Often it's necessary to build instruments for the calibration or characterization of the system actually under development. Here are a few examples.
In order to calibrate the ASAP position tracker we needed a calibration motion control system. We had been using a hexapod from Physik Instrumente, but it didn't have a big enough range of motion (especially angular), and it broke every time we used it, costing \$2K to fix. I found a Daedal XY theta stage on Ebay for \$700, got another Daedal stage to use for the Z axis, and assembled a 6DOF motion system with as much angular motion as you want and about 10x as much translation. I replaced the gearhead servo motors with microstepping stepper motors and 4K count encoders. I oversized the motors to maximize the open-loop stiffness. Because there is no gearhead, there is no backlash between the encoder and the lead screw, but the lead screw itself has 10 or 20 microns backlash, so I always approach a setpoint from the same direction, overshooting and reversing if we are coming the other way. This results in a position which is repeatable on-axis to within 2 microsteps, or about 1 micron.
The assembly weighs over 100 pounds, and has quite nicely reduced the resonant frequency of my go-kart inner-tube suspension on our optical bench breadboard to well below the predominant 10 Hz vertical vibration in NSH (probably due to HVAC equipment.)
I also had to develop a fair amount of Labview code to integrate into the rest of the system and implement the desired kinematics.
I wanted to investigate vibrations in the 6DOF Micron manipulator, and it occurred to me to try a stroboscope. I looked at industrial strobes, but the weren't well suited to microscopy, and were rather expensive, so I decided to make a LED strobe. The input is just a digital on/off signal, and analog electronics in the box regulate the peak and average current to give good brightness for both short and long pulses, while not destroying the LED. Pulses as short as a microsecond are at the peak 10A current. This is set by what the LED will tolerate. Strobe use is well outside the specs of most power LEDs, so I experimented to find what they would tolerate. Most would withstand short pulses of up to 30A, but the increase in brightness was not very much, and the high forward voltage (over 10V) showed a great deal of energy was being dissipated in the internal interconnect resistances. The strobe interface is in Labview, so the strobe can easily be synchronized to stimulus being numerically injected into the Micron servo loop.
The controlled current source used to drive the piezos in 3DOF Micron had such a high output impedance (a good quality in a current source) that it was difficult to measure. Rather than just saying it was “real high”, I persisted. The problem is that I needed to measure a very small modulation in the output current as the load voltage was varied. While a current transformer is not the obvious technology for making a low-frequency high impedance measurement, it occurred to me that the highpass characteristic of a current transformer was useful in this case because it rejected the DC component (and low-frequency drifts) of the signal. I wound a transformer with 10:1 current gain (usually a current transform is used to attenuate current), and I interfaced it to a transimpedance amplifier, rather than a resistive load. Making the secondary effectively shorted in this way considerably extends the low-frequency range of the transformer, but the secondary inductance affects the compensation of the transimpedance amplifier. Allowing for the turns ratio, the sensitivity is 100 nA/V, and this can be measured riding on top of a mA DC current. If you used a 10 megohm sense resistor, the DC current would drop KV across this resistor, which would hardly be practical.
Most of what you see is magnetic shielding to reduce interference pickup (primarily powerline harmonics.) There's a steel pipe, copper pipe, multi-layer permalloy sheilds, and then a ferrite pot core. Mostly because of the pipe it weighs about 20 pounds. After all that shielding, there was still some pure 60 Hz, which I polished off using a notch filter. I did get my measurement.