Creating Resilient Operations Through Machine Learning

Creating Resilient Operations Through Machine Learning

If there is one true constant in oil and gas, it’s that prices never stay the same for long.

Analysts have been frantically arguing about the future of oil prices likely for as long as the industry has been around, with no signs of stopping. Does lower for longer, in fact, mean lower forever? Or does it mean an upward trajectory is just around the corner? How should oil and gas operators plan for the next big shift?

The real answer is that short of installing a crystal ball, oil and gas companies need to focus less on predicting the future of prices, and more on establishing resilient and productive operations that will remain stable regardless of what tomorrow holds. This means making sure operations are running as efficiently as possible, while minimizing extraneous and unforeseen costs and unexpected loss in potential revenue.

How Do You Ensure Stable Operations?

Maintenance is a key area where drilling operations can lose millions of dollars per year in lost revenue and extra costs, negatively impacting both the top and the bottom line. A critical asset failure on an offshore rig could run up a tab of almost $8.5 million in just one week.

Predictive maintenance, wherein asset data is used to model typical asset behavior and predict impending failures before they occur, is the best way to avoid these unnecessary costs. Research by the Electric Power Research Institute compared the annual cost of scheduled maintenance, reactive maintenance, and predictive maintenance, and their findings were striking: Scheduled maintenance costs an average of $24 per horsepower per year. Reactive maintenance is $17 per horsepower annually, though that’s before considering the additional costs that asset failures may incur, such as safety hazards of operational damage. By contrast, predictive maintenance generally costs oil and gas operators no more than $9 per horsepower each year.

The other crucial way for oil and gas companies to future-proof their operations is with process optimization. Research by McKinsey & Company has uncovered that on average, offshore platforms only realize 77 percent of their fill production potential, and the industry loses about $200 billion annually to operational inefficiencies.

As important as they are, though, both predictive maintenance and process optimization are difficult to achieve without machine learning. Predictive maintenance has substantial barriers to implementation. The asset behavior models it uses are difficult and time-consuming to create, requiring hard-to-come-by data science talent. The amount of labor involved also means that scaling predictive maintenance across an entire operation, and the vast number of individual assets involved, is often unfeasible. And these models require constant upkeep and tuning for even the slightest change in asset conditions, and struggle to capture edge cases that may occur under unusual or extreme operating conditions.

Process optimization is equally challenging. Oil and gas operations are inherently complex, involving nuanced interactions between thousands of variables. Discovering the source of inefficiencies, and deciding which controls should be adjusted to maximize production, is nearly impossible in the midst of so much statistical noise, even for experienced subject matter experts.

Automated Model Building

While humans are unable to deal with the massive amounts of data, analysis, and upkeep required for predictive maintenance and process optimization, machines can fill in the gaps. In particular, automated model building (AMB) has the potential to be a massive boon for the oil and gas industry. AMB solutions are able to create, deploy, and maintain machine learning models across an entire organization, even in the hands of users without any data science expertise. An AMB platform can ingest sensor data and automatically build a model capable of predicting asset behavior or flagging inefficiencies far more accurately than manual models can, and can accomplish this feat in far less time.

Case study: Improving processes by identifying downhole drill state

In one case study, a major oil and gas company was attempting to refine the operation of a critical subterranean drill by using machine learning techniques to infer its current operating state at any given time.

In partnership with SparkCognition, an AI solutions provider, the company made use of sensor and operating data from the drill head, including time-series electric drilling recorder data from the drill’s operation. An AMB platform made use of this data in a classification approach to discover and label seventeen drill states.

The AMB platform was then able to develop a deep learning model capable of perfectly differentiating between different drill operating modes. This model has allowed the company to greatly improve production, as it updates operators on the performance of the drill and rig, enabling them to set and update KPIs in real time. By adjusting their approach on the fly, the company has been able to better maximize their efficiency.Creating Resilient Operations Through Machine LearningNatural Language Processing

The other machine learning technique that will be key to oil and gas operations looking to thrive in the years ahead is natural language processing (NLP). NLP transforms unstructured natural language content into structured data, which can then be used for process automation, decision support and analytics, and predictive modeling when paired with AMB software.

In the case of oil and gas, NLP can ingest maintenance logs and user manuals, and use this information to remove bottlenecks in organizational workflows, quickly retrieve optimal repair solutions for maintenance issues, and preserve, codify, and continuously contribute and improve the tribal knowledge from and for the subject matter experts.

Case study: Identifying non-productive time and invisible lost time on oil rigs

An E&P operator was struggling to identify and reduce non-productive time and invisible lost time, but the work of categorizing and analyzing rig activities required prohibitive amounts of time and labor—the equivalent of one full-time job for the categorization itself, plus the work required by QA teams to check over and validate the categorization. To make this process financially feasible, the operator needed a new approach.

The operator partnered with SparkCognition, and made use of the NLP solution DeepNLP™ to automatically analyze rig activity logs, categorizing activities for increased insight into rig work time and presenting that information to human users.

In the end, the project was able to automate the full job of categorizing rig activity. The accuracy it achieved was on par with manually categorized data that had been through at least two rounds of human QA. Using the information from DeepNLP, the E&P operator has been able to better pinpoint invisible lost time and non-productive time, as well as their causes, and will subsequently be able to maximize production efficiency in entirely new ways.

It may not be possible to predict what the days and years ahead will hold for the oil and gas industry. But we can be fairly confident that it will continue to require resilient operations. We know that the keys to that resiliency already exist, are fully accessible, and will return huge benefits to the operator willing to make the investment. Leave the predictions to the machines, and you can instead focus on ensuring your operations are the best they can be.

Related posts

Subscribe to OILMAN TodayDelivered to your inbox

Subscribe to OILMAN Today, a biweekly newsletter delivered to your inbox covering oil and gas business news, current events and industry information you need to know about.