This paper proposes an approach for locating faults in a distribution grid by utilizing data measured and gathered by distributed converters. The data, comprising grid voltages and impedances from multiple locations, is processed using multinomial logistic regression, a machine learning algorithm, to classify a fault location in the grid. The algorithm is first trained with simulation data, followed by evaluation of its predictive performance using a set of test data previously unseen by the algorithm. The fault location accuracy of the proposed approach is found resonable and encourages further studies of the unused potential in the converters.