📘 Resource Overview & Documentation Details

You are viewing the technical directory listing for Predictive Analytics: Turning PI Data Into Asset Insight. This engineering resource is curated under our Predictive Analytics division to support automation specialists, control systems engineers, and manufacturing technicians looking for verified reference materials.

🔍 Contextual Abstract & Technical Scope

Share Summary:LinkedInEmail

Predictive analytics can help power generators and other industrial operators turn historian data into earlier insight about equipment condition. At Genesis Energy, the approach combined PI System data, a generation-focused data warehouse and in-house models to monitor assets and support maintenance decisions. The aim was not to replace control-room alarms, but to identify patterns that could indicate a developing issue before it escalated.

Building a data foundation for predictive analytics

Genesis Energy began exploring its data potential in 2017, asking how existing information could be used more effectively and support decision-making. After testing its data science capability, the company developed a data warehouse for generation data. PI Integrator and PI Server Asset Framework (AF) formed important parts of that effort.

The PI Historian was already a key source, but the existing extraction process had limitations: data was available at 10- or 30-minute resolution, and requests for tags and calculations—such as averages, totals or point-in-time values—were managed by a consultant and incurred a cost. The project needed a more flexible way to select and prepare data for analysis.

AF provided the asset hierarchy required by PI Integrator. Genesis Energy aligned that hierarchy with its existing computerized maintenance management system (CMMS) structure. A partner supported deployment and mapped about 70% of tags into the hierarchy. This organization gave the team a way to select data in the context of assets, rather than treating every historian tag as an isolated point.

Moving PI System data into analytics workflows

PI Integrator uses the AF hierarchy to identify relevant tags, define the time interval and specify the data context—for example, minimum, maximum or average values. Data can be loaded once or on a schedule. Genesis Energy sent the prepared data to an Azure data warehouse, where it could be used by its analytics tool.

This architecture separated data preparation from model development. Historian data was made accessible in a consistent, asset-oriented form, while data scientists and engineers could work on analytics without having to request every extraction individually. The broader analytics environment could also draw on other potential sources, including CMMS information, software diagnostics, market data and weather.

What the models do—and what they do not do

In this program, predictive analytics meant applying machine-learning methods to asset health and maintenance management. Models learned relationships in historical and operating data, then compared observed behavior with expected behavior. An engineering team could investigate when actual measurements began to diverge from a model’s forecast or from the behavior of related equipment.

Genesis Energy described the approach as a way to move from calendar-based preventive maintenance toward analytics-triggered maintenance, while also aiming to detect defects before they develop into forced outages. It was not intended to replace DCS or SCADA alarming, nor was it designed as real-time alerting. This distinction matters: the models were intended to provide an additional condition-monitoring and decision-support layer, not to take over control or alarm functions.

Examples: generator and turbine monitoring

Generator stator temperature

For a generator stator-temperature model, users could select input variables, remove data that would distort training—such as periods when the plant was in outage—and choose training periods. A forecast period was then used to check model fit. In one example, the training and forecast comparison showed a 3.8% error.

The model became useful when actual stator temperature began rising above the expected pattern. Investigation found a particularly blocked air filter that restricted cooling. After the filter issue was addressed, measured temperature and the model aligned well again. The example illustrates how a divergence can prompt an engineering investigation; the model identifies an unexpected pattern, while people determine its cause and the appropriate response.

Turbine bearing temperature

Bearing condition was assessed in two ways: by comparing one bearing’s temperature with other bearings, and by modelling temperature in relation to machine output and ambient conditions. In the reported case, a bearing was repaired at the end of 2019. Genesis Energy stated that the repair avoided a two-week extension to an outage and recasting. The example shows why context matters: a temperature reading is more informative when considered alongside comparable equipment and operating conditions.

Scaling models and tracking outcomes

As model counts grow, teams need ways to oversee performance and manage follow-up. Genesis Energy used Power BI to manage multiple models and maintained a register of identified issues, priorities, actions and dates. Benefits were assessed in relation to the resulting failure, including the likelihood that the condition would become a genuine issue and whether it could have been detected earlier.

From February 2019, the company developed more than 800 models across thermal and hydro plants, using about 50% of a full-time-equivalent resource and predominantly internal resources. The work relied on open-source software and low-code solutions. The team reported more than 50 validated issues, proven savings and avoided outages, as well as engineering buy-in from people involved in investigating issues. A substantial backlog of potential models remained.

Practical considerations for industrial teams

The experience points to several considerations for organizations planning analytics-led maintenance:

  • Start with usable data. A historian is valuable, but accessible, appropriately contextualized data is essential for analysis.
  • Connect data to asset structure. An AF hierarchy aligned with maintenance structures can help teams select and interpret tags in equipment context.
  • Include engineering expertise. Models can flag deviations, but engineers must assess their operational meaning and determine what action is warranted.
  • Define the response process. A model finding needs an owner, priority, action and follow-up record if it is to inform maintenance decisions.
  • Plan for gaps and expansion. Genesis Energy identified additional modelling opportunities, transient operations such as starts and stops, long-term forecasting and sensor gaps as areas for further development.

Conclusion

Predictive analytics becomes practical when historian data, asset context, modelling capability and maintenance workflows work together. Genesis Energy’s experience demonstrates a path from PI System data integration to models that helped identify abnormal equipment behavior and support investigation. For industrial decision-makers, the central implication is that model development is only part of the effort: data foundations, engineering involvement and a clear process for evaluating findings are equally important to turning analytics into useful maintenance decisions.

Access Official Technical PDF