Unity AI Insights: Classification Flag for Wind Turbines - Classification Logic

Last Modified on 2025-08-01 20:46:54 +0200

In this article:

Introduction

This page provides a summary of the logic behind the different classifications of the Classification Flag and is a complement to Unity AI Insights: Classification Flag for Wind Turbines.

High-Level Logic

The classification follows a hierarchical approach:

  1. Data Quality Control: Unreliable data is flagged

  2. Data Imputation: As some inputs are optional, some missing values can be filled using defaults or logical relationships

  3. Classification: The classifications are simple rule-based logic. All rules are applied to every single data interval, resulting in several possible classifications per interval.

  4. Prioritization/Ranking: The final classification keeps only the most highly-ranked class from all possible classifications, at each interval, resulting in a single classification per interval.

Classes & summary of logic

Data Quality Control & Imputation

Missing or invalid inputs (flag value -1)

  • Logic: Identifies intervals where mandatory data is missing or unreliable (ActivePower, WindSpeedAdjusted, ActivePowerExpected

  • Purpose: Properly classify intervals that cannot be reliably analyzed

Data Outside Evaluation Limits (flag value 3)

  • Logic: Operating conditions outside reliable analysis range - very high wind speeds or very low power production

  • Purpose: Properly classify intervals that cannot be reliably analyzed

At this step, data imputation is performed to maximize the number of timestamps where the subsequent classifications can be performance, for optional inputs. The data imputations are as follows:

  • PossibleIcingFlag if not available, can be inferred from AmbientTemperature, if available (< 3°C)

  • Rated Cut In and Rated Cut Out map to defaults (3.5 and 25.0 m/s respectively)

  • ActivePowerExpected - UpperControlLimit can be inferred from LowerControlLimit, if existant

External limitations

Utility Curtailment (flag value 1)

  • Logic: Based directly on the Curtailment flag (see Unity APM: Curtailment Flag)

  • Purpose: Identifies periods where turbine performance is externally constrained with expected losses at the generator level

Overdesign: 11

  • Logic: ActivePowerExpected is below ActivePowerExpectedAsBuilt while still producing substantial power 

  • Purpose: Identifies wind farm-level limitations due to grid constraints

Downtimes

Site Down (flag value 9)

  • Logic: Entire site offline with negligible power production despite wind availability

  • Purpose: Distinguishes site-wide outages from individual turbine outages or issues

Turbine Down (flag value 8)

  • Logic: Individual turbine producing minimal power despite adequate wind conditions

  • Purpose: Identifies complete turbine outages

Downtime Icing (flag value 7)

  • Logic: Turbine downtime coinciding with PossibleIcingFlag 

  • Purpose: Attributes downtime to ice accumulation on blades/sensors

Degradations

Derate (flag value 5)

  • Logic: Consistent power production below expected levels with very little variation in power output for a defined rolling window (40 minutes). To assess that production is below expected levels, a threshold is based either on: 

    • Active Power Expected - Lower Control Limit, or

    • Performance Index  if Lower Control Limit is not available

  • Purpose: Identifies intentional power limitations or control system constraints

Degraded Icing (flag value 6)

  • Logic: Reduced performance coinciding with PossibleIcingFlag, for a certain rolling window (40 minutes). To assess that production is below expected levels, a threshold is based either on: 

    • Active Power Expected - Lower Control Limit, or

    • Performance Index if Lower Control Limit is not available

  • Purpose: Identifies potentially ice-related performance degradation

Degraded Unknown (flag value 10)

  • Logic: Power production below expected levels without clear cause. To assess that production is below expected levels, a threshold is based either on: 

    • Active Power Expected - Lower Control Limit, or

    • Performance Index if Lower Control Limit is not available

  • Purpose: Catch-all for various degradation modes

Other 

Uprated or Overpower (flag value 2)

  • Logic: Active power exceeds rated capacity by tolerance factor 

  • Purpose: Detects turbine upgrades or temporary overpower conditions

Nacelle Wind Speed Anomaly (flag value 4)

  • Logic: Power production significantly above expected levels for a certain window (at least 60  min) for given wind conditions. To determine abnormal conditions, a threshold is based either on: 

    • Active Power Expected - Upper Control Limit, or

    • Performance Index  if Upper Control Limit is not available

  • Purpose: Detects anemometer issues or measurement errors and periods where wind-dependent KPIs will be erroneous

Default (in absence of other classification)

Full Performance (flag value 0)

  • Logic: No other classification flags triggered

  • Purpose: Normal operation within expected parameters

Priority Ranking System

The system uses a priority ranking. The order is defined as follows, from highest to lowest priority. It ensures that more specific or critical classifications win over less specific or less critical ones. 

Priority 

Classes

Highest priority - this class will always be chosen even if other classification criteria are met, since, in absence of some inputs, no other classification is considered reliable

Missing or invalid inputs

Utility Curtailment

Uprated or Overpower

Site Down

Downtime Icing

Turbine Down

Overdesign 

Data outside evaluation limits

Nacelle Wind Speed Anomaly 

Derate

Degraded Icing 

Degraded Unknown 

Lowest priority - this class will only be assigned for timestamps where no other classification criterium was met

Full Performance 

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article