A framework for prognostic-based real-time life extension

Seyed A. Niknam Dr. Rapinder Sawhney

The University of Tennessee, Knoxville

Outline

Motivation

Introduction

Background

Elements of the life-extension program

Non-monotonic degradations

Decision Making

Future works

Prognostic Reports

Report 1

– The remaining life of system A is 55 days

Report 2

– System A will last 55 days with no change in the current operating conditions. However, by decreasing the load and speed around 10%, system A will be working for 95 days. The optimal time for the replacement of the degraded units would be after 70 days when the spare parts, equipment and personnel are available. Also, after 65 days the chance of adverse effect on unit B increase by 20%. Moreover, …..

Motivation

Efficiency over the nominal design life

Investment for new installations

Failure rate of 300kW & 1 MW wind turbines (failure per turbine per year) [1]

Unit 300kW 1MW

Generator 0.059 0.126

Brake 0.029 0.056

Hydraulics 0.039 0.096

Yaw system 0.079 0.152

Sensors 0.037 0.151

Pitch system 0.034 0.237

Blade 0.078 0.308

Gearbox 0.079 0.255

Shaft/bearings 0.002 0.046

Introduction

Operation and Maintenance (O&M) costs: 10-30%

O&M costs increase over the 20 years of wind turbine operating life

Minimizing the O&M costs:

– Reliability optimization

– Condition-based maintenance

– Maintenance strategies

Introduction

System durability vs. O&M costs

A multiple-criteria decision making process – Adequate information with respect to failure-

causing faults and the remaining operational life of the faulty components and sub-systems

Objective: to provide a frame for supporting a prognostic-based life-extending program – Lack of knowledge (e.g. about aging mechanism),

– Lack of tools (e.g. trust worthy prognosis algorithm),

– Lack of data and time

Background

ISO 13381: General guidelines for prognostic Prognosis is a case-dependent process, and therefore, it

is not reasonable to specify certain approaches or methodologies in prognosis standards.

Since prognosis is mainly based on data, determining the degree of certainty of prognosis process i.e. the confidence level is an essential task.

Four phases: – Pre-processing: identifying and modeling the failure modes – Existing failure mode prognosis: severity of existing and

future failure modes – Future failure-mode prognosis: estimation of time to failure – Post-action prognosis: how to prevent the initiation of

future failure modes.

Background

Applying prognosis for extending the operable life

– Damage-mitigating control system: connecting the dynamics of material degradation with the current active control technologies

– Major challenge: to characterize the fatigue damage model and make it compatible with the control system

Background

Emphasis on real-time and onboard prognostic

Business issues and the level of risk in appropriate selection of prognostic model

Background

Staged prognostics approach

Monitoring and managing passive systems, such as the nacelle in wind turbine

Prognostic-based Life Extension

Priority: to mobilize the potential residual life to guarantee the return on investment

The first and foremost step: effective and robust diagnosis – To detect and identify the impending faults

and relevant collateral damage. – Competing failures – Early fault detection (time) – Nature of the incipient faults (e.g. pitting of

spur gears) – Real-time diagnosis

Prognostic-based Life Extension

Real-time estimations of remaining useful life (RUL)

– RUL: time that a component or system is able to operate without the need for major repair and maintenance or without significant rate of minor faults

– Predict the chance that a system operates without a major failure up to a specific time

Prognostic-based Life Extension

Prognostic-based Life Extension

Effects-based or individual-based

– prognostic methods based on sensed or inferred degradation measures.

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.60

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8degradation paths of failed units (noisy data)

time

degrd

ation m

easure

s

( , , )i iy t

Prognostic-based Life Extension

Self‐healing or human interventions are desirable.

Repair activities:

– Major repairs have direct influence of the age of system. Normally major repairs require special equipment and have longer process.

– Minor repairs reduce the rate of aging and require short period of downtime e.g. routine services, oil change and alignments.

Prognostic-based Life Extension

Concerns about the end-of-life threshold:

– Adjustment: the failure threshold should be adjusted after life-extending events. This will add complexity to the prognostic algorithms.

– Using failure zone

– Ignore a predefined critical threshold (path classification and estimation model)

Parameter Selection

Three features were introduced by Coble and Hines to characterize suitable prognostic parameters [4, 5]:

Monotonicity: Underlying positive or negative trend of the parameter. It is given by the average difference of the fraction of positive and negative derivatives for each path."

Prognosability:

Trendability: Indicates the degree to which the parameters of a population of systems have the same underlying shape and can be described by the same functional form."

( )exp( )

( )

std failurevaluesprognosability

mean failurevalues startingvalues

Question

Is there any non-monotonic degradation trend?

Examples of non-monotonic degradation

Vacuum fluorescent displays

Degradation in light displays such as plasma display panels [6]

Critical characteristic of light display quality is luminosity (or brightness)

Examples of non-monotonic degradation

Fatigue Crack Closure

Excellent example of self-healing [7]

Examples of non-monotonic degradation

Fouling in Heat Exchangers

0 0.5 1 1.5 2 2.5 3 3.5 4

x 104

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6x 10

-3

Overa

ll T

herm

al R

esis

tance

Time (minute)

Overall Thermal Resistance 1/UA

Examples of non-monotonic degradation

Rotor Unbalance

0 0.5 1 1.5 2 2.5 3 3.5 4

x 104

0

200

400

600

800

1000

1200AE SIGNAL - Bearing 1

Time

RM

S

Trend Analysis

Trend: general pattern of the mean level.

Trend analysis: the process of identifying significant changes in the magnitude of a reference variable

Decision Making

Considerable research on prognostic algorithms

Attention to post-prognostic issues

Return on Investment (ROI) associated with the opportunities created by prognostic

Optimizing availability

Take actions in the period of the remaining operational life

Optimal set of actions (i.e. alternatives)

Decision Making

Challenges:

Presence of various decision alternatives, and therefore, the need for compromising – Fault accommodation (i.e. modifying control

rules or using redundancy) - Derating

– Altering operation conditions or tactical control

– Maintenance practices

Multiple criteria which usually have conflict and make the judgment of alternatives a complex endeavor.

Uncertainty associated with the available data

Decision Making

Modifying the operating conditions creates the need for an optimization procedure

Power curve for a 5 MW wind turbine

Decision Making

Multiple Criteria Decision Making/Analysis (MCDM/MCDA)

– Multiple Attribute Decision Making (MADM): focuses on selecting the best alternative among a finite set of predetermined alternatives

– Multiple Objective Decision Making (MODM): deals with creating alternatives from a large number of alternatives

Future Works

Prognostic: – Adaptive prognostic models – Real-time and onboard prognostic – Ability to handle non-monotonic trend

How well degradations measures indicate the repair activities – Degradation measures can be either a directly measured

parameter or a function of several measurable parameters

Noise smoothing and automatic trend estimation methods

Detecting and modeling the turning points Fuzzy logic-based approaches for trend detection:

– Able to identify the underlying trends (particularly during high fluctuations of a signal)

– Able to handle noisy data

General model for MODM

References

[1] Arabian-Hoseynabadi, H., H. Oraee, and P.J. Tavner, Failure Modes and Effects Analysis (FMEA) for wind turbines. International Journal of Electrical Power & Energy Systems, 2010. 32(7): p. 817-824

[2] Lee, J., et al., Intelligent prognostics tools and e-maintenance. Computers in Industry, 2006. 57(6): p. 476-489

[3] Reinertsen, R., Residual life of technical systems; diagnosis, prediction and life extension. Reliability Engineering and System Saftey, 1996. 54: p. 23-34.

[4] Coble, J.B., Merging Data Sources to Predict Remaining Useful Life – An Automated Method to Identify Prognostic Parameters, in Nuclear Engineering. 2010, University of Tennessee.

[5] Coble , J. and J.W. Hines, Fusing Data Sources for Optimal Prognostic Parameter Selection, in Sixth American Nuclear Society International Topical Meeting on Nuclear Plant Instrumentation, Control, and Human_Machine Interface Technologies NPIC&HMIT. 2009: Knoxville, Tennessee.

[6] Bae, S.J. and P.H. Kvam, A Nonlinear Random-Coefficients Model for Degradation Testing. Technometrics, 2004. 46(4): p. 460-469.

[7] Lee, S.Y., et al., A study on fatigue crack growth behavior subjected to a single tensile overload - Part 1. Acta Materialia, 2011. 59: p. 485-494.

[8] Nielsen, J.J. and J.D. Sørensen, On risk-based operation and maintenance of offshore wind turbine components. Reliability Engineering & System Safety, 2011. 96(1): p. 218-229.

THANK YOU