Big data analytics, applied in the industry to leverage data collection, processing and analysis, can allow a better understanding of production system's abnormal behavior. This knowledge is essential for the adoption of a proactive maintenance approach instead of conventional time-based strategies, leading to a paradigm shift towards Condition-Based Maintenance (CBM) since decision is now based on the usage of a huge, diverse, and dynamic amounts of data as a means to optimize operational costs. This paper reports an investigation of the emerging aspects in the design and implementation of big data analytics solutions for offshore installations in order to allow predictive maintenance practices.
Condition-based maintenance focuses on performing interventions based on the actual and future states (health) of a system by monitoring the underlying deterioration processes. One of the building blocks of a CBM design and implementation is the prognostic approach/system, which aims to detect, classify and predict critical failures. Considering the massive amounts of data available from a Stationary Production Unity (SPU), the use of techniques that properly deal with such a big data scenario became essential. The use of parallel processing to ingest, transform, and analyze different kinds of data in near real-time basis allows the construction of a valuable tool for implementing CBM.
This paper presents a comparison of different approaches for RUSBoost and Random Forest (RF) classification, in constructing a prognostic system for a specific class of turbogenerator failures from a chosen Petrobras' Floating Production Storage and Offloading (FPSO). Besides the comparison of different classifiers, a contribution of this work lies on the use of data acquired not only from machine sensors (telemetry data) but also non-structured data regarding the most critical failures acquired from official reports, e.g. operator's machine event annotations. Those reported annotations were correlated to telemetry data to identify real critical failures, and simultaneously avoid false positives.