MEL-FREQUENCY CEPSTRAL COEFFICIENTS (MFCC) FEATURE FOR PUMP ANOMALY DETECTION IN NOISY ENVIRONMENTS

Abstract

The continuity of a production process is supported by the availability of good assets. One of the efforts to support asset availability is through asset maintenance. One of the important assets in the industry is the pump. To detect anomalous conditions in the pump, the sound of the engine can be used. However, noisy environmental conditions can change the characteristics of the sound produced. This can have an impact on errors in identifying the condition of the machine. In this study, Mel Frequency Cepstral Coefficients (MFCC) is used, because the characteristics of MFCC are very attached to the sound signal and are appropriate for sound signals in the case of this noisy environment where the signal tends to be non-stationary. Support Vector Machine will be used as a method that maps input (machine features) and output (machine condition). In this study, a comparison of the use of combined features of time and frequency domains with time-frequency features (MFCC) will be carried out. Improved performance is obtained when the time-frequency domain acoustic feature in the form of MFCC is used with an average accuracy reaching 99.88% on the Medium Gaussian SVM model.

Keywords

• anomaly detection
• mel frequency cepstral coefficients
• pump
• noisy environment