Chapter 1: Introduction
Patrick Bangert
Artificial Intelligence Team, Samsung SDSA, San Jose, CA, United States
algorithmica technologies GmbH, Küchlerstrasse 7, Bad Nauheim, Germany
Artificial Intelligence Team, Samsung SDSA, San Jose, CA, United States
algorithmica technologies GmbH, Küchlerstrasse 7, Bad Nauheim, Germany
Abstract
This book will provide an overview of the field of machine learning (ML) as applied to industrial datasets in the oil and gas industry. It will provide enough scientific knowledge for a manager of a related project to understand what to look for and how to interpret the results. While this book will not make you into a machine learner, it will provide everything needed to talk successfully with machine learners. It will also provide many useful lessons learned in the management of such projects. As we will learn, over 90% of the total effort put into these projects is not mathematical in nature and all these aspects will be covered.
Keywords
machine learning
oil and gas industry
artificial intelligence
data science
industry 4.0
1.1. Who this book is for
This book will provide an overview of the field of machine learning (ML) as applied to industrial datasets in the oil and gas industry. It will provide enough scientific knowledge for a manager of a related project to understand what to look for and how to interpret the results. While this book will not make you into a machine learner, it will provide everything needed to talk successfully with machine learners. It will also provide many useful lessons learned in the management of such projects. As we will learn, over 90% of the total effort put into these projects is not mathematical in nature and all these aspects will be covered.
An ML project consists of four major elements:
- 1. Management: Defining the task, gathering the team, obtaining the budget, assessing the business value, and coordinating the other steps in the procedure.
- 2. Modeling: Collecting data, describing the problem, doing the scientific training of a model, and assessing that the model is accurate and precise.
- 3. Deployment: Integrating the model with the other infrastructure so that it can be run continuously in real-time.
- 4. Change management: Persuading the end-users to take heed of the new system and change their behavior accordingly.
Most books on industrial data science discuss mostly the first item. Many books on ML deal only with the second item. It is however the whole process that is required to create a success story. Indeed, the fourth step of change management is frequently the critical element. This book aims to discuss all four parts.
The book addresses three main groups of readers: Oil and gas professionals, machine learners and data scientists, and the general public.
Oil and gas professionals such as C-level directors, plant managers, and process engineers will learn what ML is capable of and what benefits may be expected. You will learn what is needed to reap the rewards. This book will prepare you for a discussion with data scientists so that you know what to look for and how to judge the results.
Machine learners and data scientists will learn about the oil and gas industry and its complexities as well as the use cases that their methods can be put to in this industry. You will learn what an oil and gas professional expects to see from the technology and the final outcome. The book will put into perspective some of the issues that take center stage for data scientists, such as training time and model accuracy, and relativize these to the needs of the end user.
For the general public, this book presents an overview of the state-of-the-art in applying a hyped field like ML to an old-world industry. You will learn how both fields work and how they can work together while the industry is transitioning to a new way of supplying energy to the world.
One of the most fundamental points, to which we shall return often, is that a practical ML project requires far more than just ML. It starts with a good quality data set and some domain knowledge, and proceeds to sufficient funding, support and most critically change management. All these aspects will be treated so that you obtain a holistic 360-degree view of what a real industrial ML project looks like.
The book can be divided into two parts. The first 8 chapters discuss general issues of ML and relevant management challenges. The second half focuses on practical case studies that have been carried out in real industrial plants and report on what has been done already as well as what the field is capable of. In this context, the reader will be able to judge how much of the marketing surrounding ML is hype and how much is reality.
1.2. Preview of the content
The book begins in Chapter 2 with a presentation of data science that focuses on analyzing, cleaning, and preparing a dataset for ML. Practically speaking, this represents about 80% of the effort in any ML project if we do not count the change management in deploying a finished model.
We then proceed to an overview of the field of ML in Chapter 3. The focus will be on the central ideas of what a model is, how to make one, and how to judge if it is any good. Several types of model will be presented briefly so that one may understand some of the options and the potential uses of these models.
A review of the status of ML in oil and gas follows in Chapter 4. While we make no attempt at being complete, the chapter will cover a large array of use cases that have been investigated and provides some references for further reading. The reader will get a good idea of what is possible and what is hype.
In Chapter 5, Jim Crompton addresses how the data is obtained, transmitted, stored, and made available for analysis. These systems are complex and diverse and form the backbone of any analysis. Without proper data collection, ML is impossible, and this chapter discusses the status in the industry of how data is obtained and what data may be expected.
Management is concerned with the business case that Robert Maglalang analyzes in Chapter 6. Before doing a project, it is necessary to defend its cost and expected benefit. After a project, its benefit must be measured and monitored. ML can deliver significant benefits if done correctly and this chapter analyzes how one might do that.
ML projects must be managed by considering various factors such as domain expertise and user expectations. In a new field like ML, this often leads to shifting expectations during the project. In Chapter 7, Peter Dabrowski introduces the agile way of managing such projects that has had tremendous successes in delivering projects on time, in budget and to specifications.
Many projects get stuck in a proof-of-concept phase and do not get rolled out. This state of purgatory is presented by Geoffrey Cann in Chapter 8. It can be resolved by clearly communicating needs and expectations on both sides. If the expectation on the operator side is "to learn something," then this is also legitimate and can be factored into the project so that a roll-out is not expected on the software side.
The next several chapters discuss concrete use cases where ML has made an impact in oil and gas.
In Chapter 9, Wu Qing presents many applications that ML has been put to in China National Offshore Oil Corporation (CNOOC) and focuses primarily on exploration and refining.
Environmental pollution such as the release of NOx or SOx gasses into the atmosphere while operating a gas turbine is harmful. With ML, physical pollution sensors can be substituted by models. These are not only more reliable, but they allow model predictive control and thus are able to lower pollution. Shahid Hafeez presents this in Chapter 10.
The simple algorithm of principal component analysis finds a great use in Chapter 11 where Long Peng predicts the failure of electric submersible pumps. This addresses the most famous use case in industrial ML: predictive maintenance.
A forecasting and classification methodology for rod pumps is presented in Chapter 12 where problems can be diagnosed in advance of a failure and a maintenance measure planned and scheduled in a timely manner.
The forecasting of slugging events in gas-lift wells is presented by Peter Kronberger in Chapter 13. These events are upsetting and can be mitigated if choke valves are closed at the right time. This intricate advanced process control application is powered by a time-series forecasting model.
1.3. Oil and gas industry overview
Millions of years ago, when sea creatures and plants died and collected at the bottom of the ocean or swamps, they sometimes found themselves in oxygen starved environments and covered by sediment. Over time, the deposits above grew, the pressure rose and with it the temperature. Under these conditions, the organic material was slowly converted into a collection of hydrocarbon molecules. This is, of course, a very simple presentation of a complex process and there are other mechanisms for the creation of oil.
As the continents moved and the Earth's surface changed, the structure of these sedimentary layers was changed. This forced some of these deposits to flow elsewhere. Some locations were special in that they attracted these deposits from far and wide and in that they were covered from above by an impenetrable layer of rock. These places are the storehouses of hydrocarbons and they are called reservoirs. Finding reservoirs is a difficult task as they are often far underground beneath terrain that is hard to get to. This task is called exploration.
A reservoir is not (usually) an underground lake of oil but rather it is a porous rock the pores of which contain the oil. Depending on the structure and its history, the internal pressure of the reservoir may be high enough, that if it is punctured by drilling a well, the oil will flow out of it naturally. In other cases, the internal pressure is not enough to transport the oil to the Earth's surface and the process must be assisted by a pump. This is a process known as artificial lift.
A well may be hundreds of meters deep and only half a meter in diameter. Such a fragile structure would collapse if it were left to its own devices and so it must be lined, often while being drilled. When the well has been drilled, and oil has been found at its bottom, it must be closed off at the top and attached to a system of pipelines that will transport the oil from the well to locations of further processing. This process is known as completions.
Managing the well from the start of operations, through its lifetime that may extend over several decades, is called production. During production, the equipment at the well must be monitored and maintained with changes made occasionally depending on the changing conditions in the reservoir as its pressure decreases over time. The main source of work effort and financial expense occurs during production and is due to the maintenance that must be performed on equipment that ages and breaks over the years.
Once the fluid from the reservoir is at the surface, we discover that it is a combination of gaseous hydrocarbons (natural gas), liquid hydrocarbons (crude oil), water, and particulate matter most of which is sand. The composition of the fluid, called multiphase flow, into these four groups, and into the many different forms of hydrocarbons, is particular to the individual reservoir and well. This fluid must be separated into these four components and that is often done on site.
All these stages from exploration over drilling, completions, and production are together called upstream oil and gas. Often the term oil and gas industry implicitly refers to the upstream side. This is not the end of the journey, however. For an excellent treatment of the upstream industry, see Hyne (2012).
The useful components are transported from the wellhead to processing centers by pipelines. These are pipes that may extend over hundreds of miles of rugged terrain and can transport both oil and gas without mixing them. The pipelines may end at shipping terminals where the substances are loaded onto tankers or they may end at the refinery. The process of transporting crude oil and natural gas from the wellhead to the refinery is the midstream oil and gas industry.
As crude oil is a haphazard mix of many different hydrocarbons, it cannot be used for practical purposes as it is. It must be processed and, the various hydrocarbons must be separated into categories of similarity. Particularly long hydrocarbon chains are not practically useful, and they must be split into shorter chains. Processing crude oil into categories of hydrocarbons chains by their length is known as refining and occurs in a large processing facility called a refinery. These refineries may be located close to the reservoirs or close to the major end users of the products, depending on the cost of transportation.
Some products of a refinery are useful end products in themselves. Particularly, they are gas, gasoline, and heating oil. Gas is generally used either as a fuel for producing heat or electricity. Gasoline comes in various special forms like kerosene as a fuel for airplanes, normal gasoline as a fuel for cars, and diesel as a fuel for some cars and trucks as well as ships. Heating oil is a major fuel for producing district heat in many cold countries.
Other products of a refinery are further processed to make a wide range of products. These products are generally either especially pure hydrocarbons that are used industrially in other processes to make yet other products. Or they are modified and enriched to produce useful end products. Due to the variety of products, the facilities that do this are quite varied and each one specializes in a substance. All such plants together are known as the petrochemical industry or downstream oil and gas. What follows from there before the substance arrives in a consumer's home is the chemical industry that produces an even broader range of products.
It is often underestimated how important crude oil is as a basic source material for our modern civilization. By volume, the main use is of course as a fuel for the propulsion of cars, planes, ships and so on and as a source of heat that may be further transformed into electricity. Via the petrochemical industry however, crude oil finds its way into a vast array of goods that we use every day. It would be difficult to produce most of these goods without starting with crude oil. Here is a list of some selected few products that rely on crude oil:
- • Most plastics and all products made from plastics like bottles and toys
- • Synthetic fibers used in clothing and other fabric-based products like tents, umbrellas, curtains, carpets
- • Wheels, toys, glasses, helmets, paint
- • Shampoo and many cosmetics like lipstick, toothpaste, and shaving cream
- • Dentures, heart valves, and artificial limbs
- • Linoleum and other synthetic surfaces
- • Fertilizers and pesticides
This list reveals that we deal with oil-based products multiple times a day in our normal lives. Many of the material comforts in our lives are based on oil as an essential raw material.
1.4. Brief history of oil exploration
There are few places on Earth where hydrocarbons are available freely at the surface. One place is the Dead Sea where one can obtain bitumen, a very thick form of oil, by hand. This was the primary source of bitumen for the Egyptian civilization in antiquity where it was used as sealant, adhesive, incense, pigment, and waterproofing, among others. Already in antiquity, the bitumen was heated and combined with other substances to produce some product, an early form of refining. Thus, the use of oil stretches back to the beginnings of human civilization and perhaps even further.
Oil was a valuable substance in ancient times, virtually unobtainable in most places. Naturally, people strove to find more. As we know today, a successful search depends on knowledge of geoscience that gives insight into the evolution of the Earth's crust. It is not surprising that proper exploration could not occur until modern times.
Modern systematic exploitation of oil can be traced to August 27, 1859 when Colonel Drake successfully drilled a well striking oil in Titusville, Pennsylvania, United States. Prior to this point, the production of oil was haphazard and based more on circumstantial finds rather than a systematic premeditated search. The industry quickly developed in the United States and other countries. In about 1900, the Russian empire was the primary producer while the United States quickly caught up in the early 1900s with some notable advances in Mexico in the 1920s. The countries of central and South America as well as the Caribbean started to play a role in the 1930s.
Significant discoveries were made in the Middle East just prior to the Second World War. It was at this point that the British Admiralty considered oil as a principal strategic element to win the war as oil-based fuel allowed the Navy to stay at sea longer and move faster. The countries of the Middle East, first and foremost Iran, were largely under British influence at this time creating the problematic tension that continues to the present day: Countries that do not have (enough) oil for their strategic needs, meddle in the affairs of other countries that have more oil than they can use. A few hidden agendas and betrayals later, oil is still on center stage where world politics and warfare are concerned, particularly in the Middle East that continues to play the primary role as an oil-producing region (Frankopan, 2015).
More recently, the innovations of drilling horizontally and hydraulic fracturing (artificially breaking the porous rock that contains oil to increase the flow) allowed oil companies to access reservoirs hitherto inaccessible. These measures transformed the United States from a net importer to a net exporter of oil in 2018 with important economic and geopolitical consequences.
In modern times, the primary use of oil was and remains as a propellant for vehicles. Scientific advances, particularly in chemistry, have led to more and more sophisticated products that ultimately are based on oil as a principal raw material. The petrochemical industry started in the 1930s and made major advances during and after the Second World War. Currently, about 5% of global oil production ends up in petrochemical processes and about 40% of all chemicals are made directly from oil.
1.5 Oil and gas as limited resources
Oil and gas are finite resources that eventually will be used up. There are places on Earth where we can observe the creation of future oil deposits such as large swamplands. However, these processes work on geological timescales. It is also clear that many deposits cannot be accessed in a cost-effective manner with present technology. Exploration continues and new finds of oil fields are reported every year. There comes a point, called peak oil, when oil production reaches a maximum and thereafter declines. Despite numerous attempts to declare a time for peak oil, this point has not arrived yet.
The Club of Rome published a famous study in 1972 entitled "The Limits to Growth" that analyzed the dependency of the world upon finite resources (Meadows, Meadows, Randers, & Behrens, 1972). While the chronological predictions made in this report have turned out to be overly pessimistic, their fundamental conclusions remain valid, albeit at a future time. A much-neglected condition was made clear in the report: The predictions made were made considering the technology available at that time, that is, the prediction could be extended by technological innovation. This is, of course, exactly what happened. Humanity innovated itself a postponement of the deadline. The problem remains, however. Growth cannot occur indefinitely and especially not based on resources that are finite. The ultimate finite resource is land, but this is quickly followed by oil, considering its place in our world. There is a 30-year update to the study that makes for interesting reading (Meadows, Randers, & Meadows, 2004).
The global climate is changing at a fast rate. This fact is largely related to the burning of fossil fuels by humanity since the start of the industrial revolution. Considering the present-day problem of climate change, the finite nature of oil and gas may be a theoretical problem-it is likely that the consequences of climate change will become significant for the ordinary individual far earlier than the consequences of reduced oil availability.
Both problems, however, have the same solution: We must find a way to live without consuming as many resources. In part because they are finite and in part because their use is harmful. This new way is likely to involve two major elements: Technological innovation and life-style change.
Technological innovations will be necessary to overcome some fundamental challenges such as how wasteful the process is from the reservoir to the point where a barrel of crude is made into a useful product. As this is a scientific challenge, we must figure out how complex processes work and how we can influence them. This is the heart of data science with ML as its primary toolbox.
Life-style change will be needed from every individual on the planet in that we must use our resources with more care - essentially, we must use far fewer resources. This will involve effort on an individual level, but it will also involve a transformation of societies. Public transportation must supplant individual transportation. Consumerism must be overcome, and products must be of better quality to last longer. Society must rely more on society, rather than material objects. To transform society in a manner that most people are happy with the transition requires careful analysis. In large numbers, humans are predictable and amenable to numeric description and study. It is again the field of data science that is pivotal in enabling the transformation through insight and design of the best systems for the future.
1.6. Challenges of oil and gas
These issues provide a range of challenges to the oil and gas industry worldwide. We must continue to find and produce oil during the transition to a new social norm. The process must become more efficient and less wasteful. There are many dangers along the way that must be mitigated. This starts from injury to individual workers and goes to global threats as illustrated by the Deepwater Horizon disaster in 2010. The digital transformation is a journey that will help to solve these challenges. It promises to do this by increasing our understanding and control of the process. At the beginning are sensors and communication software. At the end are physical technologies like drones, 3D printing, autonomous vehicles, or augmented reality.
The heart of it all is ML. ML is the piece of the puzzle that converts a dataset into a formula, also called a model. Once we have a model and we are confident that the model is right, we can do many things with it.
A popular challenge these days is classifying images into various categories like bird, dog, table, house, and so on, see Fig. 1.1. Any individual human being correctly classifies about 95% of images and makes 5% errors. Until 2014, computers performed worse. Starting in 2015 however, computer models based on machine learning started to outperform humans for the first time. This is an example that is consistently repeated time and again with many different datasets and tasks. Machine learning models are now more accurate than humans, calculate their output in a fraction of a second, and can keep doing it with the same accuracy without limit.
Figure 1.1 The evolution of machine learning as compared to human performance relative to a common standard dataset.
As they require little human effort to make and maintain as well as being fast to compute, ML models are much cheaper than any alternative. These two features also enable many novel use cases and business models that simply could not be implemented until now. In relation to the oil and gas industry, there are six main areas that ML can help with, see Fig. 1.2:
- 1. Is this operation normal or not?
- 2. What will the value or operational condition be at a future time?
- 3. Which category does this pattern belong to?
- 4. Can this complex, expensive, fragile, or laboratory measurement be substituted by a calculation?
- 5. How shall we adjust set-points in real-time to keep the process stable?
- 6. How and when shall we change set-points to improve the process according to some measure of success?
Figure 1.2 The six main application areas of machine learning in oil and gas.
One of the biggest topics in industrial ML is predictive maintenance, which is the combination of the first three points in this list. Is my equipment performing well right now, how long will that remain and when it fails, what is the damage type? Once we know a damage type and a failure time in the future, we can procure spare parts in advance and schedule a maintenance measure to take place preemptively. This will prevent the actual failure and thus prevent collateral damage and spillage, which is typically 90% of the total financial cost of a failure.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.