بازگرداندن خاص شاخص های زمانی برای هر یک از منابع ارتعاشی برای نگهداری پیشگویانه

The monitoring of rotating machinery is increasingly important for industry. Partly, it is made from indicators given by vibration analysis. These indicators give only the overall state of operation of the studied machine. Indeed, the piezoelectric sensors, which are used for vibration analysis, record the vibrations generated by the various components of the machine. Numerous studies have developed techniques for locating and quantifying the vibration sources from the recorded signals. Thus, it is legitimate to believe that these sources can give us an indicator which should be characteristic of the damage of each critical component and which should allow to follow the severity of a defect. This paper studies the feasibility (i) firstly, to restore an indicator (RMS value) for each component, (ii) secondly, to monitor the evolution of this indicator through the severity of the defect.

Nowadays, the maintenance of machinery is essential for the good operation of the productivity and the safety of the staff. Many studies focus on a predictive maintenance, in particular the monitoring of the component of the machine. Partly, predictive maintenance monitors the deviation of the indicators which represent the working order of the machine. These indicators are determined from vibration signals obtained with piezoelectric sensors [1]. However, the signals recorded by the sensors are often the result of a mixture of sources (for example, from several rolling bearings). Therefore, these indicators only diagnose the general working order. The computation of an indicator which characterizes each source would be a major asset for the monitoring and the follow-up of the damage. The restoration of each source from the recorded signals and the computation of indicators can improve their diagnosis. However, the inversion is an unstable problem due to the measurement errors or noise in the signals. To stabilize the problem, studies suggest using over-determined systems, i.e. systems with more observations than sources [2] and [3]. These methods require the use of parameters called “regularization parameters” which are difficult to determine. They can be evaluated according to the uncertainties of the mixture matrix [4], on the mixture matrix and observations [5], or on the principle of L-curve [6]. Even if these parameters are determined, the inversion may remove a source. Moreover, these studies are trying to restore the sources over a wide frequency band with a large number of sensors which are placed “randomly”, while the stability depends largely on the frequency and on the position of the sensors. Fabunmi [7] has highlighted the existence of a link between modal analysis and the number of determinable sources on a narrow frequency band. He shows that the number of modes participating in the response of the structure must be equal or superior to the number of sources. Lee and Park [8] use this relationship to make a selection of sensors which can stabilize the inversion. However, this selection is made after having positioned many sensors. In a previous paper [9], we propose to position a limited number of sensors which are used to restore the sources around several frequencies (which are characteristic of the presence of a defect) and to avoid regularization methods that can remove vibratory sources. Two approaches have been developed. Each of them can determine the areas where the restitution is stable. These areas are illustrated by conditioning maps. Those two approaches provide a reliable restoration due to the stability of the solution. They are compared to the methods which use the over-determined systems and the methods of regularization, including the L-curve principle [10]. The aim of this paper is to restore an indicator specific to each vibratory sources using inverse approach. We are studying the particular case of a fault in rolling bearings, because rolling bearings are always present in rotating machines.

The use of temporal indicators allows to monitor and to track the damages on rotating machines. These indicators are called global indicators. They only reflect the global working order. This paper proposes to calculate an indicator which characterizes the state of each component of a rotating machine, and which would monitor the severity of the detected defects on a component. This is achieved through the application of techniques associated with inverse problems. The indicators (RMS values) are computed from restored sources. This paper shows that the use of the measured indicators is unreliable, and the computation of indicators specific for each component improves the monitoring and the diagnosis of the defects. Also, we show that the monitoring from the restored RMS values is possible. However, one can notice that these values are computed from signals which are widely filtered around low frequencies. Our future efforts focus on a restoration with a large frequency range, which would get computed RMS values from a signal including high frequencies, because the defects in rolling bearings excite these frequencies.