1
\$\begingroup\$

I'm designing some modular sensors for my company's internal testing. Most are analogue, with a sensor (PT100, current shunt, pressure transducer etc), some interfacing (REF200 based PT100 circuit for example) and an op-amp (usually an OPA170) to send the 0-10V signal down the wire to the NI based measurement equipment.

I had our first batches of sensors made with 1% resistors, and there's more variance than I'd like on some sensors - for example the PT100 units are +/-3degC over our ten unit prototype run with the same PT100 sensor.

Other than switching to 0.1% resistors, is there anything else I can do to reduce variance between units?

Thanks!

\$\endgroup\$
6
  • \$\begingroup\$ Better Vos op-amps AND better resistors AND better voltage references. Also RF quenching to prevent unwanted DC errors in high EM fields. Higher input impedance to the PT100 as well. \$\endgroup\$
    – Andy aka
    Commented Aug 20, 2019 at 10:35
  • \$\begingroup\$ Do we talk about systematic errors (i.e. one test device always measures a temperature two degC to low) or about statistical errors (the measurements of one device scatter with a deviation of 1 degC)? Back in school I learned that one uses some wheatstone bridge setup to measure resistiance chances. What kind of setup you use / could you add a schematic? \$\endgroup\$
    – Christian B.
    Commented Aug 20, 2019 at 10:40
  • \$\begingroup\$ @ChristianB. I can't add a schematic, but my PT100 design is based on this TI application note: ti.com/lit/ug/tidu969/tidu969.pdf The errors are systematic - results are pretty stable, at least on my nice quiet test setup. \$\endgroup\$
    – Joe of Loath
    Commented Aug 20, 2019 at 10:57
  • \$\begingroup\$ @Andyaka can you expand on "RF quenching", I'm not familiar with the term? \$\endgroup\$
    – Joe of Loath
    Commented Aug 20, 2019 at 10:58
  • \$\begingroup\$ Have you the skills to develop partial-derivative models of each error source? This is the mindset you need. \$\endgroup\$
    – analogsystemsrf
    Commented Aug 21, 2019 at 6:12

1 Answer 1

Reset to default
5
\$\begingroup\$

Do a sensitivity analysis of each component in your circuit. Pt100 DIN changes about 0.385% per °C, so a 3°C error represents ~1% overall error. This will also inform your selection of components for stability with temperature, moisture, time and soldering. In this particular case, resistor tempco should certainly be considered for the critical parts. 1% tolerance does not necessarily mean they are extremely stable with temperature.

Errors can come from resistor tolerances, errors in the nominal resistance values when mapped to standard values (can be reduced by using series or parallel arrangements of standard values), amplifier offsets and bias currents, reference voltage error and so on.

A couple of calibration trimpots (span and zero) can cancel out many different sources of errors at once, or more suited to modern times, if you have the numbers in a microcontroller you can apply digital scale and offset values for calibration, and even correct for linearity and ambient temperature effects.

\$\endgroup\$
3
  • \$\begingroup\$ It is notable that putting some components through reflow can add offsets as large as the nominal tolerance; this information should be available in a test report which is usually available for precision parts. \$\endgroup\$
    – Peter Smith
    Commented Aug 20, 2019 at 13:58
  • \$\begingroup\$ Thanks, that gives me some more avenues to pursue. I've considered trimpots but I'm dreading the day someone needs 100 sensors and I have to calibrate them all individually... \$\endgroup\$
    – Joe of Loath
    Commented Aug 21, 2019 at 9:29
  • \$\begingroup\$ If a circuit is designed well (very little interaction) manual zero/span calibration takes only seconds as part of obligatory testing , and if you do it digitally it can be automated. I have personally calibrated tens of thousands of instruments. \$\endgroup\$
    – Spehro Pefhany
    Commented Aug 21, 2019 at 11:13

Not the answer you're looking for? Browse other questions tagged or ask your own question.