Fault detection system using acoustic particle velocity in noisy environments based on kurtosis

摘要

The assessment of noise and vibration is fundamentally important for the fault diagnosis of machinery. Microphone-based solutions offer limited performance due to the presence of high levels of background noise in most industrial measurement scenarios. Furthermore, the use of accelerometers is not viable for applications where it is not possible to attach a sensor to the object's surface. Laser vibrometry or other alternative non-contact measurement methods are not recommended when an affordable, flexible solution is required. In contrast, acoustic particle velocity sensors offer key advantages in such scenarios. They provide a better signal to noise ratio when the measurements are performed close to the radiating surface, due to the vector nature of particle velocity, intrinsic dependency upon surface displacement and sensor directivity. The background noise contribution can be minimized by measuring as close as possible to the target sound source. In such cases, the measured acoustic particle velocity becomes proportional to the surface vibration. This paper introduces the use of particle velocity transducers for fault detection moving and rotating parts in a noisy environment. The detection algorithm relies on the assumption that a high proportion of faults are impulsive sounds or clicks. An algorithm based on the Mel-Frequency Cepstral Coefficients (MFCC) and Kurtosis features extracted from the vibration signal, for detection and classification of the impulsive noise, is proposed. The classification is performed using Gaussian Mixture Models (GMM) and assumes that the vibration signal can be masked by cyclostationary noise. In addition, autoregressive models are used to increase overall performance. The experimental results obtained show that the detection system can successfully operate under weak signal to noise ratios and in the presence of background noise. This provides key evidence for the potential of particle velocity-based solutions for fault detection in end of line control applications.