Rail transport is one of the most important modes of land transport for both passengers and freight. Its reliability, safety and sustainability make rail the main alternative to road transport. However, the operation and maintenance of railway equipment and infrastructures generate significant operating costs that make it unattractive. In this line and in order to obtain sustainable and affordable transport infrastructures, the shift towards predictive maintenance, such as the use of new condition monitoring systems, will be of special interest given the high maintenance cost of railway equipment. Monitoring systems will make it possible to reduce the number of maintenance interventions to be carried out and thus reduce the economic cost. In this work the preliminary design of a condition monitoring system is proposed. The system uses Artificial Intelligence based on neural networks net trained, capable of identifying changes in railway axles structure from vibration signals providing of a bogie. As an example of application the system will be trained to identify a axle cracks. Data used for this purpose is obtained from a real bogie in a scaled 1/1 testbench. the results show the advantages and drawbacks of this type of systems and the new steps.