min read

Unlocking prescriptive maintenance with AI-based Detections

Jungle’s CIO & co-founder Silvio Rodrigues shares the thoughts and technology behind Detections and the difference it makes in everyday renewables operations.

Renewable energy owners and operators need robust prescriptive maintenance solutions to optimise their assets' performance and uptime. A key ingredient in developing such solutions are labelled datasets that record past failures, but these are hard to come by, making it a real challenge for the wind and solar industries.

The process of labelling datasets isn't straightforward, largely because understanding the root cause of failures can be intricate. There's uncertainty about when these problems begin and which sensors could pick up on these failures, making dataset labelling even more complex. Not to mention, the shortage of detailed historical records of past maintenance efforts adds another layer of difficulty. These challenges stand as significant barriers to implementing effective predictive and prescriptive maintenance strategies.

At Jungle, we've developed a solution to this problem: Detections. This AI-driven tool offers actionable insights that enable operators to establish effective prescriptive maintenance plans. The upcoming sections will explain Detection in three easy-to-follow steps.

1. ML models and sensor deviations

Our advanced AI models serve as invaluable components in this process, operating as digital mirrors of the assets. These models learn from each asset's historical data, studying their standard behaviour patterns. When deployed, they provide essential data on the internal pressures and temperatures of all components in relation to the turbine's energy output. The modelling extends to factors such as power production, pitch angle, and rotor speed. Interested readers can further explore our Machine Learning models in these articles here and here.

These AI models deliver predictions, which are compared to the actual sensor measurements. The deviations, or residuals, between these predicted and actual values provides valuable information for the next step in the process.

2. Hand-written rules driven by our domain knowledge

Consider the potential of joining concise yet potent rules such as: "Detect when the generator is overheating and the turbine is underperforming." Without the correct toolkit, this may seem like an almost impossible task. Identifying an overheating generator alone presents a hefty challenge since it vastly depends on the operating conditions and the asset's normal behaviour.

In pseudo-code, the above rule would take the following form:

sensor_generator_temperature > predicted_generator_temperature


turbine_production < predicted_turbine_production

Rules with such level of precision pack a powerful punch as they're based on apparent discrepancies between sensor measurements and AI models that predict the expected standard sensor behaviour. This seamless integration of manual rule programming with AI models is what provides a robust platform for early detection and mitigation of equipment breakdowns.

Adding to their distinct advantages, these rules are not limited to a specific turbine OEM, reinforcing their broad applicability. Once defined, a Detection can be deployed across numerous turbines within extensive portfolios.

3. Detections

Our system allows for the configuration of detectors specified to unique situations, including:

  • Turbine curtailment,
  • Generator overheating,
  • Low hydraulic pressure,

These tailored detectors simplify failure identification, root cause analysis and enable apt maintenance recommendations. Their specificity demystifies complex issues, thereby empowering operators to comprehend the nature of the failure and take the corrective steps needed.

Example: Detecting self-inflicted curtailment

The figure below provides a snapshot of an asset equipped with a self-curtailment detector.

Our specialised models detected peculiarities in the generator phase temperatures, noticing that they were hotter than anticipated. Simultaneously, the wind turbine pitched its blades more while producing less power than our models forecasted for the given environmental conditions. The following Canopy figure depicts the individual sensors that signalled these anomalies.


Detections are Jungle’s response to the industry need for prescriptive maintenance: systems that continuously monitor the assets and alert the relevant personnel to which assets are deviating from their expected normal operation and performance. With simple rule sets, defined once and scaled to entire portfolios, we can alert and instruct our Canopy users with highly detailed descriptions of ongoing asset malfunctioning.

Detections also empower users of varying expertise levels. The accommodation of both advanced users and those less experienced ensures that potential issues can be detected and addressed promptly, irrespective of the proficiency of the operator. This leads to a democratisation of dataset labelling, allowing wind farm owners to start collecting high-quality health datasets of their assets, and supporting wind farm maintenance crews in reporting significant findings.

Still curious?

If you still have doubts about how detections are empowering the teams behind the machines, reach out to see them in action. Send us a message to hello@jungle.ai to book a Canopy demo with our team and witness it for yourself!

Silvio Rodrigues

CIO & Co-founder

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique. Duis cursus, mi quis viverra ornare, eros dolor interdum nulla, ut commodo diam libero vitae erat. Aenean faucibus nibh et justo cursus id rutrum lorem imperdiet. Nunc ut sem vitae risus tristique posuere.

Back to blogs