Verify if the 3D-printed sensor modules are identical, or if each module needs to be calibrated independently.
Required accuracy on the volume measurement is, according to the MHRA-requirements, is ±(4,0 +(15 % )) ml.
The accuracy of the flow sensor depends on the dimensions of a small slit in the 3D-printed parts. See Fig 1.
Slight differences in this slit may result in large differences in measured flow and volume.
For version 2 two sets of 10 sensor housings were 3D-printed. The white modules were printed in a PA12 certified for medical applications. These were purchased from Materialise. The grey modules were printed on a – higher-resolution- Fusionjet, provided by Atlas Copco. Both technologies are SLS-printers. The grey modules are not certified for medical use.
Machine 2.01, with firmware 3.57, was set up to deliver a Vt of about 500ml.
A Fluke V900a gas analyser was placed in line with the sensor module, between test lung and the sensor module. Values of Vt were recorded both as measured by the flow sensor module and the Fluke gas analyser.
The sensor module was then switched for the next one in the series, and the machine was restarted with identical settings.
|Sensor ID||V fluke (ml)||V sensor module (ml)||Difference (%)|
TEST SETUP 2
The grey modules were tested as oxygen flow sensors on the same machine. The resulting oxygen percentage was then registered from the GUI, and compared to the value measured by the Fluke V900A.
While it would have been better to test the grey modules in an identical setup, this was hard to do as the I2C-adresses of the sensor modules is set in hardware.
|Sensor number||FIO2 reported by GUI||Measured O2 value*|
CONCLUSIONS & REMARKS
Measurements show the sensor modules show a variation of up to 25% of the allowed deviation, under steady-state conditions.
They all perform within design requirements.
The variation is however significant, and a source of worry.
This test should be added to the EOL-tests for each machine, and each module is to be tested individually.
After discussion, cut-off was set at 5% deviation. Sensors 3 and 9 were rejected.
The modules created on a higher-resolution Fusionjet did not seem to show a higher consistency.
Therefor the medically certified white PA12 should be used in further development.
A valuable improvement is to ship each sensor with calibration data, and implement sensor calibration in the GUI.
While we used both sets of 3D-prints as shipped to us, we know more cleanup was needed on the Materialise prints, than was needed on the Fusionjet parts.