The I2K Power Inverter

This project is old and will not be updated - this in an archived page. It was built in 1998 and uses old technology which has been superceded by newer and more efficient technologies. The PWM signal, for example, is genertaed using a software loop ..... nowadays the better approach would be to use a PIC18F252 chip which features an on-chip hardware PWM. And although the design presented is working, the design would benefit from improvements in the area of waveform synthesis, for example the generation of a sine wave with much lower THD.

I do NOT have the original PCB drawings nor the original schematics any longer - the only surviving drawings are on this page

Abstract

The project presented is a 2kVA power inverter designed for backup during utility failures. The inverter is affectionately called the 'I2K' for two reasons: It has a 2kVA output capacity and some have suggested it might have been useful when Y2K hits (remember the Y2K paranoia/fiasco :) Duh.

This inverter is of a traditional ferroresonant design using large step-up transformers however the pulse-width modulated signals which feed the transformers are generated in real-time by a PIC RISC microcontroller. The microcontroller senses AC output voltage via an 8-bit ADC then adjusts the pulse width accordingly using a closed control loop algorithm. The functions of waveform timing, pulse width adjustment, and error control (shutdown when overload occurs) are all done in software making upgrades easy to accomplish.

Many commercial inverters tested, including those used for PC UPS usages, will not provide the enormous start-up currents required by many motors -- especially single phase motors under load such as those found in the average natural gas furnace. This design provides those required currents and, during one test, was observed to convert over 2kVA of power (The input current exceeded 200A at 12VDC for several seconds until the motor started). Software control of the inverter allows the inverter to be overloaded for a period of time before shutdown occurs giving the motor time to start.

The inverter has been tested on 1/3hp single phase motor loads as well as with large resistive loads of over 1500W and has performed well. Power factor correction was found to be required with large motor loads and has been included in the design.

Efficiency of the inverter (Power Output / Power Input) driving a variety of both resistive and inductive loads ranges between 71% and 82%. With an improved sinewave algorithm it is felt that the inverter can approach the theoretical efficiency limit of 92%.

Disclaimer and Warning

 The design presented employs high currents on the primary side and high voltages on the secondary side. High currents of over 1000A were observed during circuit fault and despite the fact this design has safety systems included to deal with such problems, high currents can cause spectacular -- and dangerous -- effects such as component explosion. The output of the inverter is 120VAC and must be treated with the same respect as any other AC source. In addition, connection of the inverter to any device must be in accordance with local/national electrical rules (in Canada, the Canadian Electrical Code) which specifies grounding, neutral bonding, and other isuues. Do not attempt construction of a device such as this unless entirely familiar with high-power electronics as well as proper safety procedures. The author specifically disclaims any and all liabilities associated with the construction and use of such devices. The design is presented here in the interests of providing information on operational principles only.

DANGEROUS HIGH VOLTAGES AND HIGH CURRENTS ARE PRESENT IN A PROJECT OF THIS TYPE ... DO NOT ATTEMPT CONSTRUCTION UNLESS SKILLED IN SUCH AREAS

Power Drivers and Electronics

 300Ah Lead-Acid Cells The design is of a standard ferroresonant type using large step-up transformers to convert 12 V into 120 V. 12 VDC from large, high capacity batteries is chopped into alternating square waves which are fed to opposite windings of a transformer. The batteries can be seen in the photo to the left. I use a bank of twelve cells, each rated at 2V, 300Ah. The cells were surplus and were originally used to backup a large telephone PBX system. The cells are wired as two banks of six in series for a total capacity of 600Ah at 12VDC. For a 1500 VA output, current on the primary side of the transformers would be about 150 amps so the transformers must be physically large (i.e. A 120VAC to 12VAC transformer with an output of 150 A is required) and the driving electronics must be able to handle such large currents continually (most transistors cannot handle currents over 20A so a bank of driver transistors is required).

The most difficult elements to obtain were the transformers. Mine were scavenged from two old RCA power supplies used for TV broadcasting equipment. The original linear power supplies had a rated output of 70A each. Each transformer has a 75A, 18V C.T. primary and hence a power rating of about 1kVA. Two such transformers give the 2kVA power rating required. The transformer's primary (low voltage) windings wer found to have a DC resistance of about 0.008 ohms. This implies that a very low impedance drive circuit is required. In the inverter, the transformers are used as step-up transformers in push-pull configuration with the centre-tap connected to the +12V DC terminal and each of the outer terminals grounded alternately using large power MOSFETs.

The power drivers which switch battery current to the transformers consist of two banks of six IRL2203 high-current FETs, with each bank driving both transformers in parallel. These devices, manufactured by International Rectifier, have an on resistance of only milliohms each so that a bank of six in parallel has a low enough source impedance to drive the transformers without huge losses. It would be virtually impossible to use bipolar transistors for this application without huge losses in the drivers. Data sheets for the IRL2203 may be found at IR's web site. This device is clearly designed for such high-power uses. It features a low on resistance as well as a quick saturation point (5VDC on the gate will drive the device to saturation). The only problem with this device is the low voltage rating of 30V. To protect these devices, large zener diodes are required across the Drain-to-Source connections. 24V Zeners were chosen to give a good margin of safety in operation.


The above schematic shows one half of the main power drivers. From the far left, the waveform from the drivers (next schematic) is fed to the banks of FET drivers via series gate resistors which help share the load between the parallel FETs. The output of the power transformer, at right, is fed to the power-factor control and output sensing circuitry. The centre-tap of the transformer is connected to the battery +12V terminal via two large (70A) relays and a 150A series fuse for protection.

There are 24 zener diodes included on each driver bank as back-emf protection to prevent the FETS from ever having a reverse voltage of over 24V across them. This quantity of diodes was found necessary during prototyping to absorb the huge emf 'spikes' from the transformer when driving inductive loads such as motors. In the end though, with power factor correction applied (in the form of a capacitor across the AC line) this quantity of diodes was found to be quite excessive and their number could probably be cut in half or in third! (It was useful having this quantity of zeners available during prototyping though and they are kept in the system to protect the FETs during further development).

Note the 0.0033 ohm 75 Watt resistor used as a shunt to sense total current through the inverter. This resistor is actually composed of fifteen 0.05 ohm 5 Watt ceramic resistors in parallel. These resistors are forced-air fan cooled as they are operated at rated power when the inverter is at the 1500 VA level..


The drivers above take the TTL-level signals generated by the microcontroller and provide a low-impedance source to drive the FETs. The output swing is 0 to +12 volts which ensures complete saturation of the FETs (hence low power dissipation). Although the data sheet for the FETs states that +5V is enough to saturate the devices it was felt that transistor drivers would perform better since a loaded MCU output pin could output only 3 or 4 V.

Included on this driver board is an overcurrent limiter. When the current through the 0.0033 ohm shunt resistance reached some threshold value (usually about 200A), the voltage on pin 5 of the comparator will exceed the reference voltage set by the pot. When this occurs two TIP120 transistors will clamp the FET inputs to ground and will hence instantaneously halt current flow. This limiter was necessary during testing to avoid catastrophic failures when, say, a software bug led to both sets of FET drivers being switched-on simultaneously. During such a fault condition the system can easily consume over 1000A from the battery!! (_ enough to cause a fire _) as it did on one prototype inverter tested.

It was felt that a pure-hardware current limiter was the safest way to prevent huge overcurrents during testing. The limiter remains installed despite the fact that with debugged software it should never be required (and in fact has not been observed to be switched-into the system since prototyping). It must be realized that Fuses such as the 150A main fuse are not fast enough to protect the FET drivers from overcurrent. Besides, 150A fuses cost over $40.00 each!.


The above diagram shows the placement of hardware in the 19 inch rack-mount chassis. The large transformers weigh about 30 lbs. each and are mounted to the right side of the chassis. Behind the transformers are two 70A Potter-Brumfield relays. You will note that the high currents are carried by multiple pieces of 10 gauge wire.

The FET drivers along with associated zeners and the 15 shunt resistors are mounted on a PCB atop a heatsink in the center of the unit. A fan immediately behind the driver assembly cools the components on top as well as the heatsink which the FETs are mounted on top of. All control electronics such as the MCU board (containing the MCU, ADC, and 7805 regulator), the driver board (containing the TIP120 transistors and the LM339), and the output voltage sensing transformer are mounted on the left side of the chassis.

Control Circuitry

 AC Output voltage from the main inverter transformers is sensed using a small 9VAC transformer across the output. This attains isolation of the AC output side from the DC-side of the inverter. The 9VAC output is rectified by a 1N4001 diode then integrated by a 10uF capacitor. By sampling at the same point on the AC waveform each time, the microcontroller is able to read an approximation of the average AC output voltage.

Using average output voltage, rather than RMS, is a tradeoff which results in an output voltage which varies +/- 5 VAC from the RMS target of 120 VAC as read by a true-RMS meter when different types of loads are applied to the inverter. This is accurate enough for most applications however the performance may be improved by including an RMS-to-DC converter such as that made by Maxim between the AC output of the sensing transformer and the ADC input. Such a device would ensure the inverter always produces an accurate RMS output value (Again, this was not found to be crucial since an output variation of +/- 5 VAC was deemed to be quite tolerable).

The integrated output from the sense transformer (or the RMS-to-DC converter in an upgraded design) is then fed to an ADC0831 which allows the microcontroller to read the output voltage. The ADC0831 is an 8-bit converter and was chosen since it is a serial device (3 wires are required to connect it to the MCU) and hence consumes less I/O pins on the tiny (18 pin) microcontroller used.

The actual microcontroller is a PIC16C84 running at 3.6864MHz. The rather odd clock frequency is a multiple of 60Hz allowing accurate frequency synthesis. The PIC reads the ADC, adjusts the pulse-width for the next cycle, and synthesizes the PWM output on pins RB7 and RB6. Included are several status LEDs, a beeper, and an output to switch-in a power-factor correction capacitor of 22.5uF 200VAC for large inductive loads.

Schematics and PCB Drawings

I2K Files Complete Schematics (in DXF and DWG formats) and Code for the inverter.

Note that there is a bug in the schematic which shows incorrect wiring of the ADC to the PIC. The correct wiring is RA0 --> ADC Data, RA1 --> ADC Clock, and RA2 --> ADC Chip Select.

Please do not ask if I can convert these drawings into some other format. I use ACAD which is limited to it's own DWG format as well as a more universal DXF format. Many decent graphics packages (like PaintShop) will read DXF files.

Control Algorithms (Firmware)

 The algorithm to produce the necessary PWM waveform is simple in this initial test version: it is designed to produce a basic inverter allowing the hardware to be prototyped. The firmware functions as a pulse-width modulator (PWM) producing a variable pulse ranging in width from 4.1 mS to 8.2 mS. The width of the pulse depends on the output voltage required. A wider pulse produces a higher output voltage while a thinner pulse produces a lower voltage. By sensing the AC output voltage of the inverter the MCU can adjust the pulse width accordingly to produce a nominal output voltage of 115 VAC. The PWM is required in the first place to compensate for loading of the output. If the inverter's pulse width were originally set to give an RMS output of 115 VAC and a large load were applied, the output voltage would drop drastically essentially producing a brownout condition on the load. The PWM must compensate for this by increasing the pulse width as more load is applied. In essence the system operates as a voltage regulator keeping the output voltage constant as loads of varying degrees are applied.

The single variable width pulse is produced by producing two back-to-back pulses. The first pulse is fixed at 1/240th of a second (one-quarter of a complete cycle) width. During this relatively long time the ADC is read and the required pulse width is calculated. The pulse width is calculated by subtracting the actual voltage read from the target voltage (115VAC). An error figure results which is added-to or subtracted-from the current pulse width. This is a unity-gain closed-loop system.

Once the fixed time of the first pulse has elapsed (as set by TMR0 on the MCU) a pulse of variable width is generated. This variable pulse can be a minimum of 0 mS to a maximum of 4.0 mS. It is generated based on the pulse-width calculated previously. After the variable pulse is complete the FET bank is shut off and the program waits for the timer to expire (another 1/240th of a second). The total pulse width can now vary in width from 50% to 97% of one-half of the 60Hz waveform. This completes one phase of the waveform (time consumed for the first pulse is 1/120th of a second) and the process is now repeated with the opposite phase (other FET bank) with the exact same pulse width so as to produce a symmetric AC output with no DC offset.

Several features are included in the basic inverter algorithm such as indicators and overload detection. The later is most notable feature although it may not be clear why it is required. When a large motor starts it usually consumes up to 5 or 10 times it's rated current. You may have noticed lights on the same circuit as a motor will often blink when the motor starts. Imagine that a 1/3hp motor which normally consumes about 6A could consume over 30A when starting! Clearly this inverter will not produce that type of power. When a large motor attached to the inverter attempts to start the pulse width will rapidly increase to maximum and the output voltage will still drop below acceptable limits (as low as 95 volts have been measured). Normally we would consider this an overload condition and shut down the inverter however if we simply wait a few seconds the motor will start and will then consume its normal rated current of 6A. The algorithm therefore has an 'overload timer' which permits the inverter to operate at overload (maximum pulse width) for up to 10 seconds before being shut down. This has proven successful in allowing virtually any motor to start properly.

As well as allowing temporary overloads for motor starting the inverter also permits 'moderate' overloads. If the pulse width is at maximum, the MCU will monitor the AC output voltage and will not deem an overload to have occurred until the AC output voltage falls below 107 VAC. This extends the maximum power output of the inverter by allowing it to drive huge loads (albeit at slightly reduced output voltages).

Indicators are provided on the front panel which show when the pulse width is changing (the yellow LED) as well as when an overload is occurring and when the inverter has shut down due to an overload.

Preliminary software to run the inverter (in MPASM) format is available here. Bear in mind that this software is primarily designed to allow prototyping of the hardware of the inverter and will subsequently be improved.

Firmware

I2K Files Complete Schematics (in DXF and DWG formats) and Code for the inverter.

NOTE: This was written for an OLD version of MPASM and WILL require modifications in order to assemble on a newer version. I will NOT be updating this in the near future. Please do not ask me to update this code for a particular assembler.

Improvements and Modifications

 True Sine Wave Production

Improvements are to be made in the PWM algorithm now to effect better sine-wave approximation which will lower the harmonic content of the final output. In terms of the waveform produced, the biggest improvement would be to chop the large squarewave pulse into many smaller pulses varying in width from thin to fat then thin again. In this manner the output of the transformer will closely resemble a true sine wave rather than a square wave. There are a number of techniques to accomplish this. Originally, I planned to chop the 1/120th second period of one phase into ten smaller pulses and use a lookup table to determine the pulse width for each small pulse in the wave. To increase pulse width each small pulse would have to be scaled (multiplied). The problem here is that the waveshape will change somewhat as the overall pulse width changes and so harmonic content will vary depending on the load. To illustrate this, imagine a huge load where each smaller pulse is scaled so fat that it 'runs into' the next pulse. The output wave begins to resemble a square wave once again. The scaling technique is simple, but not optimal (still, it should produce an output considerably better than the simple squarewave currently used).

Added Note (2000/05): With new PIC processors getting faster and including more features (e.g. The 18Cxxx series which features 10 MIPS performance and includes an 8*8 multiply in hardware) it should be an easy matter to generate a true scaled sinewave with low harmonic distortion.

220V 50Hz Operation

I've had a number of E-Mails from people looking to use the inverter in Europe and other parts of the world. To change the output voltage to 220V the only real changes required should be to the output transformers: use units which are rated at 220-12 instead of the 120-12 ones employed in my design. In the case of my inverter, which has two separate transformers in parallel to deliver the currents required, these could simply be wired in series (observe the phase) to produce 220V at about 8A. Be sure the voltage-sensing transformer is still on only one transformer (120V) though or put it across the 220V output and change it to a 220V unit.

Changing from 60Hz to 50Hz operation can be done in two ways. The first requires changes to the microcontroller's firmware to increase the pulse-width of the base pulse to 5 mS from 4.16 mS as well as extra delays in the PWM code to scale the pulse accordingly. The second method would be to change the crystal from 3.6864MHz to 3.072MHz (also a common crystal frequency). This slows the processor by a factor or 5/6 so that it will now produce a 50Hz output wave.

Please do not ask me to modify the code or design for you. It is presented in the interest of experimentation - You're on your own! I am not in the business of designing custom inverters for specific applications nor do I have the time to give lessons on microcontroller programming. Unless you are already skilled at assembly-language programming, this project is not suitable for you.

Links

 International Rectifier manufacturers of the IRL2203 FET (datasheets available)

Microchip Inc. manufacturers of the PIC16C84 MCU (datasheets available)

Don Lancaster's Guru's Lair describing the concept of 'magic sinewaves'







You can E-Mail The Author from my main page but please read the comments in yellow above. I get a large volume of mail on this project and will not make modifications to the code nor provide lessons in microcontroller programming nor convert drawings from DWG or DXF format into anything else. I will NOT be implemeting a true sinewave algorithm so if that is your intent, you are on your own! This is a legacy project, archived here, which I will not be updating.