Sometimes an engineering problem can defy logic: My encounter with this phenomenon was a flow problem within the control valve of a medical respirator that showed characteristics similar to rare cases in nuclear reactor physics.
Fortunately our design team at Enfield Technologies (Trumbull, CT) had two things working in its favor: My nuclear reactor operator experience, and CFdesign, upfront computational fluid dynamics (CFD) software that gave us a good picture of flow characteristics within the control valve in the early design stages.
Enfield Technologies specializes in designing proportional pneumatic controls for automation, medical equipment, animatronics, and specialty instrumentation. In this particular case, the valve that defied assumptions was the primary control element for inhaling and exhaling within a lightweight and portable (about the size of a toaster) respirator.
Valve designs at Enfield Technologies usually deal with choked-flow pneumatic applications, where the upstream pressure is typically much higher than the outlet pressure. It’s easier to design for choked-flow applications, since the flow profiles are simplified and the equations governing flow are well behaved.
The respirator valve, however, needed to operate with a non-linear flow profile in relation to the poppet position. A very shallow and well-behaved flow control was needed for the first 50 percent of the valve stroke, followed by a rather steep flow profile for the remaining 50 percent.
Many of the details regarding the entire system and how it would interact with the control valve were obtained by working closely with the client’s engineers. These discussions helped us understand all of the different modes of operation of the end product, whether lung dynamics would come into play, whether upstream conditions varied, and other features that dramatically affect system dynamics.
Fast Design/Visualization Iterations
We created several physical prototypes to help the client specify desired performance, prove the valve concept, and hone in on the required flow profiles. Once the client’s needs were fully understood and the overall system and its details of operation were clarified, Autodesk Inventor was used to build solid models of the valve with complex poppet profiles. The Inventor models were then read directly into CFdesign. The tight integration between Inventor and CFdesign allowed us to quickly change the geometry and see the effect of those changes on streamlines, mass flows, and flow velocities.
We tested the valves at various positions and in different upstream and downstream conditions to develop preliminary flow profiles. With CFdesign, each design idea can be explored relatively quickly without the added expense of physical prototypes. Capabilities such as being able to cut a plane through visual representations of velocity and pressure to see flow characteristics at exact points, displaying bulk parameter mass flow, and viewing particle trajectory using cylinders and spheres were critical for better understanding the effects of design changes.
After several design iterations, we arrived at an acceptable flow profile that was worthy of a refined physical prototype. Follow-on testing was done on the physical prototype, sweeping the valve from minimum to maximum in 0.005" increments and measuring flow with a calibrated instrument while controlling inlet/outlet conditions.
I expected results from the physical prototype testing to be within 10 to 15 percent of CFD results, so I was amazed when the data agreed to within two to three percent at nearly every position.
The Nuclear Reactor Effect
Everything looked good until further testing and experimentation revealed a discontinuity in the flow data. Our valve was controlled by a microprocessor with an analog sensor to report valve position. Communications to the valve were digital, provided by a PC connected to a proprietary USB-to-SPI translator. The discontinuity could lie anywhere in the system; PC software, either Enfield Technologies’ or the client’s firmware, analog signals, interference, pressure anomalies, and mechanical physics were among the possibilities.
CFdesign was used extensively to study flow effects at the valve position of interest. The CFdesign simulations revealed a discrepancy: The convergence monitor window showed that the solver took longer than normal to converge and was less stable than expected around the suspect valve position.