The analyses of the underlying ANN models in this paper were written prior to any successful attempts at predicting peformance indicators. However, the program and data were reevaluated to remove any erroneous data or programmatic errors in an effort to realize the initial objective of the study. I was successful in correcting the issues and able to significantly reduce the MSE observed for each of the ANN models. That paper has an addendum added to the end that adds clarification of the new MSE observed after modifying the program.
The purpose of this study is to make predictions about some performance indicators of a turbofan aircraft by utilizing modern Artificial Neural Network (ANN) models.
The problem of determining optimum performance in today’s modern turbofan aircraft is a very old problem with real test data to back up calculated predictions. Being a very old industry, aerospace has been slow to adopt the use of Artificial Neural Networks to predict the performance of new engine designs. However, with modern ANN models the real-world test data can be utilized to predict engine and aircraft performance with a high degree of precision. By using an ANN to predict performance, time spent calculating the impact of design changes can be reduced.
This study is intended to complement the presentation that analyzed other studies that attempted to use ANN models to predict performance indicators of an aircraft based on a few measured inputs. The goal will be to expand upon the work by attempting to perform the same predictions using a simple ANN model, the Levenberg-Marquardt (LM) and Bayesian regularization (BR) algorithms, as well as the ReLU activation function. This will be accomplished by creating the ANN in Python along with utilizing PyTorch.
The inputs that will be used to train and test the model will be:
• Airspeed
• Altitude
• Air Temperature
• Exhaust Temperature
• Cruise Mach Number
The expected outputs will be:
• Thrust
• Ground Speed
• Fuel Consumption