It seems to me that a boil over occurs when the bubble holdup volume added to the liquid volume exceeds the vessel's operating volume. Surface level detection devices such as radar might indicate the swell in total volume.

The bubble holdup volume can be approximated as the bubble rise time multiplied by the volumetric vapor rate determined by the heat flux rate of heat input divided by the heat of vaporization for the process fluid and the density of the vapor. However, I am not sure how to generalize a calculation of the bubble rise time. Also, my experience is that a boil over develops very rapidly possibly indicating a drastic change in bubble path and hence rise time.

A more empirical method would be to compare an unfiltered velocity limited rate of change of vessel pressure or pressure drop across the vent valve to a baseline rate of change. The baseline value is a heavily filtered average rate of change. Any significant increase in the unfiltered velocity limited rate of change would signal an impending boil-over. The velocity limiting helps screen out noise from triggering a false indication of a boil-over. Also, passing the velocity limited signal through a dead time block and computing the rate of change as the difference between the input (new value) and the output (old valve) of the dead time block divided by the block dead time can improve the signal to noise ratio. Here, the block dead time is chosen just large enough to show a true change in pressure. Not that the block dead time does not affect the speed of response since the input to the dead time block is used in the computation of the rate of change.

Whether this method works fast enough depends upon the process dead time. If the signal is noise, some light filtering might need to be applied before sending the pressure signal through the velocity limit block. Also, on startup, it may be necessary to initially turn off filtering and velocity limiting until the vessel reaches operating conditions.