Introduction
Industrial companies spend billions annually purchasing, operating, and maintaining capital-intensive assets. Whether it’s oil platforms, manufacturing plants, power generation equipment, or wind turbines, the challenge is the same: keep your assets producing as economically and for as long as possible while driving maintenance costs down and ensuring product quality remains high. In years past that meant lots of preemptive maintenance to hedge against the possibility of asset failure, maintenance that in many cases was actually unnecessary and incurred undue costs, both in terms of work performed and productivity lost during that work.
With the arrival of AI-powered predictive/prescriptive maintenance, this state of affairs has changed dramatically for the better, with asset owners/operators now able to predict incipient failures well before they occur and to take proactive steps to mitigate these failures, conducting recommended maintenance at times that are least disruptive to operations and least damaging to profitability.
AI-enabled asset performance management (APM) is predicated on two important capabilities: prediction and prescription. The former indicates when failures are going to take place and on what specific equipment, while the latter tells operators what they can do about it. Both, however, are built upon a critical foundation, i.e., the fact that assets do not continue functioning throughout their lifetimes the way they did the day they left the factory. Assets age, parts wear out, maintenance takes place, the operating environment takes its toll. That means that maintenance procedures provided by OEMs for their new equipment cannot ensure long-term performance that will maximize the operating lifetime of assets whose lives are frequently measured in decades. That’s where AI-powered predictive/prescriptive maintenance enters the picture.
Predicting the future of asset performance
Predictive maintenance uses AI and real-time performance data to determine the likelihood of an asset failing and the estimated time until that failure occurs. This allows maintenance to be scheduled precisely when it’s needed, just before a breakdown, avoiding unnecessarily premature servicing or catastrophic failure, with all their expense, productivity, and safety implications. To support predictive maintenance, AI models are fed a continuous stream of data from sensors on critical assets like pumps, turbines, engines, etc. This data includes parameters like:
- Vibration (for rotating equipment like motors and pumps)
- Temperature, pressure, flow rates
- Current and power consumption
- Acoustic signatures
- Historical maintenance logs, images/videos
Machine learning (ML) algorithms then analyze this real-time data to determine “normal” baseline operating conditions—conditions which, critically, will vary significantly as equipment ages. The model then flags anomalies—subtle drifts, fluctuations, or irregularities—invisible to the human eye but indicative of the early onset of degradation.
Finally, advanced AI techniques, such as Normal Behavior Modeling (NBM), process these patterns and apply knowledge from historical failure cases to understand the sequence of events leading up to the forthcoming breakdown. The model then calculates the probability of failure as well as the remaining useful life (RUL) of a system or component, providing an accurate warning, often weeks or months in advance.
But what to do about impending failures
Prescriptive maintenance goes beyond prediction, telling operators not only what will happen and when, but also what to do about it and why. It recommends the best course of action to achieve a specific business objective (like minimizing cost or maximizing uptime or output). Having predicted an incipient failure, the model next integrates this predictive output with a much broader set of contextual business data:
- Inventory Levels: Are the required spare parts in stock? If not, where are they and how long will it take to get them to where they’re needed?
- Staff Availability: Are maintenance technicians available? What are their specific skills? Do they have the necessary tools?
- Production Schedule: When is the best window to perform the work in order to minimize impact on output?
- Cost Analysis: What are the trade-offs between a temporary fix, a full replacement, and letting the machine run longer, possibly to failure?
Finally, based on this analysis, the system provides a prescribed, optimized action plan, for example:
- Replace bearing ring 17 in turbine 4 immediately to minimize the three days of unplanned downtime that will otherwise occur next month.
- Reduce the speed of a manufacturing conveyor belt by 5% for the next 72 hours to delay predicted failure until the upcoming holiday shutdown.
In mature predictive/prescriptive maintenance systems, the AI can even initiate necessary actions autonomously, e.g., creating a work order in the maintenance system, ordering required parts, or scheduling technicians.
What’s the outcome of all this?
The adoption of AI-powered predictive/prescriptive maintenance provides significant business benefits:
- Lower Unplanned Downtime: By anticipating failures, organizations can schedule maintenance proactively, leading to significant reductions in unplanned downtime.
- Reduced Maintenance Costs: Servicing assets only when failures are predicted eliminates the cost of unnecessary, time-based maintenance.
- Extended Asset Lifespan: Proactive/targeted repairs reduce unnecessary strain on assets, extending lifetimes.
- Optimized Resources: Maintenance crews and spare parts inventory are managed based on actual, predicted needs rather than fixed schedules or one-off emergency repairs.
Impact of these techniques
AI-enabled predictive/prescriptive maintenance techniques have allowed operators in many asset-intensive industries to forego failures, increasing product output, revenue/profit generation, and employee safety.
Wind turbine pitch bearing failure prediction
A renewable energy company had previously used manual grease sample analysis to determine the operating health of blade pitch bearings on its wind turbines. This process was time consuming and required taking the turbine offline in order to obtain samples. Using Avathon’s Autonomy for Renewables platform, the company now obtains real-time pitch bearing performance data continuously without the need to take the turbine offline or forego power generation opportunities. Impending bearing failures are identified up to one year in advance, saving up to $150K per failure event through avoided multiple crane callouts and minimized turbine downtime.
Electric power generation
A large electric power company had previously used a rules-based pattern recognition system and OEM monitoring to measure performance on a combined-cycle gas turbine. But this approach failed to prevent two catastrophic failures within the first three years of operation. Using the Avathon Autonomy platform’s predictive/prescriptive maintenance technology to generate more than 3000 tags/sensors per second enabled the utility to measure and analyze performance variables like vibration, temperature, pressure, and operating speed. This allowed the utility to successfully safeguard the turbine—valued at more than $2B—predicting a unique failure one month in advance that averted roughly $500K in repairs.
Food and beverage manufacturing
A Fortune 50 food and beverage manufacturer used Avathon’s Autonomy for Manufacturing Operations platform to proactively monitor production line performance metrics to gain insights into overall plant health and production efficiency. The company was previously unable to identify production anomalies, but with the use of predictive/prescriptive maintenance, they were able to increase production efficiency by 2% – 5% while maintaining product quality and avoiding $600K in maintenance costs.
Conclusion
Asset-intensive industries rely on uninterrupted performance to ensure the financial health of their operations and organizations. Failures—particularly unplanned ones—reduce product outputs, compromise quality, put employees at risk, and drive down financial performance. Predictive/prescriptive maintenance can help industry managers and executives keep their assets operating efficiently and economically, maximizing asset lifetimes, and extracting the greatest value from their fleets.
To learn more about the Avathon Autonomy for Operations platform and how it can help you ensure the greatest output from your assets, visit our site.
