Analytics for Engineering Simulation

Why Use Analytics for Engineering Simulation?

As computational power has increased and simulations and testing have become more sophisticated, it has become possible to make accurate predictions for more real-world systems. But as design cycles tighten and problems become more complex, engineers and data scientists are challenged to get the most out of their simulation capabilities. Engineers and data scientists are faced with four major challenges in engineering simulation: how to put error bars on simulation results, insufficient or inadequate sampling of a design space, high computational expense of simulations, and disagreement of simulation models with physical tests. SmartUQ provides a powerful and viable approach to meet these challenges.

Challenges in Engineering Simulation

Putting Error Bars on Simulation Results

Many simulation results are deterministic in nature and only show a single point for one possible scenario. To consider the possible variability in the results, engineers and data scientists may perform a direct Uncertainty Propagation to quantify the uncertainty. But this approach can be computationally demanding if not impractical for many simulation models.

Insufficient or Inadequate Sampling of a Design Space

This has been a chronic problem for engineering simulation. Standard methods such as Monte Carlo sampling and factorial designs are not effective or impractical at higher dimensions.

High Computational Expense of the Simulation Model

Computational costs make design exploration impractical for many complex, long running simulation applications. Traditional design exploration techniques, such as random search, are prohibitively expensive for high dimensional problems.

Disagreement of Simulation Model with Physical Testing

The goal of most simulation design processes is to create a model that mimics reality. This goal is accomplished by comparing the simulation results to physical test results of the same system. A crisis of confidence occurs when physical tests fail to validate predictions from simulation models. Failing to do so forces re-design or re-testing that can be expensive and time consuming.

SmartUQ Solutions for Engineering Simulation

SmartUQ addresses these challenges using a variety of analytics and Uncertainty Quantification (UQ) tools, including design of experiments, emulation, statistical calibration and inverse analysis.

Uncertainty Quantification

SmartUQ’s Uncertainty Quantification techniques put error bars on simulation results. By performing UQ techniques such as Uncertainty Propagation and Sensitivity Analysis, decision makers can be better informed decision. By using a SmartUQ emulator, it is possible to perform advanced analytics that require a large number of predictions.

Verification and Validation (V&V) Under Quantified Uncertainty

It is often necessary to not only validate a simulation model’s deterministic solution for a design but also to validate the predicted probability distribution resulting from system uncertainty to ensure that the model accurately gauges the effects of system variability. Validation of simulation models is critical in many applications, such as those found in the aerospace, defense, and medical device industries, where expensive physical testing can be avoided or significantly reduced by using validated models.

Design of Experiments and Adaptive / Augmented Design

SmartUQ has modern space-filling DOEs that can handle the complex and high dimensional simulation models, including:

• Tailored DOEs for Specific Use Cases
• Iterative-solving methods capable of sampling for a wide range of analyses
• Designs for model calibration or involving discrete and continuous data
• Augmented and Adaptive DOEs
• Add space-filling points to an existing DOE to improve accuracy
• Maximize efficiency with iterative methods that intelligently select new design points
• Integration with Simulation Tools
• Easy to use statistical tools that integrate with simulation software effortlessly
• Minimize engineering time and statistical expertise required to solve complex problems

Emulation

An emulator (aka, a predictive model) is a virtual representation of a system that can quickly and accurately search the design space and perform advanced analytics. Implementing emulators makes the process computationally less expensive and more adaptive in the long run, should new designs be sought.

Statistical Calibration

Statistical Calibration can reduce the disagreement between a simulation model and physical testing results by performing tasks like:

• Diagnosis of Model Inadequacy
• Improving Simulation Accuracy
• Prerequisite for Model Validation

The benefit of Statistical Calibration is to avoid costly mistakes early on and minimize the number of physical tests.

Summary

SmartUQ’s analytics and Uncertainty Quantification tools such as DOEs, emulation, statistical calibration, and inverse analysis can be powerful in meeting the many challenges posed in simulations. Uncertainty Quantification can put error bars on simulation results and assess maturity and credibility of simulation models. Modern DOEs reduce the number of simulation runs required to explore a design space while emulation allows rapid optimization and uncertainty analysis. Statistical calibration methods can understand of the difference between simulation models and physical tests and calibrate simulation models to better represent physical realities. As the importance and complexity of design cycles increases, SmartUQ software fills a critical gap in simulation models.

If you have an analytics challenge with your engineering simulation, email us at [email protected].