arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.12967eess.SP
收藏

使用多边形接触方法模拟故障锥齿轮的扭转振动

Simulating Torsional Vibrations of Faulty Bevel Gears Using the Polygonal Contact Method

VTT Technical Research Centre of Finland Ltd.(芬兰国家技术研究中心);Aalto University(阿尔托大学)

Milla Vehviläinen, Aleksanteri Hämäläinen, Pekka Rahkola, Mikko Savolainen, Janne Keränen, Jari Halme, Jenni Pippuri-Mäkeläinen, Kari Tammi, Anouar Belahcen

AI总结:

提出基于多边形接触方法的多体仿真,模拟实验方位推进器试验台的扭转振动,能够处理任意故障几何形状,仿真信号在时域和频域与测量结果高度一致。

摘要:

齿轮是机电应用中的关键部件,但准确的状态监测方法(包括数据驱动的预测性维护)强烈依赖于高质量数据,尤其是来自故障部件的数据。为了解决数据稀缺问题,我们提出了一种使用先进多边形接触方法的多体仿真,以复制实验方位推进器试验台的扭转振动。关键创新在于能够模拟具有任意故障几何形状的健康和故障齿轮。仿真信号在时域和频域上与测量结果高度匹配。在时域中,平均扭矩水平和周期性波动吻合良好,尽管测量信号表现出更高的峰峰值幅度和更大的噪声,特别是在较低转速的健康条件下。在频域中,仿真准确再现了预期的故障频率和相应的边带,较大的故障产生更高的幅度。虽然仿真倾向于高估峰值幅度并低估外部噪声,但结果与测量高度可比,且符合物理预期。这些发现为增强数据驱动的状态监测方法(特别是那些使用机器学习或深度学习的方法)提供了坚实的基础。

英文摘要:

Gears are an integral component of electromechanical applications, but accurate condition monitoring methods, including data-driven predictive maintenance, are strongly dependent on high-quality data, especially from faulty components. To address the scarcity of data, we proposed a multibody simulation using an advanced polygonal contact method to replicate torsional vibrations from an experimental azimuth thruster test rig. The key novelty is the ability to simulate both healthy and faulty gears with arbitrary fault geometries. The simulated signals closely matched the measurements in both time and frequency domains. In the time domain, average torque levels and periodic fluctuations aligned well, although measured signals exhibited higher peak-to-peak amplitudes and greater noise, particularly in healthy conditions at lower rotational speeds. In the frequency domain, the simulations accurately reproduced expected fault frequencies and corresponding sidebands, with larger faults producing higher amplitudes. While the simulations tended to overestimate peak amplitudes and underestimate external noise, the results were highly comparable to measurements and consistent with the physical expectations. These findings provide a robust foundation for enhancing data-driven condition monitoring methods, particularly those employing machine learning or deep learning.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑