Take the CPU sampling when this problem occurs to understand the relevant methods causing this problem.
The overview will be like this and then you can click on the show sample to see all the methods and also the dynaTrace instrumentation CPU overhead if any.
Since the dark blue "Instrumented Processes" line is shooting up, we know the CPU is consumed by one (or more) of the processes being monitored on this server. Just look at the CPU charts for the Instrumented Processes to see which one is the culprit and take CPU Samples/Thread Dumps while the CPU is elevated. You should also check the Response Time Hotspots and the Method Hotspots during the time period to see if you can find the high consumers of CPU for the affected process(es). This should work if there are PurePaths associated with the CPU consumption. However, if the CPU is being consumed by some background threads that are not associated with PurePaths, you will need to rely only on CPU Sampling and/or Thread Dumps to track it down.