Written by Tarlan Mammadov, IM Handover Coordinator at Chevron’s TCO Future Growth Project, and Principal at DeepKnowledge Ltd.
Over the last two decades, the Oil & Gas industry has been facing a substantial increase in the volume, sources, and diversity of information. This trend is driven by the rising sophistication and technical complexity of projects which pursue new frontiers in ever harsher environments. Ultra-deepwater developments, remote locations, and complex, enhanced oil recovery initiatives are just a few examples.
The application of machine learning (ML) solutions in Oil & Gas asset development projects can provide substantial benefits in quality, cost, and schedule. ML solutions are already in successful use within the industry outside of asset development projects, as well as in other high-tech and information rich sectors such as pharmaceuticals, medical research, and finance. The industry will gain benefits from the application of both supervised and unsupervised machine learning algorithms.
The mathematical and statistical principles that underpin ML algorithms are already in widespread use in the execution of engineering and design processes across Oil & Gas projects. Therefore, it is logical to expect that the information created and disseminated during these projects will have similar scientific signatures and obey statistical patterns – patterns that can easily be analysed and exploited using machine learning toolkits.
Co-evolution of humans and machines in the workplace
Machine learning solutions offer potential to optimise the value of information by unlocking hidden insights from within the data and streamlining processes that would otherwise be unfeasible with conventional computing paradigms.
The overall objective should be to bring collective intelligence frameworks into the industry, where human experts on Oil & Gas projects and machine learning systems are engaged in collaborative working for co-evolution. For example, humans with the assistance of ML grasp and digest the critical information more readily and rapidly, thereby delivering superior performance.
The goal here is not to replicate human brains or replace human thinking with machines. Rather, in the collective intelligence setting, human experts and ML systems will collaborate to produce better results, each bringing different strengths to the partnership. People will provide judgment, intuition, decision making, empathy, a moral compass and human creativity. ML systems will be more rational, unbiased, and analytical, with the possession of encyclopaedic memories and tremendous computational abilities.
This kind of relationship takes the human-computer interface to a new level, one in which machines become collaborative assistants to human experts in the workplace. In this framework, human-machine interaction would occur through a personalised single interface using natural language exchanges in the form of audio, video and text messages – a radical departure from coding instructions in computer languages or switching between apps to make data inputs and queries. Machine learning will embed an extra layer of intelligence to transform information into an active state. The links between these pieces, from various sources and applications, are dynamically established and presented to the user through personalised interfaces.
The trials ahead for machine learning
At the beginning of 21st century, one of the key organisational issues facing the industry was the challenge of managing four generations (Traditionalists, Baby Boomers, Generation Xers, Generation Ys) in the workplace. This challenge is now even bigger with a fifth generation (Generation Z) and intelligent machines (digital workforce) added to the same equation. Achieving superior performance with five generations of human expertise and intelligent machines all at the same time – and in the same workplace – will require a novel architecture for collaboration and knowledge sharing. It will be essential for academia and industry to collaborate in the creation of transformational organisational architectures for the new era of human/machine partnership.
On the knowledge management side as a first strategic step, the Oil & Gas Industry should utilise machine learning to explore and study information semantics and taxonomy of their capital projects’ domains. The objective will be to define fundamental building blocks of the industry’s digital information. Each fundamental building block of information – let’s call it a ‘data-quanta’ – should encapsulate the semantic and interaction characteristics.
To express this another way, it’s about putting on a thin layer of machine intelligence wrapper (e.g. mathematical function) around each data-quanta that defines the semantics and taxonomy of the Oil & Gas industry. As an analogy, consider atoms in the periodic table of elements. The characteristics of each element are an inseparable part of their existence. Using the power of mathematics behind machine learning to mimic this design for information in the Oil & Gas industry will bring the benefits of consistency and scalability that we see in nature. The outcome will be a ‘periodic table of fundamental elements’ of Oil & Gas industry information. This design will serve as an input into the next levels of ML application stacks for the industry.
New architectures in Oil & Gas asset development
Oil & Gas asset development projects evolve through several stages that can span several years. Machine learning solutions will add value at each step of the stage-gate model of a project’s execution. The following diagram describes the solution architecture that spans the full lifecycle of asset development projects.
The solution architecture requires cohesive integration, sequencing and synchronisation of inputs, outcomes, and preservation of memory between a number of ML models spanning the lifecycle of a project and beyond, delivering seamless functionality and interaction with human experts.
While proven methods and algorithms are in place to address the individual components of this solution stack, there are still a number of active research areas to be explored further. These include the optimisation of memory management and the passing of output from one model as an input to another. Oil & Gas industry participation and support for this research is encouraged to ensure that undergoing initiatives are geared toward meeting the objectives of integrated ML solution delivery for the industry.
The solutions that machine learning offers to the Oil & Gas industry will be scalable and adaptable from one project to another without the need for additional coding. This is because, during the model training stage, ML systems learn, gain experience, and adapt based on interactions with human experts and information. This contrasts with traditional software where logic and instruction are programmed and set to the specific requirements of each task.
The transition of machine learning systems from one Oil & Gas project to another will resemble the assignment of the person from one discipline or project to another; the individual will get on-the-job training for the new role. Likewise, ML systems will need to be fed with the information sets of the new project to retrain the model. Reprogramming or reconstructing the model will not be required.
This being said, collective intelligence frameworks for the collaboration of human experts and intelligent machines do present several organisational and regulatory challenges for the industry. Examples of such challenges are: regulation of ownership and responsibility in critical decision-making matters (related to health & safety and technical integrity); development of effective training and competency transfer programmes to prepare human experts for a new working environment; roles and responsibilities delineation for humans and digital workforce (intelligent machines); and re-defining the principles of team dynamics for collective intelligence environments.
How university innovation can drive the revolution
‘Deep Symbolic Reinforcement Learning’ is a novel architecture developed by Imperial College London. This approach combines the strength of neural network learning with advantages of classical symbolic AI, which is well suited as an unsupervised learning algorithm to explore the fundamental compositional structures of industry information. The partnership between industry and academic institutions is encouraged to explore further the potential for utilisation of Deep Symbolic Reinforcement Learning methods in Oil & Gas applications.
Another machine learning innovation, developed by researchers at the Massachusetts Institute of Technology, aims to combat the failure of ML systems to predict or suggest algorithms for a new data set based on experiences with previous data sets. Published recently on IN-PART as a call for industry collaborators, this approach entails a system that can run multiple machine learning algorithms across many machines on the cloud to find predictive models of optimal efficiency.
In addition to the benefits from the university-developed technologies above, embracing the pathway of machine learning solutions have a potential to bring other wider benefits for the Oil & Gas industry. Examples of these include adding a new dimension of innovation into the industry, attracting a new generation of highly talented data science graduates into the sector and supporting the promotion of science and engineering education and careers.
The author of this article will be presenting a webinar on machine learning, exploring the applications for the Oil & Gas industry in greater detail. US/Canadian listeners can join him on the 15th of February 2017 at 2pm US-CT, and those in the UK can participate in a second session on the 16th February 2017 at 9 am GMT.
Tarlan Mammadov – email@example.com – @DeepKnowledgeUK
IN-PART Publishing Ltd 2017 – All rights reserved unless otherwise stated
Follow us on LinkedIn: