The High Luminosity Large Hadron Collider (HL-LHC) upgrade aims for a tenfold increase in integrated luminosity compared to the nominal Large Hadron Collider (LHC), and for operation at a leveled luminosity five times higher than the nominal LHC peak luminosity. In order to compensate the geometric luminosity loss due to the increased crossing angle, crab cavities will be used to transversely rotate the beam bunches, allowing quasi head-on collisions at the experiments. Crab cavity failures can be very fast, having time constants similar to the reaction time of the machine protection system. In such a scenario the beams cannot be immediately extracted, making the protection of the machine fully rely on passive protection devices such as the collimation system. At the same time the energy stored in the HL-LHC beams will be doubled with respect to the LHC to more than 700 MJ, which increases the risk of damaging the machine and the experiments in case of failure. Crab cavity failures have the potential to displace the beam core and create considerable particle losses around the machine, posing a machine protection challenge. Any increase in failure rates will be difficult to compensate, affecting the performance of the HL-LHC and therefore the integrated luminosity goal. This is why it is important to correctly interlock the machine and make sure that certain types of failure never happen. In order to do this advanced simulations are needed well before the crab cavity prototypes can be tested in real operating conditions. This thesis analyzes different failure scenarios for crab cavities installed around the ATLAS and CMS experiments. The selected failure scenarios are later simulated with the tracking code SixTrack thanks to a newly developed functionality. The distribution of the particle losses in space and time are analyzed for the different failure cases and a quantitative estimate of the impact in the collimation system is given. The results are analyzed from a machine protection point of view, where the time for the beam abort trigger is calculated for each failure case and mitigation techniques are proposed. These results allow identifying corner cases corresponding to the most dangerous crab cavity failure scenarios, serving as input for the design of the future interlocking system.