Contributed by: Arindam Pandit
I work as a valve segment engineer at Schlumberger. Schlumberger is the world’s leading oilfield services provider company. It supplies the petroleum industry with services such as formation evaluation, well testing, directional drilling, well cementing, artificial lift, well completions, etc. I work in the Valves & Measurements segment, wherein I’m entirely responsible for the end-to-end life cycle of downstream valves supplied to our customers, starting from designing to installation and after services.
Oilfield valves are rugged components with very fragile interiors in an industrial sense and need the proper care required. They can be as small as the ones used in our household taps, but the ones used in the oil and gas industry downstream can be immense in cases. The valve bore sizes range from 1 inch to about 60-inch sizes, made capable of handling pressures more than 30k psi of the good formations and crude in the basic sense. This requires proper handling, installation, maintenance, and operation of the valves. Any deformations, unfavorable conditions, or leaks can lead to catastrophic events, such as the one we witnessed in Macondo Disaster, made into the movie “Deepwater Horizon.” Valves can fail because of various reasons such as wear of seals, excessive temperatures or pressures, the inclusion of debris, excessive torque, to name a few.
The end-to-end product lifecycle management is maintained in SAP ERP systems. There are thousands of valve Part numbers in the system, starting from small components to entire assemblies. However, we don’t have a timeline set in place for the service repair and quality maintenance checks by the company. We used to work upon notifications arising from the fields.
By 2020, the Field Test Reports (FPRs) and maintenance problems issues that we were receiving for ball and butterfly valves turned out to be in large volumes from the end-users. The customer sales representatives were constantly bugging us with a lot of new leakage issues. Upon retrospection, it was analyzed that a lack of proper maintenance timeline by us was a major cause of the problems and was affecting our revenue, production hours as well as goodwill. The management came up with a team to address this issue. I was a part of this team from the designing department, along with a bunch of guys from sales, procurement, and value engineering.
I led the initiative and put forward the idea in front of my supervisors that we can use my knowledge of Data Science to proceed ahead and prepare a proper timeline in place for all the valves sold. This would help us in predicting beforehand when a valve would need periodical maintenance and if any components would need replacements. The idea was met with a positive response.
Thus, we embarked on our journey to get a scheduled maintenance timeline in place. The sales team helped obtain the historical data of valves used in oil wells for the past 15 years, including their place and time of manufacture, sales, end-users, issues, etc. We organized the data and made sure to list down the entire variables needed for the analysis starting from operating temperatures, pressures, fluid service conditions, atmospheric conditions, seat type, materials used, onshore/offshore, sour requirements, permitted leakage, tests conducted, failure reasons, service maintenance dates, etc. Each valve sold was tagged along with a time period (in months) when an FPR was raised from the customer end.
We built a predictive model using the Linear Regression tool on the data until 2017. The model presented us with a linear relation between the time period for first maintenance (in months) and the other variables listed above. We were able to identify some of the most influencing variables determining the service life of the valves, with the seal material topping the list. The model provided us with an accuracy of around 72% when tested over the 2-year time period data from 2018 to 2020, which was considered decent.
We created several reports, including various statistical and visualization tools on the work, and shared those with the external sales representatives and the product SMEs. Based on their inputs and suggestions, additional feature engineering was conducted, bumping up the accuracy of prediction to more than 80%. A tolerance of ±5 months was included in the model for prediction.
Currently, we are feeding real-time data to the dataset. The sales guys have been monitoring and using it and are providing us with constant updates on the same. The management is in talks with the product SMEs and field engineers regarding finalization implementation of the model in our current SAP system and inclusion in the product service manuals.
In the future, we hope to predict valves failure rate based on this project and intimate the end-users regarding the same. A timely maintenance service or on-field repair can prolong the life of a valve, thus having a significant effect on reducing customer complaints and generating FPRs. Afterall its goodwill of the company grows.
Personally, there couldn’t have been a better place to implement my learnings in a real-world scenario and gain valuable experience. I am eagerly looking forward to seeing how the future results hold out and work on improving the model further.
Wish to learn such concepts and implement them at work? Join Great Learning’s PGP Data Science and Business Analytics Course and upskill today!
