Research On Fault Diagnosis System Of Engine Based On Hilbert-Huang Transform

As power plant, engines are playing an increasingly important role in industry. However, abnormal engines will bring danger to operators, so it is particularly vital for fault diagnosis before accidents happen.In this paper, a vibration signal acquisition system of cylinder cover was established based on Lab VIEW, a virtual instrument platform by NI, and programs for Hilbert-Huang transform(HHT) as well as least squares vector machine (LS-SVM) was developed in Matlab. Furthermore, models were built for engine fault diagnosis based on HHT, and an intact engine fault diagnosis system was developed based on HHT and LS-SVM. An undetached diesel engine 295 was used for diagnosis of single cylinder misfire, valve leakage and early injection, and the main experimental results are as follows.(1)HHT theory was studied and employed in processing unstable vibration signal of engine cylinder cover. EMD is capable of decomposing original signals adaptively, and the resulting individual intrinsic mode function highlights the different local features of original signals. HHT can clearly tell the frequency distribution and its probability.(2)Hardware for engine fault diagnosis system was established, where control system FC2000, piezoelectric acceleration sensor CA-YD-106, charge amplifier YE5853A, data acquisition card PCI-6040E and connection box SCB-68 were utilized.(3)Signal measurement system for engine fault diagnosis was set up based on Lab VIEW. This system is composed of four parts, namely, title bar, control bar, parameter setting bar and figure display bar. Easy and flexible to use, this system can realize collecting engine vibration signals in a fast, continuous and precise way, and the signals are saved into the computer.(4)Cylinder cover vibration signals in different engine condition were analyzed based on HHT method, and marginal spectra which can reflect frequency information of vibration signals were obtained. Feature information was extracted from the marginal spectra from 1000 Hz to 2000 Hz, and a primary discrimination model was built based on least squares vector machine although the accurate recognition rate was low. There are two important parameters in LS-SVM, the adjustable punitive parameters of wrong discrimination gam and initial parameters of RBF function sig2. These two parameters were optimized by cross validation, and the optimized model output more accurate results in discriminating engine condition.(5)Algorithms of HHT and LS-SVM was developed in Matlab and a diagnosis system for normal operation, single cylinder misfire, valve leakage and early injection was established. The following conclusions were reached.When there was no load and the speed was 800r/min, it was indicated that the model could precisely discriminate different engine conditions of normal operation, single cylinder misfire and valve leakage, and the accuracy was 96.67%.When there was no load and the speed was 1200r/min, it was indicated that the model could precisely discriminate different engine conditions of normal operation, single cylinder misfire and valve leakage, and the accuracy was 96.67%.When the load was 25N-M and the speed was 800r/min, it was indicated that the model could precisely discriminate different engine conditions of normal operation, valve leakage and early injection, and the accuracy was 96.67%.When the load was 25N-M and the speed was 1200r/min, it was indicated that the model could precisely discriminate different engine conditions of normal operation, valve leakage and early injection, and the accuracy was 93.33%.(6)The results showed that the proposed system can efficiently diagnosis engine faults of misfire, leakage and early injection in undetached way.