基于状态空间的惯性测量组合剩余寿命在线预测

Residual life prediction based on the state space for inertial measurement units

实时准确的剩余寿命预测能够为惯性测量组合的维护策略安排提供有效的决策支持。由于反映惯性测量组合退化状态的性能指标不能直接监测或直接测量带有噪声,因此需要构建状态空间模型预测惯性测量组合的剩余寿命。考虑到惯性测量组合的性能退化指标随时间呈现非线性特征,首先采用基于非线性漂移的Brown运动(Brownian motion,BM)建模其退化状态,然后基于构建的状态空间模型,利用期望最大化(expectation-maximization,EM)算法和Kalman滤波(Kalman filter)实时估计和更新退化状态和模型未知参数。并且将状态估计的分布函数引入剩余寿命的预测过程,近似得到了剩余寿命分布的解析形式,实现了剩余寿命的实时预测与更新。最后,对惯性测量组合的剩余寿命实时预测问题进行了实验分析,结果表明该方法具有较高的预测精度与较小的预测不确定性。

Real-time and accurate residual life prediction for an inertial measurement unit （IMU） can provide effective decision support [or timely and cost e{{ective maintenance scheduling. The performance index reflecting the degradation of the IMU cannot be observed directly and direct measurements are contaminated by noise. Thus, a state space model was developed to predict the residual life of an IMU. Since the changes in the degradation state of the 1MU are nonlinear over time, this analysis was a nonlinear drift driven Brownian motion （BM） is used to characterize the degradation process, with the expectation maximization （EM） algorithm and the Kalman filter used to jointly estimate and update the state and model parameters. Furthermore, the estimated state distribution is incorporated into the residual life model using an approximate analytical form of the distribution. The approach is validated by comparison with experimental data which indicates that this method gives better prediction accuracies and lower uncertainties.