Introduction

Prompt gamma activation analysis (PGAA) [1] is a nuclear elemental composition analysis technique based on radiative neutron capture. By detecting the characteristic gamma rays emitted from the (n, γ) reaction [2], and having a high-quality spectroscopic library at hand [3], qualitative and quantitative elemental analysis can be carried out [4]. The method is non-destructive and bulk-representative, facilitating widespread applications to materials-, environmental-, and heritage sciences [5,6,7].

The analysis workflow of Prompt gamma activation analysis (PGAA) at the Budapest Neutron Centre’s PGAA [8] and NIPS-NORMA facilities [9], at the Heinz Meier Leibnitz Zentrum’s FRM II PGAA station [10], and other centers worldwide has relied on the use of a well-calibrated Compton-suppressed HPGe spectrometer [11, 12]. The prompt-gamma spectra recorded by traditional electronics have been evaluated by the Hypermet-PC gamma spectrometry software [13,14,15,16] and the ProSpeRo concentration calculation Excel macro [4]. This procedure was established in the late nineties and was compatible with the computing environment and data acquisition hardware of that era. It has been validated for several matrices [17, 18] and used to provide reliable results and uncertainty budgets for about 25 000 PGAA spectra over the years.

With these tools, however, the event count rate of the detector had to be limited to avoid the distortion of the peak shapes, either due to pile-up or due to extensive low-energy tailing. The peak fitting procedure with Hypermet-PC was still cumbersome and although the automated fit was completed within 1–2 min, the manual refinement of the region fits required up to an hour of an experienced analyst for many complicated gamma spectra, such as bronzes, metallic material-engineering samples.

The rise of interest from our international user community following the COVID-19-related shutdowns, in parallel with the reduced amount of available beam time worldwide, prolonged social distancing measures, international travel restrictions, and a push for better compliance with the ALARA principle of health physics called for more efficient utilization and more automated operation of the existing PGAA facilities. This initiated a collaborative capacity improvement effort of the two leading European PGAA knowledge centers, in Budapest, Hungary, and Garching, Germany. Our goal was to enhance productivity and establish a common baseline of the analytical practice, also easing interoperability and data exchange between the two facilities.

Experimental

Facility automation and bookkeeping

The neutron beamline facilities are unique and complex machines consisting of purpose-designed parts or components provided by several suppliers. All critical system components, such as the beam shutter(s), the neutron beam collimators, the gamma spectrometer, and the operation of a sample changer/sample stage must be synchronized and remote-controlled by a computer, which is a system integration and programming challenge. To minimize the dose rates of the analysts and allow 24/7 operation, the facility has to be able to take data of multiple samples (Fig. 1, left) or multiple measurement positions of a single sample (Fig. 1, right) in unattended batch mode, i.e. without the physical presence of the operator and the user. This necessitates the use of a computer-controlled sample changer device [19]. In the case of prompt-gamma activation imaging (PGAI) [20, 21], i.e. the spatially resolved version of PGAA, a translation/rotation sample stage [22] is used. Furthermore, many instrument parameters, e.g. the beam monitor intensity, the status of the shutter(s), the reactor, the cold source, as well as the count rate of the spectrometer, dose rates, sample-change events, time stamps of spectrum clear/start/save, and various I/O logical signals must be logged for quality assurance purposes, and warning messages shall be sent out when an error happens.

The spectra can be taken in unattended batch mode using a custom measurement control software or via ad hoc measurement scripts. In the second case, the JOB files (measurement scripts running within the ORTEC Maestro data acquisition software) [23] are generated using a user interface implemented in Excel, thereby excluding the malfunctions due to typos, while still having a high degree of flexibility. A test script is generated in parallel, similar to the real one but with 10 s of acquisition times, to verify the measurement sequence before starting the actual run.

Fig. 1
figure 1

the automated sample changer of the PGAA station (left) and the computer-controlled sample stage at the NIPS-NORMA station (right) of the Budapest Neutron Centre

Full size image

Pushing the count rate boundary of the gamma spectrometer

The count rate limit of traditional signal processing and data acquisition electronics has always been the capacity bottleneck at our PGAA stations. Due to the presence of hundreds or thousands of peaks in the PGAA spectra, the well-fittable peak shape and excellent energy resolution are two spectroscopic features that could not be compromised. For this reason, the typical count rate of the spectrometer has been maximized typically to a couple of thousand counts per second in Compton-suppressed mode, even if the beam intensity could produce a higher capture rate. The count rate of a given sample is indirectly regulated by adjusting the beam spot size with a set of circular apertures (PGAA), rectangular slits (NIPS), or by using attenuators (FRM II PGAA), which all result in loss of valuable neutrons. Under this condition, the measurement of a sample typically can take about 2–3, or in exceptional cases up to 12 h of beam time. By allowing a higher incoming count rate of the detection, the required counting statistics could be achieved in a proportionally shorter time, if the mass, volume, and composition of the analyte make it possible to benefit from the so-far unused part of the neutrons. Based on earlier experience, a factor of four improvement potential in the measurement productivity was identified at our PGAA facilities.

Spectrum histogramming

The NIM-style ADCs were capable of histogramming the spectrum into a maximum of 16,384 bins, which, considering the nearly 12 MeV energy range of PGAA, resulted in a bin width of 0.75–0.80 keV/channel. This is about four times larger than that of NAA and environmental or industrial gamma spectrometry. When fitting prompt-gamma peaks at low energies, the peak is spread over two–three channels only, which violates the basic assumptions of the peak fitting math. The counts in a spectrum bin are the integral of detected events from the lower to the upper edge of the bin, while the model function is computed at a single point within the bin. The pointwise peak shape model tends to systematically deviate from the observed bin-integrated peak shape if the bin width is wide and the peak area is large. So, even if we measure longer, the statistical uncertainty of the peak area does not improve with the measurement time for huge peaks due to these fitting quality issues. To properly represent a Gaussian-like peak with its histogram, especially if the peak is part of a multiplet, about 8–10 data points seem to be necessary.

Gamma spectrometer upgrade

To meet the challenges described in Sects. "Pushing the count rate boundary of the gamma spectrometer" and "Spectrum histogramming", we decided to focus the efforts on the signal processing electronics aspects rather than trying to improve the neutron beam flux. This seemed to be a cheaper and faster route to improve performance. After careful testing of several options, the ORTEC DSPEC 502A digital gamma spectrometer was found to cope the best with these challenges of neutron-capture spectroscopy. It demonstrated superior energy resolution (due to the low gain settings, still about 2.0 keV@1332 keV, instead of 2.2 keV of the analog system) and a more symmetric peak shape, even in the noisy environment of a reactor’s experimental hall. It can collect energy spectra in up to 64k histogram bins and can handle the non-ideal RC-feedback preamplifier signal shape of our HPGe detectors, thanks to the Tilt parameter of the trapezoidal signal shaping. The dual-input model can acquire events from two detectors concurrently, i.e. by connecting the HPGe to Input 1 and the BGO Compton suppressor to Input 2, one can implement an all-digital Compton suppression. This is a promising alternative to analog Compton suppression circuits built from many NIM modules. Most importantly, the spectrometer has about a factor of four increased count-rate tolerance, directly reducing the needed beam time per sample, while still maintaining a fittable peak shape. This was tested by adjusting the count rate of several metallic targets, such as Mn, Ti, and Cu, by stepwise increasing the neutron beam spot size. From 1 kcps up to about 64 kcps total count rate was investigated.

Data analysis methodology

The careful evaluation of PGAA spectra by an analyst may take up to an hour, therefore, we aimed at shortening this evaluation time significantly. Furthermore, the much-improved data acquisition possibilities may result in more spectra per beam day, so the increased data-taking rate had to be matched with a capable and more automated fitting procedure. To properly analyze the 64k SPC spectra [23] and the dual unsuppressed/suppressed output saved in the N42 format, the 2022.2 version of the Hyperlab software package [24] was developed. The N42 file format specification by NIST [25] is vendor-independent and human-readable, so complies with the FAIR data principle [26] required by many funding agencies. The nuclear database, efficiency-, nonlinearity-, and peak width calibrations, as well as its evaluation algorithms of the new Hyperlab release, were fine-tuned toward the needs of PGAA. With the more adequate FWHM model function \(\left(FWHM=\sqrt{a+bE+c{E}^{2}}\right)\) [27] and the newly introduced energy-dependent calibration of the peak shape parameters, the automatic fits became much better requiring much fewer user adjustments than previously with the Hypermet-PC. A fully automated batch evaluation module is also available to effectively process multiple spectra of the same kind. The maximum number of peaks in a region has been increased from 10 to 25–30, allowing for the consistent fitting of the middle part of the PGAA spectra, where a large number of small peaks are usually present as multiplets on a linear baseline. The PeakFit utility [28], intended to evaluate the Doppler-broadened boron peak and the positron-annihilation region of the PGAA spectra, two special peak shapes that are not (yet) supported by the mainstream software packages, was also revised to handle the new spectrum properties. Finally, the concentration calculation Excel utility ProSpeRo has been also technically upgraded to directly read in the peak list file and handle the efficiency-correction formulae of the Hyperlab.

Results and discussion

Spectrometer calibration

To calibrate the PGAA and NIPS spectrometers for the absolute detection efficiency, the nonlinearity of the energy measurement, and the energy-dependence of the peak width (see Fig. 2), a set of sealed radioactive sources (241Am, 133Ba, 207Bi, 60Co, 152Eu, 226Ra) and (n, γ) sources (urea-D, PVC, chromium) are measured at least twice a year. Here, we added chromium to the previously used set, to better cover the high-energy part of the nonlinearity curve and to limit the unrealistic fluctuation of the polynomial. The fitting engine for these calibration curves allows for the adjustment of the orthogonal polynomial’s degree to match the observed tendencies. The efficiency of the PGAA detector has been determined by both Hypermet-PC and Hyperlab, using the same set of spectra. The resulting curves were in statistical agreement.

The nuclear database of former Hyperlab releases was intended for radioactive decay and was inadequate for accommodating a PGAA library. A new class of nuclear data, the Reaction, has been implemented in the 2022 edition, which was populated with the data from the EGAF database [29], so the PGAA lines can now be accessed for energy calibration and qualitative peak identification. The database can contain entries from multiple data sources, so user data can be added, and subsets of the databases can be set up not to overload the user with too many choices. Notable examples of user-added reaction data were for the 15N(n, γ) [30,31,32] and 35Cl(n, γ) reactions, the capture lines of 52Cr and 50Cr, as well as for the decay chain of 226Ra [33].

Fig. 2
figure 2

The efficiency, nonlinearity, and FWHM curves of the Budapest PGAA station in September 2022 as determined using the Hyperlab 2022.2

Full size image

Dynamic range and selectivity improvements

By irradiating metal targets with a few-cm2 neutron beam spot, one can generate gamma source strength of ~ GBq even at a medium flux facility. Considering a geometric efficiency of about 0.001, this would still result in one million gamma events per second that can paralyze the detection systems due to the dead time. In our experience, total count rates over 100 kcps of events with MeV energies cannot be processed by our RC preamplifiers and the electronics. It also makes a difference how the counts are distributed in the energy spectrum. If there is one or a few dominant peaks that comprise the majority of the counts, e.g. B, H, the distortion of this main peak may become the practical limiting factor when setting the upper limit of the count rate.

Chromium is an appropriate metallic target material for general spectrum quality benchmarking, as it has well-separated and well-balanced intensity peaks at multiple energies up to 9.7 MeV. Its most critical line for the analysis is at 834.8 keV. In this test, we covered the recorded count-rate range of 2–44 kcps, which corresponded to 3.6–57% of dead time and 87–1600 cps in this peak of interest (Fig. 3).

Fig. 3
figure 3

a The broadening of the 834-keV peak of chromium as a function of the peak-specific count rate. b the low-energy skew parameter and the FWHM of the peak as a function of the peak count rate

Full size image

It was found that by the combination of better hardware and software, the overall count rate limit of the spectrum acquisition could be raised from 1000 cps to about 10 kcps, while the peak specific count-rate limit from 100 to 600 cps, without significantly influencing the spectrum quality. Raising the total count rate without severe spectrum distortion is not only relevant to productivity but also determines the dynamic range of gamma spectroscopy. In many applications, elements with low sensitivities must be measured in a high-cross-section matrix (e.g. Sn in bronzes, Li in Li–Ni–Co–Mn batteries, ppm-levels of H in metals), where the total count rate of the sample must be increased as much as possible to observe meaningful counts in the minor peaks of the poorly detectable component.

An illustrative example, the prompt spectrum of the element molybdenum is presented in Fig. 4. Although this is an uncommon element to analyze, its spectrum is challenging to evaluate. There are only a few well-analyzable peaks and several thousands of tiny peaks. Still, the minor peaks have to be correctly evaluated to correct for the complex interference patterns with other components.

Fig. 4
figure 4figure 4

The fit of the most intense Mo peak with 1.6 million counts in the peak area with 16k channels (a) and 64k channels (b). The evaluation of the middle region of the spectrum with 16k channels (c) and 64k channels (d), where the continuity of the baseline and its slope cannot be guaranteed when too narrow regions are evaluated independently, as done so far with the Hypermet-PC

Full size image

Two fitting cases were studied with Hyperlab, once the spectrum had the full 64k channel number, while in the other case, it was numerically compressed to 16k by merging every four bins, to resemble the bin width used in the past, but otherwise keep the two spectra statistically identical. We used the same peak fitting pattern not to bias the comparison, even if the automatic peak search did not place peaks here or there. The peak area determination was found to be consistent, while the quality of the fit improved when using more histogram channels. The chi-square of the region with the most intense Mo peak which had a peak area of 1.6 million counts, changed from 4.2 (16k case, Fig. 4a) to 1.3 (64k case, Fig. 4b), validating our premises. In the other case, this improvement was not so remarkable (Chi-square improved from 1.4 to 1.1, but still significant (Fig. 4c–d). More importantly, Hyperlab outperformed Hypermet-PC in this region fit exercise, as multiplets with 25–30 peaks could be properly evaluated, while Hypermet-PC is limited to a maximum of 10 peaks in a region. A practical application of this improved spectroscopy is being reported in a separate publication.

Another backward compatibility check has also been made using several lines of Mo. The 16k down-sampled spectrum was evaluated with the Hypermet-PC following our previous analysis routine, while the 64k spectrum was analyzed with the Hyperlab 2022. The efficiency and nonlinearity calibrations relied on the same set of measurements but were calculated in each program separately. The peak lists, after the efficiency and nonlinearity corrections, were fed into the ProSpeRo, where the peak energies were compared to the literature values, and the elemental masses were calculated from each line. The data tabulated in Table 1. indicate that the results are under good statistical control.

Table 1 The comparison of the old and new analysis protocols for several lines of the element Mo
Full size table

Conclusions

The productivity, spectroscopic performance, and accuracy of the PGAA analysis workflow have been significantly improved by recent hardware and software upgrades. A newly installed digital spectrometer, that has better energy resolution, peak shape, count rate tolerance and 64k histogram size, in combination with a higher degree of facility automation resulted in an increase of a factor of 3–4 in the measurement capacity of our facilities, without increasing the neutron flux. The higher data acquisition rate produces more spectra per beam day, so the gamma spectrum evaluation had to be made more efficient as well. This was achieved by transitioning from the Hypermet-PC to the Hyperlab 2022 software package. It is more user-friendly, and thanks to its recent refinements, requires less manual fine-tuning of the region fits following the automated fit. The new procedure has been verified under conditions beyond our previous operation boundaries and found to perform adequately. After careful validation, the much-improved measurement and analysis protocol is part of the regular analysis procedure since Q4/2022.