PG studies: Mechanical and Mechatronic Engineering Department Research area: Computational Engineering Dr Ryno Laubscher
What is computational engineering? • Relatively new research discipline compared to more focussed research areas such as turbomachinery, control systems and biomechanical research • Focuses on the development and application of computational models and simulations for scientific, social and financial computing • Scientific – CFD, FEM, DEM etc. • Social – criminal behaviour prediction, sentiment analysis, etc. • Financial – volatility predictions and modelling
Relevant fields: • Computational fluid dynamics: Simulation and numerical analysis of complex fluid flow problems using next generation numerical techniques such as: 1. Reduced order modelling (reducing DOFs) for in‐situ sim and optimisation 2. AI based turbulence and chemistry prediction algorithms 3. Investigation and simulation of large scale (industrial) fluid flow/heat transfer/reacting problems 4. Modelling using high fidelity models (LES, etc.)
Relevant fields: • Finite element analysis – computational structural mechanics: Numerical analysis of structural components/materials and systems: 1. Linear and non‐linear response finite element analysis 2. Study parametric design considerations of mechanical components through simulation (bicycle frames, aircraft components, laminate configurations, etc.) 3. Study heat transfer in porous/solid components and material property effects 4. Investigating mechanical responses of new materials through computation
Relevant fields: • Discrete element analysis: Computing the motion and interacting between large number of discrete particles: 1. Agricultural/ post‐harvest technologies 2. Materials handling
Relevant fields: • Big data analytics and AI (machine learning )applied to engineering: AI techniques are now being used by practising engineers to solve a whole range of hitherto intractable problems: 1. Real‐time intelligent automation and condition monitoring (2.) 2. Architectures, algorithms and techniques for distributed AI systems 3. Deep learning and real world applications – failure prediction, etc. 4. Computer perception/interpretation – cross between CV and ML 5. Big data analytics – understanding complex systems, IoT and CPS. 6. AI applied to simulation: CFD, etc. => ROM
Relevant fields: • Optimisation (structural, fluid and systems‐wide) and simulation Numerical design optimisation (structural, general purpose and multidisciplinary) using general and metaheuristic algorithms. Development and investigations into new algorithms and parallel computing. System (social, financial and scientific) simulation using statistical methods: 1. Monte‐Carlo simulation studies (eg. percolation statistics) 2. Combination with deep learning predictive modelling 3. GPU acceleration of optimisation algorithms (concurrency) 4. Parametric design optimisation
Relevant fields: • Financial and social modelling (engineering): Similar to statistical modelling of scientific problems but now focussed on financial models of businesses/plants/etc. along with AI predictive/classification modelling: 1. Use of statistical techniques and machine learning 2. Deep grounding in partial differential equation theory (Black‐ Scholes equation) 3. Use of optimisation techniques (portfolio optimisation) 4. Criminal behaviour prediction (baggage screening, etc.)
Relevant lecturers: • Computational fluid dynamics – Prof Harms, Prof Meyer, Dr Hoffmann, Dr Laubscher, Prof vd Spuy, Prof Von Backstrom • Finite element analysis – Dr Venter, Prof Venter, Prof Groenwald • Machine learning and Big Data analysis – Dr Laubscher, Prof Venter and Dr Venter • Optimisation and simulation – Prof Groenwald, Prof Venter, Dr Venter and Dr Laubscher • Financial and social modelling – Prof Harms, Prof Groenwald, Dr Laubscher • Discrete element analysis: Prof Coetzee and Dr Els
Example: Merging of CFD and AI • Problem statement Advanced turbulence‐chemistry interaction CFD modelling requires the solution of a massive set of differential equations which is usually very stiff and can have long simulation times. Not viable for industrial simulations
Example: Merging of CFD and AI • Solution Rather than using massive computer resources to solve the fine reactors for every cell in the computational domain, use a AI algorithm that predicts the reactor performance based on memory built from a supervised learning algorithm. Thus model will almost instantaneously predict reactor yield. Fine structure reactors Deep learning neural network Turbulence field Reaction rate prediction
Example: Merging of CFD and AI • Self learning Use statistics to develop distributions of species. Use distributions to create random species compositions and use as training data set. Use 1D reactor code to develop 1D reactor equations Deep learning neural network training data Reaction rate prediction
Example: Merging of CFD and AI • Results Massive reduction in computational cost with small error:
END • Thanks for listening and if there are any questions please feel free to come and see me and the other lecturers
Recommend
More recommend
