Babaglio, Danilo (2019) Técnicas de diagnostico por computadora aplicadas al monitoreo de la condición en procesos productivos / Computer diagnosis techniques applied to condition monitoring of productive processes. Maestría en Ingeniería, Universidad Nacional de Cuyo, Instituto Balseiro.
![[img]](/style/images/fileicons/application_pdf.png)
| PDF (Tesis) Español 14Mb |
Resumen en español
Se presenta, en esta tesis de maestría, una herramienta informática original que procesa los datos y ordena la información relevante proveniente de una maquina o proceso productivo, permitiendo detectar apartamientos de la condición normal o detección de anomalías. Una vez recibidos los datos colectados desde transductores, el sistema procesa los mismos en el dominio del tiempo y de la frecuencia y computa sus indicadores estadísticos mas importantes. El origen de las señales es validado, ya que debe tener un comportamiento adecuado. El sistema, una vez validadas las señales, extrae información clave para el análisis, la cuales llamamos features o características. Las features forman un espacio dimensional multivariable en el cual el comportamiento dinámico de una maquina yace sobre una región determinada. Las features tienen diferente peso al momento de clasificar el comportamiento. Por esta razón y por la complejidad que significa clasificar datos en espacios de alta dimensión las características son condensadas proyectando la información sobre los componentes principales de los datos analizados. De esta forma podemos reducir la dimensionalidad del problema y encontrar patrones que al considerar datos no relevantes podrían estar ocultos. Los patrones o clusters, deben ser identificados, para los que se utilizo un algoritmo de aprendizaje no supervisado derivado del aprendizaje competitivo y los mapas auto organizados llamado G-Stream, que utiliza redes neuronales artificiales que adaptan sus pesos sinápticos para encontrar los clusters. Este método permite encontrar clusters y clasificar nuevos datos, identificando a que grupo pertenecen y su topología. Finalmente, cuando los grupos de comportamiento están clasificados se emplean algoritmos para determinar y seleccionar las características originales mas relevantes para la clasificación, en forma de una regla que puede ser interpretada por un humano. El desempeño del sistema prototipo desarrollado es probado mediante tres casos de estudio utilizando señales reales provenientes de: Un sistema de monitoreo y diagnostico de la bomba del circuito primario de refrigeración del reactor de investigación RA6; Un banco de ensayos de rodamientos de la división vibraciones del Centro Atómico Bariloche; Un set de datos internacional de un banco de ensayos de rodamientos de la IMS (Intelligent Maintenance Systems).
Resumen en inglés
This thesis presents a computing tool that processes data and sorts out relevant information gathered from a machine or production process, which enables detection of abnormal behavior. Once transducer data is obtained, the system processes it on time and frequency domain, and computes their main statistic indicators. The origin of these signals is rst validated, as it must have an adequate behavior. The system then extracts key pieces of information for analysis called `features', which form a multivariable dimensional space where the machine's dynamic behavior lies upon a certain region. Given the complexity involved in arranging data in higher dimensional spaces and considering that features bear different weights at the moment of classifying behavior, the latter are condensed by projecting the information on the principal components of the analyzed data. Issues with dimensionality can therefore be reduced and behavioral patterns can be found which could be previously obscured by irrelevant data. In order to identify patterns or `clusters', a non-supervised machine learning algorithm called G-Stream was utilized, which is based on competitive learning and self-organizing maps and uses articial neural networks which adapt their synaptic weights to nd the clusters. In addition to finding clusters, this method enables the classification of new data by identifying the cluster they belong to as well as their topology. Finally, once behavioral groups have been arranged, other algorithms determine and select from among the original features those that are more relevant for classification, in the form of a rule which can be interpreted by a human. The performance of the developed prototype system was tested using three study cases with real life signals adquired from: A monitoring and diagnosis system for the primary coolant pump of the research reactor RA6; A bearing testbench in the vibration division of the Bariloche Atomic Centre; An international dataset of a bearing testbench from the IMS (Intelligent Maintenance Systems).
| Tipo de objeto: | Tesis (Maestría en Ingeniería) |
|---|---|
| Palabras Clave: | Vibrations (Mechanical); Vibraciones (Mecánicas); [Condition monitoring; Monitoreo de la condición; Machine learning; Aprendizaje de máquinas; Signal processing; Procesamiento de señales] |
| Referencias: | [1] Shu Tay, Lee Te Chuan, A. Aziati, and Ahmad Nur Aizat Ahmad. An overview of industry 4.0: Definition, components, and government initiatives. Journal of Advanced Research in Dynamical and Control Systems, 10:14, 12 2018. [2] Report. Global Machine Condition Monitoring Market - Segmented by Type, Monitoring Process, Components, Application, and Region - Growth, Trends, and Forecast (2018 - 2023). Mordor Intelligence, 2018. [3] B.K.N. Rao. Advances in diagnostic and prognostic strategies and technologies for failure-free maintenance of industrial assets. COMADEM, 2009. San Sebastian, España. [4] M.J. Neale and B.J. Woodley. A guide to the condition monitoring of machinery. COMADEM, 1978. Reporte TDR 223 para el Departamento de Industría Británico. [5] D.J. Sherwin and B. Al-Najjar. Practical models for condition monitoring inspection intervals. Journal of Quality in Maintenance Engineering, 1999. 5 (3), 203-220. [6] D. Childs. Turbomachinery Rotordynamics. John Wiley and Son, Inc., New York, 1993. [7] S.S. Rao. Mechanical Vibrations. Prentice Hall, Englewood Cliffs, 2005. [8] J. An Detoni and S. Braun. Blind source separation. Mechanical Systems and Signal Processing, 2005. 19 (6) (Special issue). [9] Nordmann R. Gasch, R. and H. Pfutzner. Rotordynamik. Springer, Berlin, 2002. [10] J.S. Rao. Rotor Dynamics. New Age International, New Delhi, 1996. [11] G. Genta. Dynamics of Rotating Systems. Springer, Berlin, 2005. [12] S. Lacey. An overview of bearing vibration analysis. Maintenance and Asset Management, 2008. vol. 23, no. 6, pp. 32-42. [13] J.D. McFadden, P.D.and Smith. Model for the vibration produced by a single point defect in a rolling element bearing. Journal of Sound and Vibration, 1984. 96 (1), 69-82. [14] Matteo Frigo and Steven G. Johnson. The design and implementation of FFTW3. Proceedings of the IEEE, 93(2):216-231, 2005. Special issue on "Program Generation, Optimization, and Platform Adaptation". [15] Boaz Porat. PA Course in Digital Signal Processing. John Wiley and Sons, Inc. New York, NY, USA, 1996. [16] P. Welch. The use of fast fourier transform for the estimation of power spectra: A method based on time averaging over short, modied periodograms. IEEE Transactions on Audio and Electroacoustics, 15(2):70-73, June 1967. [17] L. Marple. Computing the discrete-time analytic signal via fft. IEEE Transactions on Signal Processing, 47(9):2600-2603, Sep. 1999. [18] Simon S. Haykin. Neural networks and learning machines. Pearson Education, Upper Saddle River, NJ, third edition, 2009. [19] M. Bishop. Neural Networks for Pattern Recognition. Oxford University Press, Inc. New York, New York, USA, 1995. [20] F. Rosenblatt. The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain. Psychological Review, 1958. 65, 386-408. [21] K. Theodoridis, S.and Koutroumbas. Pattern Recognition. Academic Press, 2009. [22] C. Sammut and G.I. Webb. Encyclopedia of Machine Learning. Springer, 2011. [23] Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer, 2006. [24] Teuvo Kohonen, Manfred Schroeder, and Thomas Huang. Self-Organizing Maps. Springer, 01 2001. [25] Thomas Martinetz and K. Schulten. A "neural-gas"network learns topologies. Artificial neural networks, 1:397{402, 01 1991. [26] Mohammed Ghesmoune, Mustapha Lebbah, and Hanene Azzag. A new growing neural gas for clustering data streams. Neural Networks, 78, 02 2016. [27] Varsh Patel and Drashti Vashi. A survey review on concept drift. International Journal of Sciences and Applied Research, 4:20, 05 2017. [28] Komkrit Udommanetanakit, Thanawin Rakthanmanon, and Kitsana Waiyamai. Estream: Evolution-based technique for stream clustering. International Conference on Advanced Data Mining and Applications, 4632:605-615, 08 2007. [29] D. Babaglio M. Garrett M. Marticorena R. Mayer J. Vignolo García Peyrano, O. CONDITION MONITORING AND INCIPIENT FAILURE DETECTION ON ROTATING EQUIPMENT IN RESEARCH REACTORS. IAEA, Coordinated Research Project, Viena, Austría, 01 2019. [30] R.; Wang T.; Li H.; Song G. Lv, Y.; Yuan. Health Degradation Monitoring and Early Fault Diagnosis of a Rolling Bearing Based on CEEMDAN and Improved MMSE. Materials, 2018. 11 (6), 1009. [31] R. Rubuni and U. Meneghetti. Application of the envelope and wavelet transform analysis for the diagnosis of incipient faults in ball bearings. Mechanical Systems and Signal Processing, 2001. 15 (2), 287-302. [32] Quan Wang. Kernel principal component analysis and its applications in face recognition and active shape models. Rensselaer Polytechnic Institute, 07 2012. |
| Materias: | Ingeniería mecánica |
| Divisiones: | Gcia. de área de Energía Nuclear > Gcia. de Ingeniería Nuclear > Vibraciones |
| Código ID: | 897 |
| Depositado Por: | Tamara Cárcamo |
| Depositado En: | 29 Mar 2021 11:18 |
| Última Modificación: | 12 Abr 2021 12:20 |
Personal del repositorio solamente: página de control del documento
