A neural network for on-line tool monitoring

The practical application of a neural network to tool condition monitoring in seal-groove cutting of ball bearings is introduced and evaluated on time-domain data such as cutting forces. During the cutting process, the force signals from the turret of a lathe are detected and then the recognition of tool condition is accomplished through the rapid analysis of the neural network. A multi-layer feedforward network with the backpropagation learning algorithm has been developed to perform this task. The maximum cutting force for each workpiece is selected as the identification feature of tool breakage. Then, the features of 10 consecutive workpieces constitute the identification pattern of the tool condition. A set of simple simulated patterns together with the corresponding tool condition are used to train the network. Consequently, it has been shown that for on-line monitoring, the trained network can successfully identify tile tool breakage by using the real-time cutting force signals.