Anomaly Detection and Fault Diagnosis of EV Gearbox Using NVH Data

As part of a master thesis with Fraunhofer IDMT and Schaeffler AG, I developed Multilayer Perceptron and LSTM Autoencoder models for anomaly detection in Audi e-tron NVH data, achieving 82–90% accuracy. The project involved dimensionality reduction using T-SNE and clustering with DBSCAN for identifying patterns and faults. The goal was to detect gearbox anomalies using acoustic and structural vibration signals. This work demonstrated the effectiveness of machine learning in automotive fault diagnosis and R&D environments. [2]