Reactive to Predictive Operations Transformed by Data at Shell

By Amruta Kanagali


ARC Advisory Group analysts attended the OSIsoft EMEA Users Conference in Berlin, Germany in September 2016.

reactive to predictiveJohn de Koning, Technology Manager, Foundation Services, Shell Global Solutions, spoke at the conference about the company’s goal to transform its enterprise operations from reactive to predictive by treating data as an asset and deriving maximum value from the data.  For this massive project, Shell leveraged OSIsoft’s PI Collective Manager and Asset Framework (AF) to collect and structure data for further analysis. The company also carried out a proof of concept project (POC) of applying business analytics to operational data on a use case of carbon capture and storage at its facility in Canada, for which it utilized OSIsoft’s PI Integrator for Business Analytics to integrate PI System data with Microsoft Power BI.

Business Case for Predictive Operations

Shell, an innovation-driven global enterprise, invests more than $1.0 billion globally in R&D each year.  One aspect is to reactive to predictiveimprove asset performance by leveraging the huge volume of data generated in plants and achieve a gradual transition from a reactive to a “predictive” organization.

“The business needs for operations or maintenance are where and how to quickly access information about the operation,” Mr. de Koning explained.  “Is the operation on target? How does it compare to similar operations?  What is the health status of the equipment? How can downtime be prevented?” These are questions that need to be answered quickly and efficiently at the local and enterprise levels.

In the past, Shell addressed these types of questions by implementing islands of systems, including several brands of historians for data collection.  Other software applications such as ERP or LIMS were added and used on a local basis by local Shell staff, who included further functionalities and analyses needed to do their tasks.  But with the increased interest in data reactive to predictiveat the enterprise level, it became necessary to create interfaces between all these local systems.  This turned out to be a very complex undertaking.  Realizing that it was not effective for each plant to design and implement systems, which then had to be integrated, the company changed its approach about five years ago and devised a strategic response to data usage.

Shell recognized that, along with physical assets, data is a vital asset that, if managed properly, would help it achieve its goals at the plant and enterprise levels.  According to Mr. de Koning, it’s important for personnel to easily access local and global data across the enterprise, without “reinventing the wheel” each time.  Efficient data retrieval and analysis is especially critical in times of depressed oil prices.  Reducing delays in taking action can help prevent unplanned equipment downtime, and improve asset performance.  Comparing equipment across plants gives opportunities to improve overall plant performance.  The power of predictive analytics would further add to the effectiveness of responses.

The Foundation:  Data Integration and Access

The first step in the series of strategic responses by Shell was to collect operational data and make it available globally. The company standardized on the OSIsoft PI System and partnered with OSIsoft for real-time data collection.  Shell created PI System Super Collectives.  These are PI Servers holding all operational data for a line of business globally. ARC’s understanding is that these are replicated from the local PI Servers. These collectives currently process close to 7 million tags together, with an update frequency of 1 to 60 seconds.

To make it easier to access and interpret data, a uniform data model was created and implemented in Asset Framework (AF).   For each type of equipment, a standard data structure was defined, capable of modeling all possible instances of similar equipment type.  With this in place, it becomes straightforward to compare equipment performance worldwide.  Local tags are now organized per a uniform global standard and are well-understood by all users in the organization.  According to Mr. de Koning, data modeling provides a basis for smart solutions.  “If the same asset definition and data modeling are used for equipment throughout the organization, the benefits are immense,” he said.  It took several years for several teams at Shell to create the data model.  Mr. de Koning recommended designing this step carefully and executing it systematically and thoroughly to avoid inviting a serious risk of failure.

reactive to predictive

Other components found in the “Data Integration and Access” layer are event detection and advanced calculations (also using OSIsoft’s AF), and alarm management using Honeywell’s DynAMo.  Finally, OSIsoft PI Coresight and PI ProcessBook plus Telerik’s KendoUI are used for data visualization.

reactive to predictiveMr. de Koning commented that putting all data types in a data collective for time-series is not effective, which is why the data layer connects with SAP HANA for relational data. This enables real-time asset-related information to be efficiently combined with events and maintenance records from SAP to provide meaningful input to predictive analytics solutions.

Shell also uses AF to reference data in stores other than PI Systems.  One example would be for static data, such as limits, found in separate “limit stores” and LIMS.  Shell’s ODATA query tool can access the layer and find data across these sources based on the standard data model structure in AF.

To reduce development and maintenance costs, Shell leveraged off-the-shelf solutions and software to the greatest possible extent.  This required integrating the software components.  In the case of the OSIsoft PI System, integration required minimum customization. Mr. de Koning stressed that strong partnerships with suppliers, and their alignment with Shell’s vision and expected outcomes are the key success factors in implementing a roadmap.

Smart Applications for Business Value

The company built portals with advanced capabilities such as proactive monitoring, exception-based surveillance for reservoirs, and condition- based maintenance monitoring.  When adding complex calculations with MATLAB to monitor rotating equipment or fouling of heat exchangers, for example, Shell found that it could create one calculation with input from a standardized data structure for thousands of pieces of equipment of this type across many facilities.  According to Mr. de Koning, this reduced the required workload from years, to days.

Shell currently adds complex model-based analytics using MATLAB and R programming languages as the first step toward “predictive operations.”  The company aims to make use of those tools for a new, asset-oriented way of working.

Business Intelligence Tools

The most recent step Shell took in its transition to predictive operations was to test the use of Business Intelligence tools, to provide “free format data analytics and reporting” as a self-service for a broad range of users.  Shell wanted this reporting to be easy to use, including for data cleansing and preparation, performing analyses, and sharing the results. 

To test predictive operations, Shell undertook a POC applied to carbon capture and storage at its Albian Sands bitumen production operations in Canada. When upgrading oil sands by hydrogenation, CO2 is emitted as a byproduct of the hydrogen production. To reduce contribution to global warming, the captured and compressed CO2 is injected and stored in a geological formation about 2,300 meters underground.  For regulatory and environmental reasons, field instruments monitor the gas concentration at the wellhead.   Using concentration measurements and weather information, MATLAB calculations estimate whether CO2 concentrations remain within allowed limits.

reactive to predictive

Shell installed OSIsoft’s PI Integrator for Business Analytics to transfer selected data from the PI Supercollective to Microsoft’s Power BI analytical tool.  The company made, amongst others, graphs of CO2 concentrations over time.  These compared the results from different instruments and concentrations day by day. 

Shell installed the Integrator in less than four hours and could build reports within another four hours.  The results could easily be displayed in Power BI and shared in HTML5 format on any portal.  Shell appreciated the flexibility of setting refresh rates and selecting data using filters, without any coding.  As the test finished ahead of schedule, the company implemented a supply chain use case with the same success. 

Shell considers that the POC to bring data analytics to a broad audience was successful. The company aims to use the application for stock management and logistics, as well as margin visualization.  Shell will conduct further tests with other Business Analysis tools such as Element Analytics.


Shell’s strategy to transform from a reactive to a predictive mode of operation initially aims to reduce downtime and dramatically increase the efficiency in retrieving and analyzing operational and maintenance data.  Mr. de Koning’s presentation indicated, that despite the clear vision, the journey was hard work, “Shell has examined and sifted through numerous ways to achieve the desired results and selected the best option that worked for it.”  ARC believes that other companies may need to find their own optimal solution, adapted to their own business objectives and strategies.


Operational and maintenance data can serve as an enabler for organizations to respond proactively to both internal issues and external challenges. To efficiently leverage this asset in a global organization, ARC recommends that companies:

  • Define a digital roadmap consistent with the company’s strategy.  This includes a solution architecture that can accommodate a “develop once, deploy global” approach to smart add-on solutions;
  • Use commercial off-the-shelf software to the greatest possible extent
  • Make data globally accessible and interpretable based upon a uniform asset-based data model; and
  • Leverage data in smart applications to operate with foresight and deliver business benefits.

ARC Advisory Group prepared this report based on a presentation at the OSIsoft Users Conference for EMEA.  While we made every effort to accurately capture the messages delivered, the report may not completely reflect Shell’s opinion.


If you would like to buy this report or obtain information about how to become a client, please   Contact Us

Keywords: Business Analytics, Data Model, Model-based Analytics, OSIsoft, Predictive Operations, Shell, ARC Advisory Group.

Engage with ARC Advisory Group