TNY280GN 50-100Hz rippel output

Hello,

We have some problems with a circuit with TNY280GN, that we have some very hard time solving.
We have done a PCBs with this solution before (same main core), but now suddenly something strange have come in.
It runs on 230VAC 50Hz and the output is constant 700mA current over 5 LEDs (around 17V).

The problem is that we get a 50Hz ripple on the output (around 200mV and 100mA). From the beginning it's 50Hz but after a while (1-5min) the devices get hot (around 40-60 Celsius) and the ripple goes over to 100Hz. It looks like one of the 50Hz (top) is getting smaller and "moves" closer to the next top, and then "eats up" of it, and a 100Hz ripple is made.

In the attached schematic the most critical components are there. I have also made a red X over wires that are cut and components that have been removed (one by one). The MOSFET is also shortcircuited.

In the other picture is over the osc. The blue is the primary ripple (30VDC, goes from around 280V to 310V).
The yellow is the current through the LEDs, (100mA ripple, 600mA to 700mA).
The pink one is the ripple over the LEDs, it’s around 2-300mV (the noise disappear if the osc-ground is taken away from the primary side).
A Tektronix TDS2024 with 4 isolated channels has been used.

More info:
OP U5 measure the voltage over a R4 and compare it to a fixed value.
Both inputs and outputs on the OP are not OK, they have the same ripple.
2 main difference from the old (working PCB) is that we now have 5LEDs instead of 4, (now 13W out and before 10W). Plus that another 22uF was on the primary side (32uF together).
We dismounted the LEDs and made it run on constant voltage instead (Z1 act as an overvoltage protection). Then we added some load, making it go up to around 18V and 750-800mA (R4 was in this case shortcircuit). Now the ripple was gone!
We also added a 68uF on the primary side (using the 5 LED on the output), this also took away the ripple!
We tested with changing the input voltage (through an adjustable trafo), but the ripple was there more or less the hole time. Going down to around 170VAC did make it bigger.

One thought that we had was that the primary voltage went down to much (empty the capacitor), but if so, how come that it worked with an external load. And it only goes down to 270V, shouldn’t the TNY280 fix this with longer/shorter pulses? It works on around 130KHz, so I don’t think the 50/100Hz should go through the trafo, (to the secondary side.

Please help us, we don’t know what to do any more!

NEW NOTES !!!
With the current sense OP supplied from an external power supply:
If I cool down the TNY280 (with cooling spray), then the worst ripple will go away!
If I look at the outputs, I can see that the switches from TNY280 happens at around 66-67kHz. Why aren't they around 130kHz?
When I measure on the top of TNY280 the (black) case is around 68-69 degrees Celsius. If I cool it down to around 65-66 degrees the ripple more or less goes away. If I heat it up (tried to maximum 73 degrees), then the ripple got much worse (still around 100-130Hz but up 3-400mV).