Data Refinement and Analysis of Cutting Force Signals for the Improvement of Chip Form Identification Accuracy

Dealing with sensor monitoring and control of machining operations and getting insight details and inferences from sensor signal features have always been challanging tasks. A key issue to obtain inferences from a sensor signal is the achievement of clear classification and clustering of the signal characteristic parameters. This paper reports on a Self-Organising Map (SOM) based approach for classification and identification of chip form utilising cutting force sensor signals obtained from turning operations. SOM neural networks (NN) provide for an unsupervised NN methodology used here to classify the chip form features in clearly separate clusters. This approach supports Kohonen maps based visualisation of chip form data clusters. The SOM NN performance in chip form identification was examined by considering either single cutting force components separately or all cutting force components together. Moreover, SOM maps were utilised as a powerful tool to identify ambiguous data vectors lying in two or more chip form clusters. Identification of ambiguous data samples plays a crucial role in refinement and analysis of data sets by providing additional inferences about sensitive points of cluster overlapping.