My guess is that SMSExec or Hierarchy Manager , components of CM, are restarting SQL if stopped. There are logs for that in <installed location>\Logs, smsexec.log, or hman.log. I believe (but don't know, check the documentation), that "about every 15 minutes", a health check is done.
I think the better question is why you are stopping SQL when CM wants/needs it to be there? If you need to do maintenance on the SQL instance, either also stop the SMS services (and restart them after SQL is back), or look into the high availability options for sql that CM can leverage: