% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@INPROCEEDINGS{Sovetkin:890204,
      author       = {Sovetkin, Evgenii and Pieters, Bart and Weber, Thomas and
                      Achterberg, Elbert Jan and Weeber, Arthur and Bjoern, Rau
                      and Rennhofer, Marcus and Theelen, Mirjam},
      title        = {{PV}-{AIDED}: {P}hotovoltaic {A}rtificial {I}ntelligence
                      {D}efect {I}dentification. {M}ultichannel {E}ncoder-decoder
                      {E}nsemble {M}odels for {E}lectroluminescence {I}mages of
                      {T}hin-film {P}hotovoltaic {M}odules, {PEARL} {TF}-{PV}.},
      publisher    = {WIP},
      reportid     = {FZJ-2021-00793},
      pages        = {1520 - 1527},
      year         = {2020},
      abstract     = {The Solar-Era.net project PEARL TF-PV, [1], aims to reduce
                      the uncertainties in the operation of thin-film solar power
                      plants. To this end, one of the main parts of the project is
                      the gathering of performance data and electroluminescence
                      (EL) images of different types of thin-film solar cells and
                      modules (see abstract of Mirjam Theelen et al, this
                      conference). Detailed, local information on the module
                      performance is obtained using EL imaging, which may provide
                      early warning signs of degradation. A large number of
                      samples (over 6000 modules) are analyzed, ranging from cells
                      and modules produced in the different laboratories of the
                      project partners to industrially produced modules used in
                      power plants. Measurements are performed in laboratories as
                      well as outdoor directly at the power plants location. All
                      gathered data is stored in a database that in turn is used
                      to develop a failure catalogue for thin-film modules that
                      describes typical defects, visible with EL in various
                      technologies, and their influence on the solar modules
                      reliability and lifetime. In this work we present a novel
                      image segmentation approach, aiming to identify commonly
                      occurring defects in thin-film modules. We are building on
                      top of the encoder-decoder neural networks framework, that
                      have established itself as a standard tool in many other
                      image processing applications. We demonstrate our software,
                      PV-AIDED, is capable of fully automatic and fast EL image
                      processing of full-sizes modules. We are able to reliably
                      identify frequently occurring defects in thin-film modules,
                      such as shunts and so called “droplets”. The framework
                      is general and applicable to other types of defects, other
                      types of PV images, as well as other types of PV
                      technology.},
      month         = {Sep},
      date          = {2020-09-07},
      organization  = {37th European Photovoltaic Solar
                       Energy Conference and Exhibition,
                       Lisbon (Portugal), 7 Sep 2020 - 11 Sep
                       2020},
      cin          = {IEK-5},
      cid          = {I:(DE-Juel1)IEK-5-20101013},
      pnm          = {121 - Solar cells of the next generation (POF3-121)},
      pid          = {G:(DE-HGF)POF3-121},
      typ          = {PUB:(DE-HGF)8},
      doi          = {10.4229/EUPVSEC20202020-5CV.3.13},
      url          = {https://juser.fz-juelich.de/record/890204},
}