Research On Fault Diagnosis System Of Internal Combustion Engine Based On EMD, Correlation Dimension And ANN

Internal Combustion Engine is one of the most popular power machinesin China. It is fatal to the whole system's security and reliability whether theengine is normal or not. So how to carry on fault diagnosis of internalcombustion engine fleetly and without disassembly has vital significance inpractical production.Based on the NI Company's virtual instrument design tools, the softwareand hardware of the DAQ system was constructed to acquire the vibrationsignal of the cylinder-head and engine valve, the LabVIEW programs ofEmpirical Mode Decomposition (EMD) and calculating correlationdimensions of the signal were designed. And the artificial neural network(ANN) model was established to classify the faults based on Matlab. Thefault diagnosis system was constructed and designed for inspecting thestatus and fault diagnosis of internal combustion engine based on EMD,Correlation dimension and ANN. The diagnosis experiment was done with490BPG model engine when it worked at three statuses: one cylinder misfire,leaking and the abnormal clearance range without disassembly. The mainconclusions are as follows.1) The software and hardware of the fault diagnosis system wasconstructed and designed for inspecting the status and fault diagnosis ofinternal combustion engine based on EMD, Correlation dimension and ANN.the system is easy to be reconfigured and constituted with the Data acquiremodule including Fault eigenvalue defining module and the Fault classifiedModule, and the hardware module including CA-YD-106 accelerationsensor, photoelectric sensor, YE5853A charge amplifier, PCI-6133DAQcard and PC.2) The misfire and valve leaking fault diagnosis experiment was donewithout disassembly and the cylinder-head vibration signal was analyzed.The results indicate that:①Cylinder-head vibration signal has the fractal character and themisfire and valve leaking faults can be classified by taking the correlation dimension as faults eigenvalue. The correlation dimension is highest whenthe cylinder is normal, the next is when the valve leaks gas and the lowest iswhen the cylinder misfires.②The correlation dimension increases as the increase of the rotatevelocity of the engine, which indicates that the source of the cylinder-headvibration is more complex.③The misfire and valve leaking faults can be classified rapidly andcorrectly via the correlation dimension database of the typical engine statuesuch as idle with no load in the production.3) The clearance fault of engine valve experiment was done withoutdisassembly and the valve vibration signal was analyzed. The resultsindicate that:①Engine valve vibration signal has the fractal character, but thedifference of correlation dimension in different statues is too unapparent toclassify the engine valve clearance statue via the changing tendency diagramof correlation dimension.②For this case, the valve vibration signal was decomposed based onEMD method, and the correlation dimensions of the top four intrinsic modefunctions(IMF1～IMF4) was calculated. The correlation dimensions of theIMF1～IMF4 were taken as the input parameters of the artificial neuralnetwork, the network model was trained with 80 training samples of fourwork status (the clearance is small, normal, big and biggest), it indicates thatthe test results of 20 test samples of the experiment conforms to the practicalstatus and the correct accuracy was 100%.4) The experiment result indicates that the system can be online toinspect and diagnose the faults of misfire, leaking and engine valveclearance rapidly and correctly without disassembly.