SCADA Data Interpretation improves Wind Farm Maintenance

Transcription

1 SCADA Data Interpretation improves Wind Farm Maintenance Professor Kesheng Wang Knowledhe Discovery Laboratory Department of Production and Quality Engineering Norwegian University of Science and Technology EERA DeepWind 215 1

2 Outlines Introduction Predictive Maintenance Framework of WINDSENSE Project SCADA Data Based CMS Case study Conclusions KDL, IPK NTNU 2

3 Introduction Renewable energy sources are playing an important role in the global energy mix, as a means of reducing the impact of energy production on climate change. Wind energy is most developed renewable energy techniques. The management of wind farms is challenging because it involves several difficult tasks, such as wind forecasting and the operations and maintenance of turbines. The maintenance of wind turbines has received attention in recent years due to its impact on the cost of generating power from wind. The main tendency of maintenance policy is changing from Preventive Maintenance (PM) and Corrective Maintenance (CM), to Predictive Maintenance (PdM). KDL, IPK NTNU 3

4 Classification of Maintenance Policy NTNU KDL, IPK NTNU 4

5 Fault Diagnosis and Prognosis Systems on Wind Turbines Major Failures on Wind Turbines: Mechanical component Failure percentage Main Gearbox 32 % Generator 23 % Main Bearing 11 % Rotor Blades* <1% *Blade resonances lead to fatigue failures KDL, IPK NTNU 5

6 Framework for WINDSENSE - (Add-on instrumentation system for wind trubines) NTNU Degradation Process Wind Turbines Maintenance Management System Maintenance Scheduling / Maintenance Optimization Bee Colony Algorithms (BCA) Ant Colony Optimization (ACO) Particle Swarm Optimization (PSO) Key Performance Indicator (KPI) KPI Leading Fault Prognosis Auto-regressive Moving Averaging (ARMA) Fuzzy Logic Prediction ANN Prediction Offshore Wind Turbine Gentic Algorithms (GA) Meta-Heuristic approaches KPI Logging Match Matrix Prediction Fault Diagnosis Support Support Machine (SVM) Onshore Wind Turbine Wireless Data Collection Networks Data Acquisition Fiber Bragg Grating Sensors Signal Pre-process Denosing Compression Extract Weak Signal Feature Extraction Time Domain Time-Frequency Domain Frequency Domain (FFT, DFT) Data Mining (Decision Tree & Association rules) Artificial Neural Network (SOM & SBP) Statistical Maching Collapsed Wind Turbine Acoustic Emission(AE)Sensors Ultrasonic Sensors + AE Vibration sensors Filter Amplification Wavelet Domain (WT, WPT) Principal Component Analysis (PCA) KDL, IPK NTNU 6

7 SCADA Data Based CMS for WT SCADA System: First Generation SCADA Architecture Second Generation SCADA Architecture Third Generation SCADA System Typical SCADA System KDL, IPK NTNU 7

8 SCADA Data Based CMS for WT Sensor 1 RAW DATA mounted on device Micoprocessor SCADA DATA 1 mins average data DATA MINING PAST/PRESENT DATA Sensor 2 Micoprocessor remote location RTU and/or PLC WAN SCADA master Centralised SCADA System Fault diagnosis Fault prognosis Algorithms and/or models PRESENT DATA Sensor N Micoprocessor HMI Operator Workstation Corroborate Maintenance Strategy Raw Data and SCADA Data (proposed frame work) KDL, IPK NTNU 8

9 SCADA Dataset Description Wind parameters, such as wind speed and wind direction; Performance parameters, such as power output, rotor speed, and blade pitch angle; Vibration parameters, such as tower acceleration and drive train acceleration; and Temperature parameters, such as bearing temperature and gearbox temperature. Examples Active power output (1 min max/min/average) Anemometer-measured wind speed (1 min max/min/average) Turbine speed (1 min max/min/average) Nacelle temperature (1 min max/min/average) Turbine rear bearing temperature (1 min max/min/average) Turbine rear bearing vibration (1 min RMS max/min/average) Turbine front bearing temperature (1 min max/min/average) Turbine front vibration (1 min RMS max/min/average) WORLD CLASS - through people, technology and dedication Page 9

10 ANN-based Modeling of SCADA Parameter Normal Behavior Procedure of Fault Detection based on SCADA Data KDL, IPK NTNU 1

11 Case Study Parameter Selection Indicator: Temperature Possible parameters influence or relate to indicator: Rear bearing temperature (t-1) Active power output (t) Nacelle temperature (t) Turbine speed (t) Cooling fan status Rear Bearing Input and Outputs of ANN Model Model Output Input Rear Bearing Temperature Rear bearing temperature (t-1) Active power output (t) Nacelle temperature (t) Turbine speed (t) WORLD CLASS - through people, technology and dedication Page 11

12 ANN Model Training Turbine Rear Bearing Temperature Temperature (Celsius Degree) (a) Turbine Speed Turbine Speed (RPM) Temperature (Celsius Degree) (b) Nacelle Temperature (c) Active Power Active Power (KW)) (d) Time (yyyy-mm-dd) Rear Bearing Temperature Model Training Data KDL, IPK NTNU 12

13 ANN Model Testing Temperature (Celsius Degree) Turbine Rear Bearing Temperature (t-1) (a) Test Data: to Turbine Speed 45 Real Temperature and Estimated Temperature EstimatedTemp Turbine Speed (RPM) BearTemp 3 Active Power (KW)) Temperature (Celsius Degree) (b) Nacelle Temperature (c) Active Power (a) Difference Between Actual and Estimated Temperature (b) Time (yyyy-mm-dd) Rear Bearing Model Output in Normal Condition (d) Time (yyyy-mm-dd) Rear Bearing Model Testing Input Data WORLD CLASS - through people, technology and dedication Page 13

14 Detection of Rear Bearing Fault 1: The first important deviation from the model estimates occurred from the start of October EstimatedTemp BearTemp Actual and Estimated Temperature 4 3: the turbine was stopped because of overheating. The operator try to solve the problem two times in point 3 and 4 but not successful (a) Difference Between Actual and Estimated Temperature 15 5: the turbine was completely stopped because of the overheating months early warning 1 days close alarm (b) Time (yyyy-mm-dd) Fault Detection Results of Rear Bearing WORLD CLASS - through people, technology and dedication Page 14

15 Discussion Whether the model established from the SCADA data of one turbine, can be applied in fault detection to another turbines? Turbine Bearing Temperature (t-1) Actual and Estimated Temerature EstimatedTemp BearTemp (a) 35 Turbine Speed (b) Nacelle Temperature Difference between Actual and Estimated Temerature (c) Active Power (d) Time (yyyy-mm-dd) Rear Bearing Model Testing Input Data of New Turbine Yes, with the same type of turbines WORLD CLASS - through people, technology and dedication Page 15

16 Conclusions (1) The accuracy of the diagnosis model depends upon the input data. Thus a careful selection of variables and the quality of the data (free from noise) are the prime factors affecting accuracy. Hence adequate pre-processing models are desirable. Though the comparison results of various models is mentioned but no one clear cut perfect modeling technique could emerge. So this area needs further explorations. Hence a lot is to be done in this area, in order to obtain a generic model for CMS using SCADA data. The most of the research has been carried out using low frequency SCADA data (1 minutes average) and for a better prediction higher quality data is required (high frequency, noise free and long duration). This is another point which needs to be explored, i.e. what frequency SCADA data gives optimal results in case of WTs. It is proposed to diagnose and prognosticate with both conventional data and SCADA data and with the comparison of the two results. The suitable maintenance strategy could then be worked out. KDL, IPK NTNU 16

17 Conclusions (2) Furthermore the relationships between faults have not been explored. It has been demonstrated that with combination of appropriate data mining algorithms and computational intelligence concepts the accuracy and robustness of model is enhanced. Most of the researchers have pinpointed that the data sharing by the wind engineering industry was major hindrance in this type of research. Hence there is a strong need for a pooling of common data base for researchers across the globe. Based upon the CMS of WT s using SCADA data appropriate maintenance strategy could be worked in order to achieve the goal of PdM. KDL, IPK NTNU 17

18 NTNU KDL, IPK NTNU 18