Introduction
At ICE Process Management, we recently encountered a fascinating challenge that required a blend of cutting-edge technologies and creative problem-solving. Our client faced a critical situation: they needed to report the annual gas consumption, but due to a migration project they had lost some crucial data. With time ticking away, we embarked on a mission to estimate gas usage by leveraging the remaining information that was available to our team. We will dive into this success story and explore how we eventually cracked the code.
The Scenario
Our journey begins with a seemingly straightforward migration project. A boiler house was transitioning to DeltaV version 14 with DeltaV Live. It was both unexpected and unfortunate, that during this process some module tag renaming led to the loss of data. In fact, this minor migration hiccup cost the client approximately six weeks of vital data that proved impossible to recover. As the reporting period drew near, the pressure mounted—the client had to report gas consumption figures to state regulatory authorities.
The Data Dilemma
Despite consulting every expert in our network, including Emerson Corporate and the local Impact Partner’s Senior Engineering staff, there was no direct way to retrieve the missing data. Time was of the essence, and we needed a solution. Here is the challenge we faced:
- Data Set Overview: Our dataset included various time-series data points: gas flow, air flow, exhaust flows, temperatures, pressures, and valve/damper positions.
- Critical Gaps: The missing data fell into specific time windows:
- January to October: The boiler house was inactive during these months, rendering the data irrelevant for our estimation.
- November 1 to December 15: Gas-flow data was lost in this period when there were boiler start-up activities.
- December 15 to March 1: We possessed a complete dataset for this period.
The Solution: Chemical Engineering Meets Artificial Intelligence
Our challenge resembled an undergraduate senior thesis problem. Given a complex data set loaded with physical and chemistry relationships, we need to develop a model and then make predictions based on that model. Unlike the typical thesis problem though, we were not confined to classical process modelling techniques. The days of resolving Eigen values and linear systems was too far in the past for efficient recall (in other words, I haven’t done that since I was, myself, and undergrad!)
In real life, here is how we tackled it:
- Data Aggregation: We used ICE proprietary scripts to extract the data needed for loading into the Azure Machine Learning engine.
- Machine Learning: Using the complete data set available from Dec 15 to Mar 1, we were able to train the model to predict gas flow. Further, we were able to test the model on independent data and prove that it was over 95% accurate.
- Data Imputation: We leveraged the model we had built to estimate missing gas-flow values during the critical weeks. By analyzing the available data, we filled in the gaps intelligently.
ICE’s Novel Innovation
This project showcases 2 relatively novel innovations employed by ICE.
- DeltaV’s Continuous Process Historian and the DeltaV Excel Addin and VBA Object Library can be used for much more than the surface level visualization. Interaction with underlying databases and data services are exposed for developers’ use in the Development Environment. ICE leveraged the object library in this project and others, making the DeltaV Continuous Historian a data source for other process analytics.
- By accessing the databases using the VBA Object Library, the data is made available for analysis using Microsoft Azure’s Machine Learning Studio. The Machine Learning tool set was able to accomplish in about 2 days, what would have otherwise taken at least a dedicated week of modelling and analysis.
Results and Impact
What was shocking was the ease with which we were able to develop the model on the Azure Machine Learning platform from the DeltaV History.
Within a week, we delivered an accurate estimate of the annual gas consumption. Our client met regulatory requirements, and the success story spread across the industry. By combining existing tools in novel ways, we transformed tough task into a triumph.
Conclusion
At ICE Process Management, we thrive on such challenges. Our ability to blend technology, expertise, and creativity allowed us to crack the gas consumption mystery. As we continue to push boundaries, we are reminded that innovation knows no bounds—whether it is in a boiler house or an R&D laboratory, we are always using our collective experience to deliver success.
To explore how we can leverage our skills and tools for your business needs, contact ICE Process Management today. You can also visit iceprocessmanagement.com to learn more about our integrations and how they can benefit your operations.
