A Condition Monitoring Concept Studied at the MST Prototype for the Cherenkov Telescope Array
Victor Barbosa Martins, Markus Garczarczyk, Gerrit Spengler, Ullrich Schwanke
AA Condition Monitoring Concept Studied at the MSTPrototype for the Cherenkov Telescope Array
Victor Barbosa Martins ∗ , a Markus Garczarczyk, a Gerrit Spengler, b and UllrichSchwanke b for the MST-STR project of the CTA consortium a Deutsches Elektronen-Synchrotron (DESY)Platanenallee 6, D-15738 Zeuthen, Germany b Humboldt-UniversitÃd’t zu BerlinNewtonstr. 15, D-12489 Berlin, Germany
The Cherenkov Telescope Array (CTA) is a future ground-based gamma-ray observatory that willprovide unprecedented sensitivity and angular resolution for the detection of gamma rays with en-ergies above a few tens of GeV. In comparison to existing instruments (like H.E.S.S., MAGIC, andVERITAS) the sensitivity will be improved by installing two extended arrays of telescopes in thenorthern and southern hemisphere, respectively. A large number of planned telescopes (>100 intotal) motivates the application of predictive maintenance techniques to the individual telescopes.A constant and automatic condition monitoring of the mechanical telescope structure and of thedrive system (motors, gears) is considered for this purpose. The condition monitoring systemaims at detecting degradations well before critical errors occur; it should help to ensure long-termoperation and to reduce the maintenance efforts of the observatory. We present approaches for thecondition monitoring of the structure and the drive system of Medium-Sized Telescopes (MSTs),respectively. The overall concept has been developed and tested at the MST prototype for CTA inBerlin. The sensors used, the joint data acquisition system, possible analysis methods (like Oper-ational Modal Analysis, OMA, and Experimental Modal Analysis, EMA) and first performanceresults are discussed. ∗ Speaker. c (cid:13) Copyright owned by the author(s) under the terms of the Creative CommonsAttribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). http://pos.sissa.it/ a r X i v : . [ a s t r o - ph . I M ] A ug ondition Monitoring at the MST Prototype for CTA Victor Barbosa Martins
1. Introduction
The Cherenkov Telescope Array (CTA) will be the next-generation gamma-ray observatory,capable of measuring gamma-rays from astrophysical sources with energies ranging from 20 GeVup to more than 300 TeV. [1] The observatory is composed of Imaging Atmospheric CherenkovTelescopes (IACTs), which measure the nano-second pulses of Cherenkov radiation emitted by thesecondary particles of the gamma-ray shower.[4] This detection technique is already well estab-lished by current gamma-ray observatories H.E.S.S., MAGIC, and VERITAS.[2] CTA will providea larger energy range and sensitivity in comparison to the current observatories.[2, 3] To cover thewhole energy range, a design with different sizes and distances between telescopes was defined fortwo sites, one in the south (European Southern Observatory (ESO) Paranal site, Chile) and one inthe north (Instituto de Astrofisica de Canarias (IAC) Roque de Los Muchachos Observatory site inLa Palma, Spain). The design foresees the construction of four large-sized telescopes (LSTs) forthe lowest energies, 40 Medium-Sized Telescopes (MSTs) for the energy range from 100 GeV to10 TeV and 70 small-sized telescopes (SSTs) to cover the highest energies.
Figure 1:
Illustrative description of the MST design with the main components.
The MST is based on a modified Davies-Cotton design with a reflector diameter of 12 m anda focal length of 16 m as shown in Fig. 1.[5] A prototype was developed, built and is under testin Berlin since 2013. The requirements for the optical, electrical and mechanical components ofthe MSTs must be fulfilled throughout the 30 years of operation of the observatory. Due to thecontinuous operation of the telescope, effects such as abrasion, material fatigue, and environmental1 ondition Monitoring at the MST Prototype for CTA
Victor Barbosa Martins conditions may affect its performance and reliability. Periodic maintenance in remote sites suchas the Roque de Los Muchachos and Paranal is impracticable because of the large number oftelescopes and the large area covered by the array. Therefore, a concept for the monitoring of thetelescope condition was developed for the MST prototype.The Condition Monitoring System (CMS) aims at detecting small changes in the behavior ofthe telescope, identifying trends in the monitored data and automatically warning the local crewwhen actions should be taken to avoid the worsening of the telescope performance and preventmajor failures to occur. The CMS is divided into Structure Health Monitoring (SHM), detailed inSection 2 and drive system monitoring (DMS), detailed in Section 3. The acquisition system iscentralized for both systems and is, therefore, described in Section 4.
2. Structure Health Monitoring (SHM)
The goal of the SHM is to detect any change in the structure of the telescope. Every structurewhich is excited by an external force corresponds to a resonating system and can be, therefore,characterized by its modal frequencies, mode shapes, and damping. A change in the geometry,material or stiffness leads, consequently, to a change in these three features. The overall vibrationpattern of the structure is a composition of its individual mode shapes, modal frequencies, anddamping. Two methods are very well known and applied nowadays in the industry to derive themodal information from accelerometers datasets: the Experimental Modal Analysis (EMA) and theOperation Modal Analysis (OMA).The prerequisite to use the EMA is to know exactly the spectrum of the input force acting onthe structure and the output force. The frequency response function (FRF) is then calculated bydividing the measured function by the input function. This method is used, for example, in theautomotive and aeronautic industry for validation of the manufactured mechanical structures. Theinput force is usually provided by electronic hammers or electronic shakers. During the experiment,the structure must be first isolated, for example by hanging it, then excited, and measured in manydifferent positions to assure that all the modes are excited and detected.When the input force is unknown, the OMA must be applied.[6] In this method, the input forceis assumed to be a white noise i.e. a broad spectrum excitation. The only way to assure that themethod works is to assure: 1) the input spectrum is broad, 2) the force must be applied all over thestructure, and 3) the sensors must also be distributed throughout the whole structure. The OMAis applied for large structures, such as bridges, buildings and now telescopes, where it becomesimpossible to isolate the structure from external forces, for example, seismic vibration and windexcitation.In addition to the number and position of the sensors on the structure, the sampling ratio andtotal acquisition time must be defined according to the frequency range of interest: f s ≥ . f max , (2.1a) t total = / f min , (2.1b)where f s is the sampling frequency, f max is the maximum frequency of interest, t total is the totalacquisition time for one dataset, f min is the minimum frequency of interest.2 ondition Monitoring at the MST Prototype for CTA Victor Barbosa Martins
From Finite Element Method (FEM) simulation, it was found that the first modal frequenciesof the telescope are between 0 and 10 Hz. According to this frequency range and Eq. 2.1, theminimum sampling ratio f s was defined as 100 Hz and the total acquisition time about 30 minutes.The data must be acquired with the telescope standing still to avoid any narrowband excitationfrom the motors of the drive system, which would contradict the OMA assumptions. Besides, thetelescope is actually a different structure with different modal parameters at every azimuth andelevation angle configuration. Therefore, a specific configuration (azimuth and elevation equal tozero) was defined to be set up before data acquisition.After the data acquisition, the OMA is based on the following procedure:[6] detrending, CrossSpectral Density (CSD) calculation, decimation, and Singular Value Decomposition (SVD). Duringthe decimation, every trend on the time series data is eliminated and the mean set to zero. Next,the CSD is calculated between every two measuring channels. Every three channels correspondto the three cartesian coordinates of one sensor. During the decimation, the sampling frequency isdecimated from the Nyquist frequency f Ny = f s / =
50 Hz to the frequency range of interest of 10Hz. This procedure (oversampling + decimation) helps to reduce the noise level in the time series.The SVD is an algebraic decomposition applicable to every matrix to extract the eigenfrequenciesand eigenmodes of the system. The CSD matrix for each frequency bin is decomposed in threematrices according to Eq. 2.2:
CSD f = U . S . V , (2.2)where U is a matrix n x n , n being the number of channels, which contains the information aboutthe mode shapes; S is a n x n diagonal matrix, which contains the singular values of the system indecreasing order; and V is usually U ∗ . If the analyzed frequency is a modal frequency, the firstvalue in the diagonal matrix s will be much larger than the next values. In this case, the frequencyis a modal frequency and the corresponding modal shape is the first column of the U matrix. Figure2 shows the result of a test made with only one 1-axis force balance sensor (see Section 4) placedon the camera frame. Since there is only one measuring channel in the test, the CSD, U, S andV matrixes are all unidimensional. The analysis code was developed by the author in Python andsuccessfully compared to the result from the commercial software Artemis Modal (Svibs). [8]If more sensors were used there would be more curves for the higher degrees of freedom i.e.other singular values in Figure 2 (for details see [6]). A peak detection technique is then appliedto the first singular values to extract the potential modal frequencies of the system. After that,a validation method is applied to the selected peaks, the Modal Assurance Criteria (MAC). Themethod evaluates for each pair of peaks how linearly independent their mode shapes are. If theMAC value for two peaks is higher than a defined threshold (usually >0.85) the peaks correspondto the same mode shape, therefore one of them must be discarded from the selection. The maximumnumber of linearly independent peaks is equal to the number of channels used in the measurement(degrees of freedom). In this test, for one channel, we could estimate only one modal frequencyand mode shape, because all the other peaks would be linearly dependent on the first one.The next step is to estimate the damping ratio for each modal frequency. For each peak, aMAC value is calculated between the peak frequency and the frequency bins nearby it. Wheneverthe MAC value is larger than the threshold, the frequency bins correspond to the same mode shape.3 ondition Monitoring at the MST Prototype for CTA Victor Barbosa Martins
Figure 2:
Singular value results from a test of the OMA method using only one sensor on the structure. Thepeaks indicate potential modal frequencies selected by a peak detection technique.
All the other frequencies should be set to zero. The resulting bell shape curve is then transformedback to the time domain by an Inverse Fourier Transform and the damping is estimated from thedecay curve. [7]The monitoring system must be basically based on these three features: modal frequencies,mode shapes, and damping ratio. There are different possibilities to monitor them in time: varia-tion in the modal frequencies, or in the difference between two modal frequencies, and variationin the damping ratio. It is also possible to estimate and monitor the distance variation betweentwo sensors, to investigate the Operation Deflection Shape (ODS), which shows how the structurevibrates in each modal frequency, and to use methods for damage detection.Many of these analysis methods are offered by commercial softwares such as Artemis Modalfrom Svibs.[8] Despite these available solutions in the market, none of them offers the automatiza-tion needed for the constant monitoring of 40 telescopes and an effective warning system.
3. Drive Monitoring System (DMS)
The monitoring of the drive system is a complementary approach to study the status of thetelescope. While the SHM studies the telescope when it is standing still, the DMS studies thetelescope when it is moving. Any rotating machine vibrates in specific frequencies, which dependson the geometry of the motor and gears and on the rotation speed. The vibration is transferred to the4 ondition Monitoring at the MST Prototype for CTA
Victor Barbosa Martins housing and can be also measured by accelerometers. The time series data from the accelerometersis analyzed and the excitation frequencies of the motors can be identified.The goal of the DMS is to monitor these frequencies in time to identify trends and changes ofthe identified frequencies and of the noise level. Severe damages such as total failure of a motorcould end up making the telescope unable to park in after a night shift, possibly pointing to thesun, burning the camera and permanently disabling the telescope. It is expected that severe failurescome as a consequence of small failures. Damages such as wear, free play, and broken drive teethcould be detected through this method. Since all types of damage in rotating machines result inimpact impulses, an increase of the tail of distribution in the spectrum diagram would also indicatean increase in the number of impacts. An automatic warning system will be developed to inform thelocal crew when some change on the noise level or on the identified frequencies exceeds a thresholdlevel. After that, a maintenance service would be planned to check the status of the motors and theycould ultimately be replaced by new motors if needed.The number of sensors and their positions play an important role in the quality of the data.The sensors must be placed in regions where the vibration is larger, which does not always meandirectly on the motor. Experimental tests on the prototype are ongoing. A reproducible procedurefor data taking was defined to cover the whole range of possible azimuth and elevation angles andoptimized such that the telescope reaches maximum speed at some point during the run.Automatic data taking and offline analysis are already running at the prototype telescope inBerlin. Studies on the sensitivity of the sensors are ongoing. Other approaches to investigatethe status of the drive are also being explored such as the monitoring of the current, torque andtemperature in the motors.
4. Data acquisition
The health of the telescope structure is monitored through accelerometers distributed through-out the Camera Support Structure (CSS), which measure its vibration. Different designs for thenumbers, distribution and types of sensors were tested during the last years at the prototype. Theforce balance sensors were found to be the best choice for the SHM due to their high sensitivityand dynamic range. They are able to measure up to ng vibrations (1 g = m / s ) although thefrequency range is limited to some hundreds Hz. This limitation is not prejudicial for the measure-ment of large structures, once in these cases the modal frequencies are usually of the order of someHz or even in the sub-hertz regime. These sensors are vastly used in the industry, for instance inthe monitoring of bridges, buildings and also the control of seismic vibration.On the other hand, the sensors used in the DMS must not be as sensitive as the SHM onesbut they must be able to sample at kHz rates to monitor the rotation of the motors. Piezoelec-tric accelerometers usually fulfill this requirement and are used in the prototype for tests: twoaccelerometers in each of the two azimuth drives (motor and gearbox), and four in each of the twoelevation drives. The measuring range of such sensors are of hundreds of g and the frequency rangereaches tens of kHz. Figure 3 shows on the left four of the DMS sensors and on the right one SHMtest sensor with its voltage/current converter.The data acquired by the sensors are converted from analog to digital and processed by mod-ular devices from the company Gantner Instruments.[9] The devices are composed of one pro-5 ondition Monitoring at the MST Prototype for CTA Victor Barbosa Martins
Figure 3:
Sensors for the condition monitoring system: four accelerometers on the left for the elevationdrive monitoring and one sensor on the right for the structure monitoring on the camera frame.
Figure 4:
Picture of the acquisition system from Gantner Instruments.[9] The 6 modules and the Q.stationare visible on the upper part of the image from left to right. grammable and self-sustained operating controller Q.station 101, four Q.bloxx A108 modules with24-bit analog input each, 8 differential channels and 10 kHz sample rate at each channel, and twomodules Q.bloxx A111 with four analog inputs each and a sample rate of 100 kHz per channel.The A108 modules are used to process the data from the SHM sensors and the A111 to process thedata from the DMS sensors. The controller has an internal hard disk with a storage capacity of onegigabyte for logger functions and offers data transfer via Ethernet. The data is pushed to an FTPserver and is then stored in a database for further offline analysis. Part of the analysis runs alreadyautomatically in a local machine. Figure 4 shows the acquisition system, which is housed inside6 ondition Monitoring at the MST Prototype for CTA
Victor Barbosa Martins the telescope tower.
5. Conclusions and prospects
The monitoring system for the MSTs is essential to assure that the telescopes are able to fulfillthe requirements throughout the whole lifetime of the observatory. By monitoring the structure(SHM) and the drive system (DMS), a complete picture of the status of every MST can be obtainedon a daily basis. The analysis is developed based on methods already used in the industry andadapted to our purposes. Besides, the automatic analysis and warning system will spare the needof manpower for the data analysis. Although the results achieved so far are promising, the setup isstill experimental and should only reach its final version by the end of 2019. Further developmentof the analysis code is ongoing and damage tests are also planned for the end of 2019.
References [1] The CTA Consortium, represented by Rene A. Ong,
Cherenkov Telescope Array: The NextGeneration Gamma-ray Observatory , in proceedings of , PoS(ICRC2017)1071 [2] J. Holder
TeV gamma-ray astronomy: A summary , Astroparticle Physics (2012 61-75) [
DOI10.1016/j.astropartphys.2012.02.014 ].[3] The CTA Consortium
Design concepts for the Cherenkov Telescope Array CTA: an advanced facilityfor ground-based high-energy gamma-ray astronomy , Exp Astron (2011 32:193-316) [ DOI10.1007/s10686-011-9247-0 ].[4] A.M. Hillas
Evolution of ground-based gamma-ray astronomy from the early days to the CherenkovTelescope Arrays , Astroparticle Physics (2013 19-43) [ DOI10.1016/j.astropartphys.2012.06.002 ].[5] M. Garczarczyk et al. the CTA Consortium
Status of the Medium-Sized Telescope for the CherenkovTelescope Array , in proceedings of , PoS(ICRC2015)959 [6] R. Brincker, L. Zhang and P. Andersen
Modal identification of output-only systems usingfrequencydomain decomposition , Smart Mater. Struct. (2001 441-445) [ DOI10.1088/0964-1726/10/3/303 ].[7] L. Zhang, Y. Tamura
Damping estimation by Frequency Domain Decomposition , Proceedings ofIMAC 19 : A Conference on Structural Dynamics : februar 5-8, 2001, Hyatt Orlando, Kissimmee,Florida, 2001 (2001 698-703)[8] [9]