Knock Detection Based On The Variational Mode Decomposition And Higher-order Cumulant

In order to improve performance of engine,the down-sizing of engine develops quickly,which leads to the shape increasing of knock in engine.Strong knock will reduces the performance and useful of engine greatly.However,slight knock will results in the combustion in cylinders close to constant volume combustion,which is helpful in improving the combustion in engine.So it's necessary to detect the engine knock and make engine working in slight knock condition.This has important theoretical significance and application value for the engine field.Because of its low-cost and convenience,detecting knock by extracting block vibration features is widely used.In this paper,when bench test is carried out,the pressure signals in cylinders and block vibration signals are collected and the signals are analysed deeply.There is so much noise and complex information in block vibration signals that accurate algorithms are required to processing the signals in order to extract knock features.After analyzing traditional algorithms,the Variational Mode Decomposition(VMD)algorithm is introduced in the field of knock detection in this paper.There are various control parameters in the VMD need to be set up,such as the number of modes K and the balancing parameter of the data-fidelity constraint ? are both important for detect knock in vibration signals.For the number of modes K,two methods are proposed in this paper.The first one is that using the maximum and minimum of the difference between modes' center frequencies,an optimized VMD algorithm which can obtain the K adaptively is acquired.The second one is that with analyzing of knock mechanism,a simplified VMD algorithm whose initial center frequencies are optimized is acquired.The first one has better adaptability and the second one has higher computational efficiency.Both of them show ideal results and get narrowband signals,which make the knock features more obvious and knock in different intensities can be distinguished more easily.The data-fidelity constraint ? is optimized by the amount of information in modes,that is also helpful in accuracy.In order to distinguish knock in different intensities,the modes decomposed by optimized VMD algorithm are restructured.Reconstructed signals' natural logarithm of the four-order cumulant are used as the knock intensity index.Results show the index can distinguish knock in different intensities correctly which is close to the results obtained by MAPO.In this paper,the knock intensity index is also corrected in slight knock to make the index has more practical significance for precise control.