Training MAJN: Predicting Turbine Failures with Deep Learning and Bare-Metal C++

The article describes the author's journey in using the NASA Turbofan Engine Degradation Simulation Dataset to train neural networks for predictive maintenance of turbines, starting with a simple perceptron and moving towards more complex LSTM models.

Details

The author, an industrial engineer and self-taught programmer, was looking for a project to apply both his engineering and programming skills. He decided to use the famous NASA Turbofan Engine Degradation Simulation Dataset to train neural networks for predictive maintenance of turbines. Initially, the author tried using a simple perceptron multilayer classifier, which had achieved 99.9% accuracy on a simpler dataset. However, when applied to the NASA dataset, the classifier only reached 50-59% accuracy, even after adding more layers and neurons. Realizing the need for a more sophisticated approach, the author decided to prepare the dataset by identifying and removing sensors with constant values, as they represented noise that could negatively impact the neural network training. The author then moved towards using LSTM (Long Short-Term Memory) networks, which are designed to work with time series data and can better capture the temporal evolution of the dependent variable. The article documents the author's journey in transitioning from a simple perceptron to more complex LSTM models to tackle the challenges posed by the NASA dataset, which simulates the degradation of turbine engines over time.