PEC has a deep knowledge of how to produce physics-based models and simulations that replicate real-world events. Often these problems do not have the “big data” that allows normal machine learning techniques to be successful, and data modeling is instead hampered by issues of “sparse data.” Our staff has decades of engineering design and research experience along with a healthy knowledge of data science techniques, and we believe that the best models capture the underlying physics of the problem.
Our unique blend of engineering and data science backgrounds allows PEC to deliver accurate models that perform best where it counts most, leveraging available data, augmenting with synthetic data, and capturing the underlying physics of the problem. We employ techniques that include deterministic fast-running models (FRM), time-marching transient modeling, single and multi-degree-of-freedom (MDOF) modeling, statistical modeling, and stochastic process modeling.
Protection Engineering Consultants has dedicated data scientists that are familiar with traditional statistical approaches. We have developed statistical solutions for a variety of problems such as fragmenting debris, human injury risk, high-speed video tracking, and armored vehicle vulnerability. Our team has experience developing statistical models for sparse, real-world test data as well as large databases containing thousands of simulations. We choose and deploy models to fit the problem at hand, ranging from basic probability distribution fitting to complex Markov chain Monte Carlo algorithms in computational Bayesian frameworks.
Fast Running Models
When time is of the essence, military decision makers often cannot wait for engineers to construct and run high-fidelity finite element predictions. Fast running models (FRMs) are often used instead. These FRMs are constructed beforehand, by fitting experimental and FEA-generated data. PEC has developed a range of such models for explosion loading of buildings, bridges, vehicles, and human occupants. Our clients place as much weight on model accuracy as they do on model efficiency, with a premium placed on model robustness and extensibility. We have years of experiences developing fast running models that deliver high-fidelity with low overhead, while capturing the underlying physics of the problem. Our artificial intelligence solver, EMBER, makes quick work of such challenges and helps us create satisfying, physics-based models.
Running real-world test programs is time consuming and costly. Research engineers frequently create finite element models of physical tests before conducting them to ensure the test design is sound. Sometimes questions arise regarding the cameras, gauges, and other instrumentation involved. Our team has developed methods to effectively emulate real-world equipment, such as high-speed cameras, laser interferometry, passive vehicle guidance, etc. Our emulation approach enables clear decisions to be made about the best equipment, as well as providing ground-truth scenarios for testing using our computer vision and signal processing algorithms. We not only simulate the physical test event; we simulate the way it will be recorded and analyzed.