What's New in ODYSSEE CAE 2024.1

ODYSSEE Lunar – GUI

• In animation tab: Possibility to open and predict EnSight animation format (*.bin.case) dedicated to fluid animations.
• Clarification between the working directory and the ODYSSEE CAE project folder notions:
  • The ‘working directory’ now refers to the location of the original files.
  • ‘CAE project folder’ is a new term introduced to define the CAE project workplace (where project files and a copy of the original files are stored).
  • Thanks to the update of the start window, it is now possible to easily create a CAE project in the working directory or to leave it in the current default directory: Documents\ODYSSEE\CAE\Lunar.
• A new User Script available named Adaptive sampling methods loop proposes to add in loop new points in the sample based on two approaches to automatically enrich the database:
  • First one, based only on X, to enrich areas where sampling points are missing.
  • Second one, based on the X and Y data, to enrich areas where there is less predictability.

When the new X are found, a batch file can be executed to update the input FEM model file, next run the simulation, and finally extract the results of the simulations. At each loop, the database is updated with the new points and new results. Please note, this script requires a .batch file from the user defining his own execution process to run simulations and extract results.

• Several improvements in data management
  • If there is already an existing input dataset (X) in the project, this one is automatically loaded when opening the doe tool.
  • Input (X-XN) and output (Y–YN) files are now saved in csv format in the project folder in addition to bin format.
  • The solver setting is now saved when you change some data in the data management.
  • Autocorrelation data saved in csv in project folder.
• Sensitivity improvement
  • In adjust parameter set tab from Sensitivity menu, automatic preview of the prediction curve in orange when moving the slider if the model has already been trained a first time.
  • Multi dataset parameters effects saved in project folder.
• Bug Fixes
  • Data manager:
    • When an output file (Y–YN) had only one column, this column was interpreted as a header, and it was impossible to load it in the GUI.
    • When the parser was run via data manager, and next the input file in the data tree was delete, a Lunar interface crash occurred.
  • Parser tool via the data manager:
    • It was not possible to execute the extraction step without run the substitution step.

Known Bug

• Kriging interpolation (direct or with POD, FFT, Clustering) with cubic basis function is returning the error
  “LU decomposition failed. at “

ODYSSEE Quasar – Solver

This version includes the new following functions:

• ARIMA means Autoregressive Integrated Moving Average. This function is suitable to predict the future from time-dependent data series (time series). The future values we want to predict like y(t+1), y(t+2), … depends solely on the previous values and the error made, and not on any other variable other than y.
• AutoARIMA is an ARIMA that automatically estimates most of its parameters.
• Markov Chain Monte Carlo (MCMC) integration: This is a new DOE method that combines the Markov chain principle and the Monte Carlo algorithm to provide an independent and uniform distribution in the design space, while following the desired distribution law. The Markov chain action consists of proposing one by one a new sample dot based only from the last location of the previous one. In this 2024.1 version, this method is only available by scripting.
• GLSTM function GLSTM is a generalisation of LSTM Quasar function: Long Short-Term Memory (Recurrent Neural Network) model proposing the training and the prediction steps for the time-dependent data. In this new function, this is possible to have multiple layers of LSTM stacked on top of each other. This may help model more complex phenomena. GLSTM function can be sued either for training or prediction.
• predictGLSTM function is a useful function when the training part has already been done with GLSTM. Thus, it can return results very quickly by skipping training.
• Markov Chain Monte Carlo estimation: This is a function to find parameters P of a model ƒ such as Y=ƒ(P, X) given the user. MCMC algorithm is used to ensure the convergence to find the best estimation of parameters P using a normal distribution law. It is quite close to the reverse engineering except the parameters P proposed are a mean and standard deviation and not the exact values.
• QuasarSound library: A new library containing the functions for processing audio files (mp3, wav) and audio signals.
  • loadFile("example.mp3") or Sound file = Sound("example.wav") load a sound file.
  • duration() returns the time duration of the sound signal.
  • at() returns the magnitude of the sound signal frame by frame.
  • channelCount() returns the number of channels found in the sound file.
  • sampleRate() calculates the sample rate of the signal.
  • samplesCount() counts the number of samples into the sound file. If the file contains many channels, the samples of all channels are added.
• Bug Fixes
  • Intermediate messages pollute the display when LSTM is executed.
  • In MLP function, the past implementation of the learning rate with a plateau scheduling didn’t give any different result from the learning rate with a step scheduling, because it didn’t consider the loss.

Nova: The optimizer

This version includes the following:

• Bug Fixes
  • For the SLSQP algorithm, additional lines starting by the word ITERATION polluted the log file. It has been removed.
  • For all optimization methods except MultiObjectives (n°16 (NLPJOB)), additional columns with 0.0 were displayed at the presenting results at the end of the log. It has been removed.
  • In nova command file, GENDOEF keyword that allows to run a DOE and find the best point inside it was not working. It is now fixed and a new example “Nova-example6” has been added.

Known Bug

• If you use ODYSSEE Solver directly, for a problem with constraints H or G, iterations are correct but the summarize printed at the end of the log presents the best point for only a minimum of F without take care of H or G. Please open the CSV file optimization.csv to find the optimal point among the tested points where F is min and H is positive or G is closest of 0. This issue does not affect the optimal point presented in ODYSSEE CAE.