Estimativa de amplitude de vibração de máquinas rotativas

Abstract

This thesis addresses vibration amplitude estimation errors in rotating machines due to the elliptical shape of the vibration orbits and the operating frequency variation and proposes a methodology to reduce these errors. Rotating machines can have high acquisition, installation, and intervention costs in industrial production, especially in oil production in subsea wells. This is the case of the Electrical Submersible Pump system (ESP), the second most used method of artificial lifting of oil in the world. The ESP system is applied to pump large volumes at high pressure and has higher costs than alternative methods. Vibration analysis of these machines is carried out for approval or rejection of the equipment before installation, thus seeking to reduce the risk of significant financial losses. The vibration analysis process for approval before installation is done by comparing the estimated vibration amplitudes with vibration limits established in standardized vibration severity standards. The estimation of industrial vibration amplitude is typically done using the discrete Fourier transform (DFT). The DFT has widely known properties that can lead to estimation errors that can be mitigated by different techniques to ensure the estimate's accuracy. However, at least two other sources of errors are overlooked and can lead to uncertainty in estimates and financial losses. The first source concerns the vibration measurement not necessarily occurring in the orientation that presents the higher radial vibration. As the vibration orbit can occur in an elliptical shape, and this shape can vary during operation, the installed vibration sensors will not necessarily be oriented towards the ellipse's major axis. The second source of errors comes from the fact that rotating machines are subject to minor speed variations that can lead to amplitude estimation errors due to the specialization of the analysis methods in stationary signals. There are techniques to reduce both estimation errors; however, these techniques are complex and imprecise to deal with vibration signals with the noise level and speed variation typical of ESP-type rotating machines. The methods currently found in the state of the art for vibration orbit analysis are significantly different from the individual signal frequency spectrum and can be considered counterintuitive. Their results cannot be directly compared to standard vibration limits for rotating machines. On the other hand, the methods found for accurate estimation of amplitude with frequency variation: either are dependent on the installation of a tachometer, which is not feasible in installations of some machines such as the complete assembly of ESP systems; or are dependent on the instantaneous velocity estimate, which is subject to more uncertainty; or they are imprecise with the noise level and frequency variation found in ESP systems, as demonstrated in this Thesis. This thesis presents a methodology for estimating the vibration amplitude of rotating machines, invariant to the shape of the vibration orbit and robust to variations in speed and presence of noise. The methodology involves the proposition of two methods: the Orbit Semi-major Axis Spectrum for estimating the spectrum invariant to the elliptical shape of the vibration orbits; and the EHRS method to perform the vibration amplitude estimation with robustness to the presence of noise and variation of speed, without depending on the instantaneous frequency as is necessary for other methods. To demonstrate its efficiency, the methodology proposed in this thesis was compared with typical methods of amplitude estimation on vibration signals from rotating machines of ESP systems. Evaluations were performed with simulated and experimental signals of vibration orbits of ESP systems. Analysis of simulated signals can allow the calculation of the estimation error because the generated amplitude value is known. On the other hand, it is impossible to know the correct amplitude in the analysis of experimental signals. However, it is possible to demonstrate that the differences in vibration amplitude between the evaluated methods follow the pattern of the simulated signals. The performance evaluation results showed that the proposed methodology, using the proposed methods EHRS and Orbit Semi-Major Axis Spectrum, is invariant to the shape of the vibration orbit, has greater robustness to frequency variation and the presence of noise than the other methods evaluated, and is recommended for the analysis of vibration amplitude of rotating machines.