Neural-Network-Based Approaches for Optimization of Machining Parameters Using Small Dataset

Area of Science:

• Materials Science and Engineering
• Manufacturing Processes
• Computational Intelligence

Background:

• Surface quality is critical for machined part performance, but traditional roughness prediction methods are complex or limited.
• Arithmetic mean roughness (Ra) is a key surface quality indicator.
• Empirical models for Ra prediction are often restricted to specific machining parameters.

Purpose of the Study:

• To design and develop artificial neural networks (ANNs) for predicting arithmetic mean roughness (Ra).
• To evaluate the reliability of ANNs trained on a reduced dataset obtained via the Taguchi method.
• To optimize machining parameters for minimizing Ra in AA7075 aluminum alloy.

Main Methods:

• Experimental machining of AA7075 aluminum alloy under varied conditions.
• Application of the Taguchi method to reduce 81 full factorial experiments to 27 runs using an orthogonal array.
• Development and training of artificial neural networks (ANNs), specifically backpropagation multilayer feedforward networks with the BR algorithm.

Main Results:

• ANNs can be successfully trained using small datasets to predict arithmetic mean roughness.
• The Taguchi method effectively reduced experimental runs while maintaining data for reliable ANN training.
• Accurate prediction of Ra was achieved, enabling parameter optimization for surface finish.

Conclusions:

• Artificial neural networks are a viable and reliable tool for predicting surface roughness in machining.
• The combination of Taguchi method and ANNs offers an efficient approach for experimental design and analysis in manufacturing.
• Optimized machining parameters can significantly minimize arithmetic mean roughness, enhancing part quality.