基于SVD-UKF的车用锂离子动力电池传感器故障诊断研究

针对传感器故障诊断问题，提出通过分析模型预测电压与传感器观测电压的残差来诊断传感器是否发生故障的方法。使用无迹卡尔曼滤波（Unscented Kalman Filter,UKF）算法估计电池的端电压，并提出使用奇异值分解（Singular Value Decomposition,SVD）代替平方根法分解，以解决协方差矩阵非正定导致的算法无法正常运行的问题。提出使用累积和（Cumulative Sum,CUSUM）法对残差进行分析，通过监测CUSUM的变化来判断传感器是否发生故障。以动应力测试（Dynamic Stress Test,DST）工况作为验证工况，用3类常见的传感器故障对提出的方法进行验证。结果表明，提出的传感器故障检测方法在一些微小故障的检测中，相比于传统的设定残差阈值的方法更灵敏，能更快检测出微小的数据偏移情况。

Fault Diagnosis for Sensors in Automotive Li-Ion Power Battery Based on SVD-UKF

A sensor fault diagnosis method is proposed which analyzes the residual difference between the predicted voltage and the observed voltage. Firstly the Unscented Kalman Filter (UKF) algorithm is used to estimate the terminal voltage of the battery. Singular value decomposition (SVD) is applied instead of Cholesky decomposition so that the algorithm will work normally even though the covariance matrix is not positive definite. Then the method of cumulative sum is used to analyze the residual error, monitoring cumulative sum changes and determining the malfunctioning sensors. Finally the dynamic stress test (DST) condition is carried out to verify the proposed method for three common types of sensor faults. The results show that compared with the traditional residual-based methods with certain thresholds, the proposed sensor fault detection method is more sensitive to identify some small faults and the small data offsets can be detected more quickly.