Process data analysis: Using a Bayesian Network approach to model processes in the Marine Contracting practice

Process data analysis: Using a Bayesian Network approach to model processes in the Marine Contracting practice

Lievens, R.A.

Jonkman, S.N. (mentor)
Van Thiel de Vries, J.S.M. (mentor)
Minns, T. (mentor)
Jäger, W.S. (mentor)

Civil Engineering and Geosciences

Hydraulic Engineering

Coastal Engineering

2014-12-01

Marine contractors deal with processes that are understood qualitatively but are hard to quantify. The amount of data available for these processes is ever growing, and so too the intrinsic value that lies within this data. Several data driven model approaches can be used to analyse this process data, one being a Bayesian Network (BN) approach. A BN is a statistical tool that has been used in previous research for predicting processes in the area of Marine Contracting. The BN approach is further explored in this thesis as a way to analyse process data. The objective of the research was to determine how a data driven model approach like the BN approach can be used to effectively analyse process data that drive everyday Marine Contracting practices. To this end, first a theoretical study was performed. Then the BN method is applied to three processes that are dealt with on a regular basis in the marine contracting practice. The first process concerns the wear of cutter suction dredger tools. The pickpoints that are used for cutting soil wear down depending on various parameters. Being able to predict this wear process is of high importance for any soil cutting project. The BN proved to have some predictive skill though the errors were quite large. This research it could not conclude if the predictive skill is an improvement on the currently used methods for geology- or wear prediction. The second process is the cliff erosion at the island of Lolland (Denmark). As part of a large tunnel construction project an erosion cliff is to be constructed with the purpose of supplying adjacent coast with a certain amount of sediment. The erosion cliff will be constructed in two phases. Using a data driven model approach, the first phase can be used as a pilot in which various construction configurations are monitored. This process data can then be used to feed a data driven model, like a BN, to optimise the construction of the second phase. The third process is the wave overtopping of coastal structures. A database with quite a lot of well-defined overtopping cases is readily available for feeding a data driven model approach. Several other calculation methods, based on the same data, are readily available as well. Due to time constraints, the actual validation of an overtopping BN and the comparison with other overtopping methods is has not been completed for this research. Concluding, the benefit of using a data modelling approach like the BN approach for the Marine Contracting industry lies mostly in the collection, storage and management of data. Well-defined instances (sets of process parameter values related to one single case) needed for feeding a BN is the most vital aspect of its benefit. These instances give a good handle on large quantities of available data. A BN’s main strength is that its structure in combination with the properly managed data is capable of holistically capturing highly complex processes.

Bayesian Network
process data analysis
data driven model
marine contracting
big data
cutting tool wear
overtopping
cliff erosion

http://resolver.tudelft.nl/uuid:b1ba7b2a-c9e9-47f7-b289-973d32206a1b

Student theses

master thesis

(c) 2014 Lievens, R.A.