Fatigue Life Predition Of Magnesium Alloy Based On Neural Network

Magnesium alloy components were widely used in automotive sector,communication and aircraft industries for its prominent advantages such as low density, high strength, rich resources and so on. Magnesium alloy would subject to cyclic loading and led to fatigue failure like other engineering materials. There were more studies on uniaxial fatigue than on multiaxial fatigue. When study the fatigue life of magnesium alloy, researchers always use different fatigue models to predict its fatigue life.In this study, Strain-controlled fatigue experiments were conducted on extruded AZ31 B and ZK60 magnesium alloy. Except those experimental date, experimental data of AZ61 A magnesium alloy from literature were also adopted. Multiaxial fatigue criteria based on critical plane and neutral network were applied to predict the fatigue life, respectively. In addition, these two approachs were evaluated based on the experimentally obtained fatigue results. The main work and conclusions are as follows:(1) Stain-controlled multiaxial fatigue experiments were conducted on AZ31 B and ZK60 magnesium alloy under four loading paths, including fully reversed tension-compression, cyclic torsion, 45°in–phase axial-torsion and 90°out-of-phase axial-torsion. Strain-controlled fatigue experiments were conducted on an extruded AZ31 B magnesium alloy at three strain ratios(R=-1,0,-∞). Result shows that the strain-life curve has a kink under each loading path. The fatigue life of magnesium alloy was influenced by strain amplitude and loading path.(2) SWT model, F-S model, modified SWT model and CLX model were applied to predict the fatigue life of AZ31 B and ZK60 magnesium alloy under four loading paths. SWT model was used to predict the fatigue life of AZ31 B magnesium alloy at three strain ratios. It is observed that CLX model is the best model to predict the fatigue life of AZ31 B magnesium alloy under four loading paths. All predict results are within factor-of-five lines and 78.9% results fall within factor-of-two lines. F-S model has advantages in fatigue life predition of ZK60 magnesium alloy. 98% predict results are within factor-of-five lines and 77.6% results fall within factor-of-two lines. Modified SWT model is better than other models in predicting the fatigue life of AZ61 A magnesium alloy under four loading paths. All results are within factor-of-five lines and 68.1% predict results fall within factor-of-two lines. All predict results of AZ31 B and AZ61 A magnesium alloy at three strain ratios are within factor-of-five lines. 93.9% and 93.2% predict results fall within factor-of-two lines, respectively.(3) Four BP neural networks were adopted to predicted the fatigue life of magnesium alloy, including steepest descent back propagation(SDBP), variable learning rate momentum back propagation(VLMOBP), VLMOBP use sum squared relative error as the performance function(VLMOBP-SSRE), optimized BP by genetic algorithm(GABP). Results shows that in fatigue life predition of AZ31 B magnesium alloy under four loading paths, GABP is bertter than other BP neural network. 97.4% predict results fall within factor-of two lines. Also, GABP still has the best performance in predicting the fatigue life of ZK60 magnesium alloy. All predict results are within factor-of-two lines. In fatigue life predition of AZ61 A under four loading paths, all predict results by four BP methods are within factor-of-two lines. In fatigue life predition of AZ31 B and AZ61 A at three strain ratios(R=-1,0,-∞), all those methods still have good performance and their predict results are within factor-of-two lines.(4) The traditional fatigue models and neural networks were evaluated based on the predict results. Result shows neural networks are better than fatigue models in predicting the life of magnesium alloy. In addition, the adventages and disadventages in fatigue models and neural networks methods were investigated.