A neuro-fuzzy approach to the weight estimation of aircraft structural components

This paper explores the issues related to the application of fuzzy logic techniques to aid the process of weight estimation for aircraft structures at preliminary design stages. The focus lays on the design of a neuro-fuzzy system for the weight analysis, through the use of the Neuro-Fuzzy Function Approximator (NEFPROX) algorithm. The paper introduces a three-level process designed around Mamdani fuzzy systems derived through NEFPROX and analyses its application to the sizing and weight estimation of spoiler attachment ribs. Problems such as structure parameterisation, variable selection and model optimisation are explored as part of the process validation phase on the selected structural case study. The model performance is evaluated with respect to modelling accuracy, generalisation capabilities and the interdependencies between the variables the model is able to derive from the analysis of the given structural examples. The results are then compared to the performance obtained from the application of Takagi-Sugeno-Kang (TSK) fuzzy models derived using Adaptive Network-based Fuzzy Inference System (ANFIS) on the same sample of spoiler attachment ribs. Results highlight the benefits of adopting NEFPROX for the derivation of fuzzy systems to be applied in structural weight estimation problems, from the point of view of both accuracy of the approximation provided and the quality and interpretability of the final rulebase derived by the system.