Development of Wear Condition Diagnosis Model for Piston-Liner with Multisource Information

Abstract

As a crucial component of internal combustion engines, the piston-liner assembly operates under high temperatures and pressures, making it susceptible to failures. To ensure the engine’s normal operation and maintenance, it is essential to conduct condition monitoring of the piston-liner. In this study, the variational mode decomposition algorithm was applied by identifying key parameters based on the characteristics of the block vibration. The algorithm decomposed the block vibration into 6 intrinsic mode functions (IMFs). Continuous wavelet transform was employed for time-frequency analysis of the block vibration. The results of the time-frequency analysis revealed a close correlation between IMF1 and combustion, as well as IMF6 and piston slap. Building upon this finding, multiple evaluation criteria were utilized to determine the characterization parameters of the IMFs associated with combustion and piston slap. Subsequently, a support vector machine model was established by confirming the input vectors, constructing training and test sets, and selecting an appropriated kernel function. To optimize the support vector machine model, a genetic algorithm was employed to fine-tune the key parameters of penalty factor and the width parameter of kernel function. The optimized support vector machine model was then trained and tested. The diagnosis model achieved a remarkable classification accuracy of 97.9%, meeting the requirements for piston-liner monitoring effectively.

Keywords: piston-liner, Fault Diagnosis, characterization parameter extraction, Support vector machine, Genetic Algorithm