Experimental validation of an ANN model for random loading fatigue analysis

Abstract

The use of artificial intelligence especially based on artificial neural networks (ANN) is now prevalent in many fields of data analysis and interpretation. There have been a number of papers published in the literature on the use of ANN for fatigue characterisation. Most of these have however been developed for rather focussed application with limited capability for fatigue life prediction for a broad scope of material and loading conditions. The authors recently presented a uniquely generalised ANN model that is capable of making fatigue life prediction for a broad range of material fatigue properties and loading spectral forms. The model was developed using simulated data albeit subject to conceivable constraints between possible materials properties and load forms. This paper presents a validation of the ANN model using a Society of Automotive Engineers (SAE) random fatigue loading experimental test data. The capabilities and potentials of the model are demonstrated by comparison with the SAE random load fatigue test results and with results obtained from other predictive methods. The performance of the ANN is highly encouraging as a general tool for random loading fatigue analysis.

Fatigue Analysis of the Microturbine Rotor Disc Made of 7075 Aluminium Alloy Using a New Hybrid Calculation Method

“…This software can be used to perform fatigue analysis of different machine components such as notched cantilever beams [47], engine connecting rods [48] or tubular welded assemblies [49]. This program also successfully passed the experimental verification, which was carried out for the notched specimen [50]. The predicted results agreed well and consistently with experimental data.…”

Artificial neural network for Gaussian and non-Gaussian random fatigue loading analysis

“…It should also be remarked that the ANN model which includes skewness and kurtosis effects still predicts good results without noticeable reduction in performance compared to previous ANN models that used less number of inputs. 30,31,43…”

“…The training, validation and testing process followed in this work is similar to those in previous works. 30,31,43 Generally, the feedforward backpropagation multilayer perceptron (MLP) model was followed. The training starts by initialising the weights W ij in the neuron connections illustrated in Figure 1.…”

Fatigue modeling using neural networks: A comprehensive review

“…The model proposed in Durodola et al 173 is tested only on simulated data. This model is further validated using experimental data in Ramachandra et al 174 and is developed to consider the mean stress effect in Durodola et al 175 This model is also extended for non‐Gaussian loadings by considering skewness and kurtosis in Durodola 176 . Another fatigue damage model 177,178 is developed for online fatigue damage modeling based on linear damage accumulation rule and FNN.…”