Unsupervised domain adaptation for dynamic in weighing motion system of freight rail carriages under varying ballast conditions

We have recently developed a novel Dynamic Weight In Motion (DWIM) system for freight carriages, which is based on a combination of Discrete Fourier Transform (DFT) and ELastic NET (ELNET) linear regression. To improve this method, we propose an unsupervised domain adaptation method based on subspace alignment. This learns a mapping function to align the features extracted from data of carriages of known load in a given DWIM setting with those extracted from carriages of unknown load in another positions. The application of the proposed method provides promising results.