Adaptive unscented Kaiman filter with a fuzzy supervisor for electrified drive train tractors

Fuzzy-logic assisted power management for electrified mobile machinery

“…In the class one, the normalized root mean square error (NRMSE) between the measured and actual power consumption lies in 2-4%. This NRMSE is in agreement with previously obtained results on tractor single wheel drives Osinenko et al [21]. In the class two, the NRMSE is within 4-6% and 8-12% in the class three which is rather extreme.…”

Experimental verification of an online traction parameter identification method

“…The root mean square error is normalized by the difference between maximum and minimum value of the observations. The overall performance is comparable to earlier works, see (Osinenko et al, 2014) and (Osinenko et al, 2016).…”

“…The empirical model from (Schreiber and Kutzbach, 2007) is used as a basis in this work. The equation was first modified in (Osinenko et al, 2014) by replacing a linear term with an exponential one und further refined in (Osinenko and Streif, 2017), by reducing the number of parameters by one:…”

“…The choice of noise covariance Q is crucial -a poor choice of Q may result in divergence issues or noisy estimates (Fitzgerald, 1971). The standard UKF was hence modified by adding an adaption matrix A k which replaces the state noise with A k Q (for an extensive description, please refer to (Osinenko et al, 2014)). The adaption matrix A k minimizes the difference between true measurements and measurement estimate, which helps avoiding divergence issues, however the estimates may become too noisy.…”

“…First of all, in a previous work (Kobelski et al, 2020) an UKF-based identification algorithm was used to estimate traction properties of a tractor driving over different ground types in a simulation. It was based on the methodology developed in (Osinenko et al, 2014(Osinenko et al, , 2016Osinenko and Streif, 2017). In (Osinenko et al, 2016) the adaptive unscented Kalmanfilter with a fuzzy supervisor was used to assist traction control, while (Osinenko and Streif, 2017) improved the model for the adhesion slip ratio characteristic curves.…”