I am using a similar circuit to speed control a 60 W AC fan using phase control. Unlike a TRIAC, supply to fan is given at the start of the cycle. I thought it would minimize the switching noise usually heard in TRIAC controls.
PWM is from 0 to 10 milliseconds. In low PWM, the MOSFET heats a lot with an inductive load, but not with a resistive load. A snubber using a 0.1 µF capacitor and 100 or 39 ohm resistors is connected across the MOSFET's source and ground pins.
What should I do?
See this AC PWM Dimmer for Arduino article on instructables, which says:
problems start, because he is feeding the gate from the MOSFET, with a voltage that is shorted by that same MOSFET. In other words, if the MOSFET is fully opened the DC voltage coming from the rectifier is completely shorted. Therefore there will be no voltage anymore to put on the gate and the MOSFET will block again. This effect might not be so outspoken by a low dutycycle (= lamp on a low intensity), because of the presence of C1, that will retain its charge for a while and will be receiving new charge thanks to the low dutycycle, but at 25-80% dutycycle the voltage on C1 just cannot be sustained anymore and the lamp may start to flicker. What's worse is that at moments that the voltage on the gate drops, for a while the MOSFET will be still conducting, but not be fully saturized: it will slowly go from its nominal 0.04 Ohm resistance to infinite resistance and the slower this goes, the higher the power that needs to be dissipated in the MOSFET. That means a lot of heat. MOSFETS are good switches but bad resistors. They need to be switched ON and OFF fast. Currently the circuit heavily relies on D1 to keep the voltage on the gate of T1 at acceptable limits while the voltage is swinging between 0 Volt and Full peak At peak the rectified voltage is 230x1.4=330V The average rectified voltage is 230x0.9=207V
If we forget about the smoothing effect of the capacitor for a while and presume the optocoupler to be fully open the average voltage on the capacitor would be 22/88 * 207 =52 Volts and in peak 22/88 * 330= 83 Volts. That it is not is because of D1 and the fact that the MOSFET will short the Voltage.
If the optocoupler is not in saturation and its impedance therefore infinite, the capacitor C1 would charge up to full rectified voltage if not for D1. On average 3mA will flow through R3,R4 and R5 (207-10)/66k which equals a power consumption of 0.6 Watt in the resistors R3,R4, R5
First of all, this circuit cannot be used to control inductive loads. T1 is switched asynchronously with the mains frequency and this can cause DC current to flow. The reason that you can see this effect in low PWM is that the voltage across D1 remains the same (10 V) to about 90% of the duty cycle span. So T1 conducts a little longer than you would expect from PWM. At a higher duty cycle the voltage drops and T1 starts to conduct sufficiently.
In addition, the snubber dissipates power as heat. The snubber will have a different effectiveness at different frequencies. You need to choose the values for R and C to suit the frequencies you want to work with.
For inductors,
$$V = L \frac{di}{dt}$$
PWM is an on-off style switch and cutting the supply current from the inductor instantaneously will generate a tremendous reverse voltage that will most likely break your MOSFET.
simulate this circuit – Schematic created using CircuitLab
Figure 1. Schematic stripped down to essential current path with MOSFET represented by a switch and re-organised for clarity.
Figure 1 may help to understand the problem.
In a DC circuit Figure 1b would have a snubbing diode wired in parallel with LAMP2 and pointing upwards (anode to N). Figure 1c would have it pointing downwards (cathode to N). It should be clear that we can't have the diode pointing both ways and so we can't use a snubbing diode for an inductive load.
Your options would be to use an RC snubber but we don't have enough information to help you with that.
If you use this to drive an inductive load, you're very likely to fry T1.
When the PWM signal goes low, T1 will attempt to interrupt the current while the load tries to maintain it. Result: high voltage will be induced till something breaks.
You could use a big ass zener (avalanche diode actually) across the transistor as a snubber. This will limit the back-EMF voltage from the load to safe levels.
Having some capacitance in parallel with the inductive load would be nice too.
This dimmer outputs a rectified AC voltage which is basically unfiltered DC. An inductive load current with a DC source is only limited by the Resistance of the coil. This creates a high current through the components which causes an overheating and ultimately the destruction of the motor and the Mosfet.
