A Diagnostic Technology of Powertrain Parts that Cause Abnormal Noises Using Artificial Intelligence



11th International Styrian Noise, Vibration & Harshness Congress: The European Automotive Noise Conference
Authors Abstract
In general, when a problem occurs in a component of powertrains, various phenomena appear, and abnormal noise is one of them. The service mechanics diagnose the noise through analysis by using their ears and equipment. However, depending on their experiences, analysis time and diagnostic accuracy vary greatly. To shorten the analysis time and improve the diagnostic accuracy, we have developed a technology to diagnose powertrain parts that cause abnormal noises. To create the best deep learning model for our diagnosis, we tried to collect many abnormal noises from various parts. The collected noise data was measured under idle and various operating conditions from our vehicles and test cells. This noise data is abnormal noises generated from engines, transmissions, drive system and PE (Power Electric) parts of eco-friendly vehicles. From the collected data, we distinguished good and bad data through detailed analysis in time and frequency domain. As a result, good data was used for training the model of the deep learning. This step is very important in the development process for the diagnostic technology. Features were extracted from the good data so that the deep learning can train to build the diagnostic model. The final diagnostic model was composed of various methods of deep learning such as RNN (Recurrent Neural Network), Attention Mechanism and DNN (Deep Neural Network). This model has been applied to the equipment for the mechanics in our service centers. Therefore, they can see the diagnosis results within seconds when a single noise is entered into the equipment.
Meta TagsDetails
Jung, I., Lee, D., Yoo, D., Lim, K. et al., "A Diagnostic Technology of Powertrain Parts that Cause Abnormal Noises Using Artificial Intelligence," SAE Technical Paper 2020-01-1565, 2020, https://doi.org/10.4271/2020-01-1565.
Additional Details
Sep 30, 2020
Product Code
Content Type
Technical Paper