Speaker
Description
We present the experimental characterization of a fast spectroscopic imager for real-time room-temperature detection of low-density contaminants in food production lines. The presented imager is part of XSpectra®, an innovative inspection technology which combines a fast X-ray detection hardware and neural network processing techniques to improve the current limits in detection systems for food quality, material recycling, pharma safety and security applications [1,2].The detection unit consists of a 1-D array organized in four detection modules: each module is composed of a 32-pixel CdTe crystal, which have demonstrated promising results in the field of fast spectroscopy [3], coupled to four 8-channel read-out ASICs. The analog pre-amplified signals are sampled by an off-chip A/D converter and processed numerically by an FPGAs in the back-end electronics board. Due to the real-time requirements of the target application, the incoming events must be processed in the deep sub-microsecond range. A new version of the front-end ASIC has been recently designed to improve energy resolution and the dynamic characteristics of the system at short shaping times (<100 ns), allowing a consistent stability of spectroscopic performance in presence of high incoming photon fluxes.
The complete detection unit (128 channels in total) has been characterized to assess both low-rate spectroscopic performance and the ability to withstand high incoming photon fluxes. An average line width of 3.6 keV FHWM has been recorded on the 59.5 keV peak of a 241Am calibration source at a peaking time of 60 ns, with the low-energy threshold lying at 6 keV. Spectral resolution has been measured on all tested channels, with a deviation of ±10% with respect to the average value. The spectroscopic imager can cover an energy range of 200 keV, while keeping non-linearity error below ±1%.
A tungsten target X-ray tube, with a voltage working point of 30 kV, has been used to assess the imager performance in a realistic operational environment using contaminated food samples, showing a stable operation of the system up to an input count-rate of 2.1 Mcps.
[1] B. Garavelli et al., IEEE NSS/MIC, (2019), 1-3
[2] B. Garavelli et al., IEEE BioCAS, (2017), 1-4
[3] M. Sammartini et al., IEEE TNS, 68 (2021), 1, 70-75
