Simulated Annealing Based Optimised Neural Networks for Manufacturing Process Modelling

International Journal of Artificial Intelligence and Knowledge Discovery

Sudipto Chaki¹ 

[1] Associate professor, Dept. of Automobile Engineering MCKV Institute of Engineering, Liluah, Howrah-711204, West Bengal, India

关键词: Artificial Neural Networks (ANN);    Simulated Annealing (SA);    Genetic Algorithm (GA);    Hybrid model;    optimisation;    LASOX cutting;   

DOI  :  

学科分类：建筑学

来源: RG Education Society

【 摘 要 】

In the present era of computer based automation in manufacturing industry, Artificial Neural Networks (ANN) have been widely applied for process modelling and prediction of process parameters. But best prediction performance of ANN can only be achieved through optimisation of the ANN architecture and training parameters. In absence of a theoretical formula connecting ANN parameters with prediction performance, present work proposes a model of simulated annealing(SA) and ANN (SA-ANN) to determine optimised ANN parameters like hidden layer neurons, learning rate and momentum coefficient for minimum prediction error (MSE). Performance of the model has been validated using an experimental dataset on Laser Assisted Oxygen (LASOX) cutting process. Performance of SA-ANN model is compared with a Genetic Algorithm (GA)-ANN hybrid model. Minimum MSE obtained by SA-ANN (0.0019) model is smaller compared to GA-ANN model (0.0024). It is observed that, during prediction with optimised SA-ANN model, Mean absolute % error for SA-ANN is below 6% only, compared to around 7% error of optimised GA-ANN model. More over, optimisation time of the SA-ANN model is almost 15 times less than GA-ANN model. Therefore, relatively better prediction capability with negligible optimisation time compared to GA-ANN has made SA-ANN as a powerful tool for optimisation of ANN architecture.

【 授权许可】

Unknown