高级检索

    基于含工况影响维纳过程模型的剩余寿命预测方法

    Remaining Useful Life Prediction Method Based on the Wiener Process Model with Condition Influence

    • 摘要: 机械设备运行工况的变化不仅更容易诱发零部件性能退化,还会影响退化过程的稳定性,为零部件的剩余寿命预测带来了挑战。针对工况因素对机械设备退化的影响,本文提出一种含工况影响的维纳过程模型,通过引入幂律形式的工况相关项捕捉退化过程与工况之间的关系,并使用单元极大似然估计算法估计模型参数;然后,使用卡尔曼-粒子滤波算法更新单元可变参数;根据构建的模型,推导剩余寿命概率密度函数的解析表达式。最后,利用合金和轴承退化数据集进行实验验证。结果表明,所提出模型在两种数据集中的赤池信息准则为−185.64和−537.76,优于传统模型;所提出预测方法在两种数据集中的平均累积相对精度为0.93和0.86,平均总体均方误差为25.74和28.34,均优于基于粒子滤波和基于贝叶斯原理的预测方法。

       

      Abstract: Changes in the operating conditions of mechanical equipment not only increase the likelihood of performance degradation in mechanical components, but also impact the stability of the degradation process. This introduces challenges in predicting the remaining useful life (RUL) of mechanical equipment. Traditional Wiener process models are typically limited to the RUL prediction under a constant operating condition, leading to reduced accuracy in practical engineering scenarios. To address this problem, this paper proposes a Wiener process model that incorporates the influence of operating conditions. An exponential condition-dependent function is introduced to capture the relationship between the degradation process and operating conditions. Model parameters are estimated using the unit maximum likelihood estimation algorithm. Then, the variable parameter within the model is dynamically updated using an ensemble Kalman filter combined with a particle filter. Additionally, an analytical expression for the probability density function of the RUL is derived based on the constructed model. Finally, the effectiveness of the proposed prediction method is validated using a fatigue crack propagation dataset and a bearing degradation dataset. The experimental results show that the Akaike Information Criterion values of the proposed model are −185.64 and −537.76 for the two datasets, which are better than those of the traditional model. Additionally, the proposed prediction method achieves average Cumulative Relative Accuracy values of 0.93 and 0.86, and average Total Mean Square Error values of 25.74 and 28.34 for the two datasets, both of which outperform the prediction methods based on the particle filter and Bayesian principle.

       

    /

    返回文章
    返回