Can Artificial Intelligence Using Machine Learning Provide Real Value in Industry?

By Ebele Maduekwe

Category:
ARC Report Abstract

Overview

Artificial intelligence (AI) has been a domain of the academic world since the 1950s. In recent decades, AI gained some traction in industry, but has remained largely project- and application-specific.  This was due mainly to the high cost of computing and inadequate data systems. Now, with the advent of more robust computing systems and the ever-declining cost of computing power, AI is moving into industrial applications, from the plant floor to the oilfield.  Clearly, automation and other technology suppliers are embracing AI, as evidenced by the number of new AI products and solutions seen at trade shows.

However, many potential industrial end users are still in the discovery phase, asking questions like: “Is AI worth the effort?” and “Can AI help us make better decisions?”  This ARC Insight provides a brief overview of the potential for AI applications in discrete and process industries and then gives a simple example of how AI with machine learning (ML) could be applied to simple upstream oil & gas data to provide actionable insights.

Key Findings:

  • AI is gaining traction in industry
  • Many early adopters report rapid initial return on investment (ROI), but then experience challenges scaling up
  • Adoption rate is highest in the discrete industries (automotive, electronics) and lowest in process industries (chemical, oil & gas)
  • ML can accurately highlight or extract features that support actionable decisions (as shown in our simple oil & gas example)
  • Outcomes depend on customer pain points and can be scaled up or down

Applying AI in Manufacturing

AI’s breakthrough in industry is mostly due to the growing popularity of machine learning (ML), a subset of AI.  Techniques used in manufacturing and other industrial applications include supervised learning, unsupervised learning, deep or reinforcement learning, and ensemble learning.

Most learning techniques can be combined - the techniques used depends on the type of data, complexity of objective, and desired outcome. The table below lists common techniques and their applications in industry.

Artificial intelligence ML%20Techniques%20for%20AI%20Applications%20in%20Manufacturing.JPG

For industrial purposes, these ML techniques are used in maintenance, quality assurance, optimization, reliability, and advanced process control (APC) applications. A 2018 McKinsey study on AI use cases in manufacturing showed a 10 percent yield improvement for integrated circuit products when AI was used in R&D and a 30 percent improvement in material delivery time when ML was used to predict timing of goods transfer.

Discrete industries like automotive and semiconductors & electronics are early adopters due to the use of AI technologies in autonomous vehicles and computer vision for quality control or pick-and-place assignments. But research indicates that process industries such as oil & gas have a relatively low adoption rate of AI technologies.  This is surprising due to the high potential ROI. In these industries, AI can be used to support asset management (including prescriptive maintenance), process optimization, and root cause analysis to improve costs, improve productivity, and enhance asset utilization.

To illustrate the potential for ML-enabled AI for decision support in industrial applications, ARC created a simple case using publicly available data about oil wells in Florida in which we ask the following hypothetical questions:

  • What features support oil well performance in our use case (e.g., specific operating company, years in operation, etc.)?
  • What features support the location for better performing wells in our use case?

It is important to note here that for integrating ML-based AI solutions, sufficient data is needed and complex relationships will arise.  Large data makes it difficult to extract information on performance with a degree of reliable accuracy due to human error.  However, the goal here is to get you thinking about ML-enabled AI application for general decision support and to show that ML models can help in decision making with an accuracy level that humans alone may not guarantee.

Predicting Oil Well Production and Performance

In this scenario, a company needs to decide if and/or where to invest in new drilling assets or close existing ones. Investors with no production information want to get insights on how the operators rank against each other. We use an ensemble learning algorithm called “Random Forest” to extract the first layers of this information.  An ensemble learning algorithm combines predictions from multiple decision tree models to make predictions that are more accurate than would be possible using an individual decision tree model.  One of the most popular and most powerful machine learning algorithms, Random Forest, uses a tree aggregation method known as “bootstrap aggregation” or “bagging.”  Going forward, we try to determine the features that influence the performance of oil wells as well as answers our hypothetical questions above using oil production-volumes-by-county data from the state of Florida in the US.

Incorporating Data Sources

For the analysis, we downloaded and pre-processed data from the Florida State Department of Environmental Protection on oil well operations from 2012 to 2017 (seven years’ worth of data with approximately 8,500 observations).  The dataset contains features such as name of oilfield, operator and oil well, county location, oil, water, and gas production per month; and the number of days a well operated in each month. Using these data, we found features that support cumulative oil production (cumoilpd) and where a well is more likely to be located (county).

Application Results and Insights

The analysis shows that cumulative oil production is most influenced by water and gas volume with wells located in two counties: Collier and Santa Rosa, and for oil wells operated by two companies: Breitburn Energy Partners (now Maverick Natural Resources, LLC) and Quantum Oil & Gas.  For a decision maker at Breitburn, this simple exercise helps highlight that the company’s most productive oil wells are those that produce the most water and gas and are located in the two named counties.

ARC Advisory Group clients can view the complete report at ARC Client Portal 

If you would like to buy this report or obtain information about how to become a client, please Contact Us 

Keywords: AI, Manufacturing, Automation, Predictive Decisions, Oil & Gas, ARC Advisory Group.

Comments

Engage with ARC Advisory Group