Reverse Engineering Approach for System Condition Monitoring under Big Data and Advanced Analytics
1. Reverse Engineering Approach for System
Condition Monitoring
under Big Data and Advanced Analytics
Lokukaluge Prasad Perera
MAINTENANCE ANALYTICS SUMMIT 2018
Stockholm
2. Industrial Digitalization
" Poor data quality costs…
the US economy around US$ 3.1 trillions per year"*
"1 in 3 Business Leaders don't trust…
the information they use to make decisions "*
"27% of the respondents in a survey…
were unsure of how much of their data sets are
inaccurate"*
*IBM, The four V's of Big Data, URL:http://www.ibmbigdatahub.com/infographic/four-vs-big-data, 2017.
3. Ship Engine Data
– Data from a bulk carrier with ship
length: 225 (m) and beam: 32.29 (m)
– Engine speed and power
– Combined kernel density estimation
– Engine data are clustered around
three distributions
– Three engine modes in this vessel
– Data sets can often be clustered in a
high dimensional space
– Those clusters relate to vessel
navigation and ship system
operational conditions
– That introduce the discreteness (i.e.
digital-ness) into the models
4. Data Clustering
– Three data clusters, i.e. system states, with the respective mean vectors
– Three-dimensional vector space with the right-hand coordinate system
– Each data cluster consists of local navigational and operational information of the
vessel and ship systems
– The structure of each data cluster is denoted by several vectors: singular vectors
(i.e. associated with the respective singular values)
5. Digital Models
– Each cluster is a linear model, i.e. Piecewise linearization.
– The best approximation of a nonlinear function with several system states.
– Systems can jump from one state to another state in a data space
– Some data clusters may relate to data anomalies or system abnormal events
7. Singular Values & Vectors
– The structure of each data cluster is denoted by several vectors: singular
vectors (i.e. associated with the respective singular values)
– Singular values and vectors represent the building blocks of electrical and
mechanical systems
– That represent systems behavior
8. Reverse Engineering Approach
– From components to sys. of systems
– Various model levels
– High dimensional data driven models
– From big data to low level models
9. Advanced Data Analytics
– Descriptive analytics identifies various data anomalies
– Diagnostic analytics recovers/removes such data anomalies
– Predictive analytics forecasts vessel and ship system behavior
– Visual analytics visualizes the same information
– The information creates Advanced Knowledge and that will lead to Industrial
Intelligence
– Both advanced knowledge and industrial intelligence support Decision Analytics.
– Decision analytics consists of appropriate Key Performance Indicators (i.e. KPIs)
10. Descriptive & Diagnostic Analytics
Data Anomaly Detection and Recovery Procedure
– Digital models interact with Descriptive and Diagnostic analytics to improve the data
quality
– Data Anomaly Filter 1: Missing data points and preliminary data anomalies (i.e. Min-
Max values)
– Data Anomaly Filter 2: Additional data anomalies (i.e. the outliers of digital models)
11. Data Anomaly Filter 1
– Data Anomaly Filter 1: Missing data points (NaNs) and preliminary data anomalies
(i.e. beyond Min-Max values)
– Data anomalies send to separate groups where the data anomalies against known
and unknown sensor and DAQ faults and system abnormalities are compared
12. Data Anomaly Filter 2
– Data anomaly filter 2: Additional data anomalies (i.e. the outliers of digital models)
– Data anomalies send to separate groups, where that will compare with known and
unknown sensor and DAQ faults and system abnormalities
14. Data Recovery Filter
– Data sets from anomaly group 1 and 2 transfer through the data recovery filter and
digital models
– A considerable amount of data anomalies can be recovered by this step
15. Visual Analytics
– Digital models should be visualized to extract relevant parameter relationships
– The covariance values of the data sets are represented by singular vectors
– Each singular vector represents the correlation information among the respective
parameters
16. Visual Analytics
– Singular Vectors in a high dimensional space should be represented,
appropriately
– The vectors are represented by circles with colored circles
– Relative correlation should also be defined, adequately
17. Visual Analytics
– Multiple high dimensional vector space is presented
– The top singular vector is presented in the outer circle.
– The bottom singular vector is presented in the inner circle.
20. Predictive Analytics
– Digital models are connected in parallel
– Observers introduce some model flexibility
– A global nonlinear model can be developed
21. Predictive Analytics
– Posterior probability evaluation decides which model (i.e. data cluster) to excite
during this process
– Predictive analytics to observe future vessel and ship system behavior
22. Predictive Analytics
– Predicted vessel navigation and ship system behavior supports visual analytics
– Visual analytics facilitates decision analytics
– Appropriate KPIs in decision analytics should be investigated
23. Decision Analytics
– The KPIs will be the structure of the data sets
– The structure variations relate vessel and ship system energy efficiency
and system reliability considerations, i.e. system condition monitoring
24. Conclusions
– Novel mathematical framework to support industrial digitization of shipping
is presented: from Industrial IoT to Predictive Analytics.
– The proposed data analytics can…
• self-learn (i.e. the data structure can learn itself)
• self-clean (i.e. data anomalies can be detected, isolated and recovered by the data
structure)
• self-compress and expend (i.e. the respective parameters in the data sets can be reduced
and expanded by considering the same data structure)
• self-visualize (i.e. the respective data structures can be visualized to extract relevant
information)
– Developed from the respective data sets toward system components: A
reverse engineering approach
– Intelligent Analytics for industrial applications, that also provides important
solutions to the big data challenges