Around 40? Around 180?
Fuzzy words like that are difficult to make work in a mathematical computation. You need to quantify exactly what the word means, and then write code that identifies the event.
I might suggest writing a code that identifies those regions where the data is well represented by a constant, so any new point added to a constant model falls outside of a prediction interval around that model. But even there, you will find regions where that will be problematic. It would help if you knew the noise in your data. My estimatenoise tool (on the file exchange) might be of service to you, as it tries to compute the noise level in a time series.