I have never used this but I saw somewhere that an option to get more details is to add a logging to systick interrupt. In the systick interrupt, you can record program counter of instructions it interrupts. Store these addresses somewhere to RAM buffer in a loop. Then you can add short code somewhere to startup files (before it touches the RAM) which will check source of reset. If it is watchdog, you can enter endless loop, so you can attach a debugger and review the RAM buffer. Here you should see if the problem is related to some tight loop.
