PM Optimization:
Using Data-driven Analytics for
Life Centered Maintenance
Edzel Lapira, PhD
David Siegel, PhD
Rodrigo Vieira
WE KNOW
WHAT HAPPENS NEXT
OUR HISTORY & CORE TECHNOLOGY
•
Predictronics was started by senior researchers and developers from
the National Science Foundation Industry/University Cooperative
Research Center for Intelligent Maintenance Systems (IMS), which has
been a leader in predictive maintenance technologies since 2001.
•
At the core of Predictronics’ solutions is the Watchdog Agent
®
Toolbox: A collection of intelligent, rapidly deployable software agents
that can enable users to realize worry-free uptime of critical assets.
INDUSTRY VALIDATIONS
3
PREDICTRONICS’ VALUE PROPOSITION
4
REC
ON
FI
G
UR
A
BL
E
EVOLUTION OF MAINTENANCE PARADIGMS
5
Reactive
Maintenance
Trend of Maintenance Paradigm for Engineering Systems
Preventive
Maintenance
Condition-
Based
Maintenance
Prognostics
and Health
Management
Precision & Optimal Maintenance
•
Fail & Fix
•
Time based
•
Reliability centered
maintenance
•
FMECA – usage based
(unit, hours, …)
•
Need based
•
Condition
monitoring and
assessment
•
Predict & Prevent
•
Predictive maintenance
•
CBM+
•
ISHM, IVHM
“PHM is a system engineering discipline focusing on detection,
prediction, and management of the health and status of complex
engineered systems.”
-- the First International Conference on PHM (2008)
PM OPTIMIZATION
6
REACTIVE
Maintenance
and/or
PREVENTIVE
Maintenance
CONDITION-BASED
Maintenance
and/or
PROGNOSTICS
& HEALTH
MANAGEMENT
LIFE
CENTERED
MAINTENANCE
EVENT-BASED
(Maintenance
Records, alarms,
and fault logs)
DATA-DRIVEN
(Controller or process
data, sensor signals and
measurements)
IMS’ SYSTEMATIC PHM SYSTEM DEVELOPMENT
7
Normal Behavior Most Recent Behavior
Health Assessment
Performance Feature
TorqueData
Performance Confidence ValueHealth Information
Health Radar Chart
Health Visualization
Joint 1
Joint 2
Joint 3
Joint 4
Joint 5
Joint 6
ROBOT TORQUE MONITORING
TORQUE DATA DURING NORMAL/HEALTHY CONDITION
•
The RMS torque data for each of the six robot axis consists of a
single value for one complete cycle of the robot and the value is
acquired once per hour.
•
After each cycle, the disturbance torque values in the positive and
negative directions are also taken for each joint
9
TORQUE DATA DURING A KNOWN FAULTY CONDITION
•
In this example, the joint profile for axis 3 starts to deviate from the
torque profiles for the other 5 axis, indicating that it is experiencing
degradation and eventual failure.
•
The failure signature is visually noticeable; however detecting the
early signs of this problem requires more advanced analytical
methods.
DATA-DRIVEN ROBOT TORQUE MONITORING
11
ROBOT MONITOR – SHOW TREND OPTION
Life Centered Maintenance (LCM)
Rodrigo Vieira
Visiting Scholar
Center for Intelligent Maintenance Systems (IMS)
University of Cincinnati
Greater Cincinnati-Northern Kentucky
Chapter Meeting - SMRP
14
Decision Making Process
Decision
Making
Support
De
cis
ion
s
(Eve nts)Data
/ Info
rmat
ion
(Measure men ts and Ev ents)De
cis
io
ns
S
up
po
rt
Ex.:
Maintenance Strategy Models
“Agent”
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
15
Life-Centered Maintenance (LCM)
►
Methodology and Practical Example:
Asset:
Wind Turbine
Component:
Gearbox
Maintenance Action:
Gearbox Preventive Maintenance
(MTBM = 6 Months)
“Theoretical Maintenance Plan”
Last Maintenance Action “Now” Maintenance Action Maintenance Action Maintenance Action Maintenance Action
...
“Real Maintenance Plan”
Last Maintenance Action “Now” Maintenance Action Maintenance Action Maintenance Action...
Why?
- Asset Condition
- Production Windows
- Maintenance Team
availability
- Spare Parts
availability
- Work Tools
availability
- Weather Constraints
, etc
…
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP 16 0 0.05 0.1 0.15 0.2 0.25 0 10 20 30 40 50 PrevMaintCostHour($/h) # A c ti ons
Gearbox Preventive Maintenance
Life-Centered Maintenance (LCM)
Last Maintenance Action “Now” Maintenance Action Maintenance Action Maintenance Action...
PrevMaintCost
CorrMaintCost
TotalDownTimeCorr
TotalNoProdCorr
Maintenance Action Interval Maintenance Action Interval Maintenance Action Interval Maintenance Action Interval>> Preventive Maintenance Cost
>> Corrective Maintenance Cost
>> Total No Produced Energy due to Corrective Maintenance
>> Total Downtime due to Corrective Maintenance
/
TBM
=
PrevMaintCost/h
CorrMaintCost/h
TotalDownTimeCorr/h
TotalNoProdCorr/h
Label Description
-2 Unsatisfactory Life-Requirement
-1 Below Life-Requirement Reference
0 Achieve Life-Requirement Reference
+1 Above Life-Requirement Reference
+2 Greatly Above Life-Requirement “Ref.
V
er
y
B
ad
“-2
”
B
ad
“-1
”
O
k
“0
”
G
o
o
d
“+1
”
V
er
y
G
o
o
d
“+2
”
Life-Requirements
(Very Bad)
(Bad)
(Ok)
(Good)
(Very Good)
“Life-Requirements Class”
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
17
Life-Centered Maintenance (LCM)
►
Wind Turbine Examples:
-2 -1 0 1 2 0 20 40 60 80 100 120 140
WindFarm Class for PrevMaintCostHour($/h)
#
A
c
ti
ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 PrevMaintCostHour($/h)
# A c ti ons -2 -1 0 1 2 0 20 40 60 80 100
WindFarm Class for CorrMaintCostHour($/h)
#
A
c
ti
ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 CorrMaintCostHour($/h)
# A c ti ons -2 -1 0 1 2 0 10 20 30 40 50 60 70
WindFarm Class for TotalDownTimeCorr(h)
#
A
c
ti
ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 TotalDownTimeCorr(h)
# A c ti ons -2 -1 0 1 2 0 20 40 60 80
WindFarm Class for TotalNoProdCorr(kW)
#
A
c
ti
ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 TotalNoProdCorr(kW)
# A c ti ons
Ve
ry
B
ad
Bad
Ok
Good
Ve
ry
G
o
o
d
Ve
ry
B
ad
Bad
Ok
Good
Ve
ry
G
o
o
d
Ve
ry
B
ad
Bad
Ok
Good
Ve
ry
G
o
o
d
Ve
ry
B
ad
Bad
Ok
Good
Ve
ry
G
o
o
d
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP 18
Life-Centered Maintenance (LCM)
Multi-Dimensional Scaling (MDS)
“4x” Life-Requirements Class
“4x” Life-Requirements
Maintenance Action Maintenance Action Maintenance Action IntervalMDS 1
MD
S
2
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
19 -2 -1 0 1 2 0 20 40 60 80 100 120 WindFarm GLR Indicator # A c ti ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2 WindTurbine 002 GLR Indicator # A c ti ons -2 -1 0 1 2 0 20 40 60 80 100 120 WindFarm GLR Indicator # A c ti ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2 WindTurbine 004 GLR Indicator # A c ti ons
Life-Centered Maintenance (LCM)
MDS 1
GLR Indicator Evaluation
Maintenance Action Maintenance Action Maintenance Action Interval 𝐺𝐿𝑅(𝑡) = ( 𝛾𝑟∗ 𝐿𝑅𝑟(𝑡) -2 -1 0 1 2 0 20 40 60 80 100 120 140WindFarm Class for PrevMaintCostHour($/h)
#
A
c
tions
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 PrevMaintCostHour($/h)
# A cti ons -2 -1 0 1 2 0 20 40 60 80 100
WindFarm Class for CorrMaintCostHour($/h)
#
A
c
tions
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 CorrMaintCostHour($/h)
# A cti ons -2 -1 0 1 2 0 10 20 30 40 50 60 70
WindFarm Class for TotalDownTimeCorr(h)
#
A
c
tions
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 TotalDownTimeCorr(h)
# A cti ons -2 -1 0 1 2 0 20 40 60 80
WindFarm Class for TotalNoProdCorr(kW)
#
A
c
tions
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2
WindTurbine Class for 002 TotalNoProdCorr(kW)
# A cti ons -2 -1 0 1 2 0 20 40 60 80 100 120 WindFarm GLR Indicator # A c ti ons
Gearbox Preventive Maintenance
-2 -1 0 1 2 0 0.5 1 1.5 2 WindTurbine 002 GLR Indicator # A c ti ons
MD
S
2
Very Bad “-2”
Bad “-1”
Ok “0”
Good “+1”
Very Good “+2”
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
20
Life-Centered Maintenance (LCM)
Decisions
“Clustering”
…
Decision (“Launch a Maintenance Action”)
…
For Similarity
…
Normal v.s. Abnormal
Decisions
“Label”
…
Based on “System/Machine” Objective
…
Better v.s. Worse
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
21 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 002
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 004
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Life-Centered Maintenance (LCM)
►
Wind Turbine Examples:
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
22
Life-Centered Maintenance (LCM)
Decisions
“Clustering”
…
Decision (“Launch a Maintenance Action”)
…
For Similarity
…
Normal v.s. Abnormal
Decisions
“Label”
…
Based on “System/Machine” Objective
…
Better v.s. Worse
Decisions
“Understanding”
…
Help/Support the “Decision Making Staff”
to understand its “decisions impact” over
the “machine objective”
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
23 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 004
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 002
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Life-Centered Maintenance (LCM)
►
Wind Turbine Examples:
-3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 002
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 004
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
24
Life-Centered Maintenance (LCM)
Decisions
“Clustering”
…
Decision (“Launch a Maintenance Action”)
…
For Similarity
…
Normal v.s. Abnormal
Decisions
“Label”
…
Based on “System/Machine” Objective
…
Better v.s. Worse
Decisions
“Understanding”
…
Help/Support the “Decision Making Staff”
to understand its “decisions impact” over
the “machine objective”
Decisions
“Learning”
… From Previous Decisions
… Expert System
(Adaptive Neuro Fuzzy Inference System) - ANFIS)
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
25
GLR Real
f(Life-requirements)
GLR Estimation
(Based on “Health Indicators”)
INPUT VARIABLES
“Measurements”
AHRV11
AHRS11
.
.
.
AHRVij
AHRSij
TBM
OUTPUT VARIABLES
(Training Phase)
GLR
“Life-Requirements”
Expert
System
GLR Estimation
f(“Measurements”)
GLR Rules
Decision Making
Support
GLR Estimation
f(“Measurements”)
GLR Prognosis
f(“Health Indicators
Prediction”)
“Now”
Life-Centered Maintenance (LCM)
Maintenance Action Maintenance Action Maintenance Action Interval “Now”Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
26
Life-Centered Maintenance (LCM)
… Help/Support the “Decision Making Staff”
to understand its “decisions impact” over
the “machine objective”
… Decision (“Launch a Maintenance Action”)
… For Similarity
… Normal v.s. Abnormal
… Based on “System/Machine” Objective
… Better v.s. Worse
… From Previous Decisions
… Expert System
(Adaptive Neuro Fuzzy Inference System - ANFIS)
Decisions
“Label”
Decisions
“Clustering”
Decisions
“Understanding”
Decisions
“Learning”
…
Predict/Prognosis the “decisions impact”
over the “machine objective”
Decisions
“Prognostic”
“Now”
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
27
Life-Centered Maintenance (LCM)
►
Wind Turbine Examples:
-3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 002
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 WorDec BetDec MDS 1 MD S 2
Gearbox Preventive Maintenance - GLR Indicator Evaluation - Wind Farm vs Wind Turbine 004
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
1 Week
GLRp = -0.3
3 Weeks
GLRp = +0.2
2 Weeks
GLRp = +0.6
1 Week
GLRp = -1.0
3 Weeks
GLRp = +0.1
2 Weeks
GLRp = -0.6
Greater Cincinnati-Northern Kentucky Chapter Meeting - SMRP
28
Life-Centered Maintenance (LCM)
“Now”
… Help/Support the “Decision Making Staff”
to understand its “decisions impact” over
the “machine objective”
… Decision (“Launch a Maintenance Action”)
… For Similarity
… Normal v.s. Abnormal
… Based on “System/Machine” Objective
… Better v.s. Worse
… From Previous Decisions
… Expert System
(Adaptive Neuro Fuzzy Inference System - ANFIS)
Decisions
“Label”
Decisions
“Clustering”
Decisions
“Understanding”
Decisions
“Learning”
…
Predict/Prognosis the “decisions impact”
over the “machine objective”
Decisions
“Prognostic”