LECTURE 12 MAINTENANCE: BASIC CONCEPTS Politecnico di Milano, Italy piero.baraldi@polimi.it 1
LECTURE 12 PART 1: Introduction to maintenance PART 2: Condition-Based and Predictive Maintenance 2
PART 1: INTRODUCTION TO MAINTENANCE 3
4 DEGRADATION FAILURE MAINTENANCE Equipments, however well designed, will not remain safe or reliable if they are not maintained 4
Maintenance expenditures in some industrialized countries 5 Derived from M. Garetti 5
6 PART 2: MAINTENANCE STRATEGIC PLANNING 6
Maintenance Strategic Planning 7 WHEN to act- Before or after the fact : maintenance intervention approach; ON WHAT BASIS- Reliability, Availability, Cost, Safety, Environmental-centred : maintenance decision-making strategy 7
8 MAINTENANCE INTERVENTION APPROACHES 8
Types of maintenance approaces Maintenance Intervention Unplanned Planned 9
Planned Maintenance Maintenance Intervention Unplanned Planned Corrective Replacement or repair of failed units Scheduled Perform inspections, and possibly repairs, following a predefined schedule Conditionbased Monitor the health of the system and then decide on repair actions based on the degradation level assessed Predictive Predict the Remaining Useful Life (RUL) of the system and then decide on repair actions based on the predicted RUL 10 10
Corrective maintenance 11 Failure Maintenance No maintenance action is carried out until the equipment or structure breaks down. Upon failure, the associated repair time is typically relatively large large downtimes Efforts are undertaken to achieve Small Mean Times to Repair (MTTRs) Logistics 11
Corrective maintenance: when is it applied? 12 Failure Maintenance Equipments: No safety critical No crucial for production performance Spare parts easily available and not expansive 12
Planned maintenance 13 Decision Failure Maintenance Why? Production and safety benefits Costs of performing Maintenance 13
Maintenance Philosophies (2) N.S. Arunraj, J. Maiti / Journal of Hazardous Materials 142 (2007) 653 661 14 14
Scheduled Maintenance Planned Scheduled Condition based Predictive Maintenance is carried out at scheduled intervals Intervals can be given in terms of: calendar time running time number of start and stop their combination Equipments may be repaired or replaced 15
Scheduled Maintenance: Objectives To rejuvenate the equipment = to decrease its failure rate Planned replacement (e.g. Planned replacement of the bearing in a rotating equipment) To slow down degradation (wear, fatigue) = to limit the increase of the failure rate Lubrication Routine maintenance (tightening of connectors) 16
Scheduled Maintenance: Pros and Cons Pros: Reducing number of failures Maintenance can be planned when it production or availability of the systems has the lowest impact on Cons: A scheduled maintenance approach generates maintenance tasks after a specific time interval which can result in a too early replacement of components, which is unprofitable. Failure 17
Scheduled Maintenance: Pros and Cons Pros: Reducing number of failures Maintenance can be planned when it production or availability of the systems has the lowest impact on Cons: A scheduled maintenance approach generates maintenance tasks after a specific time interval which can result in a too early replacement of components, which is unprofitable. Scheduled Maintenance Scheduled Maintenance Failure 18
Scheduled Maintenance: Pros and Cons Pros: Reducing number of failures Maintenance can be planned when it production or availability of the systems has the lowest impact on Cons: A scheduled maintenance approach generates maintenance tasks after a specific time interval which can result in a too early replacement of components, which is unprofitable. Scheduled Maintenance Scheduled Maintenance Failure Maintenance induced failures 19
Scheduled maintenance: decision Optimize the Decision: Intervals between PM maintenance actions Action rules Model: Failure/degradation process Maintenance effects, time to repair Costs of planned maintenance, corrective maintenance, production unavailability Decision Intervals between PM actions Action Rules Failure/degradation Failure times Degradation evolution Maintenance Effects on future failure/degradation behavior Time to Repair 20
Scheduled Maintenance: Decision Optimize the Decision (intervals between maintenance and action rules) Model: Failure/degradation process Maintenance effects, time to repair Costs Unavailability Costs interval between maintenance interval between maintenance 21
Condition-Based Maintenance Planned Scheduled Condition based Predictive 22
Maintenance Philosophies (2) N.S. Arunraj, J. Maiti / Journal of Hazardous Materials 142 (2007) 653 661 23 23
Condition-Based Maintenance (CBM) Decision Monitoring Failure Maintenance Equipment degradation monitoring: Periodic inspection by manual or automatic systems d failure x x d failure 0 d detection Inspection time 24
Condition-Based Maintenance (CBM) Decision Monitoring Failure Maintenance Equipment degradation monitoring: Periodic inspection by manual or automatic systems Continuous observations Ultrasonic Monitoring (regularly used in the oil and gas industry) 25
Condition-Based Maintenance (CBM) Decision Monitoring Failure Maintenance Equipment degradation monitoring: Periodic inspection by manual or automatic systems Continuous observations Equipment degradation level identification by: Direct measure (crack depth of a mechanical component) Indirect observations (symptoms related to the degradation process, e.g. quality of the oil in an engine, partial discharges in electrical cables, vibrations frequencies and amplitudes in rotating machinery) 26
CBM: Conclusions Identification of problems in equipment or structures at the early stage so that necessary downtime can be scheduled for the most convenient and inexpensive time. Scheduled Maintenance Scheduled Maintenance Condition Based Maintenance Failure Failure 27
CBM: Conclusions Identification of problems in equipment or structures at the early stage so that necessary downtime can be scheduled for the most convenient and inexpensive time. Machine or structure operate as long as it is healthy: repairs or replacements are only performed when needed as opposed to routine disassembly and servicing. Availability Unscheduled shutdowns of production Reduced costs Improved safety 28
Predictive Maintenance Planned Scheduled Condition based Predictive 29
Maintenance Philosophies (2) N.S. Arunraj, J. Maiti / Journal of Hazardous Materials 142 (2007) 653 661 30 30
Predictive Maintenance Decision RUL PROGNOSIS Monitoring Failure Maintenance Equipment degradation monitoring: Remaining Useful Life (RUL) prediction Maintenance Decision 15 10 5 0 500 1000 PROGNOSIS RUL 20 10 0 0 500 1000 31
Predictive Maintenance: Ex. 1 t=300: perform maintenance now or postpone it to the next planned outage at t=400? Degradation level d failure past degradation observations degradation model RUL PREDICTION t=300 Present Time t=400 time postpone maintenance to the next planned outage at t=400 32
Types of maintenance approaches Maintenance Intervention Unplanned Planned Corrective Replacement or repair of failed units Scheduled Replacement or Repair following a predefined schedule Conditionbased Monitor the health of the system and then decide on repair actions based on the degradation level assessed Predictive Predict the Remaining Useful Life (RUL) of the system and then decide on repair actions based on the predicted RUL 33 33
PART 2: CONDITION-BASED AND PREDICTIVE MAINTENANCE 34
Prognostics and Health Management Equipment (System, Structure or Component) Measured signals x 1 t x 2 t Detect Diagnose Predict Anomalous operation Normal operation c 1 c 2 c 3 Malfunctioning type (classes) Remaining Useful Life (RUL) 35
Data PHM & INDUSTRY 4,0 36 Digitalization 2.8 Trillion GD (ZD) generated in 2016 Available data Analytics 2012 2017 Time PHM Data Analytics 36
Maintenance Intervention Approaches & PHM Maintenance Intervention Unplanned Planned Corrective Scheduled Conditionbased Predictive Detection X X Diagnostics X X Prognostics X 37 37
38 Fault Detection
Fault Detection: what is it? 39 Equipment Measured signals 39
Fault Detection: objective 40 40 f 1 Automatic algorithm Normal condition f 2 Equipment f 1 Forcing functions Measured signals f 2 40
41 Methods for Fault Detection: Limit-based Model-based Data-driven 41
Data & Information for fault detection (I) 42 Normal operation ranges of key signals Example: Pressurizer of a PWR nuclear reactor Water level 10.2 m 3.8 m Abnormal condition Abnormal condition time Normal operation range 42
Methods for fault detection (I) 43 Normal operation ranges of key signals Limit Value-Based Fault Detection Example: Pressurizer of a PWR nuclear reactor Water level 10.2 m 3.8 m Abnormal condition Abnormal condition time Normal operation range 43
Methods for fault detection (I) 44 Example: Normal operation ranges of key signals Limit Value-Based Fault Detection Pressurizer of a PWR nuclear reactor Drawbacks: No early detection Control systems operations may hide small anomalies (the signal remains in the normal range although there is a process anomaly) Not applicable to fault detection during operational transients Water level 10.2 m 3.8 m Abnormal condition Abnormal condition time Normal operation range 44
Methods for fault detection (II) 45 Normal operation ranges of key signals Physics-based model of the process (used to reproduce the expected behavior of the signals in normal condition) Example: Pressurizer model Signal reconstructions 80 75 70 65 0 500 1000 20 10 0 0 500 1000 45
Methods for fault detection (II) 46 Normal operation ranges of key signals Physics-based model of the process (used to reproduce the expected behavior of the signals in normal condition) Example: Signal reconstructions Real measurements Pressurizer model 80 75 70 65 0 500 1000 20 80 75 70 65 0 500 1000 20 10 10 0 0 500 1000 0 0 500 1000 Abnormal Condition 46
Methods for fault detection (II) 47 Normal operation ranges of key signals Physics-based model of the process (used to reproduce the expected behavior of the signals in normal condition) Example: Signal reconstructions Real measurements Pressurizer model 80 75 70 65 0 500 1000 20 80 75 70 65 0 500 1000 20 10 10 0 0 500 1000 0 0 500 1000 Typically not available for complex systems Long computational time Abnormal Condition 47
Data & Information for fault detection (III) 48 Example: Normal operation ranges of key signals Physics-based model of the process in normal operation Historical signal measurements in normal operation Liquid Steam Pressuretemperattemperat ure ure Spray flow Surge line flow Heaters power Level 150.2 321 362 539 244 0 7.2 Pressure 150.4 322 363 681 304 0 7.5 150.3 323 364 690 335 1244 7.7 Water level 48
Methods for fault detection (III) 49 Pressure Normal operation ranges of key signals Physics-based model of the process in normal operation Historical signal measurements in normal plant operation Water level Empirical model of the process: Auto Associative Kernel Regression Principal Component Analysis Artificial Neural Networks 49
Methods for fault detection (III) 50 Normal operation ranges of key signals Physics-based model of the process in normal operation Historical signal measurements in normal plant operation Example: Signal reconstructions Real measurements EMPIRICAL MODEL OF PLANT BEHAVIOR IN NORMAL OPERATION 80 75 70 65 0 500 1000 20 80 75 70 65 0 500 1000 20 10 10 0 0 500 1000 0 0 500 1000 Abnormal Condition 50
s 1 The fault detection approach Real measurements MODEL OF COMPONENT BEHAVIOR IN NORMAL CONDITIONS ŝ 1 ŝ 2 t 51 Signal reconstructions s 2 t Pb. 1 t COMPARISON t s 1 ŝ 1 s 2 ŝ 2 Residuals Pb. 2 t DECISION NORMAL CONDITION: No maintenance ABNORMAL CONDITION: maintenance required 51
52 Modeling the component behavior in normal conditions The Auto Associative Kernel Regression (AAKR) method 52
Auto Associative Kernel Regression (AAKR)
What is AAKR? Auto-associative model x 1 x 2 x n Auto- Associative Model ˆx 1 ˆx 2 xˆn xˆ i f x1, x2,, i 1,, n x n Empirical model built using training patterns = historical signal measurements in normal plant condition Signal x 2 obsnc obsnc x 11 x1 j x1 n obsnc X x k1 xkj xkn obsnc obsnc xn1 xnj xnn Observation x 1 54
55 How does AAKR work? Training pattern Test pattern: input Output X obsnc x obsnc 11 xk obs xn1 1 nc x x 1 j kj x Nj obs obs obs x ( x 1,, xn ) nc nc nc xˆ ( xˆ,, ˆ 1 x ) X n obs nc x obsnc 1n x kn x obsnc Nn = historical signal measurements in normal plant condition = measured signals at current time = signal reconstructions (expected values of the signals in normal condition) obs x 1 obs x 2 obs x n AAKR nc xˆ 1 nc xˆ 2 nc xˆn 55
56 How does AAKR work? Training pattern Test pattern: input Output X obsnc x obsnc 11 xk obs xn1 1 nc x x 1 j kj x Nj obs obs obs x ( x 1,, xn ) nc nc nc xˆ ( xˆ,, ˆ 1 x ) n x obsnc 1n x kn x obsnc Nn = historical signal measurements in normal plant condition = measured signals at current time = weighted sum of the training patterns: x 2 x 1 56
57 How does AAKR work? Training pattern X Test pattern: input Test pattern: output obsnc x obsnc 11 xk obs xn1 1 nc x x 1 j kj x Nj obs obs obs x ( x 1,, xn ) nc nc nc xˆ ( xˆ,, ˆ 1 x ) n x obsnc 1n x kn x obsnc Nn = historical signal measurements in normal plant condition = measured signals at current time = weighted sum of the training patterns: On all the training pattern x 2 xˆ nc j N k 1 w( k) x N k 1 w( k) obsnc kj x 1 57
58 How does AAKR work? Output nc xˆ ( xˆ 1 nc,, xˆ nc n ) = weighted sum of the training patterns: On all the training pattern xˆ nc j N k 1 w( k) x k 1 obsnc kj weights w(k) = similarity measures between (the test and the k-th training pattern): n w N w( k) d ( k ) 1 2 2h ( k) e 2 h x obs and obsnc x k 2 obs obsnc 2 with d ( k) ( x j x ) Euclidean distance between and j1 kj 2 x obs x 2 low weight obsnc x k high weight x 1 h = bandwidth parameter 58
Bandwidth parameter d=0 w=0.40/h d=h w=0.24/h d=2h w=0.05/h d=3h w=0.004/h w d h w( d 3h) 0.24 0.004 60 w 14 w 12 10 8 6 h=0.2 h=2 4 2 0-6 -4-2 0 2 4 6 d 59
Example 1 Signal values at current time: Signal reconstructions? Normal or abnormal condition? x obs obs obs ( x 1,, xn ) x 2 x 1 available historical signal measurements in normal plant condition 60
Example 1: Solution Signal values at current time: x x ( xˆ obs nc nc ˆ 1,, obs obs ( x 1,, xn nc xˆ n ) Signal reconstructions: based on the available historical signal measurements in normal plant condition ) x 2 x obs xˆ nc x 1 normal condition 61
Example 2 Signal values at current time: Signal reconstructions? Normal or abnormal condition? x obs obs obs ( x 1,, xn ) x 2 x 1 available historical signal measurements in normal plant condition 62
63 Example 2: Solution Signal values at current time: x x ( xˆ obs nc nc ˆ 1,, obs obs ( x 1,, xn nc xˆ n ) Signal reconstructions: based on the available historical signal measurements in normal plant condition ) x 2 x obs xˆ nc x 1 abnormal condition available historical signal measurements in normal plant condition 63
64 AAKR: Computational Time Computational time: No training of the model Test: computational time depends on the number of training patterns (N) and on the number of signals (n) d 2 ( k) n j1 obs ( x j x obsnc kj ) 2 64
65 AAKR Performance: Accuracy Accuracy: depends on the training set: N Accuracy x 2 x 1 65
66 AAKR Performance: Accuracy (2) Accuracy: depends on the training set: N Accuracy x 2 Few patterns and not well distributed in the training space Inaccurate reconstruction x 1 66
FAULT DETECTION IN NPP APPLICATION Reactor coolant pumps 67
Fault Detection: Application* 68 COMPONENT TO BE MONITORED Reactor Coolant Pumps of a PWR Nuclear Power Plant x4 MEASURED 48 signals Training patterns = historical signal measurements in normal plant condition measured for 1 year, every 30 seconds * Work developed with EDF-R&D 68
x(4a) residuals x(4a) residual x(4a) residuals residuals residuals residuals residuals residuals residuals residuals Results: reconstruction of three different sensor failures SENSOR: Temperature of the water flowing to the first seal of the pump in line 1: Failure 1 = measurement noise increase x(4a) 50 49 48 x test nc (4a) x test ac (4a) 1 0.5 0 x(4a) x(4a) x(4a) x(4a) x(4a) 47 50 46 490 10 20 30 40 50 60 70 80 90 100 50 48 49 47 50 50 47 48 x test nc (4a) x test ac (4a) Failure 2 = sensor offset Time 46 48 490 10 20 30 40 50 60 70 80 90 100 x test nc (4a) x test ac (4a) 49 46 470 10 20 30 40 50 60 70 80 90 100 Time 50 48 46 0 10 20 30 40 50 60 70 80 90 100 49 Time 50 47 48 49 46 0 47 50 10 20 30 40 50 60 70 80 90 100 Time 48 49 46 0 47 10 20 30 40 50 60 70 80 90 Time100 Failure 3 =sensor stuck Time 48 46 0 10 20 30 40 50 60 70 80 90 100 47 Time 50-0.5 1-1 0.50 10 20 30 40 50 60 70 80 90 100 1 0 0.5-0.51 0 Time -0.5-1 0 10 20 30 40 50 60 70 80 90 100 Time 1-1 0 0 10 20 30 40 50 60 70 80 90Time 100 0.5-0.51 0 0.5-1 -0.50 10 20 30 40 50 60 70 80 90Time 100 10-1 0.50-0.5 10 20 30 40 50 60 70 80 90Time 100 1 0-1 0.50-0.5 10 20 30 40 50 60 70 80 90Time 100 0-1 0 10 20 30 40 50 60 70 80 90 100-0.51 Time 46 490 10 20 30 40 50 60 70 80 90 100 48 Time 0.5-1 0 10 20 30 40 50 60 70 80 90 100 0 Time 47 46 0 10 20 30 40 50 60 70 80 90 100 Fault injection Time -0.5-1 0 10 20 30 40 50 60 70 80 90 100 Time 69
Results: seal deterioration detection 70 AUTO-ASSOCIATIVE MODEL OF PLANT BEHAVIOR IN NORMAL CONDITIONS ŝ 1 t COMPARISON MEASURED SIGNALS s 1 s 1 ŝ 1 (SEAL OUTCOMING FLOW) t DECISION t ABNORMAL CONDITION: seal deterioration NORMAL CONDITION ABNORMAL CONDITION 70