This is a constant temperature thermal anemometer built to use inexpensive ($1) F2020-100-B-100 platinum RTDs from Omega as the sensor elements. The constant temperature feedback circuit modulates the power applied to the sensors to keep them at a constant resistance (and constant temperature, since they're temperature-dependent resistors). Faster fluid flow past the sensors strips more heat away, and so the voltage that the circuit applies to keep the sensors' temperature constant is a measurement of the flow speed.
Outside of instrument.
I built a four-sensor flush-mount array (pictured below) for the sake of allowing cross-correlations between sensors.
Four sensors in diamond array.
For this kind of one-off instrument building, I usually wire everything point-to-point on general-purpose prototype board, but the fact that there are more choices for surface-mount parts these days and the fact that I had to build four identical channels for the array above made printed circuit boards more attractive. The circuit was simple enough that I did a layout (PDF) by hand. I used a laser printer toner transfer as a resist mask to etch the boards shown edge-on in the picture below.
The circuit required for each channel is shown below. A rail-to-rail dual op amp is used. U1 adjusts the current through Q1 to maintain the balance of the Wheatstone bridge consisting of R1, R2, Roh and the sensor. When the bridge is balanced, the resistance of the RTD sensor is a known fraction of the overheat resistance Roh. R1 and the sensor have a much lower resistance than R2 and Roh, so most of the current flows through the leg of the bridge containing Roh and the sensor. R1 is made from three parallel resistors so that it doesn't heat up appreciably. The overheat resistor can be adjusted to change the operating temperature of the sensor. Higher temperatures mean a more responsive measurement; lower temperatures allow measurement of higher speed flows.
Schematic: Click to enlarge or get a PDF version.
The output of the circuit is the voltage required to balance the bridge (the voltage that appears at the R1/R2 junction). In this implementation it passes through another op amp stage that allows adjustment of the zero flow offset.
The performance of this circuit is not as good as that of commercial units made for turbulent flow measurements, but it was adequate for the task at hand and vastly less expensive. I spent about $300 including extra parts on the entire build. A commercially available system with four flush-mount sensors and the required constant-temperature circuitry would have cost about $15,000, which we couldn't justify for the type of measurements we needed to do.