Abstract
Artificial Intelligence (AI) has become a largely popular topic in recent years. The existence of such a concept had surprisingly been around for decades and has been booming in the last decade. This phenomenon is related to significant progress that hardware has seen with regards to the level of performance in such systems. While AI requires large computational power and strength, its potential has shown to be far superior. Able to capture more complex relations where previous methods might have struggled. With such a horizon, we will look at how to improve the status quo and approach new areas to benchmark the systems. As AI is a very broad topic, we want to narrow the scope to an area of deep learning that is of great use and interest. We will be looking into Artificial Neural Network (ANN) and its various applications. Image processing is no new area in AI, enabled only because of the great improvement leaps in hardware and vast collections of big data. Being around for decades, the recent top end hardware has allowed image processing to truly take off to impressively human levels of accuracy. Such systems are more than capable of imitating human intellect such as classifying images, detecting them or even identifying activities. Other forms of Neural Networks can do predictive type of analysis and while there are many existing systems of machine learning that tackle real world problems, indeed ANN are a strong new entry to the suite of current solutions. Studying the potential networks and using it effectively to a specific issue might prove to produce favorable results. The field of prognostics and predictive maintenance is also a very large market of application. With maintenance being such an integral and important gear in many industries, improving its maintenance process would be exceedingly beneficial. While there have not been extensive works of artificial intelligence in prognostics applications, we would be using neural networks for our prognostics application. On top of that we will be focusing on increasing its efficiency and enabling it on small light weight hardware. As prognostics systems are large and very costly, small changes at the basic level could go a long way in improving the entire system, proving to be very beneficial. In this thesis, we will be using neural networks and exploring its applications for prognostics on large maintenance data sets where we will assess its potential in a light weight setting.