Application of the Signal Spectrum-Based Image Pattern Recognition in the Vibration Process Analysis

Abstract

One of the aims of vibration signal analysis is the assessment of the condition of technical objects at each stage of operation based on a cause-and-effect relationship between the technical condition of the device and the features of the vibration signal available through the measurement experiment. The essential features of the signal (in the form of certain measures) are available through appropriate operations on the measurement signal. The problem arises when the measuring signal is not a fully deterministic one, it shows features common to deterministic and random signals. Consequently, some simple periodic signal measures such as spectrum frequency components introduce some level of uncertainty in the before mentioned cause-and-effect relationship. Thus, the assessment of the technical condition in the selected frequency components of the spectrum of the measuring signal may not be reliable. These difficulties can be overcome by using the frequency-frequency analysis of the diagnostic signal. The entire frequency spectrum is converted into an image in which the frequency components form characteristic patterns unique to the actual condition of the technical object. The final assessment of the condition is achieved through the identification of a unique image pattern by comparison with reference image patterns created with a generation model.