Development of a diagnostic system using LPC/cepstrum analysis in machine vibration

Based on the linear predictive coding (LPC) and cepstrum analysis coefficients (LPC/Cepstrum), the implementation of LPC procedure to detect faults in engine assembly line diagnostics systems is proposed. In addition, with the purpose to gain more accurate results in the implementation of LPC procedure instead the use of Euclidean distance, Mahalanobis distance is applied. In order to prove the effectiveness of the procedure, we retrieved data sets obtained from diesel engines under normal and abnormal conditions with different RPMs. During experiments, it was possible to detect the system status and it also was proved that the Mahalanobis distance overcomes the limitations of Euclidean distance where the statistical distribution of the variables are taken into account. In this way, we achieved a clear and accurate diagnosis of the system even when the differences between system status were not possible to detect through conventional methods. It was proved that the consistent patterns of each system status can be established and also possible to determine the system status successfully. The differences and advantages of LPC/Cepstrum approach compared with common analysis are discussed in this paper.