ISSN   1004-0595

CN  62-1224/O4

高级检索

基于深度学习的超高压橡塑往复密封磨损预测方法

Wear Prediction Method for Ultra-High Pressure Reciprocating Seals Based on Deep Neural Networks

  • 摘要: 密封与人工智能交叉领域的研究是国际热点,与载人航天、核电和舰船等安全运行相关. 首先基于Rhee的经典磨损理论与往复密封流固耦合数值模型,构建了针对Rhee理论的往复密封磨损轮廓预测模型,并成功量化了6种T型圈结构的磨损轮廓变化趋势,为后续的神经网络应用提供了丰富的数据集. 然后,本论文中创新性地引入了深度神经网络(DNN)技术,建立了DNN磨损轮廓预测模型,并通过细致的超参数调优,不断优化其预测效能. 最终,试验结果显示该DNN模型的预测精度高达90%以上,召回率亦超过95%,表明了在密封元件的智能预测性能及高效的运算能力.

     

    Abstract: By integrating Rhee's wear theory with a customized fluid-solid coupling model tailored for reciprocating seals, a predictive model was developed, providing profound insights into the wear behavior of these seals. This model facilitated an exhaustive analysis of six distinct T-ring configurations, each characterized by unique features and operational parameters, resulting in a detailed dataset mapping wear progression across these configurations. To enhance the predictive accuracy and refine the model, advanced Deep Neural Network (DNN) technology was incorporated. Through meticulous hyperparameter optimization, a sophisticated wear profile prediction model was established, effectively capturing the complex interplay of factors influencing wear behavior. Experimental validation confirmed the robustness of the DNN-based model, exhibiting remarkable performance with prediction accuracy exceeding 90% and a recall rate surpassing 95%. These exceptional results highlighted the substantial potential of our model as a pivotal tool for anticipating wear profiles in reciprocating seal systems. Overall, this research constituted a significant advancement in tribology and mechanical engineering, contributing novel insights and methodologies for the design, optimization, and maintenance of reciprocating seal systems.

     

/

返回文章
返回