Simulink’s variable-step solvers automatically shrink time steps when signals change rapidly—near discontinuities this can mean many tiny steps (“chatter”) or even stepping over a brief sign flip. Zero-crossing detection watches the signals for sign changes between steps. There are two algorithms in Simulink for zero-crossing detection:
The Nonadaptive algorithm (the default) always brackets every detected zero-crossing event and uses increasingly smaller time steps to pinpoint when the crossing occurred. Because it has no built-in tolerance window, simulations with high-frequency oscillations around the crossing point can experience very long run times or even halt due to excessive consecutive events—unless you raise the consecutive-crossings limit, which still may not let the run complete.
The Adaptive algorithm dynamically turns bracketing on and off. It stops iterating once either the zero-crossing error falls within the specified Signal Threshold (creating a dead-band around zero) or the number of consecutive zero crossings reaches the user-set maximum. This guard-band and event-count cap let you trade a small amount of locating precision for much faster, more robust simulations in the presence of chatter.
The model at the following page represents an example system with Abs, Saturation and Switch blocks which are few blocks where a zero crossing can commonly occur.
The output of the scope indicates how Simulink successfully handles zero-crossing.
For a better explanation refer the example at the following page:
