Application of Improved LMD, SVD Technique and RVMto Fault Diagnosis of Diesel Valve Trains

Targeting the non-stationary characteristics of diesel engine vibration signals and the limitations of singularvalue decomposition （SVD） technique, a new method based on improved local mean decomposition （LMD）,SVD technique and relevance vector machine （RVM） was proposed for the identification of diesel valve fault inthis study. Firstly, the vibration signals were acquired through the vibration sensors installed on the cylinder head inone normal state and four fault states of valve trains. Secondly, an improved LMD method was used to decomposethe non-stationary signals into a set of stationary product functions （PF）, from which the initial feature vector matricescan be formed automatically. Then, the singular values were obtained by applying the SVD technique to theinitial feature vector matrixes. Finally, slant binary tree and sort separability criterion were combined to determinethe structure of multi-class RVM, and the singular values were regarded as the fault feature vectors of RVM in theidentification of fault types of diesel valve clearance. The experimental results showed that the proposed fault diagnosismethod can effectively extract the features of diesel valve clearance and identify the diesel valve fault accurately.