Assessment and Evaluation of an Unsupervised Machine Learning Model for Automotive and Industrial NVH Applications
* Presenting author
Abstract:
Current NVH analysing techniques involve an interdisciplinary knowledge of structural dynamics, signal processing, technical- and psycho-acoustics but most notably they require an experienced professional to analyse and assess the ever-expanding amount of industrial acquired NVH data. Recent advances in machine learning (ML) have shown the possibilities of inference on feature representations of input data without human intervention, which has helped experts to focus on actual solutions and reduce manual efforts for pre-processing and classification significantly. [1] We challenged an unsupervised deep neural network model (DNN) based on autoencoders (AE) to detect anomalies and semantic relations in different types of industrial NVH data and compared findings with an analytical state of the art approach. [2] Conclusion is drawn on applications, data types and future challenges. [1] Y. Bengio, A. Courville, and P. Vincent, “Representation learning: A review and new perspectives,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 35, no. 8, pp. 1798–1828, 2013.[2] Y. Koizumi, S. Saito, H. Uematsu, Y. Kawachi, and N. Harada, “Unsupervised Detection of Anomalous Sound Based on Deep Learning and the Neyman-Pearson Lemma,” IEEE/ACM Trans. Audio Speech Lang. Process., vol. 27, no. 1, pp. 212–224, Jan. 2019.