From Reactive to Proactive: How Predictive Analytics Software Transforms Maintenance

For example, to enable a proactive maintenance strategy, an oil and gas operator in Scotland deployed GE Vernova’s SmartSignal predictive analytics software across 14 offshore platforms.  Shortly after a routine maintenance outage, GE Vernova’s Industrial Managed Services (IMS) team, which monitors customer equipment using SmartSignal, received an alert from the software’s digital twin indicating a slight increase in a pump bearing temperature. While the reading was still well below alarm thresholds, the software’s multivariate analytics detected subtle pattern changes, flagged the anomaly, and matched it to a SmartSignal failure mode.
 
The assigned IMS engineer investigated further and uncovered a concerning trend — rising temperatures and shifting vibration patterns. Acting on this insight and the recommendations from IMS, the customer performed a thermographic inspection and discovered the situation was urgent. While it was a simple pump on the verge of failure, it was one that could shut down the entire offshore platform’s operation if it failed. Since the platforms are hundreds of miles away, the customer does not have spare parts readily available.  Fortunately, with the early warning, the team had time to order a replacement part and have it ready for the planned outage. After the pump was replaced, the assigned GE Vernova IMS engineer monitored the pump and confirmed the anomaly trend disappeared and the replacement was successful.
 
This is the power of predictive analytics: transforming a reactive maintenance strategy into proactive maintenance, improving availability and reliability. In this blog, we’ll explore what predictive analytics can do for maintenance teams, how it works, different types of predictive analytics, and what it takes to succeed in implementing them.

Before we dive into what predictive analytics can do, let’s clear up a few common misconceptions:

That said, here’s what predictive analytics can do—and why it’s a game-changer.

Read how Total EP achieved zero unanticipated failures using SmartSignal Total EP counts on zero unanticipated failures with SmartSignal-powered monitoring center

One of the most transformative technologies driving digital transformation is the digital twin, a virtual replica of a physical asset that simulates its behavior using real-time data. This data can come from sensors, historical records, control systems, and contextual inputs like operational settings. The purpose of a digital twin is to monitor and simulate physical assets or processes.

Digital twins enable predictive maintenance by detecting subtle patterns that traditional monitoring systems might miss.

There are different types of digital twins, each serving different purposes, for example:

Learn how EDP partnered with GE Vernova to create a remote Monitoring & Diagnostics Center aimed at using predictive O&M strategies with real business impact. Watch Now EDP Uses Predictive Analytics to Optimize Generation Portfolio | GE Vernova

Predictive analytics can draw from systems and datasets across the enterprise. Digital twins that power predictive analytics leverage statistical models and machine learning to identify trends, patterns, and anomalies in real-time data. Other systems that are essential for an effective predictive maintenance strategy include historical and contextual data from other systems as well as human expertise.

Depending on the sophistication of the software, these systems may also offer prescriptive insights, recommending actions to address predicted issues. While artificial intelligence is increasingly capable, human expertise remains essential for interpreting complex scenarios and making final decisions.

There are several different models that are used in predictive analytics, including:

Purpose: Every measured sensor tag contributes to estimating every other tag. It’s useful when inputs are highly intercorrelated.

Upstream Example:  identifies anomalies in drilling operations by modeling expected pressure and flow behaviors and highlighting deviations.
 
Power Generation Example: Early-stage turbine faults by learning normal operational patterns and flagging deviations in sensor data.

Purpose: Estimates unmeasured or hard-to-measure sensor values using a selected subset of key input variables. Focuses on the most influential drivers to simplify model design and improve flexibility. Only a subset of sensor inputs is used to infer other (dependent) sensor values. This allows the analytic to focus on key drivers while offering flexibility in model design.

Oil & Gas Example: Inferring product composition in a distillation column using feedstock flow rate, column top and bottom temperatures, and reboiler duty.

Power Generation Example: Estimating combustion temperature in a gas turbine using compressor inlet temperature, fuel flow rate, and turbine speed.

Purpose: Predicts a continuous outcome based on one or more input variables

Upstream Example: Estimating oil production rate from a well based on reservoir pressure, choke size, and water cut.

Power Generation Example: Predicting electricity output of a solar panel based on sunlight intensity, panel angle, and ambient temperature.

Purpose: Used for binary classification (e.g., yes/no, fail/pass)

Upstream Example: Predicting whether a well will experience a casing failure within the next 30 days based on pressure and temperature anomalies.

Power Generation Example: Classifying whether a wind turbine will shut down due to high wind speeds in the next hour.

Purpose: Analyze data points collected or recorded at specific time intervals

Upstream Example: Forecasting daily oil production from a well using historical production data and pressure trends.

Power Generation Example: Forecasting hourly power demand in a grid using historical consumption data.

Purpose: Statistical method used to model the relationship between two or more variables

Upstream Example: Understanding if temperatures rising and changes in vibration is a pump about to fail at a refinery

Power Generation Example: An alert received from analysis comparing vibration thresholds against model predicting a thrust bearing issue.

Purpose: Tree-based models that split data into branches to make predictions

Upstream Example: Identifying causes of production decline by branching on reservoir properties, completion type, and artificial lift performance.

Power Generation Example: Determining optimal generator startup sequences based on fuel cost, demand forecast, and maintenance schedules.

Purpose: Mimic the human brain to detect complex patterns in large datasets Upstream Example: Predicting reservoir permeability from well log data using deep learning models.

Power Generation Example: Predicting equipment failures in a thermal power plant using real-time sensor data from multiple subsystems or closed-loop automated gas turbine tuning to find optimal operating zone.

Purpose: Group similar data points together to identify patterns or anomalies

Upstream Example: Grouping wells with similar production profiles to identify underperforming wells or sweet spots.

Power Generation Example: Clustering wind turbines based on performance metrics to identify underperforming units.

GE Vernova uses a combination of digital twins and statistical models in its software solutions for energy industries. For example, SmartSignal is an AI/ML predictive analytics solution that uses auto-associative and inferential statistical modeling methods. This solution is used to prevent unplanned downtime and improve asset reliability.

Implementing predictive analytics isn’t just about installing software. It requires a strong foundation across several critical areas:

Predictive models are only as good as the data they’re built on. Inconsistent, incomplete, or noisy data can lead to inaccurate predictions and missed opportunities.

Predictive analytics must fit into your existing workflows. Integration with CMMS, SCADA, ERP, and historian systems support insights that are actionable.

Predictive analytics is an investment, but one that should pay off in reduced downtime, lower maintenance costs, and extended asset life.

Predictive analytics is a data-heavy solution. Having sensors available that are correlated with each other from a physical or process perspective is paramount to ensuring the analytic can draw useful conclusions in correspondences.

Predictive analytics is not a plug-and-play solution, it’s a strategic capability. Evolving from reactive/condition-based maintenance strategies requires the right data, tools, and people in place. It can transform how your organization manages risk and achieves reliability and performance excellence.

Watch Cosmo Oil explain how they achieve digital excellence with GE Vernova SmartSignal and APM solutions.GE Vernova is a purpose-built, industry leader focused on enabling electrification and decarbonization. Electricity is crucial to modern civilization and improved quality of life. We help customers create a more sustainable world by delivering products and services that generate, transfer, orchestrate, convert, and store electricity. With approximately 30% of the world’s electricity generated by our customers using our technologies, GE Vernova is uniquely positioned to lead customers through the energy transition, solving for the energy trilemma of reliability, affordability, and sustainability. Our diverse and experienced leadership team brings deep industry and public company expertise to drive long-term shareholder value. We are excited about our plans to become a more focused, independent company sometime in early 2024.