Using neural networks to model the blanking process

Abstract

This research investigates the use of neural networks (NN) to model the blanking process for predicting, blanking parameters by analysing acoustic emission (AE) and force/displacement waves. Suitable AE sensors, digital signal processing (DSP) hardware and acquisition software were used to measure the AE signals in blanking, together with corresponding force/displacement signals. The key features of the AE wave and the force/displacement signal were extracted, using specialised software. These were the basic inputs for the NN model that was used to predict changes in blanking parameters such as the die–edge profile, the punch–edge profile and the punch–die from AE waves and force/displacement signals. The AE waves were acquired simultaneously with the force/displacement curve for the entire blanking cycle. More than 100 NN models were built with different inputs and NN types, one for each blanking parameter. Various NN techniques and parameters were tried. Regression correlation tests were used to screen the potential NN models for further development and testing using actual experimental data. Some of the NN models predicted the prevailing blanking parameters with sufficient accuracy. The best results were obtained from hardness and clearance models (the average error was less than 5%). The aim of the research was to develop a model using NN and to monitor the pattern of variation in the blanking of metal, with a view to detecting variances in the performance of the machine and in the product This would facilitate the development of an integrated, real-time, expert model using NN technology to control and monitor the blanking process.