MULTI-SENSOR CONDITION MONITORING USING SPIKING NEURON NETWORKS Rui G. Silva Universidade Lusíada de VN Famalicão Largo Tinoco de Sousa, 4760-108 Vila Nova de Famalicão, Portugal rsilva@fam.ulusiada.pt Steven Wilcox University of Glamorgan, School of Technology Pontypridd, Wales UK CF37 1DL, UK sjwilcox@glam.ac.uk António A.M.M. Araújo Universidade Lusíada de VN Famalicão Largo Tinoco de Sousa, 4760-108 Vila Nova de Famalicão, Portugal ajmm.araujo@gmail.com ABSTRACT The paper presents an intelligent system for on-line monitoring of the cutting process. The monitoring apparatus is developed both in hardware and software. The system is based on a PC which is connected to a set of sensors, via a data acquisition card, for on-line post-processing and classification. The proposed monitoring system takes advantage of most attractive features of neural networks, such as abstraction of hardly accessible knowledge and generalisation from distorted sensor signals, to give a reliable prediction on tool condition. It consists of six components: data collection, feature extraction, multi-sensor integration, pattern recognition, tool wear estimation, and outlier detection. The proposed architecture has a built-in Self Organizing neural architecture component based on Spiking Neurons and it is demonstrated that these computational architectures have a greater potential to unveil embedded information in tool wear monitoring data sets, and that smaller structures, compared to sigmoidal neural networks, are needed to capture and model the inherent complexity embedded in tool wear monitoring data. KEYWORDS Spiking Neuron Networks; Machining; Condition Monitoring; Tool Wear. 1. INTRODUCTION Manufacturing industries and their customers are now demanding substantial increase in flexibility, productivity, and reliability from process machines, as well as increased quality and value of their products. Many condition monitoring systems have been used to supervise the state of different industrial processes. A condition monitoring system can be viewed as serving the following purposes: advanced machine and process fault detection; verification and protection of machine and process stability; maintenance of process tolerances by providing a compensatory method; protection of machine and process failure. Several factors have retarded advances in the development of condition monitoring systems, including inappropriate choice of sensor signals and their utilization, and their inability to perform robustly in noisy environments. Artificial neural networks of sigmoidal and McCulloch-Pitts neurons have found increasing support in industry research (Sick, 2002) because of their most attractive features, i.e. abstraction of hardly accessible knowledge and generalisation from distorted sensor signals. In recent years, accumulated experimental evidence has suggested that biological neural networks, which communicate through spikes, use the timing of these spikes to encode and compute information in a more efficient way, providing to the new neural network