Just to let the forum know, Keysight are going to send me a replacement unit and Fraser has agreed to take the broken one off my hands.
I have experimented further with the faulty unit, as the erroneous behaviour might give further insight into the exact workings of the thermal imager. Fraser told me privately (but I'm sure he won't mind me quoting him here):
The switch on sequence of the camera is the same as many.
1. The camera processor does a self check of all vital elements and reports any errors. Your camera reports no such errors.
2. The TEC control circuit begins to heat the microbolometer to +32C which is its normal operating temperature. During warm up the tilda (~) is shown next to the measured temperature to indicate that it is approximate..
3. Once stabilised at 32C the microbolometer produces its best images and accurate radiometric data for measurement.
4. Your camera initially shows a good image and correct temperature. Then the image degrades and rows of microbolometer data begin to blank. The measurement process continues to work OK and is just reading teh data that it is being fed by the microbolometer ROIC.
5. The symptoms suggest a potential issue with the microbolometer ROIC that may be thermal in nature. As the camera warms the microbolometer, the ROIC starts to fail.
I've determined that if I power the camera off and immediately on again, the problem persists. It needs a cool-down period of about 4 minutes. Shorter periods may result in a sort of half-working image which is fascinating. The fact that the problem always grows from the top left-hand pixel downwards seems to my mind to indicate a problem in the digital domain. Often the lower part of the image still bears some relationship to reality (Image 1). The Max and Min indicators correspond the the colour on the screen, so the issue is certainly before whatever processing step handles that (Image 2).
After the error takes place, the camera settles down to something like Image 3, which no longer has anything to do with reality, and is always shown to be around 36°C, possibly the internal calibration temperature?
Occasionally after a short cool-down, some lines of the image work but not others (Image 4).