Plant predictive maintenance with Oracle Autonomous Data Warehouse

July 14, 2022 | 6 minute read
Muhammad Shuja Uddin
Principal Cloud Solution Engineer
Robin Ahmed
Senior Technology Cloud Engineer
Text Size 100%:

Predictive maintenance has been used in the industrial world since 1990, but only recently can we exploit the power of artificial intelligence (AI) and machine learning (ML) to predict more accurately, quickly, and easily. For organizations in the oil and gas industry, especially downstream, AI greatly helps minimize planned downtime, maximize equipment lifespan, optimize employee productivity, and increase revenue effectively.

Using a sample data set, I conducted an event for the oil and gas industry in Indonesia. In this blog post, I explain how we can use the built-in capabilities of AI in Autonomous Data Warehouse and Oracle Analytics Cloud to perform predictive maintenance for plant machinery equipment.

Predictive maintenance for oil and gas plants

Predictive maintenance ensures that all machines are operating at maximum efficiency. Let’s start with looking at some statistics and what industry leaders say. According to a recent EY report, $450B in maintenance was spent on downstream assets between 2011—2035. $672B in maintenance was spent on upstream assets during the same period according to investment from an International Energy Agency study.

A chart comparing the contractors, production, and state budget targets in bpg.

“The lower-than-targeted production realization was attributed to technical issues,” said Wisnu Prabawa Taher, program and communications division head at SKK Migas.

When we don’t have appropriate information to predict the life of each device more accurately and the ability to corelate its impact on the overall equipment failure, the increase in equipment failures results in missing production targets and reducing revenue for the organization. To address this challenge, we need to analyze the frequency and occurrence of failure of devices in the plan and develop a predictive maintenance model.

McKinsey reported that most oil and gas operators haven’t maximized the production potential of their assets. “Digitalization in operations management will have a game-changing impact to the future of the upstream oil and gas industry,” said Muhammad Zamri Jusoh, executive vice president of Asia Pacific Upstream at Eni Franco Polo and regional president of Asia Pacific at BP Nader Zaki.

AI, ML, and Autonomous Data Warehouse

AI and ML can unleash the next wave of digital disruption in the oil and gas industry. Predictive analytics allows better use of machine cost, scheduling maintenance activities more appropriately, monitoring the machine performance, and automation of the processes with the help .

Oracle Autonomous Data Warehouse provides the best platform to store and process trillions of data-points collected every second by those device sensors throughout the plant. The ability to identify quality data and process as training data is one of the main determinants of an AI model's predictive power. Autonomous Data Warehouse can not only ingest data at high speeds but also provides scalable data processing performance for big volumes of data for better prediction and forecasting.

The AI algorithms then compare that data against the ideal performance data to find discrepancies between the current and ideal state. Using the built-in AI capability of Autonomous Data Warehouse through the Zaplin interface, you can quickly and easily use deep data modeling techniques and perform forecasting and other analysis. Autonomous Data Warehouse’s built-in security also helps with data privacy and compliance requirement of this highly regulated industry.

Using Oracle Analytics Cloud, you can rapidly build visual dashboards with simple drag-and-drop entities of interest. You can develop reports and flashy dashboards and find the failing devices, find the number of occurrences, analyze the duration and frequency of breakdowns, create charts, group by devices and years, and perform all sorts of analysis on the data, without learning any programming language or database SQL query language.

Today, wireless sensors or IoT devices mounted on machinery can monitor machines for various attributes, such as noise, voltage, temperature, air, chemicals, and weather condition. Using cloud-based technologies like Autonomous Data Warehouse and Oracle Analytics Cloud, you can develop real-time analytic dashboards, which your maintenance teams can use to monitor, alert, plan, and predict a machine’s condition more accurately.

​Using advanced AI algorithms, such as the attribute importance algorithm, we can find out the top 10 sensor factors contributing to the failure of a particular device, as shown in the following screenshot. The higher explanatory value shows the attribute’s importance for failure. Through this algorithm, we can see that cathode voltage, cast bar temperature, and lube oil temperature are some of the top attributes

A screenshot of the Performance Management Dashboard in the Oracle Cloud Console, showing the Predictive Maintenance tab, Attribute Selections, and top attributes influencing equipment failure.

After we identify the top attributes contributing to the failure, we need to find their value combinations that cause a device to fail. We can build a model to predict the failure of that equipment. We use the cart classification algorithm in the classification and regression trees and run it on those failing equipment data in Oracle Analytics Cloud to train the model.

When the model is built and trained, we run it on a larger dataset and then compare data ranges for those attributes, which contribute to failure of that equipment. All this information is visually depicted using Oracle Analytics Cloud dashboards. The green lines show conditions for nonfailure, and the red lines show conditions for failure. For example, the model predicts failure when lube oil temperature <= 49.5.

A screenshot of the example machine learning model.

​Next, we use the forecasting capability of Oracle Analytics Cloud to predict future failures and nonfailures so that timely decisions are made to avoid outages. We can see that the parameter ranges for these predictions are well within the ranges predicted by our model. The first graph shows the prediction for nonfailure condition in green.

A screenshot of the example model predictions for parameter ranges of positive cases.

The second graph shows the prediction for failure condition in red.

A screenshot of the example model predictions for parameter ranges of failure cases.


Building a predictive maintenance model becomes critical, not only to reduce failures, but to predict failures in advance, so that we can avoid them, reducing overall assets management cost, reducing assets downtime, preventing unexpected failures, increasing productivity of oil production, and improving plant safety. We achieve these goals through machine learning, providing the ability to analyze large sets of data comparing ideal conditions. ML is built into both Oracle’s Autonomous Data Warehouse and analytics tools with an intuitive GUI console, which make it easy to get started on machine learning.

Try Oracle Cloud Free Tier! A 30-day trial with US$300 in free credits gives you access to OCI services.

For more information, see the following resources:

Muhammad Shuja Uddin

Principal Cloud Solution Engineer

Shuja has 20+ years of experience, mainly as a Technical Sales Consultant and Solution Architect, with specialization in database, datawarehouse, analytics, ECM, Engineered Systems and Cloud platform products from Oracle and previously IBM. He is active in volunteering, writing, moderating Oracle related activities. He has been invited to speak at various public forums, including presented a paper at the IOUG conference in Toronto in 2004. He is a Computer Science graduate and MBA. Currently, he advises organizations moving to Cloud.  

Robin Ahmed

Senior Technology Cloud Engineer

Previous Post

Receive Web Application Firewall notifications using Service Connector Hub

Cody Brinkman | 5 min read

Next Post

Stop overspending and underutilizing in the cloud

Julien Lehmann | 3 min read