BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//16thmodelicafmiconference//speaker//7MEYGH
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-16thmodelicafmiconference-YWAZ3D@pretalx.com
DTSTART;TZID=CET:20250909T101900
DTEND;TZID=CET:20250909T102000
DESCRIPTION:This session is chaired by
DTSTAMP:20260422T204328Z
LOCATION:202
SUMMARY:Session Chair - Bernhard Bachmann
URL:https://pretalx.com/16thmodelicafmiconference/talk/YWAZ3D/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-16thmodelicafmiconference-YTGZHW@pretalx.com
DTSTART;TZID=CET:20250909T104500
DTEND;TZID=CET:20250909T111000
DESCRIPTION:We propose a novel approach for training Physics-enhanced Neura
 l ODEs (PeN-ODEs) by expressing the training process as a dynamic optimiza
 tion problem. The full model\, including neural components\, is discretize
 d using a high-order implicit Runge-Kutta method with flipped Legendre-Gau
 ss-Radau points\, resulting in a large-scale nonlinear program (NLP) effic
 iently solved by state-of-the-art NLP solvers such as Ipopt. This formulat
 ion enables simultaneous optimization of network parameters and state traj
 ectories\, addressing key limitations of ODE solver-based training in term
 s of stability\, runtime\, and accuracy. Extending on a recent direct coll
 ocation-based method for Neural ODEs\, we generalize to PeN-ODEs\, incorpo
 rate physical constraints\, and present a custom\, parallelized\, open-sou
 rce implementation. Benchmarks on a Quarter Vehicle Model and a Van-der-Po
 l oscillator demonstrate superior accuracy\, speed\, generalization with s
 maller networks compared to other training techniques. We also outline a p
 lanned integration into OpenModelica to enable accessible training of Neur
 al DAEs.
DTSTAMP:20260422T204328Z
LOCATION:Audi-Midi
SUMMARY:Efficient Training of Physics-enhanced Neural ODEs via Direct Collo
 cation and Nonlinear Programming - Linus Langenkamp\, Bernhard Bachmann\, 
 Philip Hannebohm
URL:https://pretalx.com/16thmodelicafmiconference/talk/YTGZHW/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-16thmodelicafmiconference-HLEDGX@pretalx.com
DTSTART;TZID=CET:20250910T091500
DTEND;TZID=CET:20250910T094000
DESCRIPTION:The convergence failure of iterative Newton solvers during the 
 initialization of Modelica models is a serious show-stopper\, particularly
  for inexperienced users. This paper presents the implementation in the Op
 enModelica tool of methods presented by two of the authors in a previous p
 aper\, to help diagnosing and resolving these convergence failure by provi
 ding ranked lists of potentially critical start attributes that might need
  to be fixed in order to successfully achieve convergence. The method also
  provides library developers with useful information about critical nonlin
 ear equations\, that could be replaced by equivalent\, less nonlinear ones
 \, or approximated by homotopy for more robust initialization.
DTSTAMP:20260422T204328Z
LOCATION:202
SUMMARY:Diagnosing Newton’s Solver Convergence Failures in theInitializat
 ion of Modelica Models - Francesco Casella\, Karim Abdelhak\, Bernhard Bac
 hmann\, Philip Hannebohm\, Teus van der Stelt
URL:https://pretalx.com/16thmodelicafmiconference/talk/HLEDGX/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-16thmodelicafmiconference-K8CECA@pretalx.com
DTSTART;TZID=CET:20250910T100500
DTEND;TZID=CET:20250910T103000
DESCRIPTION:Direct collocation-based dynamic optimization plays an importan
 t role in the optimization of equation-based models. With this approach\, 
 continuous problems are transcribed into sparse nonlinear programs (NLPs) 
 that can be solved efficiently. The open-source Modelica environment OpenM
 odelica provides an implementation using Radau IIA collocation\, but has m
 ajor limitations\, such as the lack of parameter optimization\, no adaptiv
 e mesh refinement\, and no support for higher-order integration schemes. T
 his paper presents (1) a comprehensive reimplementation that addresses the
 se limitations and (2) a novel $h$-method mesh refinement algorithm. Imple
 mented in the custom Python / C++ optimization framework GDOPT\, the appro
 ach demonstrates significant performance improvements\, solving typical pr
 oblems 2 to 3 times faster than OpenModelica under equivalent conditions. 
 Using the proposed mesh refinement algorithm\, the framework correctly ide
 ntifies non-smooth regions and increases resolution accordingly\, requirin
 g only a small increase in computation time. The implementation lays the f
 oundation for a future integration into the OpenModelica toolchain.
DTSTAMP:20260422T204328Z
LOCATION:202
SUMMARY:Enhancing Collocation-Based Dynamic Optimization through Adaptive M
 esh Refinement - Linus Langenkamp\, Bernhard Bachmann
URL:https://pretalx.com/16thmodelicafmiconference/talk/K8CECA/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-16thmodelicafmiconference-XJA98S@pretalx.com
DTSTART;TZID=CET:20250910T150000
DTEND;TZID=CET:20250910T152500
DESCRIPTION:Equation-based modeling that utilizes reusable components to re
 present real-world systems can result in excessively large models. This\, 
 in turn\, significantly increases compilation time and code size\, even wh
 en employing state-of-the-art scalarization and causalization techniques. 
 This paper presents an algorithm that leverages repeating patterns and uni
 form causalization to enable array-size-independent constant time processi
 ng. Allowing structural parameters that govern array sizes to remain resiz
 able during and after the causalization process enables the formulation of
  an integer-valued nonlinear optimization problem. This approach identifie
 s the minimal model configuration that preserves the required structural i
 ntegrity\, which can subsequently be resized as needed for simulation. The
  proposed method has been implemented in OpenModelica and builds upon prel
 iminary work aimed at preserving array structures during causalization\, w
 hile still resolving the underlying problem in a scalarized manner.
DTSTAMP:20260422T204328Z
LOCATION:202
SUMMARY:Constant Time Causalization using Resizable Arrays - Karim Abdelhak
 \, Bernhard Bachmann
URL:https://pretalx.com/16thmodelicafmiconference/talk/XJA98S/
END:VEVENT
END:VCALENDAR