Modeling surface roughness based on artificial neural network in mould polishing process

The mould polishing is a complex material removal process under various polishing conditions. The process parameters (polishing pressure, tool speed, feed rate, polishing times, pose angle, etc.) and material parameters (workpiece material, abrasive tool material) have effects on surface roughness. In this paper, a new surface roughness model based on artificial neural network (ANN) is presented, which consider workpiece material hardness and grit of abrasive tool. ANN model consists of three layers: input layer, hidden layer and output layer. Input layer has 7 neurons: hardness, grit, pressure, tool speed, feed rate, polishing times, surface roughness prior to polishing. Hidden layer has 12 neurons. Output layer has 1 neuron: surface roughness after polishing. The training samples are 64 and testing samples are 16. The training function is the powerful Levenberg-Marquardt (LM) algorithm. The training epoch is 29 when mean square error (MSE) is less than the goal value (3.6×10^-4). Average relative error is less than 0.05 when testing. The testing results show that surface roughness model based on ANN presents a good agreement with experimental results.

I. Introduction

Polishing technology is very critical to determine quality and performance of the final products and occupies up to 30∼50% of the whole die manufacturing time. Polishing purpose is to reduce the surface roughness to a desired amount and keep form accuracy. However, polishing is a kind of complex material removal operation, and the surface roughness after polishing is related to many factors such as machining parameters: polishing pressure, feed rate, tool speed, polishing times, grit of abrasive tool etc. There are many surface roughness models at present. The approaches are classified into as follows: