“In our oil and gas business, we have wells that produce oil. And each of these wells has instrumentation and sensors installed that report how the well is flowing, what is the temperature and pressure at the surface, and a few other parameters. Because of a simple rule of return on investment, we may not have invested in installing sensors in the low oil-producing capacity wells,” said Anand Laxshmivarahan, Chief Digital Officer at Vedanta Resources.
Vedanta had a few wells which were producing the oil but had no sensors in them to help the company with the information. Though they produced less oil, the reservoir team still had to manage a lot of things in the well for which they needed the information which only sensors could give.
“We had two options: either to invest money and then put those sensors downhole so we can get the information needed or to use the data from the other wells. We used that data from the wells which had those sensors to build a supervised machine learning model. We used the information that we had in those wells to train a model that would predict the bottom hole flowing pressure for the rest of the wells which did not have the sensors. With this, we have applied predictive analytics and data to get visibility on how wells are flowing at the bottom,” he explained.
Vedanta worked with a niche data analytics partner to work on this project. Looking at the part of the investment, Anand feels that the investment that the company thus made in this process was minuscule compared to if they had installed sensors and instruments in the wells.
“In the aluminum business, we produce billets which are the final products that go out. Sometimes because of the operating conditions, the final product which comes out may have a crack. It’s almost like saying it is a defective product. So we thought if we could create a model which predicts the right operational conditions,” said Laxshmivarahan.
Based on all the defective products that were produced in the past and looking at the operating conditions that there were, Vedanta built a model that can predict what is the right operating condition under which the billet will not come out defect-free. It’s almost like giving the operator the information about the right operating condition in which to operate the plant to.
“Further, it was not humanly possible to see every billet and check if it is defective or not. So the company has also set up a computer vision technology. All of this makes sure that the company does not ship a defective product to the customer,” he said.
Vedanta is further planning to build the predictive models and algorithms at scale. In the last 9 months, Vedanta has brought in BCG as the group’s digital partner. Along with them, Vedanta has charted three pillars to work on for this journey. Read more about their strategy here.