基于有序決策樹的故障程度診斷研究
發(fā)布時(shí)間:2019-01-06 15:09
【摘要】:近幾年,人工智能,尤其是機(jī)器學(xué)習(xí)和模式識(shí)別技術(shù)大量應(yīng)用于設(shè)備狀態(tài)監(jiān)測(cè)和故障診斷領(lǐng)域,采用智能技術(shù)檢測(cè)和分析機(jī)械故障成為一種趨勢(shì)。在故障分析中,用戶除了需要知道某設(shè)備是否發(fā)生故障以及為何種故障外,還需獲得故障的嚴(yán)重信息,從而制定適當(dāng)?shù)木S修策略和檢修方案。 故障程度的智能檢測(cè)本質(zhì)上是有序分類問(wèn)題:將故障程度用一組有序整數(shù)n表示(n=1, 2, 3,…),表示‘輕微故障’、‘中等故障’‘嚴(yán)重故障’等。相比于經(jīng)典分類問(wèn)題,有序分類的類別號(hào)(即故障程度)之間存在大小關(guān)系。由于存在這種序的關(guān)系,在分類器設(shè)計(jì)原則上以及分類器性能評(píng)價(jià)準(zhǔn)則上都與經(jīng)典的模式分類問(wèn)題有所差別。 在故障程度分析中,當(dāng)表征故障嚴(yán)重性的特征值增加(或減少)時(shí),故障程度隨之增加,即故障特征與故障嚴(yán)重性之間存在單調(diào)性的約束。稱這種能表征故障程度的特征為單調(diào)故障特征。這種單調(diào)性故障特征為故障程度診斷提供簡(jiǎn)便實(shí)用的信息,可以根據(jù)這種特征來(lái)衡量故障程度。 本文從模式識(shí)別中的有序分類問(wèn)題出發(fā),研究了有序分類問(wèn)題的本質(zhì)和從數(shù)據(jù)中歸納學(xué)習(xí)有序分類模型的方法,提出2種有序分類規(guī)則學(xué)習(xí)算法,并將其應(yīng)用于機(jī)械故障嚴(yán)重程度建模。具體內(nèi)容如下: 首先,本文系統(tǒng)地介紹了模式識(shí)別與機(jī)器學(xué)習(xí)領(lǐng)域中的有序分類問(wèn)題,指出該問(wèn)題與經(jīng)典模式分類問(wèn)題相比較,發(fā)展歷史與研究深度相對(duì)簡(jiǎn)單,還有很大的研究空間。 其次,介紹了單調(diào)分類中特征與決策之間的單調(diào)性約束,F(xiàn)有分類器只有在數(shù)據(jù)集是單調(diào)一致時(shí)才能訓(xùn)練出單調(diào)的分類模型,而現(xiàn)實(shí)中非單調(diào)噪聲廣泛存在,單調(diào)一致數(shù)據(jù)集很難得到。本文引入一種衡量特征與決策之間隨機(jī)單調(diào)約束的指標(biāo),指出特征與決策是概率上的單調(diào)。介紹了有序信息熵模型,該模型繼承了經(jīng)典信息熵的魯棒性,且能夠反映特征與決策之間的隨機(jī)單調(diào)相關(guān)性。最后基于該理論構(gòu)造了基于有序信息熵的決策樹學(xué)習(xí)算法。 再次,針對(duì)部分特征與決策單調(diào),部分特征不單調(diào)的決策問(wèn)題,構(gòu)造了隨機(jī)有序混雜決策樹算法。首先用非單調(diào)特征分裂數(shù)據(jù),再用單調(diào)特征繼續(xù)細(xì)化。為測(cè)試本文提出方法的性能,分別在人工數(shù)據(jù)和標(biāo)準(zhǔn)數(shù)據(jù)上用本文算法與其他經(jīng)典有序分類算法進(jìn)行了比較。結(jié)果顯示,本文提出的兩種算法魯棒性好、泛化誤差小。 最后,將本文提出的算法應(yīng)用于實(shí)際故障診斷。本文應(yīng)用在齒輪裂縫故障程度監(jiān)測(cè)實(shí)驗(yàn)中。在實(shí)驗(yàn)中,用位移傳感器監(jiān)測(cè)齒輪箱的振動(dòng),由此反映齒輪的故障嚴(yán)重程度。為區(qū)分故障等級(jí),人為制造不同深度的齒輪裂縫。在不同負(fù)載和轉(zhuǎn)速情況下得到振動(dòng)數(shù)據(jù),然后在頻域與時(shí)域上提取了一系列特征得到故障數(shù)據(jù)。結(jié)果顯示,本文方法分類損失低,展示了該方法的有效性。
[Abstract]:In recent years, artificial intelligence, especially machine learning and pattern recognition, has been widely used in the field of equipment condition monitoring and fault diagnosis. In fault analysis, in addition to the need to know whether or not a certain equipment failure and what kind of failure, users also need to obtain the serious information of the fault, so as to formulate appropriate maintenance strategy and maintenance plan. Intelligent detection of fault degree is essentially an ordered classification problem: the degree of failure is represented by a set of ordered integers n (nn, 2, 3, 鈥,
本文編號(hào):2402956
[Abstract]:In recent years, artificial intelligence, especially machine learning and pattern recognition, has been widely used in the field of equipment condition monitoring and fault diagnosis. In fault analysis, in addition to the need to know whether or not a certain equipment failure and what kind of failure, users also need to obtain the serious information of the fault, so as to formulate appropriate maintenance strategy and maintenance plan. Intelligent detection of fault degree is essentially an ordered classification problem: the degree of failure is represented by a set of ordered integers n (nn, 2, 3, 鈥,
本文編號(hào):2402956
本文鏈接:http://sikaile.net/kejilunwen/jixiegongcheng/2402956.html
最近更新
教材專著