Tibor Komróczki and Károly Ott presented MOL’s DROP (Danube refinery online) program, comprising OSIsoft PI Asset Framework (AF) and a data science stack built around Hadoop and Cloudera. MOL’s team of process information and automation specialists operate with IT in a minimum, supportive role. DROP reflects the need for an operations technology (OT) data infrastructure for rapid development of scalable applications to reinforce the use of data and analytics-based decision making. The aim is to be more confident in refinery decision making by ‘capitalizing on data science to statistically predict productivity’. The system is said to increase productivity and efficiency through best practices for data harmonization and provide a ‘deeper understanding of technological processes’.
A PI AF architecture is the backbone of digital transformation and advanced analytics in MOL Downstream. Data from the PI Integrator connects into a Kafka/Spark data preparation environment. RStudio also ran. Conditioned streaming data then passes off premises into the Azure cloud for real time analysis. Tools of the trade in the cloud include the Kafka event hub, Stream Analytics, ElasticSearch, Kibana Grafana and (for visualization) Power BI. The system is to integrate with the NICE inventory management system and Opralog (in 2020).
Lorenzo Lancia and Gianmarco Rossi presented ‘EDEA’, ENI’s digital energy analytics solution, an analytics dashboard that leverages machine learning models to monitor and forecast the energy efficiency of an upstream production facility. EDEA sets out to help technicians detect anomalies and suggest corrective actions. The system embeds a PI Data Archive, PI AF and a big data infrastructure built around bespoke Python programs and PI Vision. The forecasting model leverages a gradient boosting regression algorithm to predicts a CO2 emission index KPI for energy intensive equipment over the next 3 hours. The computation takes account of operational parameters, seasonal features and ‘exogenous’ constraints like temperature or humidity. Site operators receive a notification when real time data diverges from predicted values, indicating an anomalous situation. The dashboard is then used to drill down to pinpoint the root cause of the anomaly.
Data science development in ENI uses open source tools from the Python environment including Anaconda, Jupyter and Spark, leveraging ENI’s EDOF digital oilfield platform. EDOF’s standardized architecture provides ‘zero configuration’ access to the PI data archive for secure, authenticated access to time series data and PI functionality. EDEA ingests only ‘relevant’ time series data, there is no need to perform a ‘utopic’ ingestion of all PI data. After training, a serialized Pickle object is exported for deployment. Other tools used include Cloudera and Qlik. The authors report that to date, some 15 energy efficiency actions have resulted from EDEA monitoring, leading to a ‘significant reduction in CO2 emissions from a giant oil field’.
Read MOL’s, ENI’s and other OSIsoft Gothenburg presentations here.
© Oil IT Journal - all rights reserved.