How Machine Learning Influences Energy Industry

Twenty years ago, if a plant shut down, maintenance was a “necessary evil” with the bottom line taking a hit from the resulting days of unforeseen downtime. Executive of energy firms were taking decisions without the power to predict. Like a seismologist tracking the next big earthquake, there was no sure way of knowing when the next shutdown would happen, causing profits to plunge. It was a huge and seemingly intractable problem for industry to address. Nothing hurts capital-intensive business like unplanned downtime. As an example, one large mid-stream oil and gas company was recently report to be losing close to $1 million for each failure of an oil well pump.

Companies have spent millions in the past trying to address the unplanned downtime issue, but until now they have only been able to address wear and age-based failures because they lacked insight into the process-induced failures that are estimated to cause more than 80% of unplanned downtime. Today, however, through advances in machine-learning and the science of maintenance, energy firms are empowered with technology and real-time operational data that can detect breakdowns before they occur. With a stable plant and active assets, business leaders can plan, increase performance of their performance of their business and raise profitability, safe in the knowledge that plant maintenance is seen as a way of delivering value to organization, and not as a cost center and burden. It’s a complete transformation, but how exactly has it come about?

Digitalization is far from new to the energy industry, after all. Asset-intensive industries have been capturing reams of data, much of it from internet-enabled sensors, but also from data historians and other information sources, since the late 1970s. That process has accelerated significantly in recent years. Energy and other companies in capital-intensive industries now have access to growing volumes of real-time data, as sensors become more pervasive and less expensive, and as advanced analytics are fed through increased connectivity. But this high-speed access to more and more data is not by itself giving decision-makers the time or the insights they need to break through operational excellence barriers.

The tipping point comes with the practical and reliable application of machine-learning. Asset performance management (APM) has always been key in this industry in keeping assets up and running, but it had previously relied on statistical projections and rule-of-thumb estimates to define likely future performance. APM is evolving fast, driven by the catalyst of low-touch machine-learning. This represents a breakthrough in automating data collection, cleaning and analysis to provide perspective maintenance, protection for equipment. The integration of the two marks a transition from estimated engineering and statistical models towards measuring patterns of asset behavior.

Deployed coherently, with appropriate automation, low-much machine-learning enables greater agility and flexibility to incorporate current, historical and projected conditions from process sensors, and mechanical and process events. Systems become more agile and are able to adapt o real data conditions -- and incorporate the nuances of asset behavior. Now previous maintenance practices can be improved to recognize issue affecting asset degradation. Operational integrity improves when organizations implement strategies to detect root causes early and avoid unplanned downtime. The latest breed of APM solutions is ready to improve reliability, lift net product output and increase profitability, marking it clear that the power to predict is driving positive change across the energy sector.