Operation Monitoring of Rotating Rolls

Operation monitoring of paper machinery is financially essential due to improved quality of the end product and increased machine lifetime. Rotating cylinders in rolling contact are essential parts when manipulating paper web. Roll stability is becoming more and more important since the paper machine speeds are still increasing. Dynamical instability of rolls leads to contact vibrations in roll pairs, which in turn cause unexpected variations in line load and deteriorates the quality of the product. Traditionally the vibration based operation monitoring of rotating rolls is carried out by spectrumbased signal processing methods, by synchronous time averaging method and by calculating statistical key-values like root mean square-value or peak-value from the vibration signal. This paper approaches the operation monitoring of rotating rolls by neural network method. The goal of this paper is to examine the applicability of Learning Vector Quantization neural computing algorithm to operation monitoring of rotating rolls in line contact. The paper first shortly describes the traditional operation monitoring methods of rotating rolls. After that the paper comprises the principle of the Learning Vector Quantization method and how it can be applied in operation monitoring. The experimental part is carried out in a pilot roll unit in laboratory environment by running the unit in different running situations. The roll unit consists of two rotating rolls from which one is hard roll and the other is covered by soft polymer cover. The tested running situations are normal running, nip resonance caused by the delay phenomenon of the cover and dirt particles on the surface of rolls. The raw vibration signal is transformed into feature vectors by a feature extraction procedure implemented by signal processing software. This process is also described in the paper. Finally in the end the measurement results are shown in spectrum and in time signal form and are compared to the results carried out by the neural network method. The validity of the Learning Vector Quantization method is evaluated by confusion matrix method. As conclusion, based on the present study the neural network method appears a promising method for determining the running situation of rotating rolls in line contact.