20th European Photovoltaic Solar Energy Conference and Exhibition, Barcelona, Spain, 2005



                         SunDogTM STCU: A GENERIC SUN TRACKING CONTROL UNIT
                                   FOR CONCENTRATION TECHNOLOGIES

                          I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
                                    Inspira, SL, C/Chile, 10, Las Matas, 28290 Madrid.
                                                  email: iluque@inspira.es


   ABSTRACT: The present development status of the SunDogTM sun tracking control unit is here described.
   SunDogTM is an electronic controller devised to deliver the stringent accuracies and reliabilities required by high
   concentration technologies, a present offshoot of the world’s most dynamic PV markets, offering Solar a route to the
   conquest of wholesale electricity prices. Profiting from the high processing power and the low costs attained by
   today’s ubiquitous embedded systems, SunDogTM STCU is based in a sun ephemeris computation core implemented
   in a low-end microcontroller. The effect of the characteristic error sources which cut down the tracking accuracy of
   this “open-loop” approach is greatly suppressed by the inclusion of a generic error model, valid for any configuration
   of tracking axes, on which SunDogTM performs automated calibration at onset. The advantages of this computerized
   controller are further exploited providing efficient handling of management and emergency situations, versatile man-
   machine interface, and extended connectivity resources enabling Internet monitoring.
   Keywords: Concentrators, Tracking, Strategy

1. INTRODUCTION                                                   fitting of the parameters of the model at onset with some
                                                                  set of performance errors, automatically collected,
    Field experience gathered up to know with                     characterizes the error sources intervening and can be
photovoltaic concentrators consistently points out closed-        used from then on to correct the output of the preliminary
loop, sensor based, tracking controllers, as not being up         ephemeris outputs, restoring in such a way the open-loop
to the severe reliability standards required by the low           tracking scheme.
maintenance plants, which sustain the economic
feasibility plans of these technologies [1].                      2. THE TRACKING ERROR MODEL
    Open-loop controllers, based in the computation of
sun ephemeris, implemented as digital embedded                         Leaving aside the JPL style of numerical
systems, appear as a reliable and almost maintenance-             ephemeredes based on interpolation of tabulated
free alternative, able to attain the increasingly high            numerically integrated solutions of the equations of
accuracies demanded by the rising trends in                       celestial motion, which is the most accurate approach
concentration ratio, and offering the possibility of              available to date but requiring significant data storage
including added value content, such as remote control             and updating, relatively simple approximate analytic
and monitoring capabilities, at reasonable costs when             solutions can be also be obtained of these equations,
getting into industrial volumes.                                  which in the case of the sun, attain accuracies in the 0.1
    However an open-loop controller even if operating             arc minutes range, which is a firm enough basis for CPV
on the very precise sun ephemeris equations available to          tracking control. Sun ephemeredes for sun tracking
date, is affected, once connected in the field to its             purposes are usually adjusted to output topocentric
appointed concentrator, by many error sources which can           horizontal coordinates (azimuth & elevation), which are
highly degrade its final tracking accuracy well below its         then to be converted into rotation angles measured from
ephemeris’ nominal value, and to the point of even                a certain reference orientation in the one or two axes of
missing concentrator’s specifications.                            the tracker under control. However when converting to
    Among these error sources the most significant have           these coordinates into axes turns, default assumptions are
a deterministic nature and mainly result from                     to be made regarding the in-field arrangement of the
characterization inaccuracies, i.e. tolerance deviations          tracking axes and their reference orientations. These
appearing in the controller’s default assumptions                 assumptions even if specified to the tracker and
regarding the tracker’s installation or manufacturing.            concentrator manufacturer and to the installation crews
Drifts in the internal timing required for the computation        will be met within tolerance margins, and these
of the sun ephemeris, is the other critical error source          deviations will imply a significant decrease in the final
which is to be restrained. Second order error sources, and        accuracy of the ephemeris guided aiming. Process
also to some extent predictable, such as gravitational            tolerance shrinking is always a costly effort which may
bending in wide aperture trackers, the effect of mismatch         quickly override the tight cost constraints usually
in multi-secondary axis trackers, or even ephemeris               imposed on CPV technologies. So in order to relax
inaccuracies, due to the effect of local atmospheric              manufacturing and installation requirements, while
refraction, might have to be considered as well.                  limiting the tracking accuracy losses, a set up calibration
    Feedback of the tracking errors caused by the                 approach can be developed, in which a mathematical
referred sources is to be somehow integrated, in the              model characterizing the sources of departure from the
control strategy in order to suppress them. This open loop        conversion assumptions can be used, and its parameters
core strategy blended with the fed back closed-loop, is           fitted to a set of tracking errors collected at onset. Once
sometimes referred as the hybrid approach, which in past          the parameters quantifying the share of the different error
implementations have been mainly connected to control             sources in the resulting tracking errors, are obtained, the
theory [2]. Instead the method here proposed relies in the        model tuned to those values will be used to correct the
principles of automated instrumentation calibration,              ephemeris output from the on.
based in the modeling of systematic error sources. The                 The calibration model developed here assumes the


                      I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
20th European Photovoltaic Solar Energy Conference and Exhibition, Barcelona, Spain, 2005



tracker’s axes and their reference orientations have the     or again the improper assembly of aperture frame or
same reference system as the horizontal Az./El.              optics.
coordinate system used by the ephemeris, i.e. the                These six parameters appear in a 2 → 2 function,
grounded, or primary axis pointing to the local zenith       consisting in the composition of five partial transforms,
with its reference orientation pointing south, and the       and converting horizontal coordinates provided by the
secondary axis, the one which is fixed to the primary        ephemeris to pairs of angular rotations for both axes.
always remains at right angles with it, and has its          Behavior of this calibration function can be visualized
reference orientation pointing the horizon, in other         through the usual grid transform representation in
words, the ideal pedestal tracker.                           complex variable analysis (Fig.1).
    Six are the parameters characterizing the departure of
the real tracker under control form the ideally assumed:
• Primary axis azimuth ( φ ) and zenith angle ( θ ):
These two parameters are the azimuth and zenith angle
coordinates determining the real orientation of the
primary axis, which no matter the axis configuration is
always defined as the axis which is fixed to the ground.
This is mainly an installation error due to the imprecise
foundation of the tracker                                    Figure 1. A simple calibration transform applied to a
• Primary axis offset ( β ):                                 rectangular grid in the azimuth-elevation plane (rad) into
This parameter determines the location of the reference      the plane of axes angle rotations in a two axis tracker,
orientation of the primary. Reference orientation is         when θ = 15º and the remaining parameters are zero
usually determined by a specific sensor, or the index
mark when working with optical encoders. Misplacement           The fact that the assumed reference system for the
of this sensor during manufacturing or its incorrect         tracker under control is that of an ideal pedestal tracker is
alignment at installation may cause this error. When         just a convention and the model is able to correct
 φ = θ = 0 , β simply becomes the angular offset to the      horizontal ephemeris coordinates to any two axes
south.                                                       configuration, including others frequently used such as
    These first three parameters are in the referred order   the tilt-roll assembly (ideally φ = θ = π 2 ). In order to
the nutation, precession, and spin Euler angles which        maintain this generality of the model no simplifying
relate any two reference systems with a common origin,       assumption have been made regarding the transform
and only these will be required if our tracker has only      parameters.
one axis, the primary axis, such as in present polar,
azimuthal, or horizontal axis trackers. In case a            3. MODEL FITTING
secondary axis is attached to the primary, three more
parameters are required and the pointing vector is               Discarding simplifying linearization which would
introduced as that one which is oriented by the joint        decrease the generality of the model, the parameters
action of the two tracking axis, and if aligned with the     characterizing a specific tracker and its in-field
sun vector produces the maximum power output of the          installation will have to be fitted to a set of tracking error
concentrator array.                                          observations by means of iterative non-linear
• Non-orthogonality of axes ( λ ):                           optimization techniques. Provided the target has been to
This parameter takes the value of the difference to the      integrate this numerical procedure in a low cost
right angle between the primary and secondary axis. This     embedded system based in an 8 bit microcontroller,
mainly being a manufacturing error source, a non zero        programming efficiency was a must, and good
value for this angle implies the two axis tracker is no      conditioning of the maximum likelihood estimation
longer ideal and a cone of orientations around the           (MLE) function is required. Least squares, was the MLE
primary axis will remain out of reach.                       chosen which even if there are other more robust
• Pointing vector axial tilt ( δ ):                          estimators, is by far the one presenting the most effective
The pointing vector is assumed normal to the secondary       non-linear minimization techniques. The existence of
axis and contained in the horizontal plane when this axis    local minima obliged to resort to the global optimization
rotation is zero. The axial tilt of the pointing vector is   toolbox, and finally a clustered multi-start with
the difference angle to a plane normal to the secondary      Levenberg-Marquardt [3],[4] based local searching was
axis. This error can have its origin in the defective        implemented. Robustness is further conferred by means
assembly of the tracker’s aperture frame, but also in the    of a pre-fit outlier filter discarding clearly defective error
misalignment of the concentrator optics.                     observations. In order both to test the reliability of the
• Secondary axis offset ( η ):                               local search stage, once running in the microcontroller, it
The secondary axis offset accounts not only for the          was tested on the Moré set of test functions [5]. This was
difference angle between the plane normal to the primary     also useful to choose among the three compilers available
axis and the reference orientation of the secondary axis,    depending on the accuracies in the fits obtained when
but also for the difference angle between this reference     compared to this same fits in a Intel Pentium 4 processor.
orientation and the plane containing the pointing vector     First attempts on fitting the calibration transform with
and the secondary axis, i.e. a radial tilt which is the      error sets produced by the calibration transform tuned at
second value characterizing pointing vector departure        a prefixed solution, during the solstices and equinox, had
from assumptions. This error therefore derives from both     a very lengthy convergence in the microcontroller due to
the misplacement of the secondary axis reference sensor      the existence of multiple local minima. Careful analyses
                                                             showed most of the local minima were so in a wide sense


                      I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
20th European Photovoltaic Solar Energy Conference and Exhibition, Barcelona, Spain, 2005



and appeared in the frontier of discontinuities. Lacking       to 1º accurate) could be made of each solution (Table I).
physical meaning these discontinuities were isolated,
with the help of the graphical representation of all the               (Case) Module               Best fit estimate of mount
combinations of two dimensional search subspaces (Fig.         (1) “all glass” Fresnel 500X      θ = 2 º , λ = −1º , δ = 2º , η = 45º
2), and progressively smoothed by convenient fixes in                (2) TIR-R 1000X             θ = 2 º , λ = −1º , δ = 32º , η = 45º
the definition of the constituent transforms. This was a              (3) Fresnel 500X            θ = 2 º , λ = −1º , δ = 2 º , η = 5º
painstaking process which however finally achieved the
                                                               Table I: Three test mounts, modules employed, and a
acceleration of the convergence required to make the
                                                               priori estimates of the best fit parameter solution.
process feasible.
    After this revision of the calibration transform
                                                                   A dawn to dusk session collecting tracking errors for
definition the simulated tests, which where performed for
                                                               the three mounts was carried out, were the measurements
several sets of solution parameters, representing different
                                                               time stamp was obtained from a GPS and precise
types of tracker concepts or tolerance degrees in their
                                                               pointing was assumed at maximum short circuit current
production and installation, and with tracking error sets
                                                               output. Ephemeris used for the LS [6] fit had a mean
of a maximum of 100 points, mean error and error
                                                               accuracy of 0.06º when running in the microcontroller.
standard deviation in the fits obtained by the
microcontroller where in the 0.5 prad range and
convergence times in the 1 to 5 minutes range, and only a
tough case which required almost 2 hours.




Figure 2. Levenberg-Marquardt sample local searches
meeting a global (0.17,1.35) and a local minima (3.31,-
1.35) at two dimensional subspace of the search space
(φ, θ ) for LS best fit.
                                                               Figure 3. Trell Tracker at IES-UPM, showing the three
                                                               test mounts with their corresponding modules (Table I).
3. PHYSICAL VALIDATION OF THE MODEL
                                                               Case       φ       θ       β          λ        δ         η       σmin
     The calibration transform was now properly defined,
however physical validation of the model was still               1       24.19   1.84    13.58     -0.69    2.04      43.92      0.19
                                                                 2       39.33   1.70    -2.95     -1.07    33.90     43.71      0.37
required. For this purpose a laboratory two axis pedestal        3       22.38   2.13    14.65     -1.09    1.03       4.44      0.17
tracker was designed. Able to withstand up to 100kg of
                                                               Table II: Best fit solutions for the three cases, and their
payload in its 4m2 aperture, the positioning resolution of
                                                               corresponding Min Std Deviation for model fitness (deg)
its proprietary drives, the high accuracy of its encoder set
up, and its internal sun ephemeris allow for overall
                                                                    As seen in Table II in all three cases the best fit
tracking accuracies in the 0.1º (max. error) range, if
                                                               parameters obtained were within the referred 1º range
ideally installed. The structure of this lab tracker,
                                                               from its initial estimates, which is a first successful
commercially named Trell Tracker, integrates the
                                                               indication of the physical compliancy of the model. In
tracking control in its structure, and presents a user
                                                               addition resorting to LS theory we could estimate the
interface front panel which includes LCD display of the
                                                               minimum standard deviation of supposedly normally
axis turn angles in encoder counts.
     Three different mounts where provided in the              distributed measurement errors σmin in order to further
aperture of this tracker for three different prototype         ensure the model is physically grounded. The three
concentration modules, all of them housing a micro-             σmin estimates shown in Table II agree with the observed
concentrator parquet and with concentration factors            limitations of the measuring process which are
500X and 1000X (Fig. 3). These mountings where                 significantly affected by some outliers introduced by the
prepared so that their best fit parameter solution in the      manual procedure employed. The TIR-R module due to
calibration transform, highlights a certain parameter, and     its highest acceptance angle is the one introducing the
with the aid of a digital inclinometer some estimates (up
                                                               highest σmin estimate errors in the error observations.


                       I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
20th European Photovoltaic Solar Energy Conference and Exhibition, Barcelona, Spain, 2005



4. MODEL EXTENSIONS                                              acceptance angle distance, can resort to some wide angle
                                                                 sensor such as an aperture plane irradiance sensor, but
    After model validation further work was produced in          beyond some point the search is to proceed blindly and
order to introduce in our model other second order               produce extensive plane searching routines, i.e. spiral
tracking error sources. A tracker specific module is             search [7]. Thereafter maximization of the power
included which is to host a model of the variation of the        function with tracker’s coordinates is required.
pointing vector with orientation due to gravitational                It is worth noting here that the tracking error set is
bending. Also the effect of manufacturing tolerances in          not necessarily to be obtained from the Sun, in principle
the drive chain gearings when it is not possible to mount        any other light source with precise analytic kinematics,
axis turn sensors directly in the primary axis can be            and with enough emitting power to extract a measurable
introduced in the case of the most common linear drives.         output from the concentrator, would suffice. In this
Multi-secondary axis trackers, i.e. those integrating            respect the full moon proves to be a good candidate, as
several secondary axes mechanically linked in order to           far as it will enable night calibrations not interfering with
decrease aperture wind profile, are also considered in           concentrator’s daily production. Its 0.49º apparent
order to take into account the effect of the electrical          diameter is very similar to the suns’, while its irradiance
mismatch of the CPV arrays in each secondary axis as             is six orders of magnitude smaller than the sun, so its
well as their overall connection scheme, in order to             photo-generated current is still within reach of cheap
locate the pointing vector.                                      current sensing devices. In the other hand the full moon
    Even if Internet connectivity or GPS might provide           irradiance is three orders of magnitude above that of the
atomic time synchronicity to the tracking control unit           most brilliant planets and stars so it will be easily
internal clock, availability of high accuracy ephemeredes        distinguishable by the concentrator when searching the
provides an immediate and autonomous alternative to              night sky. Moon ephemeris and phase equations have
precise time-keeping. A time drift parameter has been            been analyzed and encoded for the development of a
included in the model in such a way that it can be fitted        Moon error collection procedure on which to fit the error
with a tracking error set either, jointly with the rest of the   model.
parameters or individually within periodic time                      Along with these performance enhancements, long
adjustment procedures.                                           term accuracy tests are being conducted on several
    The complexity of the calibration transform when             SunDog based systems.
introducing these new parameters, obliged to switch in
the Levenberg-Marquardt module from working on
analytic derivatives to its finite-difference counterpart,
which anyhow does not significantly decrease either
accuracy or convergence time.

7. PHYSICAL IMPLEMENTATION

    The above described capacities have been
implemented in an electronic embedded system, based on
an 8 bit microcontroller, along with the required chipset
and sensors to carry out the described algorithms and
perform the analog measurements, and also motor
driving units and an encoder decoding and interpolation
subsystem. Named SunDogTM for commercial purposes,
it is supplied with SunDog Monitor a Windows                     Figure 4. SunDogTM STCU (left) along with its SunDog
application to run in a locally or remotely connected PC         Monitor virtual interface software (above right), and
as a virtual user interface. It also integrates an               WWW remote control panel hosted by its internal web
interchangeable modem for PSTN, RF, Ethernet or                  server (below right)
GSM/GPRS Internet connectivity, which enables e-mail
reporting and web based control and monitoring.                  This work has been supported by the European
Prepared for operation in harsh environments its rugged          Commission through the funding of the project
enclosure integrates a simple control front panel (Fig. 4).      FULLSPECTRUM (Ref. N: SES6-CT-2003-502620)

5. ONGOING WORK                                                                        REFERENCES
                                                                 [1] A. Maish, Proc. of the 20thIEEE PVSC, (1988) 1309
    Manual collection of tracking errors is a tiresome and       [2] I. Luque-Heredia et al. Proc. of the 3rd Int. Conf. on
error prone task, which demands automation, in order                 Solar Concentrators, (2005)
prevent outliers creeping in and increase the accuracy of        [3] A.A, Törn, 2nd IFAC Symposium on Stochastic
the corrected ephemeris.                                             Control, 19, (1986)
    Automatic error collection has already been                  [4] D.W. Marquardt, Journal of the Society of Industrial
extensively simulated proving efficient performance. In it           and Applied Mathematics, 11, 2 (1963), 431
direct search of the alignment of pointing and sun vector        [5] J. Moré et al. ACM Transactions on Mathematical
to obtain each tracking error measurement, imposes, to               Software, 7, 1, (1981), 17
some extent, the use of the concentrators power output or        [6] T.C. Van Flandern, et al The Astrophysical Journal
some approximately equivalent variable, such as short                Supplement Series,41,(1979), 391
circuit current, as feedback signal. Preliminary search          [7] R. Baeza-Yates et al, Information and Computation,
until sun and pointing vectors get into the concentrator’s           106, (1993), 234


                        I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães

Sun dog stcu...(circ)

  • 1.
    20th European PhotovoltaicSolar Energy Conference and Exhibition, Barcelona, Spain, 2005 SunDogTM STCU: A GENERIC SUN TRACKING CONTROL UNIT FOR CONCENTRATION TECHNOLOGIES I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães Inspira, SL, C/Chile, 10, Las Matas, 28290 Madrid. email: iluque@inspira.es ABSTRACT: The present development status of the SunDogTM sun tracking control unit is here described. SunDogTM is an electronic controller devised to deliver the stringent accuracies and reliabilities required by high concentration technologies, a present offshoot of the world’s most dynamic PV markets, offering Solar a route to the conquest of wholesale electricity prices. Profiting from the high processing power and the low costs attained by today’s ubiquitous embedded systems, SunDogTM STCU is based in a sun ephemeris computation core implemented in a low-end microcontroller. The effect of the characteristic error sources which cut down the tracking accuracy of this “open-loop” approach is greatly suppressed by the inclusion of a generic error model, valid for any configuration of tracking axes, on which SunDogTM performs automated calibration at onset. The advantages of this computerized controller are further exploited providing efficient handling of management and emergency situations, versatile man- machine interface, and extended connectivity resources enabling Internet monitoring. Keywords: Concentrators, Tracking, Strategy 1. INTRODUCTION fitting of the parameters of the model at onset with some set of performance errors, automatically collected, Field experience gathered up to know with characterizes the error sources intervening and can be photovoltaic concentrators consistently points out closed- used from then on to correct the output of the preliminary loop, sensor based, tracking controllers, as not being up ephemeris outputs, restoring in such a way the open-loop to the severe reliability standards required by the low tracking scheme. maintenance plants, which sustain the economic feasibility plans of these technologies [1]. 2. THE TRACKING ERROR MODEL Open-loop controllers, based in the computation of sun ephemeris, implemented as digital embedded Leaving aside the JPL style of numerical systems, appear as a reliable and almost maintenance- ephemeredes based on interpolation of tabulated free alternative, able to attain the increasingly high numerically integrated solutions of the equations of accuracies demanded by the rising trends in celestial motion, which is the most accurate approach concentration ratio, and offering the possibility of available to date but requiring significant data storage including added value content, such as remote control and updating, relatively simple approximate analytic and monitoring capabilities, at reasonable costs when solutions can be also be obtained of these equations, getting into industrial volumes. which in the case of the sun, attain accuracies in the 0.1 However an open-loop controller even if operating arc minutes range, which is a firm enough basis for CPV on the very precise sun ephemeris equations available to tracking control. Sun ephemeredes for sun tracking date, is affected, once connected in the field to its purposes are usually adjusted to output topocentric appointed concentrator, by many error sources which can horizontal coordinates (azimuth & elevation), which are highly degrade its final tracking accuracy well below its then to be converted into rotation angles measured from ephemeris’ nominal value, and to the point of even a certain reference orientation in the one or two axes of missing concentrator’s specifications. the tracker under control. However when converting to Among these error sources the most significant have these coordinates into axes turns, default assumptions are a deterministic nature and mainly result from to be made regarding the in-field arrangement of the characterization inaccuracies, i.e. tolerance deviations tracking axes and their reference orientations. These appearing in the controller’s default assumptions assumptions even if specified to the tracker and regarding the tracker’s installation or manufacturing. concentrator manufacturer and to the installation crews Drifts in the internal timing required for the computation will be met within tolerance margins, and these of the sun ephemeris, is the other critical error source deviations will imply a significant decrease in the final which is to be restrained. Second order error sources, and accuracy of the ephemeris guided aiming. Process also to some extent predictable, such as gravitational tolerance shrinking is always a costly effort which may bending in wide aperture trackers, the effect of mismatch quickly override the tight cost constraints usually in multi-secondary axis trackers, or even ephemeris imposed on CPV technologies. So in order to relax inaccuracies, due to the effect of local atmospheric manufacturing and installation requirements, while refraction, might have to be considered as well. limiting the tracking accuracy losses, a set up calibration Feedback of the tracking errors caused by the approach can be developed, in which a mathematical referred sources is to be somehow integrated, in the model characterizing the sources of departure from the control strategy in order to suppress them. This open loop conversion assumptions can be used, and its parameters core strategy blended with the fed back closed-loop, is fitted to a set of tracking errors collected at onset. Once sometimes referred as the hybrid approach, which in past the parameters quantifying the share of the different error implementations have been mainly connected to control sources in the resulting tracking errors, are obtained, the theory [2]. Instead the method here proposed relies in the model tuned to those values will be used to correct the principles of automated instrumentation calibration, ephemeris output from the on. based in the modeling of systematic error sources. The The calibration model developed here assumes the I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
  • 2.
    20th European PhotovoltaicSolar Energy Conference and Exhibition, Barcelona, Spain, 2005 tracker’s axes and their reference orientations have the or again the improper assembly of aperture frame or same reference system as the horizontal Az./El. optics. coordinate system used by the ephemeris, i.e. the These six parameters appear in a 2 → 2 function, grounded, or primary axis pointing to the local zenith consisting in the composition of five partial transforms, with its reference orientation pointing south, and the and converting horizontal coordinates provided by the secondary axis, the one which is fixed to the primary ephemeris to pairs of angular rotations for both axes. always remains at right angles with it, and has its Behavior of this calibration function can be visualized reference orientation pointing the horizon, in other through the usual grid transform representation in words, the ideal pedestal tracker. complex variable analysis (Fig.1). Six are the parameters characterizing the departure of the real tracker under control form the ideally assumed: • Primary axis azimuth ( φ ) and zenith angle ( θ ): These two parameters are the azimuth and zenith angle coordinates determining the real orientation of the primary axis, which no matter the axis configuration is always defined as the axis which is fixed to the ground. This is mainly an installation error due to the imprecise foundation of the tracker Figure 1. A simple calibration transform applied to a • Primary axis offset ( β ): rectangular grid in the azimuth-elevation plane (rad) into This parameter determines the location of the reference the plane of axes angle rotations in a two axis tracker, orientation of the primary. Reference orientation is when θ = 15º and the remaining parameters are zero usually determined by a specific sensor, or the index mark when working with optical encoders. Misplacement The fact that the assumed reference system for the of this sensor during manufacturing or its incorrect tracker under control is that of an ideal pedestal tracker is alignment at installation may cause this error. When just a convention and the model is able to correct φ = θ = 0 , β simply becomes the angular offset to the horizontal ephemeris coordinates to any two axes south. configuration, including others frequently used such as These first three parameters are in the referred order the tilt-roll assembly (ideally φ = θ = π 2 ). In order to the nutation, precession, and spin Euler angles which maintain this generality of the model no simplifying relate any two reference systems with a common origin, assumption have been made regarding the transform and only these will be required if our tracker has only parameters. one axis, the primary axis, such as in present polar, azimuthal, or horizontal axis trackers. In case a 3. MODEL FITTING secondary axis is attached to the primary, three more parameters are required and the pointing vector is Discarding simplifying linearization which would introduced as that one which is oriented by the joint decrease the generality of the model, the parameters action of the two tracking axis, and if aligned with the characterizing a specific tracker and its in-field sun vector produces the maximum power output of the installation will have to be fitted to a set of tracking error concentrator array. observations by means of iterative non-linear • Non-orthogonality of axes ( λ ): optimization techniques. Provided the target has been to This parameter takes the value of the difference to the integrate this numerical procedure in a low cost right angle between the primary and secondary axis. This embedded system based in an 8 bit microcontroller, mainly being a manufacturing error source, a non zero programming efficiency was a must, and good value for this angle implies the two axis tracker is no conditioning of the maximum likelihood estimation longer ideal and a cone of orientations around the (MLE) function is required. Least squares, was the MLE primary axis will remain out of reach. chosen which even if there are other more robust • Pointing vector axial tilt ( δ ): estimators, is by far the one presenting the most effective The pointing vector is assumed normal to the secondary non-linear minimization techniques. The existence of axis and contained in the horizontal plane when this axis local minima obliged to resort to the global optimization rotation is zero. The axial tilt of the pointing vector is toolbox, and finally a clustered multi-start with the difference angle to a plane normal to the secondary Levenberg-Marquardt [3],[4] based local searching was axis. This error can have its origin in the defective implemented. Robustness is further conferred by means assembly of the tracker’s aperture frame, but also in the of a pre-fit outlier filter discarding clearly defective error misalignment of the concentrator optics. observations. In order both to test the reliability of the • Secondary axis offset ( η ): local search stage, once running in the microcontroller, it The secondary axis offset accounts not only for the was tested on the Moré set of test functions [5]. This was difference angle between the plane normal to the primary also useful to choose among the three compilers available axis and the reference orientation of the secondary axis, depending on the accuracies in the fits obtained when but also for the difference angle between this reference compared to this same fits in a Intel Pentium 4 processor. orientation and the plane containing the pointing vector First attempts on fitting the calibration transform with and the secondary axis, i.e. a radial tilt which is the error sets produced by the calibration transform tuned at second value characterizing pointing vector departure a prefixed solution, during the solstices and equinox, had from assumptions. This error therefore derives from both a very lengthy convergence in the microcontroller due to the misplacement of the secondary axis reference sensor the existence of multiple local minima. Careful analyses showed most of the local minima were so in a wide sense I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
  • 3.
    20th European PhotovoltaicSolar Energy Conference and Exhibition, Barcelona, Spain, 2005 and appeared in the frontier of discontinuities. Lacking to 1º accurate) could be made of each solution (Table I). physical meaning these discontinuities were isolated, with the help of the graphical representation of all the (Case) Module Best fit estimate of mount combinations of two dimensional search subspaces (Fig. (1) “all glass” Fresnel 500X θ = 2 º , λ = −1º , δ = 2º , η = 45º 2), and progressively smoothed by convenient fixes in (2) TIR-R 1000X θ = 2 º , λ = −1º , δ = 32º , η = 45º the definition of the constituent transforms. This was a (3) Fresnel 500X θ = 2 º , λ = −1º , δ = 2 º , η = 5º painstaking process which however finally achieved the Table I: Three test mounts, modules employed, and a acceleration of the convergence required to make the priori estimates of the best fit parameter solution. process feasible. After this revision of the calibration transform A dawn to dusk session collecting tracking errors for definition the simulated tests, which where performed for the three mounts was carried out, were the measurements several sets of solution parameters, representing different time stamp was obtained from a GPS and precise types of tracker concepts or tolerance degrees in their pointing was assumed at maximum short circuit current production and installation, and with tracking error sets output. Ephemeris used for the LS [6] fit had a mean of a maximum of 100 points, mean error and error accuracy of 0.06º when running in the microcontroller. standard deviation in the fits obtained by the microcontroller where in the 0.5 prad range and convergence times in the 1 to 5 minutes range, and only a tough case which required almost 2 hours. Figure 2. Levenberg-Marquardt sample local searches meeting a global (0.17,1.35) and a local minima (3.31,- 1.35) at two dimensional subspace of the search space (φ, θ ) for LS best fit. Figure 3. Trell Tracker at IES-UPM, showing the three test mounts with their corresponding modules (Table I). 3. PHYSICAL VALIDATION OF THE MODEL Case φ θ β λ δ η σmin The calibration transform was now properly defined, however physical validation of the model was still 1 24.19 1.84 13.58 -0.69 2.04 43.92 0.19 2 39.33 1.70 -2.95 -1.07 33.90 43.71 0.37 required. For this purpose a laboratory two axis pedestal 3 22.38 2.13 14.65 -1.09 1.03 4.44 0.17 tracker was designed. Able to withstand up to 100kg of Table II: Best fit solutions for the three cases, and their payload in its 4m2 aperture, the positioning resolution of corresponding Min Std Deviation for model fitness (deg) its proprietary drives, the high accuracy of its encoder set up, and its internal sun ephemeris allow for overall As seen in Table II in all three cases the best fit tracking accuracies in the 0.1º (max. error) range, if parameters obtained were within the referred 1º range ideally installed. The structure of this lab tracker, from its initial estimates, which is a first successful commercially named Trell Tracker, integrates the indication of the physical compliancy of the model. In tracking control in its structure, and presents a user addition resorting to LS theory we could estimate the interface front panel which includes LCD display of the minimum standard deviation of supposedly normally axis turn angles in encoder counts. Three different mounts where provided in the distributed measurement errors σmin in order to further aperture of this tracker for three different prototype ensure the model is physically grounded. The three concentration modules, all of them housing a micro- σmin estimates shown in Table II agree with the observed concentrator parquet and with concentration factors limitations of the measuring process which are 500X and 1000X (Fig. 3). These mountings where significantly affected by some outliers introduced by the prepared so that their best fit parameter solution in the manual procedure employed. The TIR-R module due to calibration transform, highlights a certain parameter, and its highest acceptance angle is the one introducing the with the aid of a digital inclinometer some estimates (up highest σmin estimate errors in the error observations. I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães
  • 4.
    20th European PhotovoltaicSolar Energy Conference and Exhibition, Barcelona, Spain, 2005 4. MODEL EXTENSIONS acceptance angle distance, can resort to some wide angle sensor such as an aperture plane irradiance sensor, but After model validation further work was produced in beyond some point the search is to proceed blindly and order to introduce in our model other second order produce extensive plane searching routines, i.e. spiral tracking error sources. A tracker specific module is search [7]. Thereafter maximization of the power included which is to host a model of the variation of the function with tracker’s coordinates is required. pointing vector with orientation due to gravitational It is worth noting here that the tracking error set is bending. Also the effect of manufacturing tolerances in not necessarily to be obtained from the Sun, in principle the drive chain gearings when it is not possible to mount any other light source with precise analytic kinematics, axis turn sensors directly in the primary axis can be and with enough emitting power to extract a measurable introduced in the case of the most common linear drives. output from the concentrator, would suffice. In this Multi-secondary axis trackers, i.e. those integrating respect the full moon proves to be a good candidate, as several secondary axes mechanically linked in order to far as it will enable night calibrations not interfering with decrease aperture wind profile, are also considered in concentrator’s daily production. Its 0.49º apparent order to take into account the effect of the electrical diameter is very similar to the suns’, while its irradiance mismatch of the CPV arrays in each secondary axis as is six orders of magnitude smaller than the sun, so its well as their overall connection scheme, in order to photo-generated current is still within reach of cheap locate the pointing vector. current sensing devices. In the other hand the full moon Even if Internet connectivity or GPS might provide irradiance is three orders of magnitude above that of the atomic time synchronicity to the tracking control unit most brilliant planets and stars so it will be easily internal clock, availability of high accuracy ephemeredes distinguishable by the concentrator when searching the provides an immediate and autonomous alternative to night sky. Moon ephemeris and phase equations have precise time-keeping. A time drift parameter has been been analyzed and encoded for the development of a included in the model in such a way that it can be fitted Moon error collection procedure on which to fit the error with a tracking error set either, jointly with the rest of the model. parameters or individually within periodic time Along with these performance enhancements, long adjustment procedures. term accuracy tests are being conducted on several The complexity of the calibration transform when SunDog based systems. introducing these new parameters, obliged to switch in the Levenberg-Marquardt module from working on analytic derivatives to its finite-difference counterpart, which anyhow does not significantly decrease either accuracy or convergence time. 7. PHYSICAL IMPLEMENTATION The above described capacities have been implemented in an electronic embedded system, based on an 8 bit microcontroller, along with the required chipset and sensors to carry out the described algorithms and perform the analog measurements, and also motor driving units and an encoder decoding and interpolation subsystem. Named SunDogTM for commercial purposes, it is supplied with SunDog Monitor a Windows Figure 4. SunDogTM STCU (left) along with its SunDog application to run in a locally or remotely connected PC Monitor virtual interface software (above right), and as a virtual user interface. It also integrates an WWW remote control panel hosted by its internal web interchangeable modem for PSTN, RF, Ethernet or server (below right) GSM/GPRS Internet connectivity, which enables e-mail reporting and web based control and monitoring. This work has been supported by the European Prepared for operation in harsh environments its rugged Commission through the funding of the project enclosure integrates a simple control front panel (Fig. 4). FULLSPECTRUM (Ref. N: SES6-CT-2003-502620) 5. ONGOING WORK REFERENCES [1] A. Maish, Proc. of the 20thIEEE PVSC, (1988) 1309 Manual collection of tracking errors is a tiresome and [2] I. Luque-Heredia et al. Proc. of the 3rd Int. Conf. on error prone task, which demands automation, in order Solar Concentrators, (2005) prevent outliers creeping in and increase the accuracy of [3] A.A, Törn, 2nd IFAC Symposium on Stochastic the corrected ephemeris. Control, 19, (1986) Automatic error collection has already been [4] D.W. Marquardt, Journal of the Society of Industrial extensively simulated proving efficient performance. In it and Applied Mathematics, 11, 2 (1963), 431 direct search of the alignment of pointing and sun vector [5] J. Moré et al. ACM Transactions on Mathematical to obtain each tracking error measurement, imposes, to Software, 7, 1, (1981), 17 some extent, the use of the concentrators power output or [6] T.C. Van Flandern, et al The Astrophysical Journal some approximately equivalent variable, such as short Supplement Series,41,(1979), 391 circuit current, as feedback signal. Preliminary search [7] R. Baeza-Yates et al, Information and Computation, until sun and pointing vectors get into the concentrator’s 106, (1993), 234 I.Luque-Heredia, J.M.Moreno, G.Quéméré, R. Cervantes, P.H.Magalhães