Analysis and discrimination through neuronal and PSD approximations of the vibrational monitoring of an impulsion system

  1. Javier Castilla Gutiérrez
Supervised by:
  1. Juan Carlos Fortes Garrido Director
  2. José Miguel Dávila Martín Director

Defence university: Universidad de Huelva

Year of defence: 2021

  1. María Luisa de la Torre Sánchez Chair
  2. Joaquín Bernal Méndez Secretary
  3. Tomás E. García Suárez Committee member

Type: Thesis


One of the main problems industries in general are facing is the prediction of errors or failures of the most sensitive or critical assets of their equipment. For this reason, techniques based on predictive maintenance are emerging, searching for the early detection of possible cracks, imperfections or defects that are likely to cause an accident or the partial or complete shutdown of a plant. This not only increases safety, but also improves the industry's profitability. Bearings are one of the most important mechanical elements in industrial production equipment and systems. This study characterizes the 15-year operating scheme of a set of bearings in a real mechanical system (a blower) by means of spectral analysis of the vibrations generated. We obtained the values over the years of every characteristic frequency of the bearings and the equipment itself in two sampling positions and in their typical axes. For data collection and we have followed the ISO 10816 standards, thus using the values of speed in RMS, aiming to reduce the masking of these signals that occurs depending on whether they are high or low frequencies. The study will respond to one of the most important requirements found in the predictive and preventive control of industrial sites. The problem of the predictive systems of maintenance of equipment with bearings lies in the number of monitoring and analysis points that generate a high cost in time and human resources. The result will be to determine which of all the study frequencies is the most significant and in which position and measurement axis has the biggest impact. To do this, we will analyses the rotation frequency of the blowing machine, the resulting frequency of all the frequencies, the frequency of the impulsion blades and finally the frequency of the bearing. With Artificial Neural Networks (ANNs) models, we estimated the most significant bearings, measurement points and variables in the changes of the maximum amplitude of the frequency spectra, a parameter that determines the feasibility and safety of the equipment. A sensitivity analysis concludes that the most significant frequency has been the one generated by rotation of the machine itself, followed by the second harmonics of the bearing ball frequencies and the frequency generated by the blades of the equipment. The study also shows that vertical position 4 is the most critical one, followed by axial position 4 caused by the action of the blades, and the bearing most important is the SKF6322 one.