Publication: Nuevas metodologías no invasivas de diagnosis de defectos incipientes en rodamientos de bola
Loading...
Identifiers
Publication date
2007-05
Defense date
2007-07-09
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Los rodamientos de bola forman parte del subsistema de apoyo de máquinas
rotativas, soportando una gran porción de la carga estática y dinámica del sistema, por
lo que la posibilidad de un defecto incipiente en su estructura, es inherente a las
tensiones involucradas en la conversión de energía del sistema.
En las últimas décadas se ha incrementado la demanda de la automatización de
la diagnosis de defectos en procesos industriales, por su influencia en la economía de la
empresa, al reducir costes operacionales y de mantenimiento, y mejorar el nivel de
seguridad. De las distintas formas de diagnosticar un defecto mecánico, los métodos no
invasivos aportan mediciones de bajo coste y de fácil acceso, a la vez que persiguen la
evaluación del estado dinámico de la máquina sin la necesidad de proceder a su
desmontaje; debido a tales factores, es de gran importancia su uso y puesta a punto.
La diagnosis de defectos en componentes de máquinas, toma en consideración
los datos de vibración mecánica obtenidos, que se conforman por señales complejas de
naturaleza estocástica y no estacionaria, dada la no linealidad de la dinámica de
máquinas. Esta información suele ser procesada y clasificada con distintas herramientas
para diagnosticar el estado del sistema mecánico.
En la presente Tesis Doctoral, se han desarrollado y aplicado diferentes
metodologías de diagnosis de defectos incipientes en rodamientos de bola. Para ello, se
ha adquirido un conjunto de señales de un banco de ensayos de rodamientos,
completando tres tipos de condición defectuosa y una condición normal en rodamientos.
Estas señales han sido procesadas mediante una herramienta de procesamiento de datos,
denominada Transformada Wavelet que, a pesar de su edad temprana, posee una
trayectoria amplia en el análisis de señales vibratorias, siendo capaz de extraer
información relevante del fenómeno físico en estudio, en dimensiones reducidas.
Adicionalmente, esta información ha sido clasificada por medio de tres tipos de redes
neuronales, que han demostrado ser capaces de efectuar un diagnóstico automático de la
condición de un sistema, al aprender adecuadamente con un conjunto representativo de
muestras, e imitar el proceso de aprendizaje humano.
Los resultados señalan que hay diversos factores que influyen en la precisión del
sistema clasificador, como son, la cantidad de datos utilizados, la complejidad de la red
neuronal, y diversas consideraciones de diseño que se explican en detalle para cada red
en particular. Finalmente, se introduce la aplicación de Sistemas Híbridos de
clasificación para la diagnosis de defectos en componentes mecánicos rotativos,
consiguiendo índices de éxito nunca antes alcanzados en este campo.
____________________________________________
Ball bearings are part of the rotating machinery anchor subsystem, supporting a considerable portion of the static and dynamic load, and the possibility of incipient faults is inherent due to the stresses involved in the energy conversion of the system. In the past few decades, there has been an increasing demand for fault diagnosis automation in industrial process, due to its influence in plant economy, by reducing operational and maintenance costs and the improvement of the safety level. Different methods are used to diagnose a mechanical fault, but the non-invasive schemes offer easily accessible and inexpensive measurements to predict the system condition without disintegrating the machine structure. Due to their characteristics, its use and conditioning are very important. Fault diagnosis in mechanical devices consider the mechanical vibration data obtained, which is conformed of stochastic and non-stationary complex signals due to machine dynamics are not linear. This information should be processed and classified with different tools in order to diagnose the mechanical system condition. In the present Doctoral Thesis, there have been acquired experimental data sets from a bearing test bench, obtaining three different faulted conditions and a normal operation condition for bearings. These signals have been processed using the Wavelet Transform, which has been becoming important, with a wide trajectory in the vibration signal analysis, capable to extract relevant information about the physical phenomenon studied, with reduced dimensionality. In addition, this information has been classified with three different Neural Networks types, whose have demonstrated capability to diagnose automatically the system condition, by learning properly with a representative set of samples, imitating the human learning process. The results show that several variables affect the Neural Network precision, as length dataset, net complexity and some design considerations which are detailed for each network in particular. Finally, it has been introduced the application of Classification Hybrid Systems for the mechanical rotating devices fault diagnosis, obtaining success rates never reached in this field
Ball bearings are part of the rotating machinery anchor subsystem, supporting a considerable portion of the static and dynamic load, and the possibility of incipient faults is inherent due to the stresses involved in the energy conversion of the system. In the past few decades, there has been an increasing demand for fault diagnosis automation in industrial process, due to its influence in plant economy, by reducing operational and maintenance costs and the improvement of the safety level. Different methods are used to diagnose a mechanical fault, but the non-invasive schemes offer easily accessible and inexpensive measurements to predict the system condition without disintegrating the machine structure. Due to their characteristics, its use and conditioning are very important. Fault diagnosis in mechanical devices consider the mechanical vibration data obtained, which is conformed of stochastic and non-stationary complex signals due to machine dynamics are not linear. This information should be processed and classified with different tools in order to diagnose the mechanical system condition. In the present Doctoral Thesis, there have been acquired experimental data sets from a bearing test bench, obtaining three different faulted conditions and a normal operation condition for bearings. These signals have been processed using the Wavelet Transform, which has been becoming important, with a wide trajectory in the vibration signal analysis, capable to extract relevant information about the physical phenomenon studied, with reduced dimensionality. In addition, this information has been classified with three different Neural Networks types, whose have demonstrated capability to diagnose automatically the system condition, by learning properly with a representative set of samples, imitating the human learning process. The results show that several variables affect the Neural Network precision, as length dataset, net complexity and some design considerations which are detailed for each network in particular. Finally, it has been introduced the application of Classification Hybrid Systems for the mechanical rotating devices fault diagnosis, obtaining success rates never reached in this field
Description
Keywords
Elementos de máquinas, Vibraciones, Máquinas rotativas, Rodamientos de bola, Análisis armónico