AI-Based Acoustic Quality Inspection in Automotive Manufacturing: A Case Study on the Detection of Disturbing Noises in Automated Driving Vehicles

With the transition to electric mobility, the perceived acoustic quality of vehicles is becoming an increasingly important factor in customer satisfaction. The elimination of the combustion engine in electric vehicles results in generally quieter vehicles. Consequently, potential disturbing noises are no longer masked and become more noticeable to customers. In combination with the possibility of automated driving on test tracks in a production environment, the question arises of how to automate and objectify the currently manual quality inspections for noise detection. The aim of this case study is to investigate the potential of Artificial Intelligence (AI) for detecting disturbing noises within a real production environment. For this purpose, audio recordings are captured with a microphone inside the vehicle during automated driving on a test track. These recordings are then converted into spectrograms, which visually represent human auditory perception. Based on these spectrograms, an AI model is developed in the form of a Convolutional Recurrent Neural Network (CRNN). The results of this case study demonstrate that the developed AI model can reliably classify vehicles as OK or not OK (NOK) with high accuracy. Furthermore, the model distinguishes between different types of noise, such as rattling and buzzing, indicating potential for further research in this area. These findings highlight the potential of AI-based acoustic quality inspection in automotive manufacturing.