Song Daehun, Hong Seongeun, Seo Jaejoon, Lee Kyounghoon, Song Youngeun
Research & Development Division, Hyundai Motor Group, Seocho-gu, Seoul 06797, Korea.
Department of Electrical Engineering, Hoseo University, Asan-si 31499, Korea.
Sensors (Basel). 2022 Mar 14;22(6):2226. doi: 10.3390/s22062226.
A new development process for the noise, vibration, and harshness (NVH) of a vehicle is presented using data analysis and machine learning with long-term NVH driving data. The process includes exploratory data analysis (EDA), variable importance analysis, correlation analysis, sensitivity analysis, and development target selection. In this paper, to dramatically reduce the development period and cost related to vehicle NVH, we propose a technique that can accurately identify the precise connectivity and relationship between vehicle systems and NVH factors. This new technique uses whole big data and reflects the nonlinearity of dynamic characteristics, which was not considered in existing methods, and no data are discarded. Through the proposed method, it is possible to quickly find areas that need improvement through correlation analysis and variable importance analysis, understand how much room noise increases when the NVH level of the system changes through sensitivity analysis, and reduce vehicle development time by improving efficiency. The method could be used in the development process and the validation of other deep learning and machine learning models. It could be an essential step in applying artificial intelligence, big data, and data analysis in the vehicle and mobility industry as a future vehicle development process.
本文介绍了一种利用数据分析和机器学习以及长期的车辆噪声、振动与声振粗糙度（NVH）驾驶数据，来进行车辆NVH开发的新流程。该流程包括探索性数据分析（EDA）、变量重要性分析、相关性分析、敏感性分析以及开发目标选择。在本文中，为了大幅缩短与车辆NVH相关的开发周期和成本，我们提出了一种能够准确识别车辆系统与NVH因素之间精确联系和关系的技术。这种新技术使用了全部大数据，并反映了动态特性的非线性，而这在现有方法中并未得到考虑，且没有数据被丢弃。通过所提出的方法，可以通过相关性分析和变量重要性分析快速找到需要改进的领域，通过敏感性分析了解当系统的NVH水平变化时噪声增加的幅度，并通过提高效率来缩短车辆开发时间。该方法可用于其他深度学习和机器学习模型的开发过程及验证。作为未来车辆开发流程，它可能是在车辆和移动出行行业应用人工智能、大数据和数据分析的关键一步。
