Static load estimation using artificial neural network: Application on a wing rib. (November 2018)

• Record Type: Journal Article
• Authors: Cooper, Samson B.; DiMaio, Dario
• Is Part Of: Advances in engineering software. Volume 125(2018)
• Page Start: 113
• Page End: 125
• Publication Date: 2018-11
• Subjects: Artificial neural network -- Finite element method -- Load identification -- Structural health monitoring; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2018.01.007
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Proposed a pragmatic approach to static load estimation on a wing rib. Finite Element (FE) Model of the wing rib was calibrated using measured data. Successfully trained a range of neural networks using random data and FE strains as inputs. Artificial neural network (ANN) model predicted the static load to a degree of accuracy. ANN estimated load are in good agreement with experimental data. Abstract: This paper presents a novel approach to predicting the static load on a large wing rib in the absence of load cells. A Finite Element model of the wing rib was designed and calibrated using measured data obtained from static experimental test. An Artificial Neural Network (ANN) model was developed to predict the static load applied on the wing rib, this was achieved by using random data and strain values obtained from the static test as input parameters. A number of two layer feed-forward networks were designed and trained in MATLAB using the back-propagation algorithm. The first set of Neural Networks (NN) were trained using random data as inputs, measured strain values were introduced as input into the already trained neural network to access the training algorithm and quantify the accuracy of the static load prediction produced by the trained NN. In addition, a procedure that combines ANN and FE modelling to create a hybrid inverse problem analysis and load monitoring tool is presented. The hybrid approach is based on using trained NN to estimate the applied load from a known FE structural response. Results obtained from this research proves that using an ANN to identify loads is feasible and a well-trained NN shows fast convergence and high degree of accuracy of 92% in the load identification process. Finally, additional trained network results showed that ANN as an inverse problem solver can be used to estimate the load applied on a structure once the load-response relationship has been identified.