Sintering of zirconia toughened alumina biomaterials: development of an artificial neural network model

Abstract

Higher hardness, low wear rate together with excellent biocompatibility makes ceramics suitable for biomaterial applications like total hip arthroplasty (THA). Pure alumina hip prostheses have been used for more than 30 years and have dominated the history of ceramic hip implants. At present Zirconia Toughened Alumina (ZTA) is becoming more popular than pure Alumina in these applications due to its higher fracture toughness values. The sintering behaviour of ceramic materials is generally complex and one cannot easily describe it by some linear relationships. Sintering parameters like temperature and time greatly influence the grain growth and densification of ceramic materials. For ceramic composite like ZTA it is really perplexing to determine the optimum sintering parameters to get the desired properties. The situation is even worse when nano size particles are used. It requires an awful lot of experiments to be carried out to optimise the process. Interest in simulation which replaces difficult, expensive and time consuming experiments has recently been increased. By virtue of the simulation, we can predict and verify material behaviour that is hard to obtain by experiment. Also, using simulation we can get virtual results otherwise obtained from experiments carried out at high temperature. Artificial neural networks (ANN), or neural networks (NN) for short, are relatively new computational tools and their inherent ability to learn and recognize highly non-linear and complex relationships makes them ideally suited in solving a wide range of complex realworld problems. In this research work it was aimed to develop an ANN model that can predict the relationship between sintering parameters and properties of zirconia-toughened alumina ceramic nano composites. In this work ZTA with 5-20 volume % of ZrO2 was prepared by conventional pressure less sintering technique at 1500-1650°C for 4-12 hours. Some pure alumina samples were also prepared for comparison with ZTA. The starting materials were nano-sized alumina and zirconia powders of high purity. Density, Grain size and Vickers hardness and Fracture toughness of the samples were determined. About 95-98% of theoretical density was achieved. The Vickers hardness value above 20 GPa was also found for ZTA with higher density. The fracture toughness value of 8.2 MPa m was obtained for ZTA containing high amount of ZrO2. The variation in density, hardness and alumina grain size with sintering temperature and time was observed. The addition of ZrO2 in alumina was found to have no effect on density, while the growth of alumina grain size was found to be restricted by ZrO2 addition. Artificial Neural Network toolbox of MatLab 7.4.0 was used to develop an ANN model. The effects of sintering temperature, holding time and ZrO2 content on density, hardness and alumina grain were tried to be predicted with the ANN model. 43 sets of experimental data were used to train a feedforward back propagation algorithm. Gradient decent learning rule was used in this study. After training, 5 sets of test data, which were unknown to the network, were used to evaluate the generalization ability of the network. It was found that the developed model, noted as 3-18-3 network, with 18 neurons in the hidden layer was found to have the best performance and was able to predict the above mentioned property with a high degree of accuracy (max. R2 = 0.9999 for training and 0.98 for testing). The functions ‘tansigmoid’ and ‘purelin’ were used as the transfer functions in the hidden and output layer respectively.