Empleo de algoritmos de Machine Learning para la detección de fallos en el sistema de encendido y admisión de aire en un motor Otto

In the present work, Machine Learning algorithms and models were applied within MATLAB to classify four different engine operating states: optimal conditions, faulty spark plug, obstructed air filter, and both faults combined. To achieve this, voltage data from the MAP sensor of a YESA 3133 Otto engine was collected using an NI-USB 6009 data acquisition board and LabVIEW software. Then, features were extracted from each MAP sensor signal corresponding to 720° of crankshaft rotation, the most relevant ones were selected based on ANOVA analysis, correlation analysis, and Random Forest. The selected features were trained using Artificial Neural Networks (ANN) and Support Vector Machines (SVM), with multiple configurations to optimize their parameters. Additionally, MATLAB's "Classification Learner" toolbox was used to implement various Machine Learning models for data classification. This process was applied for both the raw MAP sensor data and the MAP sensor data processed through a low-pass Butterworth filter. The work results indicated that ANN is the most suitable model for used because it has the best accuracy. After several configurations, it reached an accuracy of 96.05% without showing signs of overfitting, and also had the shortest training time compared to the other Machine Learning models. Additionally, the study demonstrated that increasing the number of training features improved ANN accuracy, raising it from 93.97% to 95.53%.

• Autor principal: Medina Namicela, Juan Pablo (author)
• Format: bachelorThesis
• Idioma: spa
• Publicat: 2025
• Matèries: MOTOR; CONDISIONES; CLASSIFICATION LEARNER
• Accés en línia: https://dspace.unl.edu.ec/jspui/handle/123456789/32413