See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251527065

HPGe detectors long time behaviour in high-resolution γ spectrometry

Article  in  Nuclear Instruments and Methods in Physics Research Section A Accelerators Spectrometers Detectors and Associated Equipment · August 2011
DOI: 10.1016/j.nima.2011.03.031

CITATIONS READS

6 870

10 authors, including:

L. Sajó-Bohus Andres Matias Sajo-Castelli

Simon Bolívar University Simon Bolívar University
177 PUBLICATIONS   1,135 CITATIONS    31 PUBLICATIONS   34 CITATIONS   

SEE PROFILE SEE PROFILE

D. R. Napoli Haydn Barros

INFN - Istituto Nazionale di Fisica Nucleare Simon Bolívar University
859 PUBLICATIONS   9,839 CITATIONS    76 PUBLICATIONS   323 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Shape Coexistence View project

Quadrupole correlations in N=Z=50 region View project

All content following this page was uploaded by Andres Matias Sajo-Castelli on 26 October 2017.

The user has requested enhancement of the downloaded file.

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
 Author's personal copy

Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138

Contents lists available at ScienceDirect

Nuclear Instruments and Methods in

Physics Research A
journal homepage: www.elsevier.com/locate/nima

HPGe detectors long time behaviour in high-resolution g spectrometry

L. Sajo-Bohus a,n, D. Rosso b, A.M. Sajo Castelli a, D.R. Napoli b, E. Fioretto b, R. Menegazzo c, H. Barros a,
C.A. Ur b, D. Palacios a, J. Liendo a
a
Universidad Simón Bolı́var, Apartado 89000, Caracas, YV-1080A, Venezuela
b
INFN Laboratori Nazionali, Viale dell’Universita 2, Legnaro (Padova), I-35020, Italy
c
INFN Sez. Padova, via F. Mazzolo 8, Padova, I-35131, Italy

a r t i c l e i n f o a b s t r a c t

Article history: A large set of data on long term performance of n-type HPGe detectors used in GASP, EUROBALL and
Received 11 January 2010 CLARA g spectrometers, as well as environmental measurements have been collected over two decades.
Received in revised form In this paper a detailed statistical analysis of this data is given and detector long term behaviour is
18 March 2011
provided to the scientific community. We include failure, failure mode, repair frequency, repair
Accepted 18 March 2011
Available online 31 May 2011
outcome and its influence in the energy efficiency and energy resolution. A remarkable result is that
the life span distribution is exponential. A detector’s failure is a memory-less process, where a previous
Keywords: failure does not influence the upcoming one. Repaired spectrometers result in high reliability with deep
HPGe detectors implications in the management of large scale high-resolution gamma spectrometry related projects.
Gamma spectrometry
Findings show that on average, detectors initial counting efficiency is slightly lower (  2%) than that
Long time performance
reported by the manufacturers and the repair process (including annealing) does not affect significantly
the energy efficiency, even after a long period of use. Repaired detector energy resolution statistics
show that the probability, that a repaired detector will be at least as good as it was originally, is more
than 3/4.
& 2011 Elsevier B.V. All rights reserved.

1. Introduction spectrometer damage was related to electronic failures, vacuum

loss or neutron damage. Partial information on maintenance and
The cooperation between the Istituto Nazionale di Fisica repair activity has been published yearly in the LNL Annual
Nucleare (INFN) and the Simón Bolı́var University (USB) began Report [5]. In agreement with literature reports [6], we have
in the late seventies when HPGe detectors became available observed that good cooling system improves the resistance of
starting a new era in high-resolution gamma spectroscopy. State the detector against the neutron damage. Most of the detectors
of the art HPGe technologies, well reviewed by Eberth and used in this statistical work belong to the GASP array [7] where
Simpson [1] and by Vetter [2], gave rise to an important number it was decided to shorten the cold finger, obtaining thereby a
of 4p g spectrometers and good opportunity for nuclear spectro- more compact configuration. The working temperature becomes
scopy studies [3]. around 5 1K lower than Exogam [8,9] or Gammasphere [10]. This
During the last 20 years we have operated at INFN-Laboratori modification provided a reduction in neutron radiation damage.
Nazionali di Legnaro (LNL) a significant number of n-type HPGe Furthermore, a radioactive gamma source was employed between
coaxial detectors acquired between1985–1995. These detectors in-beam experiments and this allowed longer detector opera-
have suffered normal damage due to in-beam operation, and most tion under high gamma counting rates. Due to manufacturer
of them have been fully recovered in our laboratory. In the few policies, full repair of HPGe detector was not possible. Detector
cases where local maintenance was unable to deal with the performance is adversely influenced by the non-release of suffi-
problem, the detector was sent to the manufacturer for servicing. cient information on the detector failures-tree, specific repair
The degradation and recovery of the HPGe have been studied (e.g. the vacuum valve), electronic adjustment procedures, and
earlier by Albernhe et al. [4] In the present work we will show our lack of the availability of special tools or spare parts in their
experience related to maintenance and repair of these HPGe standard catalogues. Manufacturers of germanium detectors fre-
detectors from 1992 to 2009. During this period of time most quently suggest replacement rather than repair. However, after
tedious negotiation most of our detector manufacturers gave us
all the additional information needed, sold us tools and gave
n
Corresponding author. Tel.: þ58 2129637179; fax: þ58 2129063590. technical support. Without this help it would have been much
E-mail address: [email protected] (L. Sajo-Bohus). more difficult to attain the skill in this field. The results presented

0168-9002/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.nima.2011.03.031
 Author's personal copy

L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138 133

in this study, confirm that a good maintenance and repair 3. Maintenance and repairs
programme can preserve the HPGe detectors. Not only preserve
satisfactory operational conditions for a long period, but Most of the maintenance we have done in our laboratories
also provide a reduction of the financial burden of costly deals with electronic, vacuum and annealing. In the present
replacements. section we will start with a short description of each type of
repair and later we will show some statistics and summary of the
repairing process.
2. Detectors selection and employment
3.1. Electronic repairs
Our experience on detector repair, annealing, etc., is based on
99 detectors. Two groups of g spectrometers are considered Electronic repairs include not only the substitution of compo-
depending on the user’s laboratory location and application. The nents in the preamplifiers external stages and the HV filter but
first group consists of 9 detectors, the second of 90 detectors. For also components like the FET or the resistances inside the vacuum
the statistical analysis, Section 4, only a subset of these detectors chamber of the cryostat. Quite early it was decided not to
was used: those which had a full record available. This subset substitute internal temperature sensors (PT100 for crystal tem-
numbered 51 detectors. perature measurement or PT500 for annealing control) since it
was preferred to add an additional PT100 inside the Dewar
2.1. Spectrometers operating under low radiation environment instead of opening the crystal housing or the use of an external
self controlled system for the annealing.
Detectors are from five manufacturers (Ortec, PGT, Oxford-
Tennelec, Canberra and Selenia) selected for their characteristics 3.2. Vacuum recovery
that suit most applications including low-level g spectrometry for
nuclear analytical applications in Venezuela. In this study, a set of It is been reported in the literature that most of detector problems
nine HPGe detectors and three SiLi for X-ray spectrometry have are related to the vacuum loss and the resulting lack of required
operated for the better part of the decade 1990–2000 with few cooling capacity. Dewar vacuum has a limited life span and degrades
interruptions due to power supply break-down or spectrometer due to its inherent manufacturing design. Several alternative tech-
failure. Four units were acquired more recently in 2006, but most niques for liquid nitrogen cooling have been developed, some are
entered in operation between 11 and 17 years ago. based on the Kleemenko mixed-gas refrigerator [13]. The related
cooler detector combination reached the market standards, see Upp
et al. [14] and Keyser and Hagenauer [15], offering low cost/low
2.2. Spectrometers for in-beam nuclear structure studies
power cooler systems.
It is worthwhile to mention that Germanium crystal devices
The second group of detectors is related to large projects such
operating under continuous cryogenic conditions, stable reverse
as the GASP [7] or EUROBALL g-arrays at LNL. Most of the
bias and a low background radiation environment, are not
experience was acquired from a group of 90 spectrometers at
expected to suffer physical deterioration in the detector active
LNL employed for in-beam experiments carried out during the
volume or in the associated electronics.
past 18 years. Each one of these detectors is accompanied by a
Supporting evidence of the above statement is that, during
technical data sheet, which describes working and performance
data acquisition, the device’s current flow is almost negligible.
characteristics with a short technical description. An electronic
Likewise degradation due to the heat generated during operation
database was made containing these data sheets. It included
is negligible in contrast with the radiation related physical
detailed information on the annealing date, refurbishment (either
damage. This means that, in principle, long term operation could
electronic circuitry or mechanical modification), repair, shelve
be attained in a low background radiation environment. On the
time and time required to recover the unit to its original
other hand, detectors operating in high radiation fields degrade,
specification or close to it. It is possible to know if any technical
as experience shows. Neutron fields, for instance, favour nuclear
intervention on any detector changed the operational perfor-
reactions in the crystal itself with subsequent lattice modifica-
mance over the factory certified specifications. The detectors
tions, which cause worsening in the detector performance [16].
were employed at the LNL tandem accelerator for about the
Detector crystals with small cryostats are prone to malfunction;
50%–70% of the yearly total available beam time [5] and were
the usual way to repair is improving the cryostat low pressure.
exposed to a low intensity fast neutron flux as a result of the
However this can cause damage to the Ge-crystal, which in turn
reaction mechanism. Table 1 shows the number of detectors that
can be repaired by thermal recycling. Our experience shows that
have been used at the LNL for each gamma-array.
the most effective technique to regain the crystal’s original
operational performance is by the annealing procedure.

Table 1 3.3. Annealing procedure

LNL detectors summary and associated project.

Number of n-type GASP [7] CLARA clovers [11] EUROBALL

Energy resolution deteriorations or pulse shape distortions are
HPGe g detectors (single crystal (four single crystals clusters [12] often the troubles caused by improper charge collection or
detector) in cryostat) (7 single crystal neutron induced damage and could be solved by annealing the
incapsulated for detector’s crystal. In this section we will detail some technical
each cryostat)
aspects on a less invasive annealing method developed at the LNL
Total available 51 24 15 compared to manufacturers’ protocol.
In operation at the 40 21 12 Once the detector’s access valve is opened at room tempera-
same time in ture (25 1C), the cryostat is connected to the vacuum system with
the array a turbo-molecular pump, as given in the block diagram of Fig. 1.
Spare detectors in 11 3 3
operation
The crystal undergoes a thermal cycle that involves heating it up
to 103.5 1C þ/  0.5 1C. The heating time is 40 h followed by a
 Author's personal copy

134 L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138

3.5. Maintenance and repairs summary

The most often encountered failures for g spectrometers and their

frequency for the set of mentioned HPGe detectors are given in
Table 3. This table indicates that the annealing was necessary in 50%
of the cases in order to restore the equipment and 20% required
Fig. 1. Block diagram for the annealing system: Heating chamber (HCh), heater (H),
heater control (HCon), temperature control (TCon), detector D, Cryostat valve (CV),
vacuum pumping. These results are indicative of the type of main-
quadrupol mass spectrometer (QMS), turbomoleculra pump (TMP) and rotative tenance and the restoration technique. After annealing, 11 detectors
pump (RP). improved up to a 10% FWHM. The loss of vacuum was detected when
water condensed on the Dewar neck, a clear indication that pressure
reached high values and the Dewar was warming up. Regarding the
Table 2 frequency failure versus time reported in Table 4, it can be noted that
Summary on the annealing technical parameters.
there are other laboratories, which have had similar experiences with
Parameter Value g spectrometers e.g. Laboratori Nazionali Gran Sasso (LNGS, Italy)
with their detectors on the LUNA experiment. In Table 5 we list the
Initial vacuum 0.15 mbar kinds of most often encountered detector faults, actions taken and the
Average temperature 103.5 1C
outcome. Finally, it is worthwhile noting that the repair process
Pumping time 3h
Annealing time 40 h (including annealing) does not affect significantly the energy effi-
Vacuum at starting the 1.5  10  6 mbar ciency of the detectors, even after a long period of use (see Fig. 2).
heating process
Final pressure 9.8  10  6 mbar
Heater resistance 195 O
PT100 resistance 110 O 4. Performance statistical analysis
PT500 resistance 540 O
During the time of operation of the detectors an electronic
data sheet was filled for each repair. This information is part of a
cooling time that lasts from 40 to 60 h, depending on the
database with which it is possible to follow the history of each
detector’s surface cleanness. Temperature and pressure tendency
detector. In the present statistical analysis we selected the data
as a function of time for the detectors under treatment are
related to 51 HPGe of the total 99 detectors available. The
monitored to provide information on how to proceed during
selection criteria were that the detector had to have a full record
annealing treatment and the cooling rate to be followed. The
on the parameters of interest. For the statistical tests regarding
initial low pressure of 10  6 mbar is maintained during the
detector counting efficiency, due to typical values in their data
thermal operation time. The out-gassing flow is monitored by a
sheet, a subset of 43 from the 51 detectors was employed. For the
coupled quadrupole mass spectrometer (Leybold-Heraeus Q 200)
rest of the statistical tests data of all 51 detectors were employed.
that measures the residual gas concentration in the system before
The data available ranges from January 1991 to September 2009.
and after the annealing cycle. This is observed in the character-
We studied the statistical behaviour of the following parameters:
istic curve produced during the detector’s pressure change. As the
temperature–pressure are varying with time it is possible that
eCm counting efficiency reported by manufacturer,
an irregular pattern may be observed initially and depending
eCLNL counting efficiency measured by LNL measured at the
on its departure from a smooth curve a longer annealing time
time of purchase,
(60 h) is advisable. However the cooling time interval selection
em energy resolution in keV reported by manufacturer at
also depends on the range of values of the residual gas concen-
Eg ¼1.33 MeV
tration. Typical values for residual water have a range around
eLNL energy resolution in keV measured by LNL (same g
2  10  3 mbar while for heavier molecules (e.g. residual oil
energy) along each detector’s life span
vapours) the range is around 4  10  6 mbar. The annealing
process is completed when the detector cools to room tempera-
Table 3
ture and the pressure is lower than 10  6 mbar. It is important to Summary of troubleshoot variables and their repair frequency.
mention that all the pumping processes are done through a
cryogenic trap, specifically in the first detectors treated when Type of maintenance Repair
oil free pumping systems were not available. Nowadays, in spite frequency (%)

of the turbo-molecular pump, we continue to use the cryo-trap

Temperature sensor PT100 and LEMO cable connection 2
for vapour cleaning and power failure protection (i.e. the vacuum H. V. shut down related electronics 2
is maintained when the valve closes the trap.) A summary of FET burn or FET related noise48 mV 5
annealing data is given in Table 2. Vacuum valve and molecular sieve saturation 17
H. V. filter due to high air humidity 9
Annealing when FWHM 3% or long left tail at preamp 57
3.4. Other maintenance activities Vacuum and valve problems related to mishandling 4
Other 4
Some other repairs like soldering of the Dewar filling feed-
through were done but we do not have enough cases to obtain
any conclusions. In 1999 the original preamplifiers of the GASP
detectors started to be substituted with a new home made circuit, Table 4
designed by Buscemi [5]. These have a faster rising time and a Summary of workshop repair time. In seventeen years there has only been one
case where the detector’s repair required more than a year.
higher resolution, based in a low noise operational amplifier with
an extremely high gain and a better bandwidth. These character- Repairing time intervals 1 day r1 Week r 1 Month r 1 Year Year
istics imply a reduction in the threshold of the CFD, improving the
Rounded percentage of 29 21 31 19 0
detection efficiency for low energy gamma rays and a more workshop time repairs
counting rate independent energy resolution.
 Author's personal copy

L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138 135

Table 5
Summary list of the most often encountered troubleshootings and repair outcomes.

Problem reported Solution applied result

Loss of inverse current Degassing by vacuum cycle Partially recovered

Faulty operation valve Local repair attempted Sent back to manufacturer
High noise FET change Good

Scatter Plot of Original and Post Repair Scatter plot of count efficiency for each detector
Measured Counting Efficiency (%)
95 90

Count efficiency (%)

90 85
Original Measured Efficiency

80
85

75
80

0 10 20 30 40
75
Detector number

Fig. 3. Scatter plot of eCm(circles) and initial eCLNL(crosses) respect the detector
number. Values are bounded between 69% and 91%. The scatter is not significant.
70
70 75 80 85 90 95
Efficiency After Last Annealing Boxplot

Fig. 2. Comparison of detector efficiencies: original vs. measured values after the 90
annealing.
Count efficiency (%)

hd heating treatment applied (boolean value, used implicitly in 85

variables Ay and An, Section 4.2.a)
t date of resolution measurement after repair (mm/yy, i.e. the
temporal data resolution is 30 days) 80

Values for em and eCm are those reported on the spectrometer’s

manufacturer certificate, and were used as benchmark reference.
75
All the numerical operations where done using the ‘‘R’’ Statistical
package [17].

4.1. Detector counting efficiency

Manufacturer LNL
This section is devoted to the statistical study of eCLNL and eCm Fig. 4. Box plot of the eCm and eCLNL variables. The median values are less than 3%
variables. As a first approximation to understand the features of apart while the mean values are less than 2% apart (not shown.) Both present
g spectrometers based on the initial counting efficiencies (eCm and almost the same dispersion.
eCLNL) a scatter plot is shown in Fig. 3 of the efficiency values for
the selected group of detectors. On the other hand, the mean efficiency (only for the smaller group of detectors). In order to
values for the above mentioned two variables (certified and LNL answer the above question, we assume normality and that the real
experimental data initial values at purchase) were determined variances of both sets are the same, additionally we define
that do not match as can be seen in Fig. 4. In fact, the mean value a ¼10%
of the manufacturer efficiency is 83% while the experimental mo ¼ meanðeCm Þ
mean reported by LNL is 81%. This raises the question of whether
mL ¼ meanðeCLNL Þ
or not these mean values can be assumed to be the same. This will
be answered shortly by following a hypothesis test on the two The hypothesis test
means. For the analysis done in this subsection, it was determined Ho ¼ mo mL ¼ 0
that the data provided by detectors 12, 34, 38, 40, 48–51 have Ha ¼ mo mL a 0
atypical values (out of the 3s range) and were removed from the
bank during this specific study. This leaves the data-set with 43 and using the following t-distribution test statistic
detectors. Table 6 shows a summary of statistical parameters for sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
mo mL 0 ðno 1Þs2o þ ðnL 1Þs2L
these two variables and Fig. 5 shows a scatter plot of the T ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , Sp ¼
manufacturer certified and the measured initial counting Sp ð1=no Þ þ ð1=nL Þ no þnL 2
 Author's personal copy

136 L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138

Table 6 Table 7
Summary of the eCm and eCLNL variables. The means differ by 1.97%. Average life span for each detector group between two consecutive repairs.

eCm eCLNL Group Annealing Average of working

days between two repairs
Min 71 71
1st Qu. 81 79 Ay yes 503
Median 84 81 An no 541
Mean 83 81 Ay U An – 519
3rd Qu. 88 85
Max 91 89
Std. dev. 5,0 4,9 (b) Does a detector repair improve or degrade the detector’s
Std. err. 0.76 (0.91%) 0.75 (0.93%) energy resolution?
(c) Is it possible to predict detector malfunction or life span
expectancy from a statistical point of view?
Scatter plot of Certified and Measured
Counting Efficiency (%) The following sections will deal with these questions.

4.2.1. Average life span and the annealing process

We would like to know how long will be working, on average,
90 a repaired spectrometer. We also would like to know if the
annealing technique applied during repair has some kind of effect
Certified Efficiency

over performance and working time. Two groups are considered,

85 in the first one we choose those spectrometers whose repair
include the annealing (Ay) and the second group those spectro-
meters that did not (An). The life span until the next repair can be
80 seen in Table 7. Over a total of 565 repairs done, 57% of them
where treated with the annealing technique. It would seem that
when the annealing is properly done the life expectancy is not
75 modified. A fast overview of the confidence interval of 97.5% of
each group’s life span average confirms the above:
CIðAy ,0:975Þ ¼ ð503 7 68Þ days
75 80 85 90 CIðAn ,0:975Þ ¼ ð541 774Þ days
Measured LNL Efficiency This result shows that the annealing process does not affect
Fig. 5. Scatter plot of the manufacturer’s certified and the LNL measured counting the average life span of the detectors, and on the other hand, it
efficiencies. The dashed line shows the LMS fit and the continuous line shows the represents a good procedure to enhance/recover some features of
symmetry axis. the semiconductor detectors (i.e. energy resolution, etc.).

where no and nL are the number of detectors, namely 43, sO and sL 4.2.2. Detector energy resolution
denote the standard deviations of eCm and eCLNL respectively. Note Since detector repair is a necessity, it would seem reasonable
that there are n ¼no þnL–2 ¼84 degrees of freedom. The rejection to know (based on previous experience) if it is worth the effort.
region is given by 9t9 4ta/2, 84. Finally obtaining t ¼1.85 and since Employing the 51 detectors data we compare the average of each
ta/2 ¼qt(0.95, 84) ¼1.66 we reject the null hypothesis with a p-value detector’s energy resolution after repair along its life span with the
of 0.0676, additionally we report a IC90% for the difference of the detector’s energy resolution published by the manufacturer. For
means each detector we use a hypothesis test (assuming normality too
mo mL A 0:20, 3:75 and with a confidence level of 97.5%). We define:

note that 0eCI90%. The data seems to show enough evidence to i ¼ 1. . .51, miO ¼ eim , miL ¼ meanðeiLNL,j Þ, j ¼ 1. . .ki
( )
support the discarding of Ho (with a confidence of 90%). Conse- HO ¼ miO miL ¼ 0
i
quently it can be assumed that the two means are not equal and HT ¼ ki : times detector i was repaired:
Ha ¼ miO miL a 0
therefore detectors initial performance is lower than the reported
by the manufacturers. in the event where Ho is not accepted, we investigate whether or
not we can assume that miO o miL . This is also evaluated with the
4.2. Long term detector statistical aspects same confidence level. The HTi were evaluated by the following
t-test:
Here we discuss some aspects that concerning the long term
miL mio
behaviour of the detector’s energy resolution (eLNL and em vari- Ti ¼ pffiffiffi
s= n
ables). The expected life span after repair of a detector is, in
average, 519 days (1y and 5m). During the ‘91–09’ period with n ¼n  1¼50 degrees of freedom and s being the standard
(18 years) there has been a total of 520 repairs made in our deviation. For the first case the reject region is 9ti9 4ta/2, 50 and
laboratory and 45 were done by the manufacturer for the selected the second optional case is ti o  ta/2, 50. That is, if the data does
group of 51 detectors. The database describes the life span of the not support evidence that the average resolution is the same as
detectors excluding the ones of the last repair, related to a total of the manufacturer’s, we would like to know if the restored
513 repairs. Some questions arise concerning detector repair. detector’s resolution is higher or lower that the certified one.
Regarding the 51 detectors and all the corresponding repairs,
(a) Does annealing change the life span expectancy of the treated the results show that the average resolution after repair, com-
detector? pared to manufacturer’s initial value, the initial efficiency is
 Author's personal copy

L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138 137

recovered in 65% of the cases, 14% improve on the initial value where Oi and Ei are the observed and expected frequencies for
and on 21% there is improvement but the regained efficiency is bin i, respectively, a standard criteria to chose the number of
lower than the initial value. This means that it is quite an categories was used, k¼8 and the only free parameter is l (l¼ 1).
attractive alternative to repair damaged HPGe detectors. There At a confidence of 95% the test statistic approximated value is
stands a chance of more than 3/4 (79%) that the repaired detector w26 ¼ 7:48 with a p-value of 0.2787. HO is not rejected as the p-value
will be at least as good as it was originally. In particular we can is rather high: p-value¼0.278740.05¼ a.
observe that, after these 18 years of operation, 33 of the 51 It is worth noting that the Pearson GOF test is not as powerful as
detectors have the same resolution of the original one reported by the Anderson–Darling GOF test [18] and while the data supports
the manufacturer, 11 detectors show slightly less (4%) resolution the null hypothesis for the Pearson GOF, it was rejected when
and the remaining 7 detectors improved by 2%. Of course, we can applied to the Anderson–Darling GOF. We consider that the rejec-
not be sure of the influence of the preamplifiers substitution tion motive for the more powerful test is caused by the handling of
(described in Section 3.4) in these figures. the date logging (resolution of the t parameter: mm/yy). This is
evidenced in the P–P probability plot [19] shown in Fig. 7, where it
4.2.3. Break-down and life-span expectancy can be noted a ‘‘grouping’’ nature of the data, with groups appearing
Detectors malfunction or get damaged and it would be useful every 30 days (1 month). This artificial grouping is a consequence of
to be able to know in advance with what probability a set of having the date recorded just monthly although each failure has the
detectors will fail during a given interval of time. It is well known same chance to occur in any day of the month. This problem could
that electronic failures follow an exponential distribution, how- be avoided by adding to the data the day in which the detector
ever the failures in a detector can be either electronic or non- failing have occurred, so it is advisable to fill a more complete
electronic. From the available data on detector’s life span, we built control sheet that includes the exact failure and repair dates.
a histogram of the total working days for the detectors (see Fig. 6), Another way to approach this lack of information is by introducing
that results in an apparent exponential distribution (at first this an artifice, in order to undo this apparent 30 days grouping.
would not seem obvious), which provides insight of the combined Specifically, some kind of noise could be introduced to spread the
effect of electronic and non-electronic failures. The life-span data, which can be an interesting subject for a further research.
expectancy can be approximated by the most likelihood estimate Based on the previous results, simulations can be done (over
for the exponential density l  1e(  t/l) with l ¼532 days (shown the failure period) as well as predictions on the amount of
also in Fig. 6). This brings all the statistical knowledge and proofs detectors needed to fulfil a given experiment, among other
from the exponential density, mainly the advantage to know that statistical estimations. As an illustrative example, Table 8 shows
it is a memory-less process, which means that assuming ageing results of a specific simulation run and data for comparison
independence, a previous failure does not influence the upcoming purposes, while Fig. 8 shows the corresponding histograms.
one. This has very deep implications in the management of large
scale projects (GASP, EUROBALL, CLARA, etc.).
The exponential approximation was validated with the following 5. Discussion and conclusions
hypothesis test and Pearson’s Chi Square Goodness of Fit statistic:
( ) The long time behaviour of a set consisting in 51 n-type HPGe
HO ¼ Sample data come from the ExpðlÞ distribution
detectors was analysed. These devices have been employed more
Ha ¼ Sample data does not come from the ExpðlÞ distribution than satisfactorily over almost two decades, but they suffered from
X ðOi Ei Þ2 degrading with time either due to radiation induced lattice defects,
w2n ¼ , n ¼ kl1 vacuum loss, electronic component failure or bad handling. However,
Ei
applying a scheduled maintenance programme and when necessary,

Histogram of Detector Failure (rate=532 days)

Detector’s life–span P–P Probability Plot

99.63

87.28
0.0015

53.12

24.29
Percent
Density

0.0010
9.82

3.69

0.0005 1.31

0.56

0.0000

0 500 1000 1500 2000 2500 3000 3500 3 7 20 55 148 403 1097 2981
Days days

Fig. 6. Histogram and density approximation for the failure time (or life span) of Fig. 7. P–P Probability Plot for the life span of the detectors, assuming that the
the HPGe detectors. The density is approximated by MLE to an exponential with corresponding density function is Exp(l ¼532 days). The grouping nature is
rate equal to 1/532¼ 0.00188. attributed to the monthly logging of faulty detectors.
 Author's personal copy

138 L. Sajo-Bohus et al. / Nuclear Instruments and Methods in Physics Research A 648 (2011) 132–138

Table 8 density l  1e  tl with l ¼532 days. Therefore, this density function

Simulation run over 100,000 samples. Using l ¼532 days. can be used to predict detectors failing times and allow estimating
the quantity of detectors needed for long experiments, e.g. for in-
Element Experimental Data (days) Simulation Run (days)
beam spectroscopy. In most cases the repair time required is less
Min 2 0 than a month and often the g spectrometer could be back in
1qt 183 153 operation after only one day. Another important observation is that
Median 401 369 under the assumption that the correct density is exponential;
Mean 532 533
3qt 730 739
a break-down does not influence the next one, suggesting high
Max 3500 5850 reliability on repaired spectrometers. Laboratory technicians super-
vising a large number of g spectrometers can organise operational
shifts efficiently, based on expected troubleshoots type (Table 5),
Histogram of a Specific Simulation Run the frequency of their kind (Table 3) and the out-of-operation time
on Detector failing Times (rate=532 days) (Table 4). These data provide the required information on the
0.0012 quantity and type of replacements that should be available for a
given set of g spectrometers in order to assure that a scheduled
experiment can be executed satisfactorily. When large sets of
spectrometers are in operation, it would be convenient to collect,
additionally to the data presented here, the exact dates of: produc-
0.0008 tion, detector failure, sending to repair and reincorporation, as well
as the energy efficiency values after each repair. It is important not
Density

only to guaranty the quality and reliability of the measurements,

but also could help to better understand the high-resolution g
spectrometers behaviour. It seems that in the future, HPGe will still
0.0004 be the basic detector material, remaining the preferred choice
where high-resolution g spectroscopy is required. We hope that
the results of this study provide useful information for the com-
munity of next generation spectroscopists.

0.0000 References
0 1000 2000 3000 4000 5000 6000
[1] J. Eberth, J. Simpson, Prog. Part. Nucl. Phys. 60 (2) (2007) 283.
Days [2] K. Vetter, Annu. Rev. Nucl. Part. Sci. 57 (2007) 363.
[3] J.F. Sharpey-Schafer, J. Simpson, Prog. Part. Nucl. Phys. 21 (1988) 293.
Fig. 8. Histogram of a simulation run on detector life span (failing times). The [4] F. Albernhe, V. Borrel, P. Frabel, G. Vedrenne, R. Coszach, J.-M. Denis, P. Leleux,
white histogram is obtained from the simulation run while the light grey one is Detectors Assoc. Equip. 492 (1–2) (2002) 91.
from the experimental life span data shown for comparison purposes. [5] LNL Annual Report, INFN,Laboratorio Nazionale di Legnaro, Italy, 1992–2008.
[6] H.G. Thomas, J. Eberth, F. Becker, T. Burkardt, S. Freund, U. Hermkens,
T. Mylaeus, S. Skoda, W. Teichert, v.A D. Werth, P. von Brentano, M. Berst,
applying major repair or modifications, they could be back in D. Gutknecht, R. Henck, Nucl. Instr. and Meth. A 332 (1-2) (1993) 215.
operation satisfactorily. Experience shows that often detector char- [7] GASP Collaboration Report, INFN/BE-90/11, 1990; D. Bazzacco, in: Proceed-
acteristic could be fully recovered or even improved. In fact, our ings of the Workshop on Large g-ray Detector Arrays, Chalk River, Canada,
AECL-10613, 1992, p. 376.
study points out that there is a 79% chance that the restored detector
[8] F. Azaiez, Nucl. Phys. A 654 (1999) 1003c.
has at least the same energy resolution as that specified by the [9] F. Azaiez, G. deFrance, R. Julin, P.J. Nolan, B.M. Nyakó, P.M. Walker, EXOGAM
manufacturer in spite of a long period of activity. Under the Collaboration, J. Simpson, J. Fouan, W. Korten, G. Sletten, J. Gerl, Acta Phys.
Hung. A 11 (1-2) (2000) 159.
assumption of normality, the mean value of detector’s counting
[10] M.P. Carpenter, T.L. Khoo, I. Ahmad, R.V.F. Janssens, T. Lauritsen, G.A. Annan,
efficiency (eCLNL) differs from the manufacturer certified values (eCm), A.M. Baxter, M.E. Bleich, S. Harfenist, E.F. Moore, M. Torelli, D.W. Visser, Nucl.
with a confidence of 90%. Instr. and Meth. A353 (1994) 234.
Detector restoration such as annealing techniques do not [11] G. Duchene, F.A. Beck, P.J. Twin, G. de France, D. Curien, L. Han,
C.W. Beausang, M.A. Bentley, P.J. Nolan, J. Simpson, Nucl. Instr. and Meth.
influence life span (time interval between failures) significantly A432 (1999) 90.
being an effective practise to regain almost its original operational [12] J. Eberth, H.G. Thomas, P. von Brentano, R.M. Lieder, H.M. Jäger,
performance. One of the most interesting results can be deduced H. Kämmerling, M. Berst, D. Gutknecht, R. Henck, Nucl. Instr. and Meth.
A369 (1996) 135.
from Fig. 2, in which manufacturer’s efficiencies values and those [13] A.P. Kleemenko, One flow cascade cycle, in: Proceedings of the X International
measured after the last annealing are compared. A good agreement Congress on Refrigeration, Copenhagen. Pergamon Press, 1959, pp. 34–39.
can be observed, suggesting that (at least for n-type detectors) on [14] D.L. Upp, R.M. Keyser, T.R. Twomey, J Radioanal. Nucl. Chem. 260 (2005) 121.
[15] R.M. Keyser, R.C. Hagenauer, J. Radioanal. Nucl. Chem. 277 (2008) 149.
the overall average spectrometers maintain their performance over [16] P. Leleux, F. Albernhe, V. Borrel, B. Cordier, R. Coszach, S. Crespin, J.M. Denis,
long operating times. Unfortunately we do not have any significant P. Duhamel, P. Frabel, W. Galster, J.-S. Graulich, P. Jean, B. Kandel,
experience with p-type detectors and it is expected that other J.P. Meulders, G. Tauzin, J. Vanhorenbeeck, G. Vedrenne, P. von Ballmoos,
Astron. Astrophys. 411 (2003) L85.
laboratories may provide data for comparison. No significant
[17] R Development Core Team, R: a language and environment for statistical
difference was found between the average long term behaviour computing, R Foundation for Statistical Computing, Vienna, Austria
among detectors produced by the different manufacturers. 3-900051-07-0, 2009 /http://www.R-project.orgS.
Average number of out-of-operation days for an individual [18] Ralph B. D’Agostino, Michael A. Stephens, Goodness-of-fit-techniques, 1st ed,
Statistics: a Series of Textbooks and Monographs, vol. 68, Dekker. ISBN:10
detector is estimated to be 23 days. The overall detector failing 0824774876.
time has an exponential distribution with an approximate (MLE) [19] F.F. Gan, K.J. Koehler, Technometrics 32 (3) (1990) 289 (July 1990).

View publication stats