Oil and Gas industry: Will Machine Learning algorithms be helpful?

Calvin and Hobbs, a famous comic strip, describes ‘transformation’ accurately – ‘Day by day nothing seems to change. But pretty soon… everything is different’. The quote is perfectly applicable for the transformation occurred in the oil and gas industry since the 1980s. In the 1980s, the oil and gas industry underwent computerization and several production processes were digitized. The trend continued in 90’s as well but then, the focus of digitization stayed on tapping into a reservoir’s production potential. Now, much has changed since.

At present, a single drilling rig at an oilfield generates an enormous amount of data every day. But the companies are making few data-driven decisions. Partly because the raw data is nothing but a heap of numbers and needs to be analyzed to derive meaningful insights from it. Machine Learning (ML) algorithms have capabilities to sift through data and draw inferences. An estimate suggests the oil/gas industry can save $50 Billion by implementing artificial intelligence solutions, including ML. Business leaders are looking for information regarding implementation of ML and tasks it can solve.

An enormous amount of data, generated by single drilling rig, offers a good foundation for crating reliably machine learning prediction model.

ML is useful in a wide variety of tasks in the oil and gas industry. Some of the commonly utilized applications are:

Drilling: Drilling industry is extremely technology dependent. However, the process complexity makes it still difficult reach high level of process optimization. Uncertainties regarding conditions, tools response, pressure and temperatures, material degradation result in high extra expenses. Artificial intelligence and machine learning approach offer the best way of improving the drilling planning, diagnosing, monitoring, predicting, processes optimization at a minimal cost.

Machine learning are widely used in the reservoir modelling to do seismic patterns recognition, permeability and porosity prediction, etc. through complex analysis of raw data. Based on collected historical data machine learning algorithms allow calculate relationships between rock characteristics and different drill bits performance, predict pressure parameters, optimize trajectory. Almost every step of drilling operation can leverage on AI and ML: starting from well planning to the real-time drilling process optimization and dangerous pattern recognition.

Anomaly detection: Anomaly detection is one of the most useful and significant applications of ML in the oil and gas industry. Based on correlations between valid readings, ML algorithms identify patterns and predict failures in the near term predict. In modern off-shore oil platforms, all equipment includes sensors which are remotely monitored. On average, single facility logs several millions measurements per day that after preprocessing and cleansing offer a good foundation for crating reliably machine learning prediction model. Anomaly detection allows recognizing defects of industrial components, such as turbomachines and motors as early as possible to prevent further losses. Machine learning techniques allow building advanced monitoring systems that perform time series analysis to reveal alarming patters. ML algorithms facilitate continuous monitoring, detect anomalies, alert humans in charge, and in some cases even stop the production to prevent emergencies. This helps in avoiding situations where physical integrity of the plant itself may be in a danger. Hence, a course correction needs to be made promptly.

Anomaly detection system for early recognition of the industrial components defects.

Predictive maintenance: In addition to anomaly detection, machine learning is widely used as an effective tool for predictive maintenance.  ML algorithms indicate times for planned maintenance activities, which prevents break-down of the system and optimize the periodic maintenance operation schedule. In terms of algorithms, predictive maintenance is considered as either a classification problem that allows predicting probability of failure in next future, or a regression problem that allows predicting how much time the turbomachine or another component is going to operate before failure.

Artificial Neural Networks (ANN), fuzzy logic, Support Vector Machine (SVM) and Genetic Algorithms (GA) are particularly useful ML applications in the oil and gas industry. For instance, ANN is a compilation of processing units that effectively mimic human problem-solving attributes such as logical or analytical techniques. ANN receives input from various sensors and has a ‘weight’ assigned to the input. As the ANN learns from its environment, the neural network adjusts its weights to yield desired output. This process is often iterative. Once implemented, ANN is initially used on training datasets to identify patterns in the dataset and generate results. These results help managers to make decisions to optimize the production.

ML applications have tremendous potential in the oil and gas industry. A recent survey, conducted in 2017, noted that along with robotics, artificial intelligence is one of the fastest areas of growth over next 3-5 years exhibiting over 7% increase in investments from the oil/gas companies. About 36% of the oil/gas companies have already invested in initiatives related to big data analytics. Company executives, who are transforming their company’s production and profits, believe that it’s time to adopt ML algorithms and data analytics in the workflow.

About 36% of the oil/gas companies have already invested in initiatives related to machine learning.