Kinematic Phase Segmentationand Energy Evaluation ofElectric Forklift Travel Tests : A Telemetry Workflow for Energy Analysis andPrediction

Sammanfattning

This thesis investigates how telemetry data from Toyota RAE250 auto reach truck travel tests canbe used to identify kinematic phases and directional test legs, analyse energy consumption, andpredict instantaneous electrical power. The industrial need is to reduce the manual effort requiredto separate repeated forward and reverse travel movements and calculate comparable energyvalues across test files. The study uses high frequency signals including speed, battery current, and battery voltage,together with metadata describing operating conditions such as payload, battery state of charge,temperature condition, and speed setting. The work covers data preparation, signalpreprocessing, kinematic phase segmentation, directional test leg extraction, energy computationfor individual kinematic phases and directional test legs, and sequence based power prediction.Several segmentation approaches were investigated, including change point detection andclustering methods. Based on the results, K-means clustering combined with physics based ruleswas selected as the most suitable segmentation strategy for the available dataset. The hybrid method divided each recording into seven kinematic phases and extracted thedirectional test legs. It was validated against a manual supervisor analysis covering 26 files andachieved a mean absolute percentage error of 1.63% for total directional test leg energy. Foursequence based neural network architectures were then trained and compared for instantaneouspower prediction: CNN1D, GRU, LSTM, and TCN. The TCN achieved the best test performance,with the lowest prediction error and the highest coefficient of determination. The results show that the segmented RAE250 telemetry provides a reliable basis for supervisedsequence modelling and that convolutional architectures are well suited to the repeated motionstructure of the travel tests. The prediction study provides a baseline for future estimation of powerdemand and directional test leg energy under operating conditions that may be tested orsimulated later.