Off-shore--On-shore Collective Analytics & Intelligence for condition-based monitoring in drilling & operations using heterogeneous networks

The PETROMAKS project SMART-RIG performs research on intelligent data systems and wireless sensor network technology, providing new opportunities for better monitoring and maintenance of both production processes and machinery on offshore oil platforms. The impact of this project will be demonstrated in terms of improving the maintenance agility and responsiveness, increasing the operational availability, performance and energy efficiency, reducing the life cycle total ownership costs and increasing the safety. The key idea of this project is to investigate and develop a new ICT solution that combines intelligence both off-shore and on-shore using an heterogeneous network with different elements, covering several tasks: a) the coordinated acquisition of multiple sensor signals in platforms and rigs, including data pre-processing to keep the relevant information for inference, b) distributed intelligence and wireless network design to perform essential data analytics such as event prediction and detection, pattern feature extraction of process behaviour, decision-making support and continuous learning, c) Isolation and cause identification for different types of machine faults, as well as the prediction of evolution of different parts of the systems. Our solution will ensure a continuous consistency between the intelligent network off-shore and the on-shore high-level analytics (i.e. collective intelligence) running at the servers of Control Centers, providing recommendations or actions if necessary to service engineers. This project will provide also building blocks towards the generation of a common standard to integrate and process equipment-related data from different suppliers, which is strongly needed nowadays. The results will be showcased using real data from facilities provided by Lundin Norway and MHWirth.

- We have designed an online data acquisition infrastructure for real-time analysis of OG data, including data acquisition, storage, user interface and machine learning modules. - Our signal resampling method has led to a novel alternative to the PI system allowing to improve the compression ratio with a better reconstruction accuracy, with the potential to be integrated in real OG data acquisition systems - The learning of dependencies among sensors have been proven to be useful for control strategies, anomaly detection and event prediction, such as prediction of tripping events or stops in OG separators or detection of different drilling activities. - The graph filtering techniques that have been designed can remove noise simultaneously from multiple data time-series and extract spatio-temporal patterns of interest, exploiting the learned dependencies among the sensor data, offering a performance that is superior to the independent denoising of each sensor data time-series.

Recent advances in ICT and MicroElectroMechanical Systems has led to devices incorporating wireless communication, processing & storage capabilities, and diverse sensing & actuation functionalities in a single unit that is compact and autonomous. This revolution appears in the form of dense Wireless networks of sensors and actuators, with enormous potential for applications that are of great interest, including real-time data analytics for predictive condition-based monitoring (CBM) in drilling & operations. However, the optimal design of robust in-network practical algorithms and associated data analytics, remains far from mature. This proposal is motivated by the grand challenge of providing a new ICT solution for collective off-shore--on-shore intelligence for predictive CBM of drilling rigs, covering: a) the distributed acquisition of sensor signals, including data pre-processing, adaptive sampling rate optimization and collaborative calibration capabilities (Network Tier 1), b) in-network cooperative processing and distributed context-aware intelligence to perform essential data analytics tasks such as local event prediction and detection, low-level feature extraction, decision-making support and learning, c) design of semantic sensor management tools at micro-server nodes (Network Tier 2), with higher resources in terms of computation and communication capabilities, providing data aggregation across different inter-related subsystems and an intermediate medium-to-high level inference about the data collection, i.e., high-level diagnostics (fault detection, isolation and cause identification) and prognostics (fault & degradation prediction), tracking continuous consistency with the on-shore high-level analytics running at the servers (Network Tier 3) of Control Centers, providing recommendations or actions if necessary. The results will be demonstrated at facilities provided by Lundin and Frigstad, interacting with real data provided by Riglogger^TM - Aker MH.