacturing,
recycling and re-use
•  Biomimetics and biomorphism,
mechanobiological fabrication
System
Services
•  Service and lifecycle requirements to support integration
and co-operation of components and services
•  Optimisation of the services for efficiency across the
lifecycle
APPLICATIONS OF EMBEDDED
INTELLIGENCE
Interconnected systems, e.g. wireless agricultural environmental monitoring •  Monitors trends for crops using sensors •  Can decide if more water/ feed needed •  Machine learning influences control ASSET MANAGEMENT
•  Significant economic investments, e.g. £6.3 billion for HMS Queen Elizabeth •  Embedded Intelligence ensures opera5onal performance and life5me of high value assets ASSET MANAGEMENT
•  CfA -­‐ Contrac5ng for Availability, for example Trafalgar class submarines •  Reduc5on of Through Life Costs •  Contractors benefit from product op5misa5on •  Navies benefit from greater availability PHM – PROGNOSTICS AND HEALTH
MANAGEMENT
•  Remaining Useful Life (RUL) predic5ons on high-­‐
value assets •  Main difference between prognos5cs and condi5on monitoring systems •  Also end-­‐of-­‐life behaviour, premature ‘infant mortality’ system failure iden5fica5ons, etc. •  US Dep. of Defense 5000.2 policy document RUL - REMAINING USEFUL LIFE
•  Data is required to predict an asset’s RUL in real-­‐5me •  System inputs: historical data, reference standards, sensors, etc. Prognos5cs Health Condi5on Monitoring (PHM) Monitoring system design (CM) system design PROGNOSTIC DIAGNOSIS METHODS
• Data-­‐driven • Canary • Physics of Failure • Fusion DATA-DRIVEN DIAGNOSIS
1.  Extract relevant features of the system 2.  Apply sta5s5cal model to determine measure of system health 3.  Sta5s5cal predic5on of failure •  Do need lots of historical data •  Don’t need total in-­‐depth understanding of system physics CANARY DIAGNOSIS
•  ‘Smart coupon’ •  Canary has same failure mechanism as system being monitored •  Mul5ple canaries with varied ra5o of accelerated canary failure to system failure Cuby Sark PHYSICS OF FAILURE DIAGNOSIS
•  Primary PHM modelling approaches used are Physics of Failure (PoF) based •  Understanding of chemical, mechanical, thermal or electrical cause of system failure •  Addresses root causes of failure such as fa5gue, fracture, wear and corrosion FUSION DIAGNOSIS
•  Uses combina5on of prognos5c methods •  Algorithmic techniques could be Dempster-­‐
Shafer fusion, Bayesian inference, fuzzy-­‐logic inference, neural network fusion, simple weigh5ng/ vo5ng, etc. •  Various architectures possible •  Adds both to system complexity and confidence levels BENEFITS OF PROGNOSTIC SYSTEMS
§  Enables auxiliary systems to take over before the primary system suffers catastrophic failure §  Increased understanding and insight of systems being monitored can highlight development opportuni5es §  Helps to an5cipate maintenance requirements, therefore reducing costs and aiding logis5cs CASE STUDY – CMAPSS DATASET
•  Prognos5c applica5on where extensive datasets are available •  Ideal but unusual in Naval plaeorm applica5ons •  Novel approach combining v Autoregressive dynamic modelling tool v State-­‐of-­‐the-­‐art classifier •  Remaining Useful Life (RUL) predic5ons WHAT IS NASA’S CMAPSS?
•  Commercial Modular Aero-­‐Propulsion System Simula5on Turbofan engine •  Simulates 90K lbs thrust turbofan engines with variable characteris5cs •  Modelled mostly with MATLAB and Simulink •  Advanced deteriora5on and fault simula5on DATASET
•  Dataset of 100 turbofan engines •  Each engine had samples from 21 sensors, and 3 opera5ng condi5ons associated •  80% engines used for algorithm training and 20% for verifica5on •  Used to train both algorithm modules ALGORITHM
•  Linear Weighted Projec5on Regression (LWPR): v Dynamic real-­‐5me learning modelling tool for func5on approxima5on v Predicts future states of the asset •  Random Under-­‐Sampling Boos5ng (RUSBoost): v State-­‐of-­‐the-­‐art classifier v Discriminates between healthy and faulty opera5on given a current state of the asset COMPARISON OF ALGORITHM TO
GROUND TRUTH
Algorithm predicts RUL with mean square error of 6 engine opera5on cycles Test results for 20 engines CMAPSS CASE STUDY SUMMARY
•  Representa5ve dataset and turbofan simula5on tool provided by NASA •  Novel 2-­‐stage algorithm approach for RUL •  RUL algorithm performance verified for manufacturing uncertainty and mul5ple faults •  Ideal scenario where lots of data available CASE STUDY – FMMEA OF POWER
RELAY
•  Electromechanical relay with no historical data available – much more common for naval plaeorms •  Failure Modes, Mechanisms and Effects Analysis (FMMEA) •  Endurance tes5ng •  Iden5fica5on of failure precursors TEST SET-UP
Relay was contained in an isola5on chamber, so test results could not be contaminated by par5cles. FMMEA
•  Failure Modes, Mechanisms and Effects Analysis •  Physics of Failure based approach •  Can document as table with headings of Failure Mode, Failure Mechanism, Failure Effect, Likelihood, Cri5cality, Repairability FMMEA methodology DESIGN OF EXPERIMENT
•  Mul5ple relays tested for endurance in controlled environment for different voltage source levels •  Sensing parameters monitored: voltage, current, case temperature and external temperature •  Failure criteria set RELAY ANALYSIS
•  Relays disassembled •  Contacts subjected to: v Topological analysis using a Zygo white light interferometer v Energy Dispersive X-­‐Ray Spectroscopy •  Data processed in MATLAB to extract features FAILURE MODE INVESTIGATION
Energy dispersive X-­‐ray spectroscopy analysis Topological analysis of contacts with white light interferometry FAILURE PRECURSORS
Pip and crater features -­‐ long arc dura5ons, shown by spikes in opening 5me & mean open current Erosion failure -­‐ spikes in resistance, which result in an increased closing 5me Unacceptable opening 5me -­‐ steady upward trend in open 5me PROGNOSTIC MONITORING
Raw opening 5mes, 30V test •  Current and voltage monitoring can detect failure precursors – no lab equipment needed •  Prognos5cs can be implemented in situ with Embedded Intelligence systems APPLICATION OF RELAY DATA FOR
PROGNOSTICS
•  Inexpensive to implement •  Safety and reliability •  Maintenance •  Logis5cs Example monitoring system CONCLUSION
•  Prognos5c systems offer many advantages v Asset/ component Availability v Embedded Intelligence func5onali5es v Development •  Case studies demonstrate: v Verifiable RUL using data-­‐driven algorithm v Extrac5ng failure precursors from PoF approach without pre-­‐exis5ng dataset