| 
           Most electronic devices involve fans, which are used to remove hot air from the system
          in order to keep temperature under control. Fan specifications are provided by the
          manufacturers with the most important characteristic for cooling being the fan (or
          performance) curve that specifies the pressure drop across the fan as a function of flow
          rate. A typical  curve, Fig. 1, shows this dependence between pressure drop and flow
          rate, where each flow rate has a unique value of  pressure drop, as one would expect.
           
          The one-to-one nature of pressure-flow rate dependence is very important for CFD codes
          modeling electronic devices with fans. This is because CFD calculates a solution by
          solving a quasi-time-dependent problem, usually starting from no-flow and gradually
          building up from specified flow and heat sources until steady state flow is reached. As
          the solution (and hence the flow) evolves, the flow rate through the fan changes, too.
          With a one-to-one pressure-flow curve, such as Fig. 1, most CFD codes will have no
          problem finding a solution because for each iteratively evolving flow rate there is only
          one pressure drop. 
           Another, still rather common, fan curve is shown in Fig. 2. The problem with such a
           curve is immediately obvious: there are sections of the curve where a single pressure
           drop corresponds to more than one flow rate. Such fan curves are discussed in the Fan
           Physics section below. If you are lucky and the operating point for your system is to
           the right of [3] on Fig. 2, you may get a solution. However, between 1-2-3 most CFD
           codes will either fail to converge or converge to a wrong solution, with the operating
           point often off the fan curve.   
           Some CFD codes recognize the problem up front and tell the user to modify the fan
           curve, so that the multi-value section 1-2-3 is replaced with a one-to-one curve
           connecting 1-3 (dashed line). This should help with convergence, but you might be
           converging to a solution of a different problem, as the modified fan curve is not the
           curve that was specified for your fan. Thus if your system is such that the fan actually
           does operate to the left of [3] on Fig. 2, your solution will be wrong. In systems with
           multiple fans, this will not even point out what is amiss, as you cannot know which
           particular fan is at fault as they all affect each other's performance and, hence, the
           operating point.  
            Daat engineers discovered that this apparently intractable problem has a unique
            solution, despite the multivalue-ness of the fan curve.  Their discovery not only
            permits the solution of problems with complex and multivalue fan curves, it also
            drastically speeds up the convergence rate. The advantage is especially noticeable for
            problems with multi-value fan curves, even in cases where standard algorithms manage to
            converge. An order of magnitude speed increases for such cases is not uncommon. Even
            with one-to-one curves, the new algorithm converges 30%-100% faster. 
           Fan Curve Physics
            Consider the case of a fan in a wind tunnel. The fan starts operating  with a fully
            unobstructed flow - point 1 on Fig. 3. As the flow resistance increases, the flow rate
            diminishes and the pressure drop rises along the curve 1-2-3. As the resistance further
            increases, the corresponding flow rate decreases beyond point [2], and the fan goes
            into an unstable regime.  There could be different reasons for this, such as flow
            separation and stall. At this point the flow rate drops sharply together with dP until
            it reaches point [5]. During the unstable regime the fan may possibly vibrate and
            generally  display signs of "distress". Once the system resistance increases further,
            the flow rates drops below point [5] and the fan regains the stable operation, so that
            dP rises along 5-6.  
            One can conduct a different experiment by starting with a fully obstructed wind
            tunnel to the left of point [6]. As the wind tunnel is gradually opened, flow
            resistance is decreased, and the flow rate is increased along 6-5-4. At [4], the fan
            becomes unstable until it regains stability at [2]. It then moves along 2-1 as the flow
            resistance is further reduced.  
            The fan operation inside the 2-3-5-4 gray box is unstable and generally unknown. Fan
            manufacturers show only a single line on their performance curve, rather than the
            uncertainty box and an explanation of its origin.  
     |