Machine Learning for Engineering Applications

Machine Learning (ML) has piqued the interests of researchers and engineers massively in recent years. Currently, our lab is focusing on implementing several machine learning algorithms in the context of uncertainty quantification and optimization. Some of the approaches currently being utilized are clustering approaches such as k-means and spectral clustering, support vector machines (SVM), artificial neural networks (ANNs) for surrogate modeling, and long short-term memory (LSTMs) for forecasting, among others. 
​
One of the implementations of LSTM for time series modeling and forecasting is provided below. In this research, an adaptive approach to detect how far the LSTMs can predict the responses accurately is being studied and is being applied to fatigue-life prediction of composite structures.

Highly Non-linear Lorentz Ocsillator

Adaptive Artificial Neural-Networks

Although ANNs are used widely in several disciplines, one of the main drawbacks is that is requires a large amount of training data to obtain the hyper-parameters accurately. This becomes challenging for expensive to evaluates responses such as responses from CFD or non-linear FEA. To this end, the sampling locations for the data should be identified appropriately such that more information gain in achieved even with few number of simulations.