TY  - RPRT
T1  - Data-Driven Fault Detection and Isolation of Industrial Fluid Transfer Systems
A1  - Neogi, Debanjan Guha
Y1  - 2018
M1  - Masters Thesisy
T2  - TU Darmstadt
N2  - One of the key promises of Industry 4.0 is to reshape the various industrial sectors through the
integration of advanced analytics and IoT into its day to day operations. A major component
of Industry 4.0 is predictive maintenance, which allows real-time remote condition monitoring
of critical industrial equipment, leading to reduced downtimes and optimal asset management.
However, the application of analytical methods into existing industrial ecosystems is largely
a non-standard problem and varies from case to case. In this thesis we provide an end-toend
solution for fault detection and isolation (FDI) in an industrial fluid transfer system, with
the primary objective of finding leaks in the reservoirs using advanced signal processing and
machine learning techniques. Optimization remains a driving force throughout the implementation,
keeping in mind the environmental constraints and the limited hardware and sensing
capabilities of the devices involved. We extract the relevant features from the available data-set
and use it to train the predictive models. The algorithms consist of a single strong learner and
ensemble of weak learners to provide comparative analysis for determining the best possible
option. We employ Bayesian statistics to further improve the predictive outcomes. For the next
part, we identify the optimal parameters for our use case through various experiments involving
the different phases of our implementation life cycle. Finally, we design a prototype of our
solution in the industrial setup to demonstrate the feasibility of the design concept. Our results
show a prediction accuracy of over 99% for the reservoirs, while at the same time being fast,
lightweight and unobtrusive to the existing test-bed. On top of that, our overall implementation
and optimization techniques provide flexibility in terms of architectural and hardware changes,
making it robust and future-proof.
ER  -