Electromagnetic compatibility testing, it is very important to find the right idea

I'm doing electromagnetic compatibility testing recently. In the actual test, it was found that the system would crash when the load was switched on and off. And the normal work of the products we make is not allowed to resume the machine, and the crash is a more serious problem. Find the cause.
First of all, I suspect that the interference is conducted from the power supply terminal. The system power supply will have a large TVS tube, add a 3300UF capacitor, pass through a common mode inductor, and then isolate it through DC/DC. As shown below:
The MCU of the system should not be affected, and the load solenoid valve will still crash when it keeps opening and closing, and it has entered a dead end. This path should be wrong.

Replace the wiring harness and test again, because our system is connected to the outside world through the wiring harness. Try to connect only the power supply and one load output so that the system will not crash when the load solenoid valve is switched on and off.
Restore the scene again, use the original harness, it will still crash, basically eliminate the interference coupled from the power supply terminal.
Since the system needs to be external, the line is also led out. Observe the C2CK pin of the line, when the load solenoid valve is opening and closing, there is 5-8V AC interference coming in.
String an inductor on the C2CK pin and connect a small capacitor to the ground, and then switch the load solenoid valve on and off. The system will not crash and the problem will be solved.
This crash has been troubled for a long time, and sometimes I feel really helpless. I feel that the engineer just finds and solves the problem, and I am afraid that the problem is not knowing where it is.
I once made an industrial control board, and the crash caused by the interference introduced by AD sampling, the cutting line, and the inductance were solved.
Another project that I did recently was also because the AD sampling ground was too close, and interference was introduced into the MCU, causing the system to crash. I used to copy Japanese industrial control boards, but I didn’t quite understand how to add inductors, capacitors and π-type filtering to the AD sampling terminal. Now come back and think about it. Mainly the problem of anti-interference.