Search

Healthy rolling element bearings are vital guarantees for safe operation of the rotating machinery. Time-frequency (TF) signal analysis is an effective tool to detect bearing defects under time-varying shaft speed condition. However, it is a challenging work dealing with defective characteristic frequency and rotation frequency simultaneously without a tachometer. For this reason, a technique using the generalized synchrosqueezing transform (GST) guided by enhanced TF ridge extraction is suggested to detect the existence of the bearing defects. The low frequency band and the resonance band are first chopped from the Fourier spectrum of the bearing vibration measurements. The TF information of the lower band component and the resonance band envelope are represented using short-time Fourier transform, where the TF ridge are extracted by harmonic summation search and ridge candidate fusion operations. The inverse of the extracted TF ridge is subsequently used to guide the GST mapping the chirped TF representation to the constant one. The rectified TF pictures are then synchrosqueezed as sharper spectra where the rotation frequency and the defective characteristic frequency can be identified, respectively. Both simulated and experimental signals were used to evaluate the present technique. The results validate the effectiveness of the suggested technique for the bearing defect detection.

En los últimos años se ha incrementado el uso de técnicas de modelamiento basado en datos para el diagnóstico de fallos en maquinaria rotativa. Estas técnicas requieren de grandes cantidades de datos que no siempre se pueden obtener pues generan altos costos y tiempo excesivo, que son difíciles de solventar desde el punto de vista económico y técnico. El presente trabajo se enfoca en el pre-procesamiento de las señales de vibración y propone un método para incrementar el número de series temporales informativas de una máquina rotativa sin el incremento del tiempo y costos en la etapa de adquisición de las señales. Como resultado se ha obtenido una ampliación de 315 señales en la fase de adquisición de datos a 429000 luego de la aplicación del método; cantidad adecuada para la construcción de modelos basados en datos, incluso de deep learning para la detección de fallos en maquinaria rotativa. Palabras clave: Adquisición de datos, pre-procesamiento, rodamientos, señales.

This paper presents an identification approach based on invertible singleton fuzzy models in order to implement a control system for a progressive cavity pump-based petroleum production system, by using the inverse model control scheme. The identification proposal uses input-output process variables measurements and an off-line genetic algorithm, which is designed for guarantying the analytical calculation of the inverse model. Regarding the application of the genetic algorithm, three selection methods are evaluated: tournament selection, roulette wheel selection and linear rank selection. Once the fuzzy singleton model is identified, the controller design is a straightforward procedure. Computer simulations show the potential application of this kind of model in real industrial control processes. Zhangjiajie

In this work it is presented a fault defection scheme using neural networks with a fuzzy preprocessing done to the input signals. The use of the fuzzy preprocessing enriches the input set given to the fault diagnosis neural network. We present an example of a three tanks system fault diagnosis using a conventional neural networks scheme and the proposed one.