自組織映射圖在風機預兆式健康管理上的應用研究

Abstract

本研究以自組織映射圖(Self-Organizing Map, SOM)方法為核心建立一預兆式健康管理技術對風力發電機資料進行預兆式診斷，預兆式診斷技術的流程，包含資料處理、特徵擷取、健康診斷及未來預測。本研究以風機正常運作狀況的規範將異常狀況下的資料進行過濾，而後以專家經驗進行特徵擷取，篩選出對風機健康診斷較有意義的特徵變數，並以主成分分析(Principal Component Analysis, PCA)方法降低特徵變數維度，再來以自組織映射圖方法，結合最小量化誤差(Minimum Quantization Error, MQE)，進行風機資料健康診斷，最後以自迴歸移動平均(Autoregressive Moving Average, ARMA)模型對風機做未來健康狀況的預測。研究成果為對風機SCADA資料訂立一健康指標MQE值，並且設立一閾值來評斷風機是否健康，若MQE值高於此閾值，則視為不健康狀態；對風機聲音資料，能藉由風機運轉時所發出的聲音診斷出葉片是否有問題及其他異常問題；對風機溫度資料，能診斷出溫度可能有出現異常狀況，需要進行維修檢查。

The study builds a prognostic and health management process with self-organizing map method to analyze the wind turbine data. The process of prognostic and health management includes “Data Processing”, “Feature Extraction”, ”Health Assessment”, and ”Performance Prediction”. The data processing excludes the unusual data according to the normal operating standard. The feature extracting extracts the well features by professional experience and decreases the orders of the data by principal component analysis. The Self-Organizing Map is used to analyze the processed data and Minimum Quantization Error as the health index of the wind turbines is set. Finally, the future health tendency of the wind turbine is predicted by autoregressive moving average model. The analysis set a health index MQE and a threshold value from the SCADA data of wind turbine. The voice data from turbine blades and temperature data from nacelle can used to detect the abnormal operation of the wind turbine. The prognostic and health management process can be used to predict the unnormal operations of the wind turbine.