本文選題：故障診斷　切入點：能量信號　出處：《華南理工大學》2014年博士論文

【摘要】：開展創(chuàng)新性的機械故障診斷技術研究，對于提升設備的安全穩(wěn)定運行品質(zhì)具有重要意義。本文基于能量視角，以論證齒輪傳動系統(tǒng)振動與輸入能量之間的相關性為開篇，挖掘潛藏于能量信號之中的故障模式規(guī)律，在研究有效的能量信號非線性處理、特征提取以及故障模式模糊識別等方法的過程中，建立一種新型的面向齒輪傳動系統(tǒng)等旋轉(zhuǎn)機械的故障診斷方法。 首先，對齒輪傳動系統(tǒng)的振動機理進行了研究，通過對齒輪靜態(tài)傳遞誤差變量的理論分析，揭示了輸入瞬時能量與齒輪傳動產(chǎn)生的振動位移偏差量之間的映射關系。同時采用頻域相干分析方法對故障時的輸入功率和振動信號的相干性做了實驗分析，證明二者之間具有強相關性、能量信號也可表征故障信息。這些論證性研究為后續(xù)工作的合理性提供了理論依據(jù)。 其次，研究了HHT的改進方法，建立了用于抑制HHT中端點效應的PSO-ARMA波形延拓預測模型。模型建模時首先提出并研究了基于粒子熵的參數(shù)自適應變異粒子群算法(EPPSO算法)，再將其運用到ARMA模型的參數(shù)優(yōu)化估計中，依據(jù)矩估計法得到的初值在參數(shù)解空間內(nèi)全局搜索，最終得到ARMA模型的最佳參數(shù)。同時應用該模型進行了端點效應抑制仿真，結(jié)果顯示EMD分解后各IMF的波形完整度較高、有效降低了譜線能量泄漏。模型為后續(xù)準確提取系統(tǒng)輸入能量信號的時頻特征提供了技術支撐。 再次，研究了齒輪傳動系統(tǒng)故障特征提取方法。開展了能量信號HHT分析實驗研究：分別針對正常和斷齒狀態(tài)下的能量信號進行EMD分解，并對振動和能量信號的Hilbert譜和邊際譜對比分析，驗證了能量信號分析結(jié)果在表征故障狀態(tài)方面的優(yōu)越性。隨之研究了故障特征向量庫建立方法，構(gòu)建了含前6階IMF的歸一能量、偏度、峰度、標準差和近似熵等參數(shù)在內(nèi)的多維故障特征向量，為后續(xù)開展故障特征識別方法研究奠定基礎。 然后，提出并研究了一種新穎的基于核主元熵模糊聚類的故障識別方法，建立了相應的KEFKM模型。模型涵蓋3大功能：KPCA特征參數(shù)降維；基于核主元熵的核特征模糊聚類；基于模糊關聯(lián)熵的故障模糊識別。1)首先研究了故障數(shù)據(jù)模糊聚類方法，，提出了主元信息熵的概念，建立了將其和FKM算法相融合的新模糊聚類方法：運用KPCA對數(shù)據(jù)降維以減少運算量，運用核密度估計和最大熵原理，先對第一核主元數(shù)據(jù)聚類以獲取最佳分類數(shù)和初始聚類中心，再針對核主元特征進行聚類。實驗結(jié)果顯示該方法可顯著提高故障數(shù)據(jù)的聚類效果。2)接著提出了針對待檢樣本的基于模糊關聯(lián)熵的故障模糊識別規(guī)則。通過實驗對比分析了最大貼近度和模糊關聯(lián)熵方法之間的差異，指出模糊關聯(lián)熵可以面向數(shù)據(jù)整體來衡量待識別數(shù)據(jù)的分布特點，對判斷兩個模糊子集的相似度具有顯著作用。3)最后應用KEFKM模型開展了齒輪實驗研究，結(jié)果證明模型可以保證訓練樣本的顯著核主元經(jīng)模糊聚類后可形成類間、類內(nèi)分布均合理的樣本空間；基于模糊關聯(lián)熵的故障模式模糊識別方法(規(guī)則)在處理故障樣本模式識別方面表現(xiàn)優(yōu)秀，驗證了KEFKM模型的有效性。 最后，圍繞齒輪傳動系統(tǒng)能量信號監(jiān)測與故障診斷系統(tǒng)的設計展開了研究。分析了系統(tǒng)的基本結(jié)構(gòu)及信息獲取、處理等環(huán)節(jié)的實現(xiàn)過程，初步開發(fā)了故障診斷系統(tǒng)，系統(tǒng)基于虛擬儀器技術，通過內(nèi)嵌Matlab實現(xiàn)能量信號分析、故障模式模糊識別等步驟，最終實現(xiàn)故障狀態(tài)實時監(jiān)測。同時研究了無線傳感網(wǎng)絡技術在齒輪傳動監(jiān)測中的應用，設計了基于WSN的狀態(tài)監(jiān)測節(jié)點，通過優(yōu)化Zigbee協(xié)議棧，實現(xiàn)了節(jié)點自組網(wǎng)絡，可面向齒輪監(jiān)測實現(xiàn)安全、便捷的遠程數(shù)據(jù)采集。
[Abstract]:Research on the mechanical fault diagnosis technology innovation, has important significance for the safe and stable operation to improve the quality of equipment. Based on the energy point of view, to demonstrate correlation between gear transmission system vibration and energy input for the opening, mining patterns hidden in fault signal energy, in the process of nonlinear signal energy effective research, process characteristics extraction and fault mode fuzzy recognition method, the fault diagnosis method for the establishment of a new type of gear transmission system of rotating machinery.
First of all, the mechanism of vibration of gear transmission system are studied by the static transmission error of gear variable theory analysis, reveals the mapping relationship between the vibration displacement deviation and input instantaneous energy generated by gear transmission. At the same time using frequency coherent coherence analysis method for input power and the fault vibration signal of the experimental analysis has been done and there is a strong correlation between the two, the energy signal can also characterize the fault information. These arguments for subsequent research on the rationality of the work provides a theoretical basis.
Secondly, the improved method of HHT, for the PSO-ARMA waveform to inhibit the end effects of HHT in the extended forecasting model is established. First proposed and studied the adaptive mutation particle swarm algorithm based on Entropy Modeling (EPPSO algorithm), and then apply it to the ARMA model parameter estimation, on the basis of moment estimation method to get the initial value in the solution space of global search parameters, and ultimately get the best parameters of the ARMA model. At the same time to restrain the end effect simulation using the model, the results show that EMD decomposition after each IMF waveform integrity is high, effectively reducing the spectrum energy leakage. The model provides the technical support for the subsequent accurate extraction of time-frequency characteristics system input signal.
Again, the gear fault feature extraction method of the transmission system. The experiment was carried out to study the energy signal HHT analysis respectively according to the energy signal of normal and broken teeth under the condition of EMD decomposition, and the vibration and energy signal Hilbert spectrum and marginal spectrum analysis, signal analysis results can verify the superiority in the characterization of fault state the. Then the paper studies the method of building fault feature vector library, constructed by the first 6 order IMF normalized energy, skewness, kurtosis, standard deviation and approximate entropy and other parameters, the multi fault feature vector, which lays the foundation for the follow-up research methods of avoidance feature recognition.
Then, this paper presents a novel kernel principal component entropy fuzzy clustering based on fault identification method, established the corresponding KEFKM model. The model covers 3 functions: KPCA feature dimension reduction; kernel feature kernel principal component entropy fuzzy clustering based on fuzzy.1; fault recognition based on Fuzzy Association entropy) first study on the fault data of fuzzy clustering method, put forward the concept of the main element of information entropy, a new method combining the fuzzy clustering and FKM algorithm: the use of KPCA to reduce the dimension of data in order to reduce the amount of computation, using kernel density estimation and maximum entropy principle, the first nuclear main metadata clustering in order to obtain the best classification number and the initial cluster center, then clustering for kernel principal component characteristics. Experimental results show that this method can significantly improve the clustering effect of.2 fault data) and then put forward the fuzzy entropy for fault fuzzy association based on the sample to be detected The rules of recognition. Comparing to the difference between fuzzy association degree and entropy method, pointed out the distribution characteristics of fuzzy entropy can be measured to identify the overall data oriented data,.3 has a significant role to judge the similarity of two fuzzy subsets) should end with KEFKM model experiment were carried out to study the gear, the result shows that the model to ensure that the main element was nuclear training samples by fuzzy clustering analysis can be formed between classes, class distribution are reasonable sample space; fuzzy recognition method of fault pattern based on Entropy Fuzzy Association (rules) with excellent performance in dealing with pattern recognition of fault samples, verifies the effectiveness of the KEFKM model.
鏈

本文編號：1672633