In the last two Units, we looked at ways in which heat is managed by conduction, although inevitably we strayed into heat transfer by convection, because the air in the environment is where heat generally ends up, even where conduction has a key role in transferring heat from its source. In this Unit, we are concentrating on
With only a few exceptions at the extreme low-volume end of the spectrum, airflow is generated by fans and blowers. These range enormously in dimensions and performance, from the small fans typical of electronics applications to giants used in air conditioning systems and wind tunnels. But the same principles apply to fans of all sizes, and you may well come across useful material on web sites that appear to concentrate on air movement for HVAC applications.
But most electronics applications are serviced by relatively low-volume units, many of which can be described as “small axial fans”, a description that covers the typical range:
Such small fans are usually powered by DC brushless motors, electronically commutated, although larger units will use AC ‘shaded pole’ motors, both types of motor being mounted in the hub of the impellor. For more substantial items of equipment, the flow range is extended by using multiple units, or by higher-powered radial fans, usually referred to as ‘blowers’.
In this section we are reviewing the types of fan available before moving to the fan performance curves and “fan laws” that determine which fan to use for which purpose.
Then we will be looking at the ways in which fans are less than ideal, creating air flow that is not uni-directional in addition to noise and vibration, and having the potential for failure. Finally in this section is a reference on the specific issues of working at high altitudes. In our next section we will be looking at some practical choices concerning motors, bearings and mountings for the fan.
The simplest kind of fan is an open propeller, but a problem is that the pressure differential across the aerofoil produces vortices at the blade tips, and some air flow in unintended directions, leading to a waste of energy. This is the reason why the most common type of fan used in electronic cooling systems is what is technically called a “tube axial” fan, which has a venturi around the propeller. It is this style of fan that is normally implied by the contraction “axial” fan. [The “vane axial” fan takes the alternative strategy of having vanes that trail behind the propeller in the airflow in order to straighten the swirling flow created as the air is accelerated.]
Many of the fans that you will come across are the small, compact fans that are often referred to as “Papst fans”, after the originator, even if they are the product of other manufacturers. Their dimensions (square cross-section, varying from 25 to 135mm across) have become widely accepted as standard industry sizes. The first fans were a discreet black; although this is a generally-accepted colour, the plastic blades of small axial fans are occasionally treated as a “fashion accessory” – but do not mistake colour for performance!
The impellor on an axial fan resembles a propeller, with rotating blades that are usually curved for improved efficiency. The exhaust angle is generally greater than the intake angle, as shown in Figure 1.
based on a Papst original
Because the rotational speed increases with the distance from the centre line, the aerodynamic requirements change, so the blade is twisted. Using moulding techniques to make these impellors allows the blades to be contoured in the same way as an aeroplane aerofoil: the blades are rounded at the intake edge, are thickest in mid-section, and taper to a sharp rear edge. Because injection-moulding allows such optimisation at no additional cost, moulded impellors will generally give better performance than those made by welding metal blades to rotors.
When supplied for electronic cooling, axial fans are usually integrated into a square outer housing, with the electric motor built into the hub of the axial impellor. This outer housing is fitted with mounting holes, and may also contain fixed guide vanes in order to increase the available pressure.
Most fans found in electronic applications are ‘axial fans’ where the airflow is in the same direction as the axis of the motor. There is, however, a second style of ‘radial fan’, also referred to as a centrifugal fan or ‘blower’. Radial fans have a large diameter intake on one side and a smaller exhaust, which gives a high rate of flow over a small area. This makes blowers particularly suitable for applications where little space is available, for example in cooling high-end graphics cards or cooling CPUs in notebook PCs.
Radial fans have an axial air intake, but the air flows through the impellor radially, and is exhausted at right angles. There are many variants of blade design, the main difference being between “forward-curving” and “backward-curving” (Figure 2), depending on whether the exhaust angle of the blade is greater or less than 90°.
There are significant differences between different types of centrifugal fans depending on whether the blades are forward or backward curving (Table 1):
|The design requires a ‘scroll’ to funnel the airflow from the impellor||No housing is needed|
|Maximum efficiency occurs near the point of maximum static pressure||The fan will operate over a greater range without encountering unstable air|
|Power consumed rises rapidly with increase in delivery rate||Efficiency is often higher than with forward-curving blades|
|Motor overloading is possible if the losses have not been calculated carefully||Overloading is less likely|
|Minimum sound is generated at maximum pressures||Minimum sound occurs at the highest efficiencies|
|The fan is noisier than the forward-curving centrifugal fan|
|Performance is very sensitive to obstacles near the outlet|
The radial blower exists in a number of forms, including one that draws in air from both sides (“double inlet unit”), and this style of fan is often used for higher-power applications.
Whilst axial fans will usually provide the best combination of airflow and noise, radial fans are commonly used for case or power supply ventilation, or in conditions where space is restricted, as in rack-mounted configurations.
Where a higher pressure is required, but the format of an axial fan is more convenient, the “diagonal” or “mixed-flow” fan has been developed. This style of fan sucks in air axially, but discharges it outwards at a discharge angle that can vary between 0° and 90°, being determined by the shape of the external housing.
We have already met a number of different types of fan, for example, the double entry blower, in which the radial fan is supplied by air from both sides. Depending on the physical layout of the application, the room available, and the airflow/static pressure requirement, a number of alternative designs have been developed. For an example of this, look at a catalogue such as http://www.airflow.co.uk/pdfs/ind_cat.pdf for:
Review typical fans that fall into these three categories, and draw up tables of benefits and drawbacks of these approaches. Did you discover any other types of fan?
One alternative to a rotating motor is to use a piezoelectric fan. First developed as a novelty item during the 1970s, a piezoelectric fan uses small pieces of piezoelectric ceramic to which is applied an alternating current. This causes the ceramic to expand and contract, a movement magnified by attaching a thin mylar or metal blade. The performance of the fan depends on the overlap between ceramic and blade, and their relative thicknesses.
The benefit of a piezoelectric fan is that it uses very little power, produces no electromagnetic noise, and the very thin blades can be attached directly to computer chips, circulating air in areas conventional fans cannot reach. However, the low output means that the piezoelectric fan is at best a supplement for conventional cooling, rather than a main cooling agent.
The team at Purdue University that has been adapting piezoelectric fans for electronic cooling applications expect eventually to develop fans that are small enough to fit on to a computer chip, with blades only 100µm long. But there is a complication that each application will need a fan to meets its specific requirements, in terms of blade movement, the airflow produced, and the local circulation patterns – an improperly designed fan could make matters worse by re-circulating hot air back onto electronic components. Suresh Garimella observed (Knapp, A nano fan for nano gadgets) that very different results were obtained when the fan in a cell phone was moved into different positions, and that intuition was a poor guide as to which location would be best.
As you will know if you have watched a period film with a large fan being flapped manually, rotating fans are not the only solution. By bonding a blade as a cantilever to a piezoelectric crystal, the blade can be made to vibrate back and forth, generating airflow. See Piezo actuators for electronics cooling by Ioan Sauciuc.
Matthew Millar’s article Puff to be cool gives an insight into yet another approach to cooling, in which an electromagnetic or piezoelectric driver vibrates a diaphragm, forcing puffs of air through openings in the cavity. The pulsating jets direct air precisely where it is needed, and are very good at mixing the air, so that this synthetic jet ejector array (SynJet) is surprisingly effective. More information about this in Synthetic jets for forced air cooling of electronics by Mahalingam et al, Electronics Cooling, May 2007.
[ back to top ]
If you consult a fan catalogue, the first parameter you see will be a CFM figure or similar. This corresponds to the air volume that the fan will provide. However, this figure will be the airflow that is generated under “free flow” conditions, in which there is no pressure difference between the inlet and outlet sides of the fan. This corresponds to a situation where there is no resistance to air flow from the system.
Of course, real systems exhibit resistance to air flow. In order to create flow within a system, the fan needs to generate an over-pressure in the volume between fan and system. This is referred to as the ‘static pressure’. An alternative way of looking at this parameter is as the pressure increase produced by the fan.
So a fan will convert the energy provided by the fan motor into both air flow and static pressure, and there is a trade-off between these two, as shown in Figure 3. Here the air flow provided by the fan is a maximum at zero static pressure, and reduces smoothly to zero at the maximum static pressure available.
The characteristic curve of a fan showing static pressure as a function of flow rate normally goes from “free delivery” where both sides of the fan are at atmospheric pressure, to a shut-off condition at maximum static pressure. At this condition, there will be no flow through the fan, although the fan may still be rotating.
The fan characteristic is not always the shape of curve shown in Figure 3. One of the reasons is that the fan blade is an aerofoil, and subject to conditions such as “stall”, where the flow of air meeting the blade cannot follow the contour of the blade aerofoil and becomes separated from the surface. A typical curve showing the effect of the commencement of stall is shown in Figure 4. Note that, as stall is related to discontinuities in the airflow, its onset has some unpredictable elements, so the fan characteristic may also exhibit some variation with the external conditions.
Detailed consideration of this aspect is beyond the scope of this module, but further information can be found in the article by Philip Burgers, Stall of axial flow fans.
The shape of the characteristic curve varies considerably with the fan design. The smooth transition shown in Figure 3 is only one possibility – in other cases the curve may have several turning points, depending on the shape of impellor blades and housing. The range of flow rates and pressure increases also depends on the fan style, and this may be the basis on which fans are chosen for a particular application. Some general recommendations are:
[ back to top ]
The air flow through the system will be determined by the static pressure provided by the fan and by the resistance to airflow of the system, referred to as the “system impedance”. Figure 5 shows a typical curve. Note that this is not linear, but follows a square law, that is, the static pressure changes as the square of the change in flow.
Note that this figure is at odds with the normal mathematical convention in which the dependent variable is shown on the Y axis, so you may need to look twice at Figure 5 in order to see that the system impedance increases with flow rate, and that increases in pressure yield ever-decreasing results. However, this choice of axes is normal practice for fans, as this makes it possible to overlay the fan and system characteristics, so as to be able to determine graphically the operating point for a system-fan combination, as shown in Figure 6.
This operating point, being the intersection between the fan characteristic and the system characteristic, is unique to the combination of fan and system. As Comair Rotron put it, “The governing principle in fan selection is that any given fan can only deliver one flow at one pressure in a given system”.
This graphical method allows us to specify a fan, given that we have established the volume of air required, and have calculated (or measured) the resistance to flow of the system to be cooled, which allows the required static pressure to be calculated.
However, the required airflow and static pressure combination could be provided by several different fan types, and any particular fan can operate with a number of different system impedances, provided that the operating point is somewhere within the normal operating range of the fan. Figure 6 shows a typical graph of fan pressure against flow within what might be considered the normal operating range of the fan.
Figure 7 shows three typical fan characteristics:
Line A is for a 120CFM fan; line B is for a 100CFM fan; line C is for a 70CFM fan;
Line 1 represents the current system impedance, line 2 is the result of making some improvements, and line 3 is the result of a total rethink.
If 50CFM of air is needed, which fans should be chosen for which system impedances? And what is the impact on the total system performance of this choice?
Attempt this question before reading our answer.
[ back to top ]
A fan is characterised by four equations, referred to as the “fan laws”, which give relationships between the airflow (expressed in terms of volume or mass), the static pressure, and the power:
where K is a constant for similar operation, G the volumetric flow rate, N the fan speed in rpm, W the power output, D the fan diameter, ρ the air density andthe mass flow rate. It is also possible to derive a similar relationship for the noise produced by the fan.
From these basic laws, a range of useful interrelationships can be derived, of which Table 2 is a typical statement.
when speed changes
when density changes
|air flow||varies directly with speed ratio
||varies directly with density ratio
|pressure||varies as square of speed ratio
||varies directly with density ratio
|power||varies as cube of speed ratio
||varies directly with density ratio
Note that these relationships contain linear, square law and even cubic law elements. Particularly important is the dependence of power on speed, which indicates that:
Nevertheless, it is common practice to run a fan motor at full power, regardless of the cooling demand!
The fan efficiency is the ratio of the air power output to the input power provided to the fan impellor, and is expressed by the equation:
Note that this is a mechanical efficiency for the fan only, and does not include the energy losses within the motor driving the impellor.
The best fan designs will require less input energy for an equivalent air volume delivered at the same total pressure rise. Kramer quotes 2% as a typical efficiency, with small fans being perhaps only 0.5% efficient, while larger fans have maximum efficiencies of over 25%.
Note also that the maximum efficiency of a fan occurs near the centre of the fan curve, which is a good reason for choosing to operate in this area, rather than at the extremes of the fan characteristic.
Each fan has its specific curve, but it is possible to represent the characteristics in a dimensionless way, based on fan designs that are broadly similar. This approach is particularly suited for comparing different designs, dimensions and speeds, so that appropriate choices can be made for an application. To do this, pressure increase and flow rate are referenced to the rotation of velocity u at the perimeter of the fan, the outer diameter D of the impellor, and the density ρ of the fluid medium. The dimensionless parameters “pressure figure” ψ and “volume figure” φ are then given by the equations:
Figure 8 shows characteristic curves for three types of fan:
The ideal blade geometry will be determined for a specific flow rate, pressure increase and rotational speed. Computer designs can produce a blade for this specific application with the highest aerodynamic efficiency, which is closely matched to the point at which the noise generated is least.
There is a good discussion on these issues in Chapter 3 of the 2005 Papst Handbook (archived file) (PDF file, 197KB)
Cooling a system with a single fan may require the use of large, noisy fans in order to achieve the desired combination of airflow and static pressure. An alternative is to combine two or more fans in series or parallel, and this may also have some reliability advantages in the event of fan failure.
Parallel operation is defined as having two or more fans blowing together side by side. However, whilst having two fans might be perceived as doubling the flow rate, this will only happen in a condition of “free delivery”, that is where there is no impedance to the flow.
As Figure 9 shows, when a system curve is overlaid on the parallel performance curve, the higher the system resistance, the less is the increase in flow that will result from operating fans in parallel. So parallel operation should only be used when the fans can operate in a low-impedance environment.
Similarly, series operation can be achieved by using multiple fans in a push-pull arrangement. With two fans in series, the static pressure capability at a given airflow can be increased. However, as Figure 10 demonstrates, the available pressure is not doubled at every flow point. Unlike parallel operation, the best results in series operation are achieved in systems that have a high impedance.
In both series and parallel configurations, particularly when multiple fans are used, certain areas of the combined performance curve will be unstable, and should be avoided. This instability is unpredictable, and is a function of the fan and motor construction and operating points. If a multiple-fan installation is required, this should be fully tested in the laboratory before implementation.
[ back to top ]
The simplest simulation for a fan is one in which the airflow is perpendicular to the face of the fan. However, you will know from your experience with fans in a domestic environment that the airflow is not just perpendicular, but leaves the rotating aerofoils in a complex motion that contains a significant tangential component, referred to as “swirl” (Figure 11).
Simulation packages allow the user to define a fan with swirl, from which the user has to specify the amount of swirl and the direction in which the fan rotates. [Note that, as when placing components such as SOT23s, there is an opportunity to look at the problem from the wrong side, and use a mirror image of the correct footprint!]
In his paper, Fan swirl and planar resistances don’t mix, Tony Kordyban shows the extent to which swirl can affect the airflow pattern in a real application. Kordyban also shows the importance of swirl when a non-perpendicular airflow is applied to a linear resistance to airflow, such as that provided by perforated plate, card guide or air filter – whilst the air filter will straighten the flow, the pressure drop across it will be higher than expected.
Furlow and his partners (Fan swirl effects on cooling heat sinks and electronic packages) investigated the effect of swirl on the thermal resistance exhibited by representative heat sinks. Their experiments showed that the presence of swirl enhanced cooling, and this was most evident at low airflow rates, and with cross-cut heat sinks. Surprisingly, they observed swirl effects as far as 15 diameters downstream of the fan.
Noise from axial fans comes from a number of sources:
Guidelines for low noise suggested by Comair Rotron are:
The noise generated by a fan is normally treated in terms of the noise coming from the source, rather than its effect on the environment. Defined in a similar way to sound pressure, ‘sound power level’ is expressed on a logarithmic scale as:
where W is the acoustic power of the source, and Wref is an acoustic reference power.
Sound power level cannot be measured directly, and has to be calculated from measurements of the resulting sound pressure, but is useful as a means of comparing fans, or different operating points of the same fan, because it is not affected by the distance between fan and hearer.
As well as the total power, the frequency and frequency distribution of the noise are important. Fan noise is predominantly wide-band in nature, with the acoustic energy continuously distributed over the spectrum. However, some fan noise consists of ‘pure tones’, where the acoustic energy is concentrated in narrow bands.
In order to capture both sets of information, the audio frequency range is normally split into octave bands, and the average sound power level plotted for each. The bands are usually designated by the centre frequency, and have an upper frequency twice that of the lower frequency.
A number of rating methods are used for describing noise levels, and catalogue information may contain some or all of these:
Glossary of sound engineering terms at this link.
In his paper Fan selection and sizing to reduce inefficiency and low frequency noise generation Jack Evans illustrates how operating at other than the maximum efficiency point increases both noise level and energy consumption. Having oversized fans in large cooling installations may result in annoying low frequency rumble, whereas smaller fans will generate mid to high frequency noise, which is relatively easier to attenuate.
What is a fan failure? Think about this, and make your own list before reading our comments and proceeding further.
A typical life span for a small fan is 40,000 hours (4½ years), but this is highly dependent on the ambient temperature, the type of bearing, and the style of motor. Figure 12 illustrates how these parameters influence life expectancy. For an explanation, see Ian McLeod, Reliability and life issues in forced air cooling applications.
adapted from Ian McLeod, op.cit.
As with most reliability calculations, the characteristic time to failure is taken as the time at which a given percentage of the batch being tested has failed. In this case, by the end of the characteristic life, 10% of the fans will have failed. Note that quoting such a figure gives no indication of the distribution of individual fan lifetimes, although a Weibull distribution is likely. Also note that, as with many manufactured products, it might be advisable to ‘burn in’ a fan by running it for a short period, in order to remove from the distribution any defective units, such as those where lubrication has been omitted in error.
In their paper How to evaluate fan life, Kim and Claassen of IBM comment on the reliability of fans, and suggest a rule-of-thumb, that the temperature rise of the bearings should be limited to 10°C. Whilst the equation that they use for grease life may be criticised, it is consistent with the other recommendations made, that bearing loads can be reduced, and fan life extended, by installing the fan with the shaft mounted vertically, and using a larger bearing.
Many systems have more than a single fan, simply because an overall cooling system for the equipment has been supplemented by local cooling using active heat sinks. As these just generate local movement of air within the enclosure, adding to the turbulence, but providing a known environment for each heat sink, they are generally beneficial in reducing the overall thermal impedance from heat source to the external environment.
However, especially in larger systems, the designer may be faced by a choice of whether to use a single fan or, a number of fans, to provide overall cooling for the internals of the equipment. Using two or more fans has some advantages:
Note that this last statement is expressed in the conditional form; this is because a stopped fan may do much more harm than just reducing the volume of cooling air available. For example, unless steps are taken to close off the fan aperture in the event of failure, a stopped fan may provide a low impedance return path to the environment, allowing much of the cooling air from other fans to leave the enclosure without performing any useful cooling function.
Denmark (Cooling down with fan-speed control) also reported that systems using multiple fans may experience an additional source of acoustic noise because of beat frequencies between the fans. This is similar to the effect in multiple-engine aeroplanes, where a beat noise is produced as a frequency related to the difference in engine speed. This noise component is much more noticeable and irritating than the normal higher-pitch fan noise. There is also the possibility that vibration at the beat frequency may lead to resonance, with implications for both noise level and potential unreliability should the energy levels be sufficiently high. For this reason it is generally recommended that multiple fan unit drives are synchronised, for example by configuring the controllers to use the same oscillator, so that all the fans are running at exactly the same speed.
At sea level, the density of air is 1.19kg/m3, but as the height above sea level increases, the density of the air reduces, because there are fewer molecules above “pressing down”. At 5,000ft, the density is only marginally lower (1.056kg/m3), but at 25,000ft the density is greatly reduced (0.549kg/m3).
Unfortunately, a fan is a “constant volume machine”, moving the same volume of air irrespective of its density and ability to transfer heat. This means that, as the density changes, so will the mass flow, and the volume needed for equivalent cooling will be greater at altitude than at sea level (Figure 13).
So a system designed for operating at altitude has to be adequately sized, and a higher airflow is likely to be required for effective operation at high altitude. However, this could result in over-cooling at sea level unless steps are taken to reduce the fan speed.
Marthinuss and Hall, Cooling electronics at high altitudes made easy.
[ back to top ]
In much of the preceding discussion, we have concentrated on what is effectively a graphical way of fitting the fan to the application, based on the measured performance of the fan and an estimate of the system impedance. This impedance is something that we can potentially measure, and can estimate to some degree, based on Flow Network Modelling and similar procedures.
Such methods form the basis of much published material and many text books, but they suffer from the need to make simplifications, and many also require considerable expertise in mathematics. Also, whilst such procedures can help us develop the overall cooling requirement, they are insufficiently sensitive to model the temperature variations in detail. In consequence, we are forced to make judgements on the likely peak component temperatures, based on experience and good practice guidelines and on a calculated value for the overall temperature rise within the enclosure.
Fortunately, CFD modelling methods have developed to the point where they can produce models that are representative of the complexities of the real situation, but can be computed within a reasonable time, and generally converge to realistic results. The first stage is to input the mechanical structure of the equipment, and then model both the flow of heat through conduction, and the flow of air through the system. To do this we need to have the models of the heat-generating components that we discussed in Unit 12, together with models of the elements that influence airflow. These will include simple barriers, such as filters and grilles, and the much more complex models of the air-moving elements themselves. How this is done will depend on the simulation package being used, in our case the Flomerics software.
A fan is a complex structure, and one where simulation programmes typically provide simplified views that enable the designer to concentrate on the important issues. As an example, the Flotherm Fan Dialog has a range of settings for geometry, flow type and flow specification, making it straightforward to create your own SmartPart that is a good representation of the real fan or combination of fans. The mechanical settings are straightforward, and most care needs to be taken in modelling aspects of flow. Within the Flomerics software, for example, there are options on flow direction and swirl, and various ways of fitting the flow characteristic of the fan. The model can be built out of cuboids and prisms, according to a level of detail required, but the blades of the fan are not individually modelled – the flow passing through the region is given, together with any expected swirl. There is also an option for modelling the situation where the fan has failed and merely presents an impedance to the flow of air.
Obviously generating a fan description from scratch is not a trivial task. Fortunately, a number of fan manufacturers have made their fan definitions openly available. To see what a real fan specification looks like, browse at smartparts3d.com (where you should already have registered as a user), or download these PDML files, which are for fans by Delta and Papst. In order to examine the detail, these need to be opened within Flotherm: use the Import, Assembly, PDML instruction to add the part to the root assembly, and then examine the properties by right-clicking on the part, and selecting Construction from the resulting Fan Menu.
[ back to top ]
Having “sized” our fan, and decided on whether an axial or radial type is most appropriate, we still have a number of choices to make, based on how the fan is to be driven, the type of bearing to be fitted, and materials to be used. Many of these aspects relate to the expected reliability of the system, the application, and the challenges presented by the external environment.
The choices that we make will also effect the cost of the fan. An important parameter that we will only touch upon, but should not be ignored, is the commercial aspect. As with all purchases, there are trade-offs to be considered, especially if the aim is a low-cost assembly. This technical brief (Fan selection – quick techniques to compare various tube-axial fan designs) is a useful, though short, resource for identifying some of the issues that impact on cost, performance and reliability.
Detailed consideration of motors is outside the scope of this module, but we need to review a few basic features.
The brushless DC motor is used in many applications, and over a wide range of powers and speeds. It has a higher energy conversion performance than a small induction motor, avoids the mechanical wear and arc generation of the brush-type motor, and does not need the very small air gap between rotor and stator core demanded by the stepping motor. Referred to as “electronically-commutated”, this motor technology has developed to provide considerable power in a small volume.
The rotor is usually fitted with high-energy permanent magnets2, and the normal motor construction for cooling applications has an rotor inside the stator. [Less frequently, a bell-shaped rotor revolves around the stator] The field windings are distributed over the circumference of the stator, and are part of the housing, so can be energised without requiring a commutator and brush system. The necessary continuing alternation and reversal of fields is carried out electronically using transistor switches, so there is no mechanical contact, no potential for dust build up, and no possible sparking – useful in an explosive environment!
The position of the rotor is used to generate switching signals for the inverter drive circuit (Figure 14). The sensors are usually Hall-effect elements placed at 120° intervals, but the current or voltage waveform may also be used to determine the angular position of the rotor, and a third method for providing positional information is to use a digital encoder or resolver.
Brushless DC motors are found in both single-phase and three-phase types, the former being most widely used for small fans, and the latter for high-performance applications. Figure 15 shows the cross-section of the typical stator of a three-phase brushless CD motor in a four-pole, six-coil configuration.
The speed and airflow of a fan powered by a brushless DC motor is proportional to the voltage supplied, and most fans can be run over a range of supply voltages in order to give the required airflow. The voltage range over which a fan will operate reliably is determined by the design of fan, and may be as little as ±2V for 12V units, although a unit designed for a nominal 48V supply might operate over a much wider voltage range of 12–56V.
Because of its method of operation, the current drawn by a brushless DC motor is not constant, fluctuating both with frequency and during each cycle of rotation. As current excursions may vary over a 20:1 range, there will be a component of AC ripple as well as a DC running current.
There will also be a peak inrush current on start-up whose magnitude depends on the circuit and the power supply characteristics. Depending on the motor size and design, the ratio of peak starting current to running current can be as much as 5:1, although many motors incorporate some form of current limiting.
When specifying the power supply for brushless DC fans, the choice will be affected by the number of fans and their motor current characteristics. The power supply used must be able to drive the motor without itself shutting down because its over-current protection has been triggered.
Whilst DC motors are a common choice for driving fans for electronics cooling applications, they are not the only way of driving fans, especially if an AC supply is available.
Reluctance motors operate on the principle that magnetic forces will tend to minimise the volume of any air gap. The rotor consists of a laminated permeable magnetic material with teeth, and needs no coil winding or permanent magnets. The stator has a number of slots containing a series of coil windings that are energised in sequence in order to generate a moving field. As each stator coil is turned on, the magnetic flux path around coil and rotor acts to move the rotor in line with the energised coil, to minimise the flux path.
A switched-reluctance motor can be designed to operate at high torque and efficiency over a wide range of speeds. Although the technology can provide a low-cost motor, the reluctance principle relies on having relatively small air gaps, so shafts and bearings need to be of a higher quality than with other motor types. Another drawback is acoustic noise and vibration caused by the variations in magnetic flux, so that there is usually a compromise to be reached between noise reduction and performance.
For AC operation, induction motors are frequently used for fan applications. The stator consists of a series of coils wound on soft iron cores and connected to the power supply to produce a magnetic field whose polarity rotates at constant speed in one direction. The rotor is made of coils wound on the laminated iron armature, or formed of bars embedded in the armature surface. The rotating magnetic field of the stator induces eddy currents in the rotor coils, and these produce magnetic fields that interact with the stator field to exert a torque on the rotor. As with the reluctance motor, there is no need to make an electrical connection to the rotating parts.
In a three-phase induction motor, consecutive coils are connected in opposing pairs to the three phases of the supply; for single phase induction motors, there are two ways to simulate this effect by creating a rotating flux:
A capacitor-run motor uses a capacitor connected between main and auxiliary windings to provide phase shift between these windings.
A shaded pole motor has a split stator with copper shorting rings placed on it so as to ‘shade’ a portion of the stator’s magnetic field, giving a phase shift between stator and shaded pole that is enough to provide starting torque.
The main practical difference between these types is that the induction motor is essentially a fixed-speed type, rotating at the frequency of the AC supply, and only slipping slightly behind the field as the load is applied. As with the brushless DC motor, the switched reluctance motor can be controlled in speed, and is generally more flexible. However, this flexibility comes at the expense of drive controllers.
For a summary of motor types, see The electric motor: Here and now.
If you can cope with a 10.4MB download, a good resource on motor technology is the Papst 2004 catalogue Drive-Technology for AC and DC (archived file).
For AC motors, see the Motorola tutorial AC induction motors
[ back to top ]
Have you noticed how irritating the noise of your computer fans can be, even though the airflow they generate is relatively low? Most of this noise comes from the bearing system, but it doesn’t necessarily mean you should use a more expensive ball bearing, as the oil-impregnated sleeve bearing is actually less noisy and obtrusive – the two bearing systems generate noise that is very different in both frequency content and amplitude.
Figure 16 shows a typical sleeve bearing assembly for a small cooling fan, where the shaft rotates, and the bearing is stationary (although the reverse is also possible). The porous bronze in the bearing is impregnated with lubricating oil, which is fed to the shaft through the pores. As the shaft rotates, a film of oil is built up on which the shaft rides. In a perfect world, the oil would prevent metal-to-metal contact and eliminate noise, but the surfaces are rough on a microscopic level, so the bearing may generate a grinding sound. Forces on the shaft from fan imbalance and from the drive motor can cause the shaft to rock in the bearings, and make contact at the bearing ends, causing a rattling sound. The thrust washers are another source of noise, as these slide relative to each other, creating a rubbing sound. Noise from a sleeve bearing is usually broad-band and intermittent, but most good designs will stay quiet until they begin to run dry.
Sleeve bearings may be found in a range of different qualities, but all will eventually fail due to lubricant exhaustion3. More expensive alternatives use Teflon or ceramic materials for the bearing surfaces, which give longer life than impregnated metal bearings, whilst retaining the lower noise level associated with a sleeve bearing as against a ball-bearing type.
Figure 17 shows a typical ball-bearing system, with an inner and outer races, a set of balls and a cage to support them. Because of its many components, a ball-bearing system is initially relatively noisy compared to a sleeve bearing, and tends to get noisier over time. The noise level will depend on the surface finish of the component in contact, the sphericity of the balls, the alignment of the bearing, and of course the lubrication. Another problem is that ball bearings can easily become damaged, particularly when the bearing is exposed to a shock load, which can lead to the ball race becoming dented. This denting, referred to as ‘brinnelling’ has little effect on life at light loads, but greatly increases the noise. By comparison, a sleeve bearing fan can sustain multiple shocks, with no effect on noise output. Ball-bearing system noise may contain both broad-band and single-frequency elements. The additional 1–3dBA difference in total airborne noise, combined with the pure tone and higher frequency, make the ball-bearing fan run at a higher “annoyance level” to the human ear.
So why bother with a ball-bearing fan? The answer is that, whilst both bearing types have similar life expectancy at low operating temperatures, sleeve bearings have a considerably reduced life expectancy at higher operating temperatures, so ball-bearing fans may be required.
Except for very small fans, one bearing is insufficient to provide a stable rotating axle. Most fans have two bearings, but be aware that many so-called “ball-bearing” fans actually have only one ball bearing, taking the majority of the stresses, supported by a sleeve bearing. Larger ball-bearing fans should have two ball-bearings!
In the same way that car tyres need balancing to compensate for heavy spots on tyre and rim, to avoid imbalance, and to reduce stress on bearings and suspension system, the same should be done to fans. Tolerances in the production of motor and fan mean that there may be an imbalance, which can be allowed for by adding weights to the fan. The downside is that the individual balancing required adds to the cost of the fan unit, so balanced fans are found only at the low-noise, high-reliability end of the market. Of course, some manufacturers would claim that their manufacturing tolerances are tight enough to make this balancing requirement unnecessary . . .
[ back to top ]
Depending on the size of fan, a wide range of plastics, metals and composites are used, ranging from small fans, where all the parts are injection-moulded, to the largest fans, which are usually complex metal fabrications. The choice of materials is intimately related to the choice of available bonding methods, and also affects the way in which the fan is assembled.
Particularly for low-cost fans, attention has to be paid to ways in which this process can be automated. Techniques, such as injection moulding, that enable complex parts to be made in one operation, avoiding the need for subsequent assembly, are particularly appropriate. The moulding process may also allow prefabricated metal parts to be included, for example building metal bearing surfaces into an otherwise plastic enclosure. Nevertheless, for small fans, metal blades still compete with injection-moulded parts.
For larger fans, fibre-reinforced plastics (FRP) are effective substitutes for metal. FRP fans are lighter, yet the materials can have enhanced mechanical properties, and the design flexibility offered allows the impellors to offer higher efficiency, lower noise levels, and lower power consumption. (More information in Srikant, Mattal and Biswas, Energy-efficient axial flow FRP fans.)
[ back to top ]
Many fans for electronics cooling applications live in an office environment, where the only potential reliability problem comes from dust pick-up, as this may clog orifices and create a non-lubricating “grinding paste” with any grease. However, it is not uncommon for fans in other applications to be exposed to more severe conditions. Adding to the effect of any moisture may be salt (in coastal areas or from road salt) and sulphur dioxide produced from the burning of oil or coal. Under such conditions, corrosion will be inevitable unless the fan is given suitable protection.
Because a fan is both a mechanical component and an electrical motor, providing protection demands a range of approaches in order to create a barrier between all the fan components and the environment. Silicones are frequently used, in varying thickness according to the degree of environmental protection needed.
Protective barriers against corrosion are not the only form of environmental protection that may be needed for fans that are to be fitted close to the external surface of an equipment. Depending on the environment, the equipment (and thus the fan) may be exposed to liquid water, ranging from splashes to driving rain, or even immersion. Most of the protection here will be provided by the equipment housing, but splash-proofing and similar requirements may affect the positioning of the fan within the enclosure. More about both these topics in the next Unit.
[ back to top ]
In this final section we are first thinking about where and how to fit the fan, before moving first to the ancillary items associated with fans, such as grilles and filters, and then to issues of fan control.
Forced convection can be carried out in two distinct ways, by pressurising the cabinet with a fan, or by placing the fan on the exhaust side so as to evacuate the cabinet. In terms of defining the air distribution within the cabinet, having a fan on the exhaust side is more flexible, because cooling ports can be placed to give direct cooling at the required locations. Also, heat from the fan itself is not dissipated into the cabinet. The downside is that filtering the fan becomes extremely difficult. This is the reason why most designers prefer to pressurise the cabinet, which can easily be achieved with filtered air. And having the cabinet under pressure helps reduce dust seepage through any cracks in the enclosure.
Another advantage of “push” as against “pull”, is that fan life and reliability are increased, because the fan ambient temperature is lower. Also, because the fan is handling cooler air, it will have a slightly higher pressure capability. There is, however, a downside, at least for those fans where the fan motor lies within the air stream. In this case, the power dissipated by the fan motor is transferred into the cabinet, inevitably leading to an increase in local temperature.
Mike Turner (All you need to know about fans) makes some suggestions on the location of fans, as shown in Figure 18.
Combining Turner’s ideas with those by Walter Angelis in General aspects on fan selection and layout, some suggestions for good practice for a pressurised enclosure are:
Of these criteria, the last is the most complex. Typical suggestions are:
Small axial fans are typically specified for “free air” delivery, with no channel connection at the outlet. Note from Figure 19 that just adding a short pipe of similar dimension to the fan can reduce airflow by 10–30% and the available pressure by 10–40% (Figure 19).
The airflow from a low pressure fan will also be impacted significantly by the presence of features such as filters and finger guards, and even the modest increase in back pressure produced by mounting a deflector plate can significantly impact on the flow rate.
[ back to top ]
Although some types are available for direct board mounting, most fans are designed to be mounted on the enclosure, or in combination with a heat sink. The mounting method will depend on the size of the fan and the type of enclosure, but should always take into account the likelihood that at some point the fan will need to be removed for cleaning, if not for replacement.
Screw fixings are most usual, but a compliant gasket is often placed between fan surround and enclosure. This helps make the joint air-tight and may also reduce noise transmission from the fan. However, where fan vibration is an issue, a proper anti-vibration mount may be needed, and these are commonly used for the larger fans employed by HVAC systems.
Again in larger systems, it is not unusual for motor and fan to be separate, with a belt drive between. In that case, attention has to be given to the individual mounting of fan and motor, and minimising any noise and vibration resulting from the coupling itself, for example when the drive belts become slack.
At the bottom end of the fan size range, where a fan is fitted direct to a heat sink, care has to be taken to ensure that screwing the fan into position does not distort either the heat sink or the device to which it is attached.
Most fans will be supplied with power through flexible leads. As with any rotating part, live leads need to be kept well clear of the rotor by appropriate ties or lacing, leaving sufficient free play to allow the fan to be removed, or at least for any in-line plug and socket to be disconnected.
For any size of fan, in any location, the usual Design for Manufacture aspects apply:
Unfortunately, as you may know from having taken a personal computer to pieces, not every manufacturer complies with ideal practice!
[ back to top ]
Many cabinets will have obstructions in the flow path created by features such as dust filters, finger guards and EMC shields.
The dimensions and construction of these screens/grilles vary very considerably, according to the use to which the screen/grille has been put. As a finger guard, spacing can be relatively coarse, but the individual wires need to be robust; for ignition protection, a fine mesh is more appropriate; as an EMC shield, the dimensions of the structure are important. Figure 20 shows some common structures: (a) is a woven screen, (b) a rod screen, (c) a perforated plate, and (d) a square-edged grille.
A screen/grille can be modelled as a “porous structure” whose porosity is defined as the ratio of the open area of the screen to the total area. When air flows through the screen/grille, there is a drop in static pressure. The losses for the structures can be calculated, as can the effect of setting the screen at an angle to the flow. Note that the transmission of a screen drops off markedly as the screen is inclined from the normal, a 25% reduction being experienced at 30° for a round wire screen.
The loss across the screen/grille will change with time, particularly if the mesh size is small, as the build-up of dust will reduce porosity. As with filters, many kinds of screen/grille will need to be removed and cleaned as part of the maintenance process.
The materials used, and the manufacturing method chosen, will depend on the application and manufacturing volume, and will be influenced by whether or not the screen/grille is cosmetically important. Particularly at high airflow rates, it is important that the grille should have good aerodynamics, so as to obstruct flow as little as possible, and minimise the noise generated.
As well as noise from the fan, the airflow itself creates noise, particularly with turbulent flow. The design of the fan grille is important here, as many grilles have poor aerodynamics. Grilles may also generate their own noise, and not just from airflow – an inappropriate design or insecure fixing may lead to a grille “flapping in the wind”. As with fitting fans, there is a compromise to be reached between security of fitting and ease of removal for maintenance.
In his paper Specifying filters for forced convection cooling, Alan Woolfolk comments that air filters are often treated as an afterthought in the design cycle, whereas the wrong filter can compromise the electrical and thermal performance of the system. He suggests three main reasons for using filters in forced convection cooling systems:
The build-up of dust and dirt can affect electrical performance, particularly in high impedance circuitry, and a gross build-up of particles can significantly reduce airflow.
The particles in air range from relatively large (50µm) particles of dust and bacteria, down to the small particles (0.5µm) more typical of smoke. The easiest particles to remove are the larger ones – if small particles need to be kept away from the circuit, a different type of filter may be needed. For applications such as clean rooms, it is not unusual to have a two-part filter, combining a coarse filter and a HEPA (High Efficiency Particulate Air) filter for smaller particles.
Filters are not 100% effective in removing particulates, and are specified in terms of the percentage of dust that they collect; typically the higher the efficiency, the higher will be the pressure drop across the filter for a given air velocity.
The aperture needed for air intake and outflow presents a discontinuity in the EMI protective shielding around the circuit. For many applications therefore a filter will be combined with a metallic barrier, and fitted to give EMI protection at the fan aperture.
Metallic barriers can take the form of a perforated sheet, a mesh of round fibres, or a woven mesh, but the most common choice is the honeycomb mesh. This has a low pressure drop at high air velocities, and acts to guide frequencies higher than a ‘cut-off’ frequency to the enclosure around the filter assembly, and thence to ground, whilst frequencies below the cut-off frequency are attenuated. The cut-off frequency and the attenuation values are determined by the dimensions of the honeycomb cells.
A coarse filter may sometimes be cleaned by removing it from the equipment and blowing air at relatively high pressure in the opposite direction, taking appropriate health and safety precautions. Other styles of filter may be washed to remove what cleaning specialists refer to as “soil”. When neither process is suitable, and this is especially true for small-particle filters, the active section of the filter will require replacement.
The effectiveness of cleaning, whether by air blast or washing, will be improved by choosing an appropriate coating. For grilles associated with filters, having a polished coating helps reduce the pick-up of dust as well as enhancing the cosmetic appearance.
[ back to top ]
We have already seen that many motors will be associated with drive circuitry. However, despite the fact that some types of motor are capable of running at variable speed, the majority of fans are just switched on, and left to run for the whole time that the equipment is operational. This is not necessarily the way to get the best performance or life from your fan!
Bruce Denmark (op.cit) points out the need to select a fan than can move enough air to keep the system cool even with worst case ambient temperature, power dissipation, fan production tolerances and fan ageing. In reality, the system will face these conditions only infrequently, so that in most situations the fan speed can be reduced without adversely affecting a system performance.
A noticeable advantage of controlling the fan speed is the reduction of noise, a significant source of annoyance in a quiet office environment. But other advantages include reduced power consumption, reduced fan wear and longer life, and reduced dust collection.
Whilst integrating temperature control in a fan is cost-effective, there are two practical issues to consider:
The most obvious way of controlling a fan is to base its switch-on and speed on the temperature of the part being cooled. However, whilst an effective way of stabilising a system at an elevated temperature, this is less effective when the aim is to cool it as much as possible.
In his article, Keep cool by sensing current to control fans, Jerry Steele suggests that fan speed should be based on the thermal load of the system, for which a readily measured substitute is the power consumption. The advantage is that this gives immediate information on the thermal load, without the lag that is experienced by temperature sensors, even if these are on-board diodes.
Figure 21 compares CPU temperature behaviour with a temperature-controlled fan and one that is controlled by supply current. Because the current-controlled fan reacts immediately to thermal load, and is run at full speed, rather than at a speed proportional to the over-temperature, the temperature of the CPU is kept as low as possible.
There are of course some problems associated with this type of control, one of which is the use of a fan at full speed, as this reduces life. A more serious difficulty is that the changes in power supply current can be both large and swift, whereas delays in temperature rise naturally integrate the power over a period. In order to get best results, practical current-controlled fan drivers need to incorporate both integration and delays, or their digital analogues.
In any forced convection situation, failure of the AHS will result at least in a change in temperature and may cause catastrophic failure if undetected. There are three approaches for monitoring the system so as to be able to take timely action:
In a typical system, a designer will usually include some over-temperature protection, to allow for a condition where the fan may be turning, but a clogged input filter or output grille is preventing free flow of air. However, more immediate feedback of the problem can be achieved by monitoring rotation of the fan or motor4.
Motor rotation is easy to monitor, using a Hall cell to sense the rotating magnetic fields generated by the rotor. The cell will generate a square wave pulse train, at perhaps two pulses per revolution. Depending on the style of fan, this tachometer signal may be supplied for processing by the associated circuitry, or else be processed by circuitry internal to the fan, in order to provide an alarm signal in the event of a blade becoming locked.
For AC fans, an isolated pickup coil can generate a signal whose amplitude is proportional to the speed of the fan. This signal needs conditioning and, on the larger fans for which AC motors are normally fitted, the signal processing will be carried out within the fan.
In many ways, the ideal combination of sensing and protection is to embed a PTC thermistor in the motor, to provide over-temperature protection for the fan assembly, and combine this with direct monitoring of the temperature of sensitive components within the assembly being cooled, giving feedback to the fan drive, so as to keep the amount of cooling to a minimum.
An alternative control method that has the merit of simplicity is to switch the fan on and off on the basis of information from a single temperature sensor, such a PTC thermistor. In that case, depending on the thermal time constant of the system, there will be some variation of temperature about a mean value. Whilst this form of control needs few components, thought must be given to the hysteresis in the system, so that the on-off cycles are not too short. Also, as the fans will be turned on and off on a regular basis, there may be reliability concerns and issues about surge current, particularly if a number of fans are switched in parallel.
Fan failure can have such a disastrous effect on big systems that it is normal for systems to be somewhat over-specified. For example, a fan tray where four fans would provide adequate airflow might be fitted with two extra fans to provide some “redundancy” for critical applications. Unfortunately, having extra fans uses extra current, and produces extra noise, as well as costing more. Also, under normal operation, it will provide too much cooling!
An obvious way round this is to fit the extra fans, but run all the fans at a reduced speed until such time as a failure is detected, when the speed of the remaining fans can be increased to compensate. Overall, this leads to quiet and reliable operation of the system, and higher reliability for each fan, because bearings will run cooler and there will be less wear.
[ back to top ]
The need for new cooling techniques, and the increasing challenge is summarised well in the paper by Lasance and Simons Advances in high-performance cooling for electronics. The paper has an excellent summary of the heat transfer coefficient attainable by different technologies, as well as some discussion on most of the technologies that have been described in this Unit and the last. And their paper includes at least one technique that we have not described. Definitely recommended reading.
[ back to top ]
Each of these lists is in the order in which the material is referenced in the Unit text. However, note that links to SAQ answers are not included!
[ back to top ]