Introduction: Getting the most performance from your industrial assets
The objective of any company is to employ assets in ways that maximize productivity and profit. Achieving these goals means managing expenses by reducing routine or unscheduled maintenance, increasing the useful life of expensive capital equipment. Historically, these goals have been pursued with condition-based monitoring (CBM) solutions or OEM-provided asset management tools. However, in a world where equipment is increasingly reliable, and failure data scarce, these approaches aren’t nearly as effective as they used to be.
Companies spend billions annually operating and maintaining industrial assets. Whether it’s aircraft, oil platforms, or manufacturing systems, the challenge is the same: keep assets producing revenue as economically and reliably as possible while driving maintenance costs down and ensuring product quality remains high. With the arrival of artificial intelligence (AI)-powered predictive/prescriptive maintenance, asset operators can predict impending failures before they occur and take proactive steps to mitigate these failures and their downstream consequences, performing prescribed maintenance at times that are least disruptive to operations and profitability.
Traditional maintenance relies on reactive models (fixing things after they break) or preventive models (replacing parts on a rigid calendar schedule, often wasting still-functional components). AI-powered predictive and prescriptive maintenance fundamentally shift the operational mindset from guessing to knowing, using real-time data to optimize both equipment health and resulting workflows.
First, let’s get our terminology straight
Predictive Maintenance (The “When”)
Predictive maintenance uses sensors that measure variables like vibration, temperature, pressure, and acoustic signatures to monitor equipment in real time. AI algorithms process this data to calculate each asset’s Remaining Useful Life (RUL), giving operators a precise window into when a machine is likely to fail.
Prescriptive Maintenance (The “What” and “How”)
Prescriptive maintenance takes predictive data a step further. It doesn’t just say, “This bearing will fail in 72 hours.” It evaluates the full operating context and says, “This bearing will fail in 72 hours, but if you reduce machine speed by 15%, you can extend its life to 96 hours, enabling you to finish the current production run and automatically order the required replacement part for the third shift to install once labor is available.”
How does predictive/prescriptive maintenance work?
Machine learning (ML) algorithms analyze immense streams of equipment performance data to determine normal operating conditions. Once this normality has been established (a condition which, importantly, changes throughout the lifetime of the equipment in response to maintenance, wear and tear, and environmental factors) the model then proactively identifies anomalies— subtle drifts or fluctuations—invisible to the human eye but indicative of potential degradation.
Once these departures from normality have been identified, advanced AI techniques, such as Normal Behavior Modeling (NBM), analyze these patterns to understand the sequence of events likely to precede the impending breakdown. The model then calculates the probability of failure as well as the remaining useful life (RUL) of the system or component, providing an accurate warning, often weeks or months in advance.
Once the incipient failure has been flagged, prescriptive maintenance takes things a step further, combining the impending failure information with contextual business data to tell operators what actions they should take in response to the failure to achieve a specific objective (e.g., minimize cost or maximize output).
NBM not only proactively identifies impending failures, but it also helps to minimize alert fatigue and false positives. In addition, it facilitates the continuous adaptation of the monitoring system, evolving its understanding of the normal state of an asset as it ages, providing alerts based on complex interactions between the many components and parameters of the system. NBM can be applied to complex systems or single pieces of equipment, such as a hydraulic pump or wind turbine. Once trained to understand the system’s normal operating state, the model continuously evaluates the sensor data stream, generating alerts whenever anomalies—even extremely subtle ones—are detected.
Predictive maintenance provides real value to industrial assets and processes
Alert Calibration: Well-timed and well-tuned alerts delivered with all-important lead time to respond proactively come from multivariate AI models aligned with input from subject matter experts—a signature advantage of Avathon’s Autonomy platform. Our semi-supervised models catch known harmful events by virtue of threshold-based deviations, but also the ‘unknown unknowns’ that can lead to unplanned downtime: costly disruptions that knock production offline for hours or days without advance warning. The Autonomy platform is also adept at mitigating noisy alerting mechanisms that contribute to alert fatigue. Smart alerts and auto-calibration of thresholds are tunable via real-time feedback to automatically categorize, benchmark, and compare alerts, further reducing false positives.
Explainability: It’s one thing—a very important thing—to be able to apply an AI model to a real-time data pipeline that’s connected to an industrial asset to call attention as early as possible to an emerging issue. But what then? What is driving the anomalies? How do we isolate progression from the moment the anomaly was noticed to the present by looking at not just trend plots but also histograms? Avathon Autonomy provides an intuitive capability for investigating trends and categorizing timely, appropriate responses.
Learning: Analyzing the performance of industrial assets as they age is just as crucial as understanding how they’ve performed in the past. Driven by user knowledge and alert scoring, Avathon’s predictive maintenance solution facilitates recurring adjustment to the ‘new normal’ of operational states as assets age and maintenance practices change. This is also a critical backstop to overcoming the challenge of maintaining technical expertise in an aging workforce, making operations knowledge readily available to a new generation of workers.
Implementing AI-enabled predictive/prescriptive maintenance practices provides significant and enduring benefits:
- Significant Reduction in Unplanned Downtime: By anticipating failures, maintenance is scheduled proactively, minimizing downtime.
- Reduced Maintenance Costs and Optimized Resources: Servicing assets only when failures are predicted lowers maintenance costs and optimizes scheduling. AI interacts directly with ERP systems to better manage parts availability, technician skills, and production deadlines. Crews/parts are managed based on actual needs rather than fixed schedules or emergency repairs.
- Extended Asset Lifespan: Predictive/prescriptive maintenance reduces unnecessary strain on assets, extending lifetimes, reducing unnecessary capital outlays.
- Enhanced Energy Efficiency and Sustainability: Degraded machinery must work harder, pulling more electrical current and wasting energy. AI analytics identify assets operating sub-optimally (e.g., an imbalanced motor), allowing engineers to tune or repair them, which directly reduces the plant’s carbon footprint and utility costs.
- Improved Worker Safety: Catastrophic machine failures (like a pressurized pipe bursting or a high-speed rotor shattering) pose severe risks to workers. By predicting high-risk failure dynamics and providing autonomous decision support, AI shields human operators from dangerous workplace hazards.
Conclusion: Predictive maintenance is a valuable tool for ensuring system performance
Asset-intensive industries rely on uninterrupted performance to ensure the financial health of their operations. Failures— particularly unplanned ones—reduce outputs, compromise quality, put employees at risk, and lead to higher costs. Predictive/prescriptive maintenance helps managers keep their assets operating efficiently and economically, maximizing lifetimes and extracting the greatest value from their fleets. By identifying the subtle “fingerprints” of system failure long before they are apparent to the human eye, predictive/prescriptive maintenance ensures maximum performance and lifetimes from expensive capital equipment.
Normal behavior modeling is the state of the art in predictive maintenance of complex systems and equipment. It simultaneously automates complicated performance data monitoring and analysis processes while minimizing alert fatigue from false positives. It facilitates the continuing adaptation of the monitoring system to the evolving notions of what constitutes the ‘normal’ state of the system as it ages. And it enables alerts to be based on the complex and frequently nonobvious interactions between the many components and parameters within (and sometimes outside of) the system.
To learn more about how predictive/prescriptive maintenance can empower your organization to extract more value from your asset base, download our white paper Predictive Maintenance Using Normal Behavior Modeling.
