full scale data handling in shipping: a big data solution
TRANSCRIPT
Norsk Marinteknisk Forskningsinstitutt
Lokukaluge Prasad PereraNorwegian Marine Technology Research Institute (MARINTEK), Trondheim, Norway.
March 2016.
Trondheim, Norway.
Full Scale Data Handling in Shipping: A Big Data Solution.
•Introduction
•Objectives
•Data Analytics & Sensors
•Energy Flow Path: Marine Engine Centered Approach
•Data Flow Path: Big Data Challenges− Sensor Fault Identification
− Data Classification
− Data Compression
− Data Expansion
− Data Integrity
− Data Regression
•Conclusion & Future Activities
Outline
Data Analytics
Data Management
Introduction
•Big Data Solutions play an important role in Future Research and Industrial Applications.
•Strategic Priority Area for the MARINTEK.
•Research and Industrial Applications:− Data Management: Appropriate actions to develop a bunch of data in a
structured collection.
− Data Analytics: The science of examining these data with the purpose of drawing meanings about the information.
•The size of these data sets may not make a big difference in these applications.
•The outcome of the Big Data, the meaning, is the most important aspect of these research and industrial applications.
•Many Fundamental Challenges.
Objectives
•To address the Fundamental Challenges in Big Data Applications in Shipping.− Large scale Data Sources => Data management
− Sensor Related Issues
− Quality of the data
− Data communication
− Data Interpretation => Data Analytics− Energy Efficiency
− Reliability
" The data has a structure and
the structure has a meaning"
Data Analytics & Sensors•The main focus point
•Empirical/Stochastic Models− Various Empirical/Stochastic Models have been developed in shipping.
− Some challenges in handling Big Data.
•Machine Intelligence− Machine Intelligence (MI) can play an important role in the outcome of Big
Data applications.
− MI Techniques are extensively implemented on current Big Data applications.
− These tools and techniques and their applicability in shipping should be investigated.
•Knowledge on the Vessel:− Ship Dynamics/Hydrodynamics
− Automation and Navigation Systems
− Localized Models in Ship Performance Monitoring
Energy Flow Path
•The possible situations of energy conservation:− Marine power plant.
− Engine propeller interaction.
− Ship resistance.
Marine Engine Centered Approach
Localized Models in Ship Performance Monitoring
Engine Propeller Combinator Diagram
Engine Propeller Combinator Diagram
Engine Propeller Combinator Diagram
Possible Region of Engine-Propeller Operations
Basis for Localized Models in Ship Performance Monitoring
Vessel Information
•The respective data set of ship performance and navigation information is collected from:
•Bulk carrier with following particulars: − ship length: 225 (m),
− beam: 32.29 (m),
− Gross tonnage: 38.889 (tons),
− deadweight at max draft: 72.562 (tons).
− Powered by 2 stroke ME with maximum continuous rating (MCR) of 7564 (kW) at the shaft rotational speed of 105 (rpm).
− Fixed pitch propeller diameter 6.20 (m) with 4 blades
Statistical distributions of Engine Speed, Power & Fuel Consumption
Statistical distributions of Engine Speed, Power & Fuel Consumption
Engine Operating Regions: Engine Power vs. Shaft Speed
Engine Operating Region vs. Rel. Wind Speed, Avg. Draft, Trim and STW
Engine Propeller Combinator Diagram
Operating Patterns
Engine propeller combinator diagram with STW
High to Low STW
Data Flow Path
Data Flow Path
Data Classification
•Engine Centered Data Flow Path
•Localized Models in Ship Performance Monitoring
•Algorithm− Multivariate Gaussian distribution
− Gaussian Mixture Models (GMMs)
− Expectation Maximization (EM) algorithm
[Source: Matlab.com]
Data Classification: Engine Propeller Combinator Diagram
Localized Models
Principal
Component
Analysis
(PCA)
Model 3
Sensor Fault Detection
Considering
a Two Sensor Situation
Fault Level 1
Considering 10 Parameter Situation
Parameter Mini. Max.
1. Avg. draft (m) 0 15
2. STW (Knots) 3 20
3. ME power (kW) 1000 8000
4. Shaft speed (rpm) 20 120
5. ME fuel cons. (Tons/day) 1 40
6. SOG (Knots) 0 20
7. Trim (m) -2 6
8. Rel. wind speed (m/s) 0 25
9. Rel. wind direction (deg) 2 360
10. Aux. fuel cons. (Tons/day) 0 8
Fault Level 1
Considering a Two Sensor Situation
e3
e1
Fault Level 2Principal Component Analysis (PCA)
Data Standardization.
Mean = 0
Variance = 1
Least Principal Components
Projected Data into PCs
e1, e2, e3, & e4
Data Distributions in PC Axes
Least Principal Components
Real-time FaultDetection
Real-time FaultDetection
Data Flow Path
Autoencoder : Deep Learning Approach
•Encoder Side: Data Compression
•Communication Network
•Decoder Side: Data Expansion
•Top Principal Components
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Autoencoder : A Deep Learning Approach
Singular Values Data Compression Information
•Top 10, 9, 8 and 7 principal components can preserve 100 %, 99.92%, 99.48%, 97.86% 94.03% of the actual ship performance and navigation information.
•The respective 99% and 95% lines are also presented in the same figure.
•Top 7 PCs Selected
•10 Parameters => 7 Parameters
•Preserve approximately 94% of the actual information.
PCA of Ship Performance and Navigation Information.
PCs for Ship Performance Evaluation ?
PCA of Ship Performance and Navigation Information.
Actual and estimated parameters of ship performance and navigation information.
Actual and estimated parameters of ship performance and navigation information.
Data can be Recovered by Regression or Smoothing
Data Flow Path
Data can be Recovered by
Regression or Smoothing
AIS Data
Data Analysis
Relative Wind Profile of a Ship
Relative Wind Profile with STW and SOG
Gaussian Type Distributions
Wind Sensor Faults
One Sided Relative Wind Profile (Cleaned Data)
One Sided Relative Wind Profile with STW (> 3 Knots)
One Sided Relative Wind Profile (Further Cleaned Data)
Ship Speeds: STW, SOG and STW-SOG
Ship Speed Power Profile with Relative Wind Speed|STW –SOG| < 5.5 (Knots)
Speed Power Profile with Rel. Wind Speed and Angle
STW-SOG vs. Rel. Wind Speed and Direction
Modified Speed Power Profile
SWT vs. SOGUnique to the Vessel ?
Ship Performance Data
Conclusion & Future Activities
•Some advanced tools are developed in this stage.•Still a Logway to go..
− Sensor Fault Identification− Data Classification− Data Compression− Data Expansion− Data Integrity− Data Regression
•Models should be further developed.•High sampling rate data •Further collaboration with appropriate partners.•Research projects.
Thank You
Questions ?This work has been conducted under the project of "SFI Smart Maritime - Norwegian Centre for improved energy-efficiency and reduced emissions from the maritime sector" that is partly funded by the Research Council of Norway.
smartmaritime.no
Publications and high resolution color images: http://bit.do/perera.