The aim of this project has been to find suitable methods for flow measurement and characterization with passive accelerometers. The objectives were twofold. Firstly, the process industry could make use of such a sensor for process surveillance. Secondly, the water utilities of today lack simple and cost-efficient alternatives to equip their ageing infrastructures with online flow meters. These kinds of efforts are necessary for the realization of smart maintenance and for the decrease of the currently increasing amount of maintenance needs water utilities of today are experiencing. Liquid flowing in a pipe generates vibrations, detectable with accelerometers fitted along the pipe exterior. The correlated sound from synchronized accelerometers experience a lag which is dependent on the flow rate. Also, if the acquired sound is further processed, there exist a possibility to extract enough features to estimate some additional characteristics, in this case temperature. Experiments were performed at two nominal temperatures, 20 °C and 40 °C. A deep neural network was constructed for non-linear regression purposes to predict flow velocities based on lag and mean frequencies of the vibrations. Further, a proof of concept for this methodology was shown which reached a root mean square deviation from 100.8 L/min to 171.1 L/min for a nominal flow range of 0 to 1500 L/min. In addition, we train a k-nearest neighbour classifier to predict the nominal temperature of our validation dataset with 83 percent accuracy. The work was performed at RISE Research Institutes of Sweden, serving as Sweden's national metrology institute for liquid flow and acoustics. © 2021 The Authors