Due to the transition to electric mobility the interest for the battery as core component of every battery electric vehicle (BEV) and its State of Health (SOH) rises. Multiple supervised machine learning methods have been applied to the regression task of SOH estimation of lithium-ion batteries from the charging curve under laboratory operational conditions [1–6]. However, little research has been done for this task using data from battery usage under real-world operational conditions [7–9]. Yu et al.  have proposed a linear regression model based on features of curves with variable discharging currents. Their features contain noise smoothed with the exponentially weighted moving average. This is bad because one sample depends on the previous. Hong et al.  work with variable discharging currents and variable charging durations. However, they only use the reference cycles for their SOH estimation, but not those cycles with variable charging duration.
Real-world operational conditions of batteries are characterized by variable discharging currents, variable duration of discharging and charging, variable deltas of the State of Charge (SOC) as well as long rest periods. These conditions complicate the SOH estimation task. This work’s objective is to examine model performance of SOH estimation on different data sets from laboratory towards real-world operational conditions. Therefore, we apply a Multi-Layer Perceptron (MLP), multi-headed CNN and a regression Transformer to estimate the SOH from the charging curves of current, voltage and temperature. Three data sets with rising similarity to real-world operational conditions are used: 1) CC-CV charging with constant duration, constant discharge current; 2) Same as data set 1), but variable discharge current; 3) same as data set 2) with variable charging duration [10, 11]. Data sets 1) and 2) have a delta SOC of 100%.
Our results illustrate the hypothesis that the change from laboratory to real-world operating conditions significantly complicates the SOH estimation task and regression models provide correspondingly poorer regressions. The mean RMSE on the validation data relatively to data set 1) increases by 389% for data set 2) and by 807% for data set 3). This is caused by the increasing overlapping of charging curves of the same ages for data set 2) and 3). Keeping relaxation times would probably ease the distinguishability of the charging curves. There is no indication that any of the applied machine learning methods outperforms one of the others. Feature engineering comparable to the state of the art has not been examined.
We conclude that the presented methods reveal challenges at SOH estimation with real-world operational data which are also faced by methods integrated in Battery Management Systems. However, battery operational data with such SOH labels is required for data-driven SOH forecasting and analysis of ageing causes.
 J. Wu, Y. Wang, X. Zhang, and Z. Chen, “A novel state of health estimation method of li-ion battery using group method of data handling,” Journal of Power Sources, vol. 327, pp. 457–464, 2016, doi: 10.1016/j.jpowsour.2016.07.065.
 Y. Li and J. Tao, “CNN and transfer learning based online SOH estimation for lithium-ion battery,” 2020 Chinese Control And Decision Conference (CCDC 2020), 2020.
 Y. Tan and G. Zhao, “A novel state-of-health prediction method for lithium-ion batteries based on transfer learning with long short-term memory network,” IEEE Trans. Ind. Electron., p. 1, 2019, doi: 10.1109/TIE.2019.2946551.
 D. Yang, X. Zhang, R. Pan, Y. Wang, and Z. Chen, “A novel Gaussian process regression model for state-of-health estimation of lithium-ion battery using charging curve,” Journal of Power Sources, vol. 384, pp. 387–395, 2018, doi: 10.1016/j.jpowsour.2018.03.015.
 J. Jia, J. Liang, Y. Shi, J. Wen, X. Pang, and J. Zeng, “SOH and RUL Prediction of Lithium-Ion Batteries Based on Gaussian Process Regression with Indirect Health Indicators,” Energies, vol. 13, no. 2, p. 375, 2020, doi: 10.3390/en13020375.
 L. Ungurean, M. V. Micea, and G. Cârstoiu, “Online state of health prediction method for lithium-ion batteries, based on gated recurrent unit neural networks,” Int J Energy Res, vol. 43, no. 14, p. 8230, 2020, doi: 10.1002/er.5413.
 J. Yu, B. Mo, D. Tang, J. Yang, J. Wan, and J. Liu, “Indirect state-of-health estimation for lithium-ion batteries under randomized use,” Energies, vol. 10, no. 12, p. 2012, 2017, doi: 10.3390/en10122012.
 S. Hong, M. Kang, H. Jeong, and J. Baek, “State of Health Estimation for Lithium-Ion Batteries Using Long-Term Recurrent Convolutional Network,” in IECON 2020, pp. 1929–1934.
 R. R. Richardson, M. A. Osborne, and D. A. Howey, “Gaussian process regression for forecasting battery state of health,” Journal of Power Sources, vol. 357, pp. 209–219, 2017, doi: 10.1016/j.jpowsour.2017.05.004.
 B. Saha and K. Goebel, Battery Data Set. [Online]. Available: https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/ (accessed: Jan. 4 2021).
 B. Bole, C. S. Kulkarni, and M. Daigle, Randomized Battery Usage Data Set. [Online]. Available: https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/ (accessed: Feb. 4 2021).