Data-Driven Lead-Acid Battery Prognostics Using Random Survival Forests

Problems with starter batteries in heavy-duty trucks can cause costly unplanned stops along the road. Frequent battery changes can increase availability but is expensive and sometimes not necessary since battery degradation is highly dependent on the particular vehicle usage and ambient conditions. The main contribution of this work is a case-study where prognostic information on remaining useful life of lead-acid batteries in individual Scania heavy-duty trucks is computed. A datadriven approach using random survival forests is proposed where the prognostic algorithm has access to fleet management data including 291 variables from 33603 vehicles from 5 different European markets. The data is a mix of numerical values such as temperatures and pressures, together with histograms and categorical data such as battery mount point. Implementation aspects are discussed such as how to include histogram data and how to reduce the computational complexity by reducing the number of variables. Finally, battery lifetime predictions are computed and evaluated on recorded data from Scania’s fleet-management system.