CASTRO, P. F.; http://lattes.cnpq.br/7998874471032724; CASTRO, Patrício Fernandes de.
Resumo:
This work proposes a diagnostic analysis of the operational data of a turbogenerator
mineral lubricating oil system, using a fuzzy inference system (FIS). The study uses real
operational data collected from supervisory monitoring sensors in four turbogenerators
installed in a Floating Production, Storage and Offloading Unit (FPSO) over a threeyear
operational period, resulting in a dataset composed of 40,456,663 input patterns.
The failure modes were established through expert knowledge, using the Failure Mode,
Effect, and Criticality Analysis (FMECA) documentation as a basis. Initially, the universe
variables of the model were constructed using the calibration range of the sensors, and
then, fuzzy trapezoidal membership functions were formulated based on the operational
limits of each measured parameter. The fault diagnosis model is based on a fuzzy inference
system that employs predefined rules derived from expert knowledge, which encapsulates
specific fault typologies of the mineral lubricating oil system of turbogenerators, using
FMECA as a basis for defining the failure modes and affected parts of this system. The
FIS employed as an Artificial Intelligence (AI) tool demonstrates effectiveness in fault
diagnosis, with an overall assessment of the system performance that produces satisfactory
results, presenting a true positive rate of 98.35% for fault classification, together with a
true negative rate of 99.99% for the classification of the normal operating condition of the
system. These results highlight the viability of the FIS model for fault diagnosis of the
mineral lubricating oil system of turbogenerators, thus showing its potential to improve
the operational reliability and maintenance efficiency of these equipments. The model
can automate the fault recording process, providing information on fault occurrences and
contributing to their recording and diagnosis.