Monitoring of shaft backlash using vibration analysis and intelligent clasiffication systems

Abstract

Usually, the industry focuses on seeking good quality and productivity of its service or product, so maintenance tasks play a relevant role. Currently, the interest in knowing the status of systems in real time and their connection with the different areas of the industry is growing, which has been called Maintenance 4.0. One of the objectives of this type of maintenance is the detection of problems or defects during the operation of the machine, which requires prior investigation. In particular, mechanical looseness is a very common defect in rotating machinery that can cause serious problems. Detecting defects in rotating machinery through condition monitoring by performing vibration analysis is becoming more and more common. The premature detection of play during its operation allows avoiding catastrophic failures in the machine, stopping it only when it is essential to solve the problem. In this work, the looseness problem of a shaft is analysed by analysing the vibratory signals that are produced during its operation. To do this, in a fault simulation machine, two shafts will be tested at different rotation frequencies; one without looseness and another machined shaft with a diameter 0.5 mm smaller, causing a looseness in the connection of the shaft with the motor through a coupling and in the shaft with the bearings. The signals will be analysed using the Wavelet Packet Transform, a tool based on the time and frequency domain. For this purpose, the optimal mother wavelet will be previously selected by applying a methodology proposed in previous works. This study involves the use of intelligent classification systems, employing trained models of linear vector support machines. In this way, those patterns that allow the prediction of the looseness problem in the fastest and most reliable way possible will be obtained.