Publication: Análisis de sistemas de clasificación de defectos en ejes
Loading...
Identifiers
Publication date
2011
Defense date
2011-01-24
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
El mantenimiento está cobrando cada vez más importancia en la industria actual por las consecuencias económicas de ahorro que conlleva una correcta elección y ejecución del tipo de mantenimiento. Por esta razón se emplean, cada día más, sistemas de seguimiento de procesos que permiten realizar un seguimiento de las máquinas sin necesidad de parar la actividad para comprobar el estado de la maquinaria. Este tipo de sistemas generalmente hacen un seguimiento de los elementos con mayor facilidad para presentar problemas y cuyo fallo tenga gran repercusión en el proceso. Un componente de vital importancia para el correcto funcionamiento de cualquier máquina son los ejes, los cuales pueden provocar graves daños en el conjunto en el que se encuentra montado o en elementos de su entorno. Los ejes pueden presentar fisuras debido a las condiciones de operación dentro del proceso. La aparición de un nuevo método para analizar las señales temporales denominado descomposición wavelet ha hecho que se pueda detectar la presencia de fisuras mediante su estudio, algo que con la Transformada de Fourier resulta difícil de llevar a cabo. Además, el desarrollo de redes neuronales artificiales ha posibilitado la implantación de sistemas de seguimiento sin necesidad de depender de factores humanos. Estas redes permiten identificar el estado del eje dentro de unas clases previamente definidas según las señales procedentes del sensor colocado en la máquina e introducidas como inputs, tomando el operario decisiones basadas en los resultados mostrados por las redes neuronales. En el presente proyecto se ha procedido a ensayar tres tipos de ejes, uno de ellos en estado óptimo y otros dos con fisuras perpendiculares a su eje, y adquirir las vibraciones sufridas durante el ensayo para estudiar posteriormente el comportamiento de tres redes neuronales que permiten clasificar las señales adquiridas según el tipo de eje al que correspondan. Previamente a la clasificación de las señales hay que realizar un análisis wavelet de las señales que será el que proporcione los inputs de las redes neuronales. ______________________________________________________________________________________________________
Maintenance is becoming more and more important for current industry due to economic of savings that it entails a correct election and execution of the maintenance type. For this reason, monitoring systems that make possible to do a machine monitoring without the need of stopping the activity to checking the machine state are actually used more and more. Generally, this kind of systems makes a monitoring of the elements with a greater facility to having problems and whose failure has a big repercussion in the process. Shafts are components of vital importance for the correct operation of any machine which can cause serious damages to the system where they are assembled or to surroundings elements. Shafts can present cracks due to operation conditions in the process. The appearance of a new method to analyze signal temporary, called wavelet decomposition, has done that presence of cracks can be detected by its study, something that, with Fourier Transform, results difficult to carry out. Furthermore, neuronal network development has made possible the introduction of monitoring systems without need to depend on human factors. These networks enable to identify the shaft state within some pre-established kinds according to the signals from the sensor put in the machine and introduced as inputs, taking the operator decisions based on results showed by neural networks. In the present project, three shafts have been tested, one of them in an optimum state and the other two with a crack perpendicular to their axe. Vibrations suffered during the test have been acquired in order to study later the performance of three kind of neural networks that allow classify the acquired signals according to the shaft type. Before the signal classification, it is necessary to make a wavelet analysis to the signals that provide the inputs to the neural networks.
Maintenance is becoming more and more important for current industry due to economic of savings that it entails a correct election and execution of the maintenance type. For this reason, monitoring systems that make possible to do a machine monitoring without the need of stopping the activity to checking the machine state are actually used more and more. Generally, this kind of systems makes a monitoring of the elements with a greater facility to having problems and whose failure has a big repercussion in the process. Shafts are components of vital importance for the correct operation of any machine which can cause serious damages to the system where they are assembled or to surroundings elements. Shafts can present cracks due to operation conditions in the process. The appearance of a new method to analyze signal temporary, called wavelet decomposition, has done that presence of cracks can be detected by its study, something that, with Fourier Transform, results difficult to carry out. Furthermore, neuronal network development has made possible the introduction of monitoring systems without need to depend on human factors. These networks enable to identify the shaft state within some pre-established kinds according to the signals from the sensor put in the machine and introduced as inputs, taking the operator decisions based on results showed by neural networks. In the present project, three shafts have been tested, one of them in an optimum state and the other two with a crack perpendicular to their axe. Vibrations suffered during the test have been acquired in order to study later the performance of three kind of neural networks that allow classify the acquired signals according to the shaft type. Before the signal classification, it is necessary to make a wavelet analysis to the signals that provide the inputs to the neural networks.
Description
Keywords
Elementos de máquinas, Vibraciones mecánicas, Detección de errores, Mantenimiento predictivo, Ejes fisurados