davidwkerr
Junior Member level 2
Well, I finished this project and have been using it. I learned quite a lot in the process and thanks in particular to FvM who provided insightful and knowledgeable comments. So, I thought I would summarise some of what I learned here. That might repeat some parts of other posts but it could benefit someone else.
Firstly, the project. I had bought an inverter generator and used it for a few hours before travelling extensively overseas. This was a 2,500W (so labelled) pure sine-wave inverter. Well, it stopped working after minimal use, just outside the warranty period. I deduced that it was the inverter module that died. So, with some difficulty, I got in touch with the supplier and they supplied a new module even though it was out of warranty. They said the new module was for a 3,000W machine, but "should fit". Well, that module lasted about 30mins before also dying. These modules a a solid "brick" of black epoxy, from which the tops of capacitors, toroids and a few other components protrude.
So, I decided to "pull apart" a module. Easier said than done. It took a 530degC heat gun to soften the epoxy which I then chipped away. Here were my discoveries:
1. The module appeared well made with good quality and recognisable components. However, the one for the 2,500W inverter had a label rating it at 1,800W max. The one for the 3KW inverter was labelled 2KW max. The design looked fairly standard with an H-bridge using Infineon IGBTs driven from a uP via TLP250 opto isolators. It appeared to run at 20KHz. However, the permanent magnet alternator can put out 1100V peak three phase and they used series output capacitors with balancing resistors and a total rating of 800DC max. Clearly they relied upon the uP keeping the revs of the motor below the point of dangerous voltages. The output toroidal inductors were another problem. They had quad windings (because of skin effect) and the insulation had broken down on one of them (but only at high applied voltage/frequency). Two IGBTs were also zapped. I really did not like the idea of a DC link voltage around 800VDC or greater. They used a three phase bridge rectifier straight from the PMA to the capacitors.
2. I think the manufacturers had taken a generic inverter module which was intended for a lower maximum input voltage and also a lower rated unit and coupled it with a different engine and PM alternator.
3. So, I decided to build my own inverter module and NOT epoxy it so that I could easily repair it should such be necessary. I designed for 3KW.
4. This proved harder than I expected. I am a very experienced uP design person and also very experienced in VHF, audio, low voltage power management, MPPT solar chargers and similar stuff. I had used FETs a lot but never IGBTs. I had never worked with high voltage power such as in this project. I knew the basics but had no practical experience.
5. The first difficulty was a pre-regulator. I wanted to use IGBT drivers which typically go to 600V. I also wanted to use TO220s but in retrospect I should have gone straight to TO247 packages. Indeed, I would use TO220s under 1KW but next time TO247s above that. My first few attempts at a pre-regulator used chopper stabilised IGBTs. I blew up quite a few with the main problem being big inductive spikes when commutating the alternator under load. I did get the IGBT design working but was uncomfortable about it. It was also full of very high voltage devices which were expensive and things to capture the inductive spikes, store them and then re-use the energy. I had started with a TRIAC pre-regulator and finally returned to that and am happy I did. No more inductive spikes, much more rugged and response time is satisfactory when I suddenly turn off a big load with the engine throttled up (by the uProcessor). My final problem with the pre-reg was response time. I found that the opto-couplers I was using (led-photoresistor) were much slower than their specification when turning off. So, I substituted a different product and reversed the sense of the optos so that their faster turn-on response was used to drop the regulator voltage rather than increase it. So, all will be fine as long as an opto never fails (this would give maximum output volts).
6. I built a fast fused crowbar for both overvoltage (in case the pre-reg fails) and overcurrent. The final inverter design can supply 45amps at 240V for a brief time to handle large electric motors. However, the over current crowbar was always too sensitive. This is because of very fast IGBT switching and track/wire inductances. So, in the end, I run with it disconnected. The inverter might or might not survive a short circuit (probably will) but in the unlikely eventuality, I will just replace a couple of IGBTs. Next time, I will know enough and do a better PCB design to handle the crowbar rather than adding it at the end of the project.
7. As you will see from earlier posts, I started with an H bridge with the low IGBTs commutated at 50Hz and the high side modulated at 20KHz. So, I had a square wave at low/no load. Other people on this board have experienced the same and the reasons are simple once light dawns. So, I changed my software to be high-speed interrupt driven rather than hardware PWM driven (I only had 2 hardware PWM devices). I also took the opportunity to change to “Magic Sine waves”. This has been successful. I am using 19 points per quadrant. The uP generates two 16bit lookup tables, one of which is in use at any particular time. In its idle loop, the uP is constantly doing other stuff such as measuring average RMS current, motor speed, control voltage, DC link voltage and controlling the engine throttle. It is also recalculating the 2 16bit lookup tables as required if a different modulation index is needed to preserve the output 240V. Switching of tables can only occur on a full cycle voltage zero cross.
8. I am very happy with the “Magic sine waves”. I was able to get my timing accurate to about +-300nSecs using a 40MHz PIC 18F2680. The only problem is the 61st harmonic at about 3KHz. It is quite large and agrees exactly with the calculations. I needed to increase my output filter inductances. This was relatively easy. Skin effects are much less at 3Kz than at 20Khz (my original PWM frequency) so I was able to use single thicker wire and fit more turns. Ideally, I would have used a different core material as well. I increased the output capacitance as well. The big plus is no more 20KHz floating around. The overall efficiency is very similar to the original design and nothing gets very hot even at maximum load. The low side IGBTs were a little more expensive than the original slower IGBTs. Speaking of IGBTs, there was a worldwide shortage of the IGBTs I was using so I used a different unit and had to slow them down somewhat as the <10nSec switching caused more inductive spikes than the original IGBTs.
Regards,
Dave
PS- I will mark this solved in a few days.
Firstly, the project. I had bought an inverter generator and used it for a few hours before travelling extensively overseas. This was a 2,500W (so labelled) pure sine-wave inverter. Well, it stopped working after minimal use, just outside the warranty period. I deduced that it was the inverter module that died. So, with some difficulty, I got in touch with the supplier and they supplied a new module even though it was out of warranty. They said the new module was for a 3,000W machine, but "should fit". Well, that module lasted about 30mins before also dying. These modules a a solid "brick" of black epoxy, from which the tops of capacitors, toroids and a few other components protrude.
So, I decided to "pull apart" a module. Easier said than done. It took a 530degC heat gun to soften the epoxy which I then chipped away. Here were my discoveries:
1. The module appeared well made with good quality and recognisable components. However, the one for the 2,500W inverter had a label rating it at 1,800W max. The one for the 3KW inverter was labelled 2KW max. The design looked fairly standard with an H-bridge using Infineon IGBTs driven from a uP via TLP250 opto isolators. It appeared to run at 20KHz. However, the permanent magnet alternator can put out 1100V peak three phase and they used series output capacitors with balancing resistors and a total rating of 800DC max. Clearly they relied upon the uP keeping the revs of the motor below the point of dangerous voltages. The output toroidal inductors were another problem. They had quad windings (because of skin effect) and the insulation had broken down on one of them (but only at high applied voltage/frequency). Two IGBTs were also zapped. I really did not like the idea of a DC link voltage around 800VDC or greater. They used a three phase bridge rectifier straight from the PMA to the capacitors.
2. I think the manufacturers had taken a generic inverter module which was intended for a lower maximum input voltage and also a lower rated unit and coupled it with a different engine and PM alternator.
3. So, I decided to build my own inverter module and NOT epoxy it so that I could easily repair it should such be necessary. I designed for 3KW.
4. This proved harder than I expected. I am a very experienced uP design person and also very experienced in VHF, audio, low voltage power management, MPPT solar chargers and similar stuff. I had used FETs a lot but never IGBTs. I had never worked with high voltage power such as in this project. I knew the basics but had no practical experience.
5. The first difficulty was a pre-regulator. I wanted to use IGBT drivers which typically go to 600V. I also wanted to use TO220s but in retrospect I should have gone straight to TO247 packages. Indeed, I would use TO220s under 1KW but next time TO247s above that. My first few attempts at a pre-regulator used chopper stabilised IGBTs. I blew up quite a few with the main problem being big inductive spikes when commutating the alternator under load. I did get the IGBT design working but was uncomfortable about it. It was also full of very high voltage devices which were expensive and things to capture the inductive spikes, store them and then re-use the energy. I had started with a TRIAC pre-regulator and finally returned to that and am happy I did. No more inductive spikes, much more rugged and response time is satisfactory when I suddenly turn off a big load with the engine throttled up (by the uProcessor). My final problem with the pre-reg was response time. I found that the opto-couplers I was using (led-photoresistor) were much slower than their specification when turning off. So, I substituted a different product and reversed the sense of the optos so that their faster turn-on response was used to drop the regulator voltage rather than increase it. So, all will be fine as long as an opto never fails (this would give maximum output volts).
6. I built a fast fused crowbar for both overvoltage (in case the pre-reg fails) and overcurrent. The final inverter design can supply 45amps at 240V for a brief time to handle large electric motors. However, the over current crowbar was always too sensitive. This is because of very fast IGBT switching and track/wire inductances. So, in the end, I run with it disconnected. The inverter might or might not survive a short circuit (probably will) but in the unlikely eventuality, I will just replace a couple of IGBTs. Next time, I will know enough and do a better PCB design to handle the crowbar rather than adding it at the end of the project.
7. As you will see from earlier posts, I started with an H bridge with the low IGBTs commutated at 50Hz and the high side modulated at 20KHz. So, I had a square wave at low/no load. Other people on this board have experienced the same and the reasons are simple once light dawns. So, I changed my software to be high-speed interrupt driven rather than hardware PWM driven (I only had 2 hardware PWM devices). I also took the opportunity to change to “Magic Sine waves”. This has been successful. I am using 19 points per quadrant. The uP generates two 16bit lookup tables, one of which is in use at any particular time. In its idle loop, the uP is constantly doing other stuff such as measuring average RMS current, motor speed, control voltage, DC link voltage and controlling the engine throttle. It is also recalculating the 2 16bit lookup tables as required if a different modulation index is needed to preserve the output 240V. Switching of tables can only occur on a full cycle voltage zero cross.
8. I am very happy with the “Magic sine waves”. I was able to get my timing accurate to about +-300nSecs using a 40MHz PIC 18F2680. The only problem is the 61st harmonic at about 3KHz. It is quite large and agrees exactly with the calculations. I needed to increase my output filter inductances. This was relatively easy. Skin effects are much less at 3Kz than at 20Khz (my original PWM frequency) so I was able to use single thicker wire and fit more turns. Ideally, I would have used a different core material as well. I increased the output capacitance as well. The big plus is no more 20KHz floating around. The overall efficiency is very similar to the original design and nothing gets very hot even at maximum load. The low side IGBTs were a little more expensive than the original slower IGBTs. Speaking of IGBTs, there was a worldwide shortage of the IGBTs I was using so I used a different unit and had to slow them down somewhat as the <10nSec switching caused more inductive spikes than the original IGBTs.
Regards,
Dave
PS- I will mark this solved in a few days.