Bayesian Network Rotor Blade Damage Prediction Model

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional statistical methods are inefficient in generating prediction models for rotor blade damages in wind turbines due to high dimensionality, requiring excessive computational time.

Innovation Solution

A computer-implemented method using Bayesian networks to process data sets containing turbine, weather, and damage variables, involving discretization, structure learning, and parameter learning to generate a prediction model that can efficiently predict rotor blade damages with low computational effort.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional statistical methods are used to build prediction models for rotor blade damages, then the models can capture complex damage patterns, but the computational time required becomes excessively long (years)

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent transforms the continuous damage data into discrete categories through discretization, changing the parameter representation from continuous to discrete. This enables the use of Bayesian networks which can process the data more efficiently while maintaining predictive accuracy for rotor blade damage patterns

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces conventional statistical methods with Bayesian network algorithms, substituting a computational approach that is better suited for high-dimensional data. This substitution reduces computational time from years to a manageable period while preserving the ability to capture complex damage relationships

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Loss of information

If high dimensionality damage data is processed using conventional methods, then comprehensive damage analysis is possible, but the computational complexity becomes unmanageable

Engineering Contradiction:
Improvedamage data completenessVSAvoidcomputational complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent applies discretization to transform continuous damage variables into discrete categories, changing the parameter representation. This reduces computational complexity while maintaining the comprehensive analysis of damage patterns through the Bayesian network framework

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent segments the high-dimensional damage data into discrete categories and variables, organizing the complex information into a structured Bayesian network format. This segmentation makes the data more manageable while preserving the relationships between different damage factors

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4127466B1A computer-implemented method for generating a prediction model for predicting rotor blade damages of a wind turbine
Publication Date: 2024.08.21 SIEMENS GAMESA RENEWABLE ENERGY AS
  • EP4127466B1 patent drawingFigure 1
  • EP4127466B1 patent drawingFigure 2

AI summary

The invention refers to a computer-implemented method for generating a prediction model for predicting rotor blade damages of a wind turbine, wherein the method processes previously acquired data (DA), said data (DA) comprising data sets (DS) for a plurality of wind turbines, where each data set (DS) comprises respective values of variables, the variables including one or more turbine variables (T1, T2, …, T10), one or more weather variables (W1, W2, …, W4) and one or more damage variables (D1, D2, …, D19), wherein the method comprises the following steps: a) discretizing the values of those variables which are numerical variables, resulting in modified data sets (DS'); b) structure learning of a plurality of Bayesian networks (BN1, BN2 …, BNN) based on the modified data sets (DS'), where each Bayesian network (BN1, BN2 …, BNN) is learned20 by another learning method; c) determining an optimum Bayesian network (OBN) out of the plurality of Bayesian networks (BN1, BN2 …, BNN) based on a performance measure (PM) reflecting the prediction quality of a respective Bayesian network (BN1, BN2 …, BNN), where the optimum Bayesian network (OBN) has the best performance measure (PM); d) parameter learning of the optimum Bayesian network (OBN) based on the modified data sets (DS'), resulting in conditional probabilities (CP), where the optimum Bayesian network (OBN) in combination with the conditional probabilities (CP) is the prediction model.