Seven essential elements of production domain analytics success

Automation and machine learning (ML) are bringing new insights and intelligent actions to the energy industry in the ‘edge’ era. The emerging consensus is that ‘domain analytics is where most success is found’, and for industry-wide success it must be made scalable for production.

Seven essential elements for success in scaling production domain analytics are outlined here:

• Ask yourself: is the answer to your problem contained in your data?

Projects fail when the data only contains a fraction of the answer, before jumping into ML/AI projects it is essential to consider how much of the answer to your problem is likely to be contained in your data and if that matches your project goals.

• Develop customizable data quality tools

Most production workflows rely on time series data. However, data quality problems such as outliers, frozen, or missing data can throw off algorithms, and all data is subject to drift without expert recalibration, leading to misinterpretations in ML models. To prevent such issues efficiently and repeatedly, customizable data quality tools must be developed.

• Generalize the concept of label in your organization

Labelling gives contextual meaning to raw data. Events, the most common label in production are already being captured in multiple forms (e.g., email exchanges, spreadsheets, databases, etc.) but are largely not reported as events, limiting their use. The potential of data will be unlocked as improvements in generalizing the capture of events are made, enabling integration with business intelligence and ML.

• Label your data in context

Real-time visualizations of time series data are constantly improving, but the lack of context in labelling data, results in labels/events being captured and seen in different systems restricting the ML enabled automated workflows they could inform.

• Consider gamification to improve your data workflows

Though useful for ML and overall business intelligence taking additional time to capture increasingly refined data may seem tedious. Leveraging a sense of reward, like that generated when engaging with social networks, through systems that reward best reporters, will encourage discussion and collaboration. This will aid diagnosis of problems and bring greater value to operations.

• Use scalable model management

Models in the form of simulators (e.g., digital twins), will get out of calibration without constant maintenance, creating issues as production teams need to manage hundreds or thousands of assets simultaneously. Therefore, successful scalability requires advancement in tools to allow simple and efficient model (re)loading, time range validation or recalculations for the end user.

• Inspire from other industries for user experience

User experience is key to the success of software workflows. Learning from other industries, including social media to create a more collaborative environment with rapid contextualized information sharing, will inform and enable the workflows of tomorrow.

Building from these essential elements will enable seamless collaboration between field operations and production engineers. It is imperative that the new generation of tools is leveraged effectively to digitize expert knowledge, liberating production actors from routine tasks, and expanding their time spent on advanced production asset and field optimization.

Read the full blog post here

Other useful links:

Sensia-Digital-Solutions

DELFI Data Science

INNOVATION FACTORI