• Based on the 2025 version, the following 4-stage report is provided. (Some configurations may change in future versions.)

- Step 1) Screening

- Step 2) Autonomous Metamodeling

- Step 3) Optimization (Improvement / Pareto)

- Step 4) Validation

• Reports for each stage are provided as a Standard Report, which serves as a summary, and a Reference, that contains insights.

• AIDesigner sim provides all result reports in Excel format.

• Therefore, Microsoft Excel must be installed, and the version must be 2007 or higher.

• You must complete the process automation in PIAnO.

• When AIDesigner sim first starts, you must set up the PRD file, which is the project of PIAnO.

• AIDesigner sim can automatically obtain the necessary data from the set PIAnO model.

• PIAnO provides various metamodeling techniques in PIAnO Metamodeler.

• For non-experts, PIAnO Metamodeler automatically generates the most suitable metamodel for the data provided by the user using AI called “Bruce.”

• For more detailed information, please utilize the training course or contact technical support.

• A total of 6 metamodeling techniques are provided as follows.

- Regression model - PR(Polynomial Regression), RBFr(Radial Basis Function Regression), EDT(Ensemble of Decision Trees), MLP(Multi-Layer Perceptron)

- Interpolation model - Kriging, RBFi(Radial Basis Function Interpolation)

• Metamodeling techniques are largely divided into two types: regression models and interpolation models.

• The regression model is a method that can express the tendency of experimental points well when the experimental points include random errors, as in an actual experiment.

• The interpolation model is a method that configures a prediction model so that it can accurately pass through the experimental points when there are no random errors, as in a computer experiment.

• You can use Augmenting Design.

• It is a method of selecting additional data to have the most even distribution possible by considering the distribution of existing data.

• The ALHD (Augmented Latin Hypercube Design) technique, which improves space filling, such as OLHD (Optimal Latin Hypercube Design) is applied.

• Provides 11 DOE techniques.

  - FFD(Full Factorial Design)

  - CCD(Central Composite Design)

  - ICCD(Interior Central Composite Design)

  - FCCD(Face-centered Central Composite Design)

  - PBD(Placket-Burman Design)

  - BBD(Box-Behnken Design)

  - OA(Orthogonal Array)

  - OLHD(Optimal Latin Hypercube Design)

  - CV(Centroidal Voronoi Tessellation)

  - Sobol Sequence

  - Augmenting Design

• DOE data plays an important role in establishing design strategies because it can provide various information as follows:

- Relationship between design variables and performance indices

- Ranking of design variable importance

- Metamodel expressing performance indices as a function of design variables

- Useful design improvement suggestions

• PIAnO provides various DOE techniques in PIAnO Sampler.

• For non-experts in DOE, PIAnO Sampler supports the "Select Automatically" function.

• This function suggests the most suitable DOE data when you provide the purpose of the DOE, the number of levels, and the number of possible experimental points.

• For more detailed information, please use the training course or contact technical support.

• There are three main ways to handle multiple objective functions.

- A method of converting multiple objective functions into a single objective function by multiplying their weights (can be handled by all optimization algorithms)

- A method of not handling multiple objective functions at the same time, but converting them into constraints and finding the optimal solution according to the boundary values of the constraints (can be handled by all optimization algorithms)

- A method of exploring the Pareto set by considering multiple objective functions at the same time (can be handled by MOGA. Increased analysis cost is essential)

• It provides 10 optimization algorithms, and MGA and EA will be deprecated as they can be replaced by CMA-ES or HMA. 

- STDQAO(Sequential Two-point Diagonal Quadratic Approximate Optimization) - Local/Efficiency perspective/Continuous design variables

- FSolver - Local/Efficiency perspective/Continuous design variables

- ePPAO(Efficient Progressive Polynomial Approximate Optimization) - Local/Efficiency perspective/Continuous design variables

- PQRSM(Progressive Quadratic Response Surface Modeling) - Local/Efficiency perspective/Continuous design variables (Recommended as a local optimization technique)

- MGA (Micro Genetic Algorithm) - Local/Continuous and Discrete design variables (To be deprecated)

- CMA-ES(Covariance Matrix Adaptation Evolution Strategy) - Global/Continuous and Discrete design variables

- HMA(Hybrid Metaheuristic Algorithm) - Global/Continuous and Discrete design variables (Recommended as a global optimization technique)

- EA (Evolutionary Algorithm) - Global/Continuous and Discrete design variables (to be deprecated)

- MOGA(Multi-Objective Genetic Algorithm) - Multi-objective/Global/Continuous and Discrete design variables (recommended when dealing with multi-objective functions)

- PADO(Progressive Approximate Discrete Optimization) - Local/Continuous and discrete design variables

• Mathematically, it is an optimization algorithm that guarantees that if you extract infinite data, you can converge to the best optimal solution.

• Since the purpose is to search for a global optimal solution, the optimal solution does not change depending on the initial value.

• Although it can search for the best optimal solution, it requires a lot of analysis, so it is usually used mainly when doing optimization with a prediction model or a fast analyzer.

• It corresponds to the traditional optimization theory, and it is a method of changing the direction by finding the gradient (sensitivity) to find the maximum/minimum value of the objective function. • The optimal solution may vary depending on the initial value, and there is no guarantee that the best optimal solution (global optimum) is searched, so it belongs to the local optimization algorithm.

• Design variables are variables that the designer can change. In mathematical terms, they correspond to independent variables. Examples: thickness, dimensions, number of items, etc.

• Objective function is a response that you want to minimize or maximize among the performance indices composed of functions of design variables. In mathematical terms, they correspond to dependent variables. Examples: efficiency, weight, life cycle, etc.

• Constraints are responses that have a range of requirements among the performance indices composed of functions of design variables. In mathematical terms, they correspond to dependent variables. Examples: stiffness, strength, temperature, crash performance, natural frequency, etc.

• The performance indices are basically composed in the form defined above, but they can be freely placed anywhere in the objective function or constraints depending on the situation.

• Here's a hint: the optimization algorithm tends to satisfy the constraints first and then the properties of the objective function.

• Therefore, if the range of constraints is narrow, there are cases where the properties of the objective function (maximization or minimization) do not work as intended.

• If the range of constraints is sufficient, you only need to be located within that range, so you can obtain a solution that focuses on improving the objective function.

• If you can directly access the Input/Output files with the Mapping function in PIAnO or implement the API with a script (ex. Python, VBscript, etc.), Then it will work..

- Examples of Mapping functions) SUB files of ADAMS/Car, BDF files of Nastran, etc.

- Examples of API implementation) Python API function of Ansys WB, etc.

• It is possible.

• PIAnO supports CAE SW interfaces in cloud or supercomputer environments with a function called “Remotejob.”

• This function automates the process of uploading data, transmitting job commands, performing analysis, and then downloading the result data in the same way that users access a remote PC.

• If the process is automated, you can easily try out various optimization strategies with one click. However, it is not required.

• If it is difficult to automate the process, you can perform various optimization strategies with the metamodel if you can obtain the data manually.

• It is to automate the process of repeatedly performing analysis to perform optimization so that it can be re-executed according to a pre-written script. (This is generally called a template.)

  - PIDO SW such as PIAnO can use the macro or API provided by CAE SW,

  - to update the values corresponding to the design variables in the analysis model,

  - and can also retrieve the objective function or constraint values from the result file after performing the analysis.

  - Once the analysis is automated, you can automatically perform repetitive analysis by changing the design variable values.

• It does not allow installation of different versions of PIAnO on the same device.

• When upgrading to the New version while the Old version is installed, you do not need to delete the Old version separately. If you install the New version, the installation will proceed sequentially after deletion.