Machine condition monitoring for consistent paper production

Abstract

This paper focuses upon the press section of a paper machine and describes a neural network vibration-based condition monitoring system for providing advance warning of faults in the felts. These are extremely important components of the paper machine as they are responsible for guiding the paper through the press nips. Any deterioration in the felts may result in paper breakage and machine shut-downs with a resulting fall in productivity. The need for tighter control of the manufacturing process is thus the main motivation for this work. The system developed makes use of spectral analysis of the raw vibration signals. This involves the technique of acceleration enveloping for isolating the spectral components that reflect the process defects. Since visual examination of the spectra for detecting subtle changes in the early stages is both difficult and impractical, a peak detection algorithm has been designed. Its purpose is to extract the relevant features from each spectrum. These are then fed into a neural network classifier for discriminating between the fault and the no-fault conditions. The system is demonstrated to be successful for detecting felt deterioration in its early stages when assessed on industrial data. Other benefits include the ease of on-line implementation and its possible extension to a wide range of mechanical and rotating equipment.