Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
FAULT DIAGNOSIS OF TURBOCHARGED SPARK-IGNITED ENGINE SYSTEM BASED ON LONG SHORT-TERM MEMORY APPROACH
Submission Author:
LUIZ EDUARDO THOMAZ , PR , Brazil
Co-Authors:
Lucas Takara, LUIZ EDUARDO THOMAZ, Viviana Mariani, Leandro dos Santos Coelho
Presenter: LUIZ EDUARDO THOMAZ
doi://10.26678/ABCM.COBEM2023.COB2023-1672
Abstract
The continuous advancement of automotive technologies has led to an increase in the complexity of automobile functions and structures. As a result, fault diagnosis has become a popular and crucial topic in automotive engine systems, given the high susceptibility of their components to faults. While previous studies have utilized machine learning (ML) models, such as the multi-layer perceptron neural network, to detect misfire and pre-ignition faults in combustion engines, these approaches have limitations. They often fail to consider multiple faults, lack the ability to isolate them, and do not involve a comprehensive comparison of ML models. This study addresses these limitations by applying and comparing fourteen ML models to detect nine distinct fault scenarios in turbocharged spark-ignited engine systems. The models were trained using features derived from a simulation testbed specifically designed for a turbocharged spark-ignited engine system. Five features were extracted from the simulation variables and used as inputs for the models. Performance evaluation was conducted, demonstrating the models’ effectiveness in identifying and classifying various fault scenarios. Among the evaluated algorithms, the Random Forest classifier hiperparameters were optimized by the Tree-structured Parzen Estimator algorithm and exhibited the best performance across all metrics. This model achieved accuracy, recall, precision, and F1 score of 97.61%, 97.61%, 97.68%, and 97.52%, respectively. These results are highly encouraging and make future investigations into the application of these methods to other engine systems.
Keywords
Fault Diagnosis, Turbocharged Spark-Ignited Engine, Deep learning, Long-Short Term Memory
