Detección temprana de desviaciones del comportamiento nominal de sistemas utilizando algoritmos de Machine Learning / Early detection of deviations from the nominal behaviour of systems using Mechine Learning algorithms

Muñoz, Uriel A. (2019) Detección temprana de desviaciones del comportamiento nominal de sistemas utilizando algoritmos de Machine Learning / Early detection of deviations from the nominal behaviour of systems using Mechine Learning algorithms. Integration Project in Mechanical Engineering, Universidad Nacional de Cuyo, Instituto Balseiro.

PDF (Tesis)
Available under license Creative Commons Attribution Non-commercial Share Alike.


Abstract in Spanish

En sistemas tan complejos como los del campo aeroespacial, los sub-sistemas se diseñan para minimizar la inferencia mutua. Sin embargo, anomalías transversales a varios sub-sistemas existen, y son difíciles de detectar y entender. Sistemas de detección de anomalías integrales juegan un papel crítico en estas situaciones. En este trabajo se presentan herramientas de Machine Learning para la detección temprana de anomalías en una plataforma del área aeroespacial. Los métodos utilizados son: Gaussian Mixture Model, Principal Component Classier y Forecasting. Este último tiene el propósito de análisis de variables individuales, mientras que los otros dos tienen un espectro de aplicación integral, donde se apunta a la detección de cambios en la estructura de los datos y en menor medida a valores extremos individuales. En todos los casos son herramientas que se pensaron para ayudar a complementar el análisis del profesional (experto de dominio) en su trabajo, y no ser utilizadas independientemente. Son herramientas que proveen versatilidad en el análisis y permiten que se puedan aplicar ágilmente a distintos conjuntos de datos. Se lograron detectar anomalías artificiales de forma satisfactoria, para casos puntuales e integrales.

Abstract in English

In complex system like those from the aerospace eld, subsystems (atomic constituents of the full system) are designed to minimize or mitigate mutual inference. However, anomalies usually emerge as a collective phenomenon which turns the detection and the isolation a difficult task. Comprehensive anomalies detection systems play a critical role in these situations. In this thesis we present Machine Learning tools for the early detection of anomalies in a platform of the aerospace area. The methods used are: Gaussian Mixture Model, Principal Component Classier and Forecasting. The latter has the purpose of analyzing individual variables, while the rst two have a integral approach, where the objetive is to detect structure changes of the data and not so much extreme values. In all cases they are tools to help complement the analysis of the expert professional in their work, and not to be used autonomously. They are tools that demonstrate versatility in the analysis and allow to be applied agilely to different data sets. It was possible to detect articial anomalies satisfactorily, for specic and integral cases.

Item Type:Thesis (Integration Project in Mechanical Engineering)
Keywords:Telemetry; Telematría; Forecasting; Previsiones; [Machine learning; Aprendizaje automático; Principal component; Componentes principales; Novelty detection; Detección de novedades; Anomaly; Anomalías]
References:[1] B. Shetty, \Curse of dimensionality," Jan 2019. vii, 21 [2] Y. S. Abu-Mostafa, M. Magdon-Ismail, and H.-T. Lin, Learning from data: a short course., 2012. 1, 13 [3] C. M. BISHOP, PATTERN RECOGNITION AND MACHINE LEARNING.SPRINGER-VERLAG NEW YORK, 2016. 1 [4] R. H. Jones, \Entrepreneurs, beware survivorship bias," Mar 2019. 14 [5] B. Clarke, F. Ernest, and H. H. Zhang, Principles and theory for data mining and machine learning. Springer, 2011. 21 [6] M.-L. Shyu, S.-C. Chen, K. Sarinnapakorn, and L. Chang, \A novel anomaly detection scheme based on principal component classier," Jan 2003. 23, 31, 48 [7] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay, \Scikit-learn: Machine learning in Python," Journal of Machine Learning Research, vol. 12, pp. 2825-2830, 2011. 23, 35, 36, 37 [8] E. Jones, T. Oliphant, P. Peterson, et al., \SciPy: Open source scientic tools for Python," 2001-. [Version: 1.2.1]. 28 [9] J. D. Jobson, Applied Multivariate Data Analysis Categorical and Multivariate Methods, vol. 2. Springer Verlag, 2013. 30 [10] S. Seabold and J. Perktold, \Statsmodels: Econometric and statistical modeling with python," in 9th Python in Science Conference, 2010. 39 [11] R. J. Hyndman and G. Athanasopoulos, \Forecasting: Principles and practice, 2nd edition," Apr 2018. [Accedido: Abril 2019]. 39 [12] R. Cleveland and W. Cleveland, \Stl: A seasonal-trend decomposition procedure based on loess," Journal of Official . . . , vol. 6, 01 1990. 40 [13] R Core Team, R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2017. 40 [14] L. Breiman, Probability and stochastic processes: with a view toward applications. Scientic Press, 1986. 44 [15] P. G. Hoel, S. C. Port, and C. J. Stone, Introduction to probability theory. Houghton Miffin, 1996. 44
Subjects:Mechanical engineering > Ciencia de datos
ID Code:843
Deposited By:Tamara Cárcamo
Deposited On:19 Mar 2021 09:03
Last Modified:19 Mar 2021 09:03

Repository Staff Only: item control page