روش نگهداری پیشگویانه با استفاده از سیگنال های فشار و شتاب از یک فن گریز از مرکز

This work describes a predictive maintenance procedure with a particular application to the diagnosis of unexpected events related to fluid-dynamics operating conditions in turbomachinery. The application of the procedure to a centrifugal fan is presented. The work includes an experimental study of the fan spectral behavior, revealing the characteristic frequencies of the different phenomena involved. Afterwards, a code has been developed which monitors on-line signals to detect any undesired event by comparing the levels of selected frequencies with those that obtain in normal operating conditions.

The strong development of electronics and computers, and its application to dynamic signal analysis, has prompted the appearance of reliable predictive maintenance systems. This type of maintenance is also known as condition-based maintenance, on-condition maintenance or condition-directed maintenance and it is conceived to detect the onset of a failure, avoiding critical damages of high cost components before they might happen, thus reducing overall maintenance costs. Possible faults are detected by monitoring representative parameters by signal analysis techniques and comparing signals during normal and abnormal conditions. In practical applications, predictive maintenance can use different techniques, like the analysis of vibrations, the analysis of the potential contaminants in the lubrication system, the control of the energy consumption, the control of the temperature in selected positions or the analysis of the noise generated by the machine; in conclusion, the measurement of the parameter or parameters that could be considered representative of the operation of the machine. Among all these techniques, the analysis of vibrations is the most frequently used. On the other hand, with respect to the signal processing, the frequency analysis is, by far, the more widely used. The FFT (fast Fourier transform) algorithm provides the signals spectra. By means of the observation and comparison of the peaks amplitude at different frequencies, it is possible, in many cases, to analyze the different malfunctions and to establish diagnosis criteria. In the case of large rotating machinery, the spectral analysis is not sufficient to identify correctly each failure [2]. In these cases, more sophisticated analysis techniques are needed, like phase spectra, holo-spectra and orbit diagrams. Among all the possible signals generated by an installation or a machine, it is not easy to associate a change of a parameter and its cause. So, before the development of predictive maintenance systems, it is always preferred to make experimental studies on the installation or the machine, in order to verify the sensibility of possible signals to typical failures. This is particularly true in the case of fluids machinery, because of the variety and complexity of the phenomena involved in the interaction between the machine components and the flow itself. Not only is the vibrational response to purely structural excitations affected by such flow-machine interaction, but in fact fluid-dynamic excitation is dominant in many cases. In particular, this work deals with a centrifugal fan. During the energy transfer between the fan and the fluid (atmospheric air in this case), non-steady fluid-dynamic forces are produced, both at single frequencies and with broad-band spectra. The former are usually associated with the frequency of rotation, the blade passing frequency and their harmonics. Excitation at the frequency of rotation may be provoked by impeller whirling, when the impeller has an orbital motion coupled to the rotation, by unbalances and by small manufacturing imperfections in the impeller. Excitation at the blade passing frequency (BPF, frequency of rotation multiplied by the number of blades of the impeller) is a consequence of the finite thickness of the blades, which causes flow disturbances in the volute associated with the passage of each blade. In addition, high turbulence levels are generated when the fan is operating at off-design conditions (i.e., flow-rates lower or higher than the best efficiency one), due to the incidence angle of the incoming flow with respect to the blades of the impeller. In centrifugal fans, instability phenomena at partial load may affect the aerodynamic performance and generate high levels of noise and vibration. Some of these partial load instability phenomena are flow separation in the blade channels, reverse flow and prerotation. Other perturbations may also exist that are not directly related to the operation of the fan, such as the presence of an obstacle in the fan aspiration or in a channel of the impeller. This kind of excitation phenomena, together with the purely structural excitation proceeding from the vibration of external elements mechanically coupled to the fan, means that its dynamic response on a certain operation event is particular to the fan and cannot be generalized. In consequence, an efficient predictive maintenance system capable of identifying a set of possible abnormal events must be designed separately for each different machine. In this work, a methodology has been developed in order to create a predictive maintenance routine for a squirrel-cage fan. Squirrel fans are forward-curved blades centrifugal fans. These fans are widely used in air conditioning, ventilation and heating of commercial or public buildings and passenger vehicles. In this way, this study can be applied in an extensive domain of great practical application: the excessive generation of noise and vibrations of this type of machines that can give rise to frequent inconveniences to the users. According to [1], the predictive maintenance may be an appropriate option for the general case of turbomachinery (fans in particular), because the following conditions apply: (a) failure prevention is not feasible because the event to failure occurs in a predominantly random manner, (b) some measurable parameters which correlates with the onset of failure may be identified, and (c) it is possible to identify a value of that parameter when action may be taken before full failure occurs. Firstly, an experimental study has been carried out, by acquiring typical signals during normal and abnormal operation of the fan. Signals of sound pressure and acceleration in different locations were acquired for different operation conditions. After the experimental study was accomplished, the analysis of these signals was incorporated in a monitoring code, which intends to detect incipient failures. Similar approaches have been presented for the cases of large-size compressors [2] and medium-size centrifugal pumps [3].

A predictive maintenance methodology has been implemented and applied to a squirrel-cage fan, with particular attention to the fluid-dynamic operating conditions of the machine. This study can be applied in an extensive domain of great practical application: the excessive generation of noise and vibrations of this type of machines that can give rise to frequent inconveniences to the users. The experimental investigation involved acoustic pressure and acceleration measurements in different locations for different operation conditions. These signals allow the detection of fan failures and to establish whether or not the fan is operating at the best efficiency point. The maintenance program developed permits the detection of laboratory tested failures, as well as the subsequent incorporation of additional defaults that could appear during the real operation of the machine; that is, the system is able to “learn” from the accumulated experience. The success of this methodology depends upon the experimental data acquired during normal operation, and on the different fan induced failures. The results obtained and diagnosis criteria are particular to the fan tested and should not be extrapolated to other machines without previous verification. However, the approach proposed to develop the diagnosis system may be considered equally valid for any other fan. In this case, a simple analysis has been implemented in which the spectra of acoustic pressure and acceleration signals in different conditions are compared. The selected signals can be enough to detect failures in the selected fan. However, in the future it is possible to incorporate more sophisticated analysis techniques in order to better identify some peculiarities observed in the normal operation of this machine.