外文資料翻譯---基于自聯(lián)想神經(jīng)網(wǎng)絡(luò)的發(fā)動機(jī)控制系統(tǒng)傳感器故障診斷與重構(gòu)-資料下載頁
2025-05-12 04:50本頁面
【導(dǎo)讀】附件：1、外文資料翻譯譯文；2、外文原文。自聯(lián)想網(wǎng)絡(luò)關(guān)鍵在于特征提取和噪聲濾波。綜合自聯(lián)想網(wǎng)絡(luò)的最優(yōu)估計與故障診斷，自。動區(qū)分估計誤差和傳感器故障。仿真結(jié)果表明這種方法不需要模型，能診斷傳感器硬、軟故障，當(dāng)發(fā)動機(jī)性能退化時也能提供很好的解析余度。傳感器的故障診斷和重建是充分實現(xiàn)發(fā)動機(jī)控制系統(tǒng)的可靠性所必須的。求給一個故障診斷系統(tǒng)的設(shè)計提出了挑戰(zhàn)。通常情況下，這種關(guān)系可以被描述為用傳感器測量值。NLPCA是一個以最小信息損。失為代價，從而將多維非線性數(shù)據(jù)映射到低維數(shù)據(jù)的技術(shù)。來進(jìn)行測量，E由少量的涉及噪聲或次要變量的成分。射層和輸出層被映射到輸出層，并且重建輸入數(shù)據(jù)。rij反映了xi和xj的獨立關(guān)系。如果傳感器測量值及其估計值之間的差異超過閾值，而其他傳感。將被錯誤地警告，導(dǎo)致不正確的故障調(diào)節(jié)。組成的閉環(huán)控制系統(tǒng)。
【正文】 sor fault, gas path fault or estimation error. Measurements such as speed, flow,temperature and pressure will vary because of the gas turbine engine ponent faults or performance degeneration. Gas path analysis, which calculate the fault coefficient matrix , can distinguish them from sensor failures. An axial directional fault signature of sensor fault occurs whenever one ponent of the error vector bees large and all the other vector ponents remain small. (3)If difference is caused by performance degeneration, then AANN is pensated online until the difference is eliminated. (4) If the difference is caused by sensor fault,then the failed sensor is cut off and replaced with the estimation of this example, the training set is prised of 84 points at various power levels, Mach numbers and altitudes which include various steady state operating points. Zero mean and normal distributed noise is added to the training data for input(not the target values) to make the work learn the correlation among the data. Also the loop is restarted several times during the training to use different noise values in each time and to change the order of the data in the training set to avoid learning any geometries that would occur due to the specific location of the data in the training is the result of soft fault detection of engine. The soft fault is initiated by setting ng sensor output increasing slowly at speed of % per second. Fault signature and 3level threshold ( maximum tolerance( MT), isolation threshold ( IT ) and fault threshold( FT )) are adopted to detect sensor faults. There may be a soft fault of ng sensor or performance degeneration occurring when the fault control gain gn is larger than 0 at the 3 the error between evaluation and sensor out put is larger than IT( 330r/ min), a probable fault is isolated. When the error is larger than FT(1350r/min), the fault control gain is 1 and the fault control gains of other sensors remain small, a fault is declared to be detected. The residual will bee small again when the sensor fault is successfully detected and acmodated. When a fault is declared, acmodation goes smoothly from sensor output to work estimation. In the whole process, in spite of the deviation of sensor output and ANN evaluation, the ng acmodation value which will be used by controller deviates little, so the engine can operate normally. Soft fault detection of AANN 5 Conclusion ANN is a kind of feed forward neural work which has special topology architecture. It has excellent capability of feature extraction and noise filtering. An example of sensor fault detection and reconstruction of engine control system using AANN is presented. Simulation results show that analytical redundancy based on AANN uses only engine sensor outputs to train AANN and does not need engine model. The integrated logic of optimal estimation of ANN and sensor fault diagnosis is developed to distinguish optimal estimation error from sensor faults. This logic can avoid the problem of ANN damaged by sensor failure, and also can avoid the problem of false diagnosis by estimation error Control system will work normally even there are sensor faults. References [1] Huang X H,Sun J G. Engine sensor fault diagnosis using main and decentralized neural works[J]. Chinese Journal of A eronautics, 1998, 11(4): 293 296. [2] Kramer M A. Nonlinear principal ponent analysis using auto associative neural works[J]. AICH E Journal, 1991,37(2): 233 243. [3] Kramer M A. Auto associative neural works[J]. Computers Chem. Engng, 1992, 16(4): 313 328. [4] Schenker B. Cross –validated structure selection for neural works[J]. Computers Chem Engng, 1996, 20(2): 175186. [5] 黃向華 . 發(fā)動機(jī)數(shù)控系統(tǒng)智能解析余度技術(shù) [D]. 南京 : 南京航空航天大學(xué) , 1998. Huang X H. Ananytical redundancy of engine control system[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 1998.(in Chinese) [6] 嚴(yán)寒松 . 航空發(fā)動機(jī)故障診斷 [D]. 南京 : 南京航空航天大學(xué) , 1996. Yan H S. Aeroengine fault diagnosis[D]. Nanjing :Nanjing University of Aeronautics and Astronautics, 1996.(in Chinese)
畢業(yè)設(shè)計相關(guān)推薦
鄂ICP備17016276號-1