CDH/ACE (ACE = Autonomous Computational Experiments) is a new, groundbreaking software framework to seamlessly integrate the precision of traditional Computer-Aided Engineering (CAE) with the speed and scalability of Data Analytics.

This is accomplished by performing Autonomous Computational Experiments in a self-learning and self-correcting process, that automatically combines complex CAE workflows with concepts from the field of machine learning and artificial intelligence.

While currently focused on Noise and Vibration Harshness (NVH) problems, its core principles are broadly applicable across engineering disciplines.

Benefits:

Reduced Time and Cost
Streamlines processes, making them faster than manual setups and dramatically accelerating response predictions compared to relying solely on physical models.

Improved Efficiency
Optimizes simulation results and facilitates the discovery of superior solutions at an accelerated pace while minimizing manual.

Enhanced Design Space Exploration
Enables the automatic exploration of a much wider range of design possibilities.

User-Friendly Approach
Offers a more intuitive approach, making advanced CAE accessible even to engineers without extensive expertise in numerical simulation.

CDH/ACE strategically combines the strengths of two distinct modeling approaches and reduces their limitations.

Direct Physical Modeling (using FEM solvers like Nastran/Optran)

Advantages:

• Unrivaled in revealing underlying physical mechanisms
• Crucial for identifying root causes of problems
• Gaining deep insights into system behavior

Limitations:

• Requires significant engineering expertise for setup
• Computationally expensive
• Often taking hours per simulation run
• Demanding substantial resources

The innovative approach of CDH/ACE lies in its ability to automate running physical simulations and efficiently modify parameters, while simultaneously training data models based on these simulation results for rapid predictions.

Not only the CAE solver interfaces to Nastran, Optran, and Optistruct FEM solvers are automated but in addition a wide-range of result plots as well as other useful results files is being generated automatically during the iterative learn-cycles.

Response history and convergence over self-learning cycles (batches):

Optimized response and its robustness percentiles for 10’000 samples

Advanced CAE Processes
Manages easy usage of complex simulation workflows, including a plugin concept to account for customer specific functionality.

Automated Design of Experiments (DOE)
Systematizes the design of experiments to efficiently explore the impact of various features on system performance.

Parameter Optimization and Identification:
Accurately determines optimal parameter values and identifies key parameters influencing physical system responses.

Sensitivity Analysis and Robustness Studies
Conducts in-depth analyses of system sensitivity to parameter changes and assesses overall robustness, leveraging fast machine learning model evaluations.

Model Reduction
Creates simplified, real-time predictive data models using trained machine learning algorithms.

Python Integration and Automation
Provides a flexible, open Python based platform for integrating custom tools and automating workflows.

Performance Optimization
Fine-tunes simulation and solver settings for maximum efficiency.

• Seamless integration of machine learning algorithms with robust CAE solvers
• Its autonomous nature, capable of generating its own CAE simulation data
• Continuous self-learning and self-correction through iterative cycles, further enhancing accuracy over time
• Specialized focus and efficiency for vibration and acoustics analysis
• Fully automated interface with Nastran, Optran, and Optistruct FEM solvers
• User-friendly design requiring minimal manual intervention
• High scalability across multiple CAE servers and compute nodes, supporting up to 64 CPU cores per machine
• Integration with existing CAE queuing systems for efficient job management
• Automated generation of all relevant data and post-processing visualizations
• An open, Python-based architecture for easy customization
• An ideal solution for optimization and robustness studies when combined with Optran, AMLS and FastFRS

Currently CDH/ACE is limited to noise and vibration related problems using one of the following FEM solvers:

• MSC/Hexagon Nastran version 2019 or newer
• NX/Simcenter Nastran version 2020 or newer
• Altair Optistruct version 2019 or newer
• CDH/Optran version 5.3 or newer

For reasons of efficiency and stability in large-scale simulations, it is highly recommended to use CDH/AMLS and CDH/FastFRS in combination with the above listed FEM solvers.

• CDH/ACE officially supports all major Linux 64bit operating systems (preferred RedHat >= 8).
• Recommend hardware consists of Intel Xeon or AMD EPYC based CPU’s (ARM is currently not supported) with at least 8 CPU-cores and 256Gb of ECC-enabled memory.
• Suitable GPU hardware may be utilized for machine learning training acceleration.