Chapter 1. Introduction : Prognostics & Health Management

23
Seoul National University Prognostics and Health Management (PHM) Chapter 1. Introduction : Prognostics & Health Management Byeng D. Youn System Health & Risk Management Laboratory Department of Mechanical & Aerospace Engineering Seoul National University

Transcript of Chapter 1. Introduction : Prognostics & Health Management

Page 1: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University

Prognostics and Health Management (PHM)

Chapter 1. Introduction: Prognostics & Health Management

Byeng D. YounSystem Health & Risk Management LaboratoryDepartment of Mechanical & Aerospace EngineeringSeoul National University

Page 2: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University

CONTENTS

2019/1/4 - 2 -

Motivation1PHM Overview2FMECA, ALT, and NDE3Key PHM Edges4

Page 3: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 3 -

Chapter 1. Introduction

Still Physical Systems Fail…

8100TEU containership sinking, June 17, 2013

- Due to Buckling of shell plating & Fatigue in welded structure- Consequence: about $500 million property & business loss

LNG plant explosion, Jan. 19, 2004- Due to LNG Leak in Pipe- Consequence: 27 killed, 72 injured, $100 million loss

Consequence: U.S. solely spends $250 billion/year on reliability and maintenance in 2012

Wind turbine failure, Feb. 22 2008- Due to Maintenance error (brake failure)- Consequence : Collapse of whole wind turbine

Page 4: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 4 -

Maintenance

Corrective Maintenance (Reactive maintenance)• Corrective maintenance (also known as "breakdown maintenance") is a maintenance action executed in the

event of equipment breakdown.• Corrective maintenance focuses on restoring the equipment to its normal operating condition. The breakdown

equipment is returned to normal within service specifications by replacing or repairing faulty parts and components.

• Pros: Reduce unnecessary preventive maintenance action, relevant to the non-critical componentCons: Unable to prepare for the breakdown in advance, not applicable to the critical components

Preventive Maintenance• Preventive maintenance is any maintenance that is designed to retain the healthy condition of equipment and

prevent failure.• Scheduled maintenance: Maintain based on a planned schedule• Pros: Prevent the critical accident in advance and reduce the maintenance and downtime cost

Cons: Require accurate condition diagnosis system

Proactive maintenance• Proactive maintenance is the maintenance philosophy that supplants “failure reactive” with “failure proactive”

by activities that avoid the underlying conditions that lead to faults and degradation.• Condition-based (or predictive) maintenance: Maintain when anomaly condition is detected or predicted• Pros: No failure on the system, prolong system lifetime,

Cons: Require thorough understanding of the system, difficult to predict the states of the system.

Chapter 1. Introduction

Page 5: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 5 -

What PHM Can Do?

• Health degradation by aging• System reliability reduce• Failure rate increase

• Lack of system verification• Harsh operating condition for

high efficiencyInfant failure detectionPrognostics of random failures

Anticipation of failures due to agingSmart O&M decision making

PHM(Prognostics & Health Management)

System Failure New SystemDegradation

Chapter 1. Introduction

Page 6: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 6 -

Concept of PHM

- Health degradation due to aging- Prevention of health worsening by

proper diagnosis and treatment

- Performance degradation - Failure prevention and life extension

with PHM technique

Human lifetime

Haz

ard

Rat

e o

rH

ealth

con

ditio

n

Cost of medicineProbability of death

Human PHM Action

Sensing ... Management

Human Engineering System

Chapter 1. Introduction

Page 7: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 7 -

Chapter 1. Introduction

PHM Procedure

Sensing module

- To ensure high damage detectability by designing an optimal sensor network

Reasoning module

Prognostics module Management module

Failure Analysis PHM Solution

Validation

- To extract system health relevant information in real-time and to classify system health condition

- To predict remaining useful lives (RULs) of engineered systems in real-time

- To enable optimal decision making on maintenance of engineered systems

- Understanding system failure

- FMECA- ALT- NDE

- Validation metric for detection, diagnosis, and prognosis

Page 8: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 8 -

1. Failure Analysis

FMECA – Failure Modes Effects and Criticality Analysis • Failure Modes and Effects Analysis (FMECA) is methodology for analyzing causes of

failures and understanding their frequency and impact. • Failure mode: Effect by which a failure is observed on failed parts or systems.

– Ex. Bearing failure mode: excessive wear, breakage, etc.• The FMECA result highlights failure modes with relatively high probability and severity

of consequences, allowing remedial effort to be directed where it will produce the greatest value.

• The FMECA includes,– Failure mode identification– Failure effects analysis– Severity classification– Failure detection methods– Criticality ranking– Critical item/failure mode list– Maintainability analysis

• Understanding system failure is important to develop the reliable PHM algorithm.

Chapter 1. Introduction

Page 9: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 9 -

1. Failure Analysis

FMECA – Failure Modes Effects and Criticality Analysis

• Understanding system failure is important to develop the reliable PHM algorithm.

Chapter 1. Introduction

Page 10: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 10 -

1. Failure Analysis• Failure matrix

Chapter 1. Introduction

Page 11: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 11 -

1. Failure AnalysisALT – Accelerated Life Testing• Testing a product by subjecting it to conditions (stress, strain, temperatures, voltage,

vibration rate, pressure etc.) in excess of its normal service parameters in an effort to uncover faults and potential modes of failure in a short amount of time.

• By analyzing the product's response to such tests, engineers can make predictions about the service life and maintenance intervals of a product.

NDE – Non-destructive Evaluation• NDE is a wide group of analysis techniques used in science and technology industry to

evaluate the properties of a material, component or system without causing damage.• The frequently used NDE methods are

– Eddy-current– Magnetic-particle– Liquid penetrant– Radiographic– Ultrasonic– Visual testing

Chapter 1. Introduction

Page 12: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 12 -

1. Sensing Solution 2. Reasoning Solution

3. Diagnostics and Prognostics Solution 4. Management Solution

2. PHM Solutions

: Design sensor network maximizing the health information

: Estimation and classification of current health state of a system

• Data acquisition planning• Selection of sensor type & number• Sensor network design

• Health data & health index• Time- & frequency analysis• Time-frequency analysis• Signal processing• Physics-based approach• AI-based approach

• Rule-based diagnostics• Physics-based diagnostics• AI-based diagnostics• Remaining useful life (RUL) prediction• Physics-based approach• Data-driven approach

: Predictive diagnostics and prediction of future state of system

• Definition of health classes (normal/warning/fault/failure) and corresponding maintenance actions

• Corrective/preventive/predictive maintenance

• Maintenance scheduling strategies• Resilience-driven design• Cost analysis

: Optimal system operation and maintenance

Chapter 1. Introduction

Page 13: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 13 -

Chapter 1. Introduction

Domain Knowledge

Data-driven

Rule-based Physics-based

Data-driven + Physics-based

Dat

a Si

ze

Health Features

Raw Data

Feature Engineering

Preprocessing

Diagnostics/Prognostics

Health Prediction

Power & Energy

Hydro-system

Electrical &Electronics

Power Trans.

Machining

Driving

Cleaning Filtering

EnvelopingScalingStatistical Modeling

Organization

Domain Knowledge

Data-driven

Rule-based Physics-based

Data-driven + Physics-based

Dat

a Si

zeDomain Knowledge

Data-driven

Rule-based Physics-based

Data-driven + Physics-based

(Hybrid)

Dat

a Si

ze

Page 14: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 14 -

2.1 Sensing SolutionSelection of failure modes and sensor types

• Sensor network design and IT solutions– System characteristics – Operating and environmental conditions– Sensor location– Number of sensors– Sampling rate & duration– DAQ, gateway & communication– Server and cloud

Acquisition of normal and failure data • Real system– Hard to acquire failure data

• Testbed – Easy to implement failure modes

• CAE model – To help understanding of system physics

Failure modes

Sensor types

HIP LP-A LP-B

Brn-1 Brn-2 Brn-3 Brn-4 Brn-5 Brn-6 Brn-7

Chapter 1. Introduction

Page 15: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 15 -

2.2 Reasoning SolutionData Preprocessing & Processing

Time- & frequency analysis Time-frequency analysis

|Y|

Frequency

Time

Y(t)

Spectrogram

1 2 3 4

Time (secs)

0

0.5

1

1.5

Freq

uenc

y (k

Hz)

-60

-40

-20

0

20

Pow

er/fr

eque

ncy

(dB/

Hz)

Chapter 1. Introduction

Preprocessing

- Filtering (High/low/band, AR, MED, etc.)

- Denoising (TSA, resampling, etc.)

- Enveloping- Scaling

Deep learning-based feature engineering

Page 16: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 16 -

2.3.1 Diagnostics Solution

Chapter 1. Introduction

Diagnostics

Supervised classification – with labeled data

X1

X 2

Class IClass II

X1

X 2

Cluster ICluster II

Unsupervised classification – without labeled data

Rule-based classification

Page 17: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 17 -

2.3.2 Prognostics SolutionPhysics-based Approach• Pros

– Possible to assess RUL in early stages– Understanding of PoF

• Cons– Unreliable prediction accuracy– Applied to component level– Physics of failure (PoF) knowledge

required

Simulation Estimation

Identify Model

Simulate with Loading

Loading Signals

Response Signals

Update Parameters

Update & Project HI

Predicted RUL

Offline Process Online Process

Extract Offline HI

Build Health Knowledge

Training Signals

Testing Signals

Extract Online HI

Project or Interpolate

Predicted RUL

Data-driven Approach• Pros

– Applicable to system level– Management of uncertainty

• Cons– Failure data necessary– Massive run-to-failure data required

Chapter 1. Introduction

Page 18: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 18 -

2.4 Management SolutionMaintenance Management• To guide maintenance activities and prepare maintenance resources• Corrective maintenance, preventive maintenance, predictive maintenance, etc.

Asset Management• Asset management involves the balancing of costs, opportunities and risks against the

desired performance of assets, to achieve the organizational objectives. (ISO 55000)• LCC (Life Cycle Cost) Problem

Maintenance cost Downtime for maintenance Efficiency of the asset Operating cost Quality of operation or service Safety to staff and the public

Resilience• An ability to sustain functionality by resisting and recovering from adverse events

Chapter 1. Introduction

Page 19: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 19 -

Chapter 1. Introduction

3. ValidationValidation Metric• To check the validity of the diagnostics and prognostics algorithms, the validation metrics are

needed.• Model validation metrics have been developed to provide a quantitative measure that

characterizes the agreement between predictions and observations.• Classical hypothesis testing

t-test statistic F-test statistic Anderson-Darling test Kolmogorov-Smirnov (K-S) test

• Bayes factor: is based on the Bayesian hypothesis testing. It is the ratio of posterior distributions of the null and alterative hypothesis.

• Frequentist’s metric: quantifies the agreement from a different perspective by measuring the distance between the mean of the predictions and the estimated mean of the physical observations.

• Area metric: measures the differences between the entire distributions from the observations and predictions. It can be used when only a few data points from predictions or experiments are available

Page 20: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 20 -

Chapter 1. Introduction

Health diagnostics and prognostics of power transformers against mechanical faultsVi

brat

ion

Mea

sure

.He

alth

Gra

de S

yste

m

0 10 200

20

40

60

80

100

120

Oper. time [year]

120H

z m

ax.v

el. [

mm

/sec

]

0 10 200

1

2

3

4

5

6

7

8

Oper. time [year]

240H

z m

ax.v

el. [

mm

/sec

]

whole sensorsdesigned SN

α β

1 2 3 654 1 2 3 654replaced

RUL

(yea

r)

Transformer unit #He

alth

Pro

gnos

tics

Opt

imal

Sen

sor P

ositi

onin

gTime-domain

0 500 1000 1500 20000

0.5

1

1.5

2

2.5

3

3.5

X: 120Y: 3.33

X: 240Y: 0.7207

X: 360Y: 0.3368

Freq.-domain

PHM Examples

Optimal sensor positioning, Health grade system, Remaining useful life prediction

Page 21: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 21 -

Analyzed major failure mechanisms of machine tool spindle and applied them to the accelerated life testing, identified detection technique of major causes of failures

Accelerated life testing

0 100 200 300 400 500 600 7000

1

2

3Envelope Detection : Hilbert Transform

Frequency (Hz)

|X(f)

|

0 100 200 300 400 500 600 7000

0.5

1

Frequency (Hz)

|Y(f)

|

112.5Hz

218.8Hz

112.5Hz

218.8Hz

BSF_1x

BPFI_1x

BSF_1x

BPFI_1x

Data acquisition of Testbed

Failure analysis

Diagnosis for main components

Acce

lera

ted

Life

Te

stFa

ult D

iagn

osis

Zero failure reliability test

Anomaly detection

Threshold

Qualification Life

Design Life

X XX

X

Performance Evaluation

PHM Examples

Chapter 1. Introduction

Machine tool spindle failure mechanism-based accelerated life testing method and diagnostics algorithm

Page 22: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University

THANK YOUFOR LISTENING

2019/1/4 - 22 -

Page 23: Chapter 1. Introduction : Prognostics & Health Management

Seoul National University2019/1/4 - 23 -

Reference

[1] Stamatis, Dean H. Failure mode and effect analysis: FMEA from theory to execution. ASQ Quality Press, 2003.

[2] Spencer, F. W. (1991). "Statistical Methods in Accelerated Life Testing“. Technometrics. 33 (3): 360-362.

[3] Introduction to Nondestructive Testing, The American Society for Nondestructive Testing.