Are Machine Learning Benefits Worth Cyber-Security Risks at Chevron?
Machine Learning can provide immense value in plant maintenance and turnarounds but can a traditionally conservative industry lead in this space?
Chevron is a multinational, fully integrated energy company which primarily explores for, produces, refines, and transports oil and natural gas. Chevron, as most of the oil and gas industry, is hoping to identify a competitive advantage by using machine learning in its production facilities for a variety of goals including predictive analytics on equipment maintenance and to make faster decisions with less human interaction while also increasing safety [1]. Given the costs and risks of being shut down by faulty equipment or shutting down to preemptively fix equipment, as well as the tightening of margins due to lower oil prices and increased competition, machine learning which can accurately determine when to perform maintenance or turnarounds will provide tremendous value to Chevron’s operations.
Chevron faces two key challenges in deploying machine learning, 1) determining what data matters for decision-making and 2) physically and cost-effectively obtaining that relevant data. Most equipment when first installed was fitted with sensors to monitor and automate the process but did not have sensors to retrieve data that could be used for maintenance. Historically maintenance is performed on a scheduled basis per the OEM’s (original equipment manufacturer) recommendations which are typically conservative and not specific to an end user’s situation. Thanks to IIoT (Industrial Internet of Things), the cost of deploying sensors, especially on existing equipment, is getting significantly more affordable, allowing Chevron to acquire and store more performance related data [1]. Once that data is being acquired, it can be analyzed to drive decision-making. Chevron is currently working on getting as much data as possible to begin testing and developing various algorithms. It also has partnered with Microsoft to standardize on the Azure cloud globally so that Chevron can leverage its scale and operations across various geographies to advance its analytical capabilities [2]. Using one shared data platform will allow Chevron to share data and learnings and to more quickly develop algorithms.
In the medium-term, Chevron is acquiring data and deploying sensors but is also looking more closely at the business case. Current applications, such as heat exchanger health monitoring, have clear value-add but to maximize the opportunity and ensure the best decisions are being made, scale is critical [3]. Chevron is working to identify the equipment that most critically affects revenue so that it can be fitted with new sensors by 2024 in a truly business driven prioritization [1]. The goal is to service equipment exactly when needed, no earlier or later. Long-term, Chevron plans to move from analyzing individual pieces of equipment to using machine learning to look at equipment life more broadly at the plant and business unit level [1].
Chevron has focused on the cloud and sensor infrastructure but should also be looking more closely at the algorithms being used to analyze the data. Most of the analysis is via basic visualization or software-as-a-service model, whereas Chevron should develop this internally or acquire strategic partners to develop and maintain the competitive advantage. Chevron also needs to determine how much of this is additional information being provided to engineers and operations for higher quality decision-making or if it plans to use this data and analysis to increase automation. As automation has increased in the industry, so have efficiency and safety but the industry has historically been a laggard in this space given the health and safety risks [3]. If Chevron wants to truly lead in this space, it must be deliberate about which data it acquires. Chevron must also be patient as the machine learning will get better with more data. It would be easy to begin deploying this widely to accumulate any and all data which would strengthen the machine learning, but given the work that will be required to maintain this infrastructure and new organizational capabilities, an intentional deployment is critical.
Given the criticality of this equipment, data quality and cyber security are of paramount importance. IIoT sensors have different standards and are low-cost compared to existing industrial sensors. As Chevron expands the use of IIoT sensors and incorporates them into decision-making, ensuring the data is accurate and has not been manipulated will be a significant challenge. Should Chevron provide strict global quality and security standards or does it want to maintain itself agile and low-cost in this space? As hackers look to target energy infrastructure, Chevron must be sure the benefits of machine learning are worth the additional risk.
Word Count: 729 words
[1] Sara Castellanos, Wall Street Journal. “Chevron Launching Predictive Maintenance to Oil Fields, Refineries”. https://blogs.wsj.com/cio/2018/09/05/chevron-launching-predictive-maintenance-to-oil-fields-refineries/?ns=prod/accounts-wsj. September 5, 2018. Accessed November 2018.
[2] PEDC News. “Chevron Partners with Microsoft to Fueld Digital Transformation from the Reservoir to the Retail Pump”. http://www.pedc.ir/news-media/news/id/231/chevron-partners-with-microsoft-to-fuel-digital-transformation-from-the-reservoir-to-the-retail-pump. September 12, 2018. Access November 2018.
[3] Dan Hebert and Alex Misiti, IEEE GlobalSpec. “The Growing Role of Artificial Intelligence in Oil and Gas”. https://insights.globalspec.com/article/2772/the-growing-role-of-artificial-intelligence-in-oil-and-gas. June 9, 2016. Accessed November 2018.
[4] Annop Srivastava, Digitalist Magazine. “Artifical Intelligence: The Future of Oil and Gas”. https://www.digitalistmag.com/digital-supply-networks/2017/08/07/artificial-intelligence-future-of-oil-gas-05259467. August 7, 2017. Accessed November 2018.
I really enjoyed this post. As economically valuable as machine learning might be for a company like Chevron, I think your concerns about national security are warranted. I wonder, could Chevron use machine learning to improve safety? You mention that it has the potential to improve safety, but also acknowledge that Chevron and other companies in the industry tend to be slow adopters of new technology given safety concerns. I wonder if machine learning could help to predict hazardous situations? Would Chevron and other major oil and gas companies ever consider sharing data to avoid safety incidents in the future? Naturally these firms are concerned about their employees’ well-being, but should also be incentivized financially to minimize accidents that would result in lawsuits or fines.
I think this post brings up a real concern that managers face day to day when implementing or extending legacy systems. One of many interesting applications of blockchain is how disaggregation of information into small enough pieces where any one piece is not important enough to pose a security risk, but where a secure ledger can provide all of the information needed will be hugely valuable.
The two key challenges Chevron is facing regarding machine learning – 1) determining what data matters for decision-making and 2) physically and cost-effectively obtaining that relevant data – are actually challenges that virtually all companies entering the data analytics universe are facing. The same applies for the cyber-security risk. However, as we move to a more and more data driven business world each day, I believe companies have to focus on being ahead of its competitors to gain competitive advantage – or at least not to lose it – and using machine learning technology to do predictive maintenance can be a great source of competitive advantage for Chevron, so I believe the benefits here do out-weight the risks, specially when you consider the risk of falling behind its competitors.