Smart Factory Expo Industry 4.0 과 스마트발전소 장범찬선임연구원 서울대학교기계항공공학부 Laboratory for System Health and Risk Management (SHRM) (shrm.snu.ac.kr) OnePredict Inc. (onepredict.com) 2017-04-04 Seoul National University - 1 -
OnePredict Overview 2017-04-04 Seoul National University - 2 -
- 3 -
- 4 -
- 5 -
- 6 -
Industry 4.0 Briefing 2017-04-04 Seoul National University - 7 -
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2017-04-04 Seoul National University - 8-2012 2013 2014 Background (%) 10 경제성장률 (GDP 증가율 ) 전망치 5 0 2014 년이후 총매출액 6 조 4,966 억원 순이익 11% 총매출액 15 조 3,053 억원 순이익 3% WARTSILA ( 핀란드 ) 대우조선해양 ( 대한민국 )
- 9 - Background 저출산, 고령화로인한생산가능인구감소 2012 년 경제활동 1 인당 2.0 명 2030 년이후 경제활동 1 인당 2.5 명
- 10 - Industry 4.0 = Hope or Real? Machines Electricity Computer IoT, AI, & CPS (power) (mass production) (automation) (smart factory) 고부가가치산업화
Values Created by Industry 4.0 - Contemporary automation, big data analytics and manufacturing technologies - Technologies and concepts of value chain organization which draws together Cyber-Physical Systems, the Internet of Things and the Internet of Services 고객주문맞춤형유연생산 CPS 기반최적품질, 물류, 안전관리 설비보전최적화 - Condition-based maintenance (CBM) 2017-04-04 Seoul National University - 11 -
- 12 - O&M Now & Future in Power Industry O&M 비즈니스트렌드 예측기반정비 상태기반정비 Future 예방정비 Global Leading Companies 사후정비 Most Small Business Middle Companies The cost is main problem according to the survey in Nov. 2014 사후정비예방정비상태기반정비예측기반정비 Management method Rely on laborer s experience Online data acquisition Online system monitoring & control Automatic management of equipment & system Technology level Repair and management by field laborer Anomaly detection, Life estimation based on failure history Failure classification, Failure cause analysis Remaining Useful Life prediction Enterprise Level Most Small Business Middle Companies Global Leading Companies Future
Industry 4.0 Power of 1% Power 가스, 증기, 원자력발전소를위한디지털파워플랜트소프트웨어개발 가동정지시간 5%, 운영및유지보수비용 25% 감소 발전소가동능력에대한정보를적시에제공하여매출증대효과창출 Brilliant Factory( 생각하는공장 ) 데이터기반공정및작업실시간최적화를통한생산성및효율극대화 공정주기 30%, 가동정지시간 (Down time) 20%, 비용 15% 감소 Wind PowerUp 데이터수집 / 분석을통한풍력터빈운전최적화, 전력생산량증대 운영최적화를통해발전량 20% 이상향상 Power of 1% 항공 전기, 발전 오일 & 가스 철도 연료효율 1% 증가연간 2-3 조원절감 연료효율 1% 증가연간 4-5 조원절감 가동시간 1% 증가연간 5-7 조원절감 속도 1.5km/h 증가연간 1-2 조원절감 2017-04-04 Seoul National University - 13 -
- 14 - Power Industry Vision by Industry 4.0 빅데이터통합취득및관리 x 불시자율범용설비고장으로인한예측진단전력생산솔루션장애 데이터개별취득및관리 빅데이터 ( 상시, 주기적취득 ) 터빈, 발전기 : 진동, 온도, 압력등변압기, 리액터 : 진동, DGA 등 x Management M O Operation X Knowledge K B Business 사이버전문인력전문가노령화및 AI기반및암묵적형식적지식화 낮은고부가가치의부가가치로신개념인한비즈니스비즈니스도입을통한계한계극복
- 15 - Industry 4.0 Introduction to Prognostics and Health Management (PHM)
- 16 - Unscheduled Failures Outbreak of fire in Dangjin power plant(1000mw), Dec 05, 2015 Due to defect in Last Stage Blade (LSB) of turbine Consequence: property & operating loss of hundreds of millions of dollars worth Explosion during operation in Samcheonpo power plant (560MW), Dec, 2008 Due to water absorption of stator winding insulation Consequence: Collapse of whole power generator Accident in Adani group's coal-fired power plant (4600 MW) in Mundra, Gujarat, India, Apr 20, 2016 Due to boiler pipe burst, hot water spilt Consequence: 21 workers suffered burn injuries
- 17 - PHM Overview Multivariate Sensor Signals Health Health Health PHM BLACK BOX Data Classification Prediction (Reasoning) (Diagnostics) (Prognostics) Health Features Classification Index 90% Real Time Health State Diag-/Prognosis Feature 1 Feature 2 Feature 3 Normal Caution Danger Industrial IoT + Big Data Analytics + Physics (Domain Knowledge) + Artificial Intelligence
- 18 - PHM Overview Multivariate Sensor Signals Timedomain Health Health Health PHM BLACK BOX DataFrequency- domain Classification Prediction? (Reasoning) (Diagnostics) (Prognostics) Health Features Life Feature Extraction Health Classification Health Prediction Classification Index t 90% Real Time Health State Diag-/Prognosis Feature 1 Feature 2 Feature 3 Normal Caution Danger Industrial IoT + Big Data Analytics + Physics (Domain Knowledge) + Artificial Intelligence
- 19 - 발전소주요구성설비 발전기 Windings Rotor Bearing 로터의회전에너지를전기에너지로변환 축계진동데이터, 권선절연상태측정 터빈로터 Blade Shaft Bearing 변압기 Bushing Core Windings 유체의에너지를기계적동력으로변환 축계또는베어링진동데이터수집 인가전압의크기를목적에맞게변환 유중가스데이터, 외함진동데이터수집
발전소설비별데이터취득 Rotor Bearing Misalign, Rubbing, etc. GAP sensors Realtime Turbine Blade Corrosion, creep fatigue Visual inspection; Portable indentation Overhaul Winding Insulation Water absorption Capacitance measure Overhaul Rotor Bearing Misalign, Rubbing, etc. GAP sensors Realtime Insulation Oil Partial discharge Dissolved Gas Analysis Periodic Core & Winding Joint loosening Acc. sensors Periodic 2017-04-04 Seoul National University - 20 -
- 21 - Industry 4.0 1. Steam Turbine PHM Episodes in Power 2. Power Plants Generator 3. Power Transformer 4. Shunt Reactor 1. Steam Turbine 2. Power Generator 3. Power Transformer 4. Shunt Reactor
- 22 - Steam Turbine 1. Steam Turbine 2. Power Generator 3. Power Transformer 4. Shunt Reactor
- 23 - Sensing from Power Plant Steam Turbine Rotor - 5 tilting pad, 4 tilting pad, elliptical bearings HIP LP - A LP - B GEN EXC Journal Bearing - Optimized settings for accurate diagnosis & prognosis Amplifier DAQ Module CH 1 Diagnosis Module Proximity sensor CH 2 CH 3 Gap sensor (Eddy current) Proximitor Result Bearing housing Clearance (with oil) Rotating shaft
Reasoning & Diagnostics Reasoning for 500MW Power Plant RK4 Data Power Plant Training Testing Normal + 1 Unbalance O 2 O Rubbing O O 3 Misalign O n/a 4 Oil Whirl O n/a RK4 Predicted Class Original Class 3 Health State Health State Accuracy = 92.2% Scale Up 2 Power Plant 1 0 20 40 60 80 100 Samples Power Plant A 2017-04-04 Power Plant B Seoul National University - 24 -
- 25 - Generator 1. Steam Turbine 2. Power Generator 3. Power Transformer 4. Shunt Reactor
- 26 - Fault Diagnostics of Power Generator Health diagnostics of power generator stator windings against water absorption - Target Component: Power generator stator winding - Failure Mode: Water absorption of winding insulation Structure of power generator Boom Catastrophic failure
Fault Diagnostics of Power Generator 2017-04-04 Seoul National University - 27 -
- 28 - Fault Prognostics of Power Generator Health Prognostics of power generator stator windings against water absorption - Target Component: Power generator stator winding - Result: RUL prediction of power generator stator windings Copper strand Fick s second law Degradation mechanism m = concentration of water in insulation D = diffusion coefficient X = position in sample x Analytical solution: m( x, t) m0erfc( ) 2 Dt
- 29 - Industry 4.0 1. Steam Turbine 2. Power Generator Frontier O&M in Power 3. Power Plants Transformer 4. Shunt Reactor
- 30 - Deep Learning Approach to Steam Turbine 1. Steam Turbine 2. Power Generator 3. Power Transformer 4. Shunt Reactor
- 31 - Deep Learning in PHM Data acquisition Feature extraction Feature selection Classification Feature extraction and selection are the key for the success. DEEP LEARNING 다량의데이터의추상화과정을통해주요특성인자를뽑아내는자율기계학습 (machine learning) 알고리즘 Random Image Input layer x 1 x 2 x N Hidden Layer 1 Pixels Edges Shape s Hidden Layer N
- 32 - Case Study: Deep Learning in Rotor PHM Vibration Image Generation RK4 test-bed (Labeled) Power plant (Labeled) Power plant (Unlabeled) Vibration images using ODR High-level Feature Abstraction Greedy Layer-wise unsupervised pretraining: restricted Boltzmann machine in deep belief network (DBN) v h 1 h 2 h N 1 h N Health Reasoning Classification: multi-layer perceptron (MLP) Clustering: self-organizing map Clustered_con1 Clustered_con2 Clustered_con3 Clustered_con4 Clustered_con5 Clustered_con6 Clustered_con7 Clustered_con8 Normal Rubbing Misalignment Oil whirl U_Anomaly 1 U_Anomaly 2 U_Anomaly 3 Known Unknown
Case Study: Deep Learning in Rotor PHM Time & frequency features* + MLP HOG + RBM + MLP The deep learning approach with vibration image processing outperformed the conventional-feature-based approach. * max, mean, RMS, skewness, kurtosis, crest Factor, shape Factor, impulse factor, frequency center, RMS frequency, root variance frequency, (0~0.39x) / 1x, (0.4x~0.49x) / 1x, 0.5x / 1x, (0.51x~0.99x) / 1x, 2x /1x, (3x~5x) / 1x, (3x,5x,7x,9x) / 1x, (1x~10x) / 1x 2017-04-04 Seoul National University - 33 -
Clustered Labels Case Study: Deep Learning in Rotor PHM Showing image patterns (Semi-)Labeled data Test-bed Normal Rubbing Misalign. Oil whirl Normal Misalign. Oil whirl Unlabeled data Power Plant Unknown conditions Test-bed (Known) Power Plant (Known, not validated) Power Plant (Unknown) 9 Unknown (oil whirl) 8 7 6 5 4 Unknown (rubbing) Normal Rubbing Misalign. Oil whirl Unknown 3 2 Unknown (misalign.) 1 Unknown (normal) 2017-04-04 Seoul National University - 34 -
Industry 4.0 Future Power Plant 2017-04-04 Seoul National University - 35 -
- 36 - Power Industry Vision by Industry 4.0 빅데이터통합취득및관리 x 불시자율범용설비고장으로인한예측진단전력생산솔루션장애 데이터개별취득및관리 빅데이터 ( 상시, 주기적취득 ) 터빈, 발전기 : 진동, 온도, 압력등변압기, 리액터 : 진동, DGA 등 x Management M O Operation X Knowledge K B Business 사이버전문인력전문가노후화및 AI기반및암묵적형식적지식화 낮은고부가가치의부가가치로신개념인한비즈니스비즈니스도입을통한계한계극복
- 37 - Solutions by Industry 4.0 (1) 대용량데이터의발생 다양한전력설비에의적용 Boiler Transformer Coal Feeder Stack 센서기술고도화 데이터관리기술향상 데이터중요성인식 Turbine Generator 인공지능기술 전문가지식기반기술
- 38 - Solutions by Industry 4.0 (2) Goal 운영및유지보수비용최소화 Corrective Maintenance Preventive Maintenance Condition-Based Maintenance 1980 s 1990 s 2020~ 기대효과 불시고장방지 시스템노화에대한대비 상태기반감시 (Condition-based maintenance) 운영및유지보수비용최소화 자율상태진단 범용상태진단
- 39 - Solutions by Industry 4.0 (3) 빅데이터관리 상태진단및예측솔루션 실시간빅데이터처리프레임워크 E 빅데이터 전력 운전 다계층빅데이터관리 통계적패턴분석 자율특성인자추출기 진단 정비 빅데이터송수신 환경 안전 범용상태진단기 반지도학습상태예측기 건전성등급기반예측정비 데이터압축 데이터정제 데이터분산저장 빅데이터특화처리 현장전문가 시각화진단결과아키텍쳐예측결과
- 40 - THANK YOU FOR LISTENING 1. Steam Turbine 2. Power Generator 3. Power Transformer 4. Shunt Reactor ANY QESTION?