Eventos Anais de eventos
CONEM 2018
X Congresso Nacional de Engenharia Mecânica
IN-CYLINDER PRESSURE RECONSTRUCTION USING VIBRATION MEASUREMENTS AND DEEP CONVOLUTIONAL NEURAL NETWORKS
Submission Author:
Amaury André , SP
Co-Authors:
Amaury André
Presenter: Amaury André
doi://10.26678/ABCM.CONEM2018.CON18-1207
Abstract
In-cylinder pressure signal, during the combustion process, is an important parameter for fault detection in internal combustion engines. It impacts in reliability and efficiency, as far as consumption and pollutant emission. Directly measure in-cylinder pressure has many drawbacks due to the harsh environment inside the combustion chamber: it needs a costly transducer, which has a very limited life-time. The use of indirect measurements have great potential, and many methodologies were proposed in the past. But they all need pre- or post-processing in vibration data, or depend on specific operational conditions. In this work, a deep Convolutional Neural Network (CNN) fed with raw vibration data is used to reconstruct the in-cylinder pressure signal. A single accelerometer per cylinder can be used for constantly monitoring the combustion process for large internal combustion engines used in a power generation plant.
Keywords
Vibration Analysis, pressure, Internal combustion engine, convolutional neural network
