CASE STUDY: Considerations in the Thermal Design of an Electronic Load
For a Power Supply Designer, deliberately maximising the power dissipation in a switching transistor is anathema.
However – Power supplies, supply power and they need to be tested before deployment so some form of load is necessary and for most modern PSUs, that requires rapidly changing the load during a test process.
Here we look at some of the considerations during the definition and thermal design of an Electronic Load with pulse-load capability. We take a top-down approach, starting with the factory-level decisions.
The Environment / Energy Costs
As responsible manufacturers, we need to consider our impact on the environment – we strive for ever-better efficiencies within our products and see reductions in load requirements and source materials as equipment gets smaller.
A test load which will be running for extended periods (“burn-in” for example) will naturally be wasteful – in the example used here, we have a 2kW power supply which supplies 2kWrms with either DC load or a pulsed load at 10% duty cycle so a 12-hour burn-in at 2kW will dump 24kWh. If we are producing 10 units per day 5 days per week; that’s 15,600kWh per quarter which will cost around £1,000!
For that situation, we could look at energy recovery systems where the PSU output is converted back into AC and re-used. With “Titanium” efficiencies and a good inverter, we might only lose 150W per PSU – now we’ll use a bit over 1,000kWh per quarter, saving over £900 per quarter.
Of course, there is an increased cost in setting up such a system, but the return-on-investment can be readily seen.
The example used for this discussion is based on a Production Test requirement, however – In this case, the PSU will only be loaded for a matter of seconds during an automated test so the increased cost of an energy recovery system is unlikely to be justifiable.
So our requirement is for a dissipative load box, capable of 2kW steady-state and 20kW, 10% pulses at 1kHz.
Getting the energy out of the box
We will end up with an enclosure of some form, with the dissipative bits safely mounted inside; both electrically and thermally insulated from the walls of the enclosure.
At this level, we basically just have to get 2kW out of the box – simple!
Let’s assume that we want the equipment to be fan-cooled and to operate in a typical test environment of 30°C maximum. We will also assume that we don’t want either the enclosure walls or the exhaust air to exceed 40°C (metal at 50°C is painful to touch, so 40° is a good limit).
So we need to shift enough air through the enclosure to remove 2kW of waste heat with a 10° delta between inlet and outlet.
The Specific Heat Capacity of air (typical room conditions) is 0.00121 J cm3 / K.
So, for 10° rise: 0.0121 J cm3 or 12.1 J/litre.
That is; 12.1W will heat 1litre of air by 10°C every second.
∴ 2000 / 12.1 ) * 60 l/min required :- Approximately 10,000 l/min (350cf/min).
A couple of decent 120mm axial fans will do this. You might want to consider either a bi-metallic switch or an electronic switch on the fans so that they only run when required. This will reduce noise and massively increase the fan’s life.
Handling the energy inside the box
We are going to need some form of heatsink inside our enclosure to collect the energy from the load elements and transfer it to the airflow.
It is not easy to model the exact transfer of heat from a finned surface into an airflow – turbulence and impingement have significant effects. So we will need to make a reasonable estimate and then test it once we have our final assembly. Being conservative with all approximations will generally give a good result though…
To start with, and based on experience, we imagine a 4U 19” case and we’ll pick a readily available extruded aluminium heatsink (in this case x 4 of ABL’s 177AB at 400mm length).
From the manufacturer’s datasheet this gives a thermal resistance of 0.06°C/W for each 400mm length with 2m/s airflow.
Assuming that we baffle the airflow from our two fans into the space occupied by four of these heatsinks, then we have an approximate area – looking at the end of the heatsink assembly – of 600cm2. This is allowing ~40% fill with a gap between the two heatsinks. Thus, the heatsink assembly will contain 24 litres of air and at 10,000 litres/minute (167litres/second) the air will be changed 7 times per second which implies an airspeed of over 2.8m/s.
Even allowing for reduced flow due to backpressure, this will give plenty of air to meet the 2m/s used in the datasheet; so we can use the 0.06°C/W with confidence.
Time for a bit of an assumption… We will mount the fans on the inlet side to keep them as cool as possible, and we will allow that the metal of the heatsink will be 30°C above the local air so at 30°C maximum inlet ambient the leading edge of the heatsinks will be 60°C and the exhaust end 70°C. Based on this delta of 30°C, we can dissipate 500W per heatsink (30/0.06) which suits our requirements perfectly.
Dissipating the energy in the first place
Now we have our enclosure, fans and heatsinks defined. There are 4,800cm2 to mount the dissipating elements on and we will work on the basis of the heatsinks not exceeding 70°C.
What are we going to use as the actual dissipaters? Resistors might be an obvious choice, but they are large and will need to be rapidly switched. Relays are too slow to give a good pulse edge so transistors will need to be used.
As switch-mode PSU designers, arranging for a FET to be switched on and off is rather our stock-in-trade. As is getting rid of the (minimum) heat it produces. But how about using the FET itself as the prime means of dissipating heat?
Rather than fully-enhancing the channel, we can use a source-follower circuit to control the drain-source current by the insertion of a small resistance into the source circuit.
The TO-247 package is a good choice for mounting to a large metal heatsink so we’ll start by looking at FETs in that package. Our Power Supply generates 56V at 36A so a 75VDS rating would be okay based on a 75% derating policy. We can look at a device such as the IRFP4368 but recall that we are not fully-enhancing the part so RDS(ON) is not relevant – IRFP4468 will do the same job with more margin on voltage.
With multiple source-followers in parallel – all the drain connections are common, so we can bolt the devices straight down onto the heatsinks without insulators. This means that the heatsinks will be at the +ve supply potential (56V) but are enclosed and isolated so that’s fine. This saves a significant thermal impedance which would need to be included if using insulating washers.
From the datasheet of either device, we see that the RθJC is 0.29°C/W and the RθCS is 0.24°C/W (with heatsink compound). TJ(MAX) is 175°C and we would derate the junction operating temperature by at least 20°. Therefore we allow 150°C junction temperature on a 70°C maximum heatsink temperature meaning that we can dissipate. (150 – 70) / (0.29 + 0.24) = 150W in the device! That’s less than 3A per device.
Before we jump to the conclusion that we only need four FETs per heatsink, however – let’s consider the pulsed load requirement: To maintain the 2kW rms power we need a pulse current of 3.33 * IDC at 10% duty i.e. 120A.
Looking at the Effective Transient Thermal Impedance curves in the device datasheet we see that at a 1ms 10% pulse, the effective thermal impedance RθJC is 0.09°C/W. Now we can calculate (150 – 70) / (0.09 + 0.24) = 240W maximum dissipation during the pulse.
Based on driving the gate of each source-follower with ~10V maximum and therefore having a source voltage of ~8V we see that each FET will have (56 – 8) V across it so can only pass 5A to dissipate 240W. Therefore we need 24 circuits in parallel to give the 120A pulse desired. A resistance of 1.6Ω rated 5W in each source circuit will give the necessary performance.
This discussion has not covered the control, nor the monitoring functions of a practical load unit but has described the process of designing a high power load assembly with fast control possibilities.
The actual implementation of this has been working on one of our products for the last decade with nine such units controlled in parallel to give fast pulse control up to 1100A.
- CASE STUDY: DIGITAL INFLUENCES ON POWER SUPPLY DESIGN
- CASE STUDY: VIBRATION PROOFING RUGGED POWER SUPPLIES
- CASE STUDY: MINIMISING THE SIZE & WEIGHT OF A POWER SUPPLY
- CASE STUDY: HOW MILITARY POWER SUPPLY CONTRACTS HAVE CHANGED IN THE 21ST CENTURY
- COMMISSIONING A CUSTOM POWER SUPPLY