In a previous post, we had the opportunity to review some validation concepts, like the U test model. In it, the Unit Under Test receives well controlled stimuli and the system analyzes its response.
In this one we will cover some examples of it.
Power Supply Unit (PSU)
Validating the design of a PSU (board or module) can be easy… or a very complex tasks. Follows some guidelines with no desire of being exhaustive.
Even the simplest AC/DC PSU based on a transformer, diode rectifier and filtering capacitor requires a large analytical worst case support. This is something that you are never going to be able to validate by testing.
Do not forget thermal aspects when validating even the simplest PSU. Once again, an analytical worst case analysis will help much to identify hottest conditions and to set temperature increase expectations.
Testing has to be done in nominal and in corner cases. For example, you have to test a switched mode PSU when the input voltage is in the low range and the current demand is in the high range.
An electronic PSU is always a feedback system and you need to check its stability. One of the best way of doing is by giving stimulus in the form of transient load current demands.
Having the capability to use an electronic load is of great help because with it, you can easily test different load conditions and also transient behavior. But beware that the instrument has also a servo loop. Because of that, you need to take extra measurements (voltage and current) to check everything works as expected.In linear circuits, it may happen that testing with a DC load is not enough. Only under pulse current demands the bulk decoupling capacitor's internal resistance (ESR) will stand up. However, in most capacitors of a switched mode PSU there will be AC current circulation. The aspect you have to care in these devices is the voltage ripple and the temperature increase due to ESR.
Switched PSU: it is very convenient to measure coil current. Look for non-linearity (symptom of saturation) or cross-coil mutual coupling. However, for this later, be best way is to attack the root of problem and use shielded coils.
Memory test
This test may be very different if you are validating the design or doing the assembly test.
Design validation: your focus will be on the functionality at maximum transfer rate in the full temperature range.
There have been many times in which I have coded low level memory test in assembler (when external memory is still unused) or an specific memory test in an FPGA in order to speed test execution and to increase test coverage as much as possible. If such a test can also be executed for production or initial self-test is strongly dependent on application but it is always a blessing when you have to be sure about memory functionality.Assembly test. It is reasonable to assume that the memory manufacturer is providing a functional device and your main focus will be to test if the device is properly assembled. The possible faults are shorts or opens in data, address and control lines. The most exhaustive test is the one that writes deterministic data in every memory position and after doing it, reads it back and compares to expected value. Such a test tends to be lengthy for today's memory sizes.
There are many possible tests that are much faster to code and execute but its coverage may be very poor. For example, writing the same content to the whole memory or writing a value and reading after does to check address lines are working properly because you do not know if the data has been written to desired position.
Testing at full clock rate is not a requirement of the test coverage (it is expected that the design validation covers it) but it is an advantage because speeds up the test time. However, this is not always the case: a well designed boundary scan test can have a large coverage although it will be executed at a much slower speed that operational one.
Temperature sensors
Let us consider resistive temperature sensors, RTD (Resistance Temperature Detector).
It you need to test a circuit that measures temperature by measuring the resistance of an RTD device, just use high precision and low temperature coefficient resistors whose values are in the middle and in the two extremes of the measurement range and forget about keeping sensors in a well controlled temperature environment.
Beware that these sensing resistors precision should be much better than objective one unless you are able to make some sort of calibration of them with very high precision instrument and its stability can be guaranteed.
Signal conditioning
When your circuity has to process precise analog signals the best approach is to validate the design by an analytical worst case analysis. Never forget to include in the equation the measuring instrument internal uncertainty, the one in the test bench. No measurement instrument is perfect.
Do functional test for manufacturing validation and whenever possible, store obtained results, post-process them and compare against calculations. Look for statistical distributions, ranges and tendencies.
Maybe, when you have a large historical date, you could to use it to do some sorts of calibration. However this can only be done properly it the source of error is well understood and does not depend on external and uncontrolled conditions (like temperature, voltage, ageing…) You have to be very sure calibration does not worsen things. Whenever possible recheck that the calibrated system improves system precision.
Quadrature encoder
Once I did and encoder emulator based on a very low cost MCU board (Nucleo STM32 L432).
This emulator has the same interface of the emulated board and had two boards, the Nucleo and a very simple one that included the electrical drivers of the original board and some jumpers and push buttons. With them I was able to configure 1000 cycles in one direction or the reverse one, a basic sweep emulation or continuous movement in both directions. Pure gold: it has helped me much many times despite its short development time and low cost.
Remotely controlled instrumentation
Specially for production testing, having the possibility to do a remote monitoring and control of electronic instrumentation adds a lot of value because can save time, tests in a repeatable way and avoid errors in the captured data.
If you have never done it, I suggest you to fight fear and to start from something easy: program remote control of a PSU, a multimeter,… in you favorite programming language.
In this respect I give you a final advice: always you test something the first time, do it manually and automate after. Automated testing is very convenient if you measure correctly but very inconvenient when it measures improperly. In the first time you measure something, apply as much (AI) Artificial Intelligence (HI) Human Intelligence as possible to see if everything is as expected.
Anyone that has measured (or follows this newsletter) knows that «measuring right is very difficult». Validating in the right way is also very difficult :-)