In the case of a railway vehicle running on track, irregularities in track geometry can lead to complex dynamic interaction between the vehicle and track. The resulting dynamic forces accelerate the track deformation process and may lead to a derailment situation. Early detection of undesirable vehicle-track interaction allows for timely intervention before the situation presents a serious derailment risk. Accelerometers mounted on the vehicle body produce signals which may provide enough information to autonomously detect and classify dynamic interactions of this nature. This paper develops an understanding of acceleration signals and introduces an autonomous analysis method suitable for low power embedded applications using a time-frequency signal representation. The method is demonstrated on a field data sample collected from a ballast hopper wagon with three-piece bogies running on low-grade track.