本研究計畫主要目標在於運用希爾伯特-黃變換(HHT)之瞬時頻率概念結合無因次單位頻率(dimensionless frequency)計算來檢測與診斷旋轉機械在變轉速情況下的各部件損傷、瑕疵以及異常情況之可行性。由於旋轉機械各種動態行為之特徵與主軸之轉速密切相關，當旋轉機械在轉速變動情形下，其振動訊號將呈現高度非穩態情形，如此增加了分析之困難度。為了解決這類問題，同時改善傳統階次追蹤(order tracking)技術，本研究計畫將提出一個結合無因次單位頻率變換與希爾伯特-黃變換時頻分析之新方法。在主軸轉速變動之情形下，先經由經驗模態分解(EMD)或是集成經驗模態分解方法(EEMD)，將複雜振動訊號解構成若干個內稟模態函數(IMFs)，而每個內稟模態函數代表一個特定頻寬之單一震盪的訊號分量。至此，各內稟模態函數之瞬時頻率可被清楚定義並計算，並進而計算其無因次單位頻率。因此，非穩態之振動訊號將轉變為穩態，且主軸轉速變動的因子也將被移除。本研究計畫將設置一實驗測試平台，藉由實驗模擬旋轉機械在變轉速的運轉狀況下各項部件損傷以及運轉異常，例如:滾珠軸承的內、外圈損傷；軸承之滾珠損傷；軸承保持架故障；部件鬆脫；齒輪之齒磨損、齒斷裂；轉子不平衡與軸不共線等。擷取不同故障情形下運轉之機械振動訊號並進行分析，將結果呈現於無因次單位時頻分佈、邊際無因次單位頻譜、與無因次單位包絡譜等。運用不同之分析方法，可識別並提取各類旋轉機械之故障特徵。對於旋轉機械在變轉速的運轉情形下發生的部件損傷與運轉異常的故障程度，亦可對頻譜上之幅值、訊號振幅調制之大小和內波頻率調制的大小等故障特徵進行量化，來進行程度上之診斷。透過本研究計畫所提之方法，其結果將可提供更具物理意義之洞察分析。把經由本研究計畫所提之方法抽取出的各種故障特徵予以量化以形成特徵向量。運用不同的距離量測方法進行數據間之相似度分類，將可把代表各類故障之特徵向量分類，進而診斷識別其所對應之旋轉機械的故障類型與程度。人工智慧方法，例如類神經網路與模糊邏輯，亦將用來對不同類別與程度之故障情形所對應的特徵向量進行分類。透過這些分類過程，旋轉機械在變轉速情況下發生的各種部件損傷與運行異常等故障，將可以整合成一知識庫，並提供工業應用上之維修保養建議指標。The objective of this research proposal is to investigate the feasibility of utilizing the instantaneous frequency concept of Hilbert-Huang Transform (HHT) technique combined with dimensionless frequency computation to diagnose the defects and malfunctions of rotating machinery under variable speed operation conditions. Since the characteristics of rotating machinery is strongly governed by the shaft speed, the vibration signals of variable speed rotating machinery present high degree of non-stationarity and thus increase the difficulties of analysis. In order to resolve this problem and improve the traditional order tracking techniques, a novel approach combining the dimensionless frequency transform and HHT time-frequency analysis is proposed. The complicated vibration signals of rotating machinery under variable speed conditions are decomposed by the Empirical Mode Decomposition (EMD) or Ensemble Empirical Mode Decomposition (EEMD) into numbers of Intrinsic Mode Functions (IMFs) which represent the mono-oscillation components of certain frequency bands. The calculation of instantaneous frequency is then employed to obtain the dimensionless frequencies of IMFs, so that the non-stationary vibration signals are transformed to stationary ones and the influence of variable rotating speed is removed. The test rig will be installed in this research proposal to experimentally simulate versatile defects and malfunctions of variable speed rotating machinery, such as inner and outer races fault of roller bearing, ball defect of bearing, cage damage in bearing, component looseness, worn teeth of gear, broken teeth of gear, rotor unbalance and shaft misalignment. The signals captured from the experiment with different operation conditions will be analyzed by means of dimensionless time-frequency distributions, marginal dimensionless frequency spectra, and dimensionless envelope analysis. The faulty features will be identified and extracted by different analysis methods. The levels of defects and malfunctions in the variable speed rotating machinery will be diagnosed through quantifying the fault characteristics of the magnitude in spectra, the amplitude modulation (AM) as well as the intrawave frequency modulation (FM) phenomena. Through the proposed approaches, the analysis results will provide more meaningful physical insights. The different fault features of the rotating machinery that are extracted by the proposed methods will be built to form the characteristic vectors. The similarity classification will be utilized to distinguish the characteristic vectors as well as the corresponding fault types through different distance determinations. The artificial intelligence methods such as artificial neural networks and fuzzy logic, will be also employed to classified the characteristic vectors of different faults. Through such classification processes, the fault diagnosis techniques of variable speed rotating machinery will be integrated into a knowledge database and provide the maintenance suggestion for industrial indication. 研究期間：10008 ~ 10107
