Deviations between High Voltage (HV) current measurements and the corresponding real values provoke serious problems in the power trains of Electric Vehicle (EVs). Examples for these problems have inaccurate performance coordinations and unnecessary power limitations during driving or charging. The main reason for the deviations are time delays. By correcting these delays with accurate Time Delay Estimation (TDE), our data shows that we can reduce the measurement deviations from 25% of the maximum current to below 5%. In this paper, we present three different approaches for TDE. We evaluate all approaches with real data from power trains of EVs. To enable an execution on automotive Electronic Control Unit (ECUs), the focus of our evaluation lies not only on the accuracy of the TDE, but also on the computational efficiency. The proposed Linear Regression (LR) approach suffers even from small noise and offsets in the measurement data and is unsuited for our purpose. A better alternative is the Variance Minimization (VM) approach. It is not only more noise-resistant but also very efficient after the first execution. Another interesting approach are Adaptive Filter (AFs), introduced by Emadzadeh et al. Unfortunately, AFs do not reach the accuracy and efficiency of VM in our experiments. Thus, we recommend VM for TDE of HV current signals in the power train of EVs and present an additional optimization to enable its execution on ECUs.

Deviations between system current measurements and real values in the power train of Electric Vehicles (EVs) can cause severe problems. Among others, these are restricted performance and cruising range. In this work, we propose a fleetbased framework to correct such deviations. We assume that the real value is the mean of all identically constructed EVs' measurements for the same input. Under this assumption, we decide for each vehicle whether it displays hardware errors with the help of a binary classifier. Depending on the classification, if no hardware errors are detected, we recover the parameters of an assumed measurement error model via Linear Regression. Otherwise, we combine the regression with a convex optimization problem and sparsity constraints. We achieve an overall recovery rate of up to 90 %, allowing the full automation of the measurement correction procedure with no need to add more sensors, or computational units on-board of the EV.