On-line robust identification of tool-wear via multi-sensor neural-network fusion

Real-time identification and monitoring of tool-wear in shop-floor environments is essential for the optimization of machining processes and the implementation of automated manufacturing systems. This paper analyzes the signals from an acoustic emission sensor and a power sensor during machining processes, and extracts a set of feature parameters that characterize the tool-wear conditions. In order to realize real-time and robust tool-wear monitoring for different cutting conditions, a sensor-integration strategy that combines the information obtained from multiple sensors (acoustic emission sensor and power sensor) with machining parameters is proposed. A neural network based on an improved backpropagation algorithm is developed, and a prototype scheme for the real-time identification of tool-wear is implemented. Experiments under different conditions have proved that a higher rate of tool-wear identification can be achieved by using the sensor integration model with a neural network. The results also indicate that neural networks provide a very effective method of implementing sensor integration for the on-line monitoring of tool abnormalities.