Adaptive Self-Organizing Map Applied to Lathe Tool Condition Monitoring

Condition monitoring is a fundamental part of machining as well as other manufacturing processes where generally there are parts that wear out and have to be replaced. In turning or milling tool wear level is fundamental for proper machine operation and is often accessed through the use of multiple sensors combined with a reasoning method. The unique modeling capabilities of artificial neural networks in the presence of noisy information make them best candidates for condition monitoring. This paper presents enhancements to the self organizing map neural network where a modified learning algorithm is used to control the effectiveness of memory usage through the reduction of the learning rate in raw memory areas. Further modeling ability tests of the neural network were produced to test the influence of time as a feature in classification performance. Results show significant improvements in tool wear classification through the use of time as a feature as well as gained performance with memory learning rate control. The modified learning algorithm suggest that online continuous learning is possible by avoiding over fitting from previously seen data suggesting therefore the accommodation of new data patterns without forgetting previous ones.