Microdrills have a short and unpredictable tool life. It is also very difficult to monitor the condition of the microdrill tip with the unaided eye. A new approach is proposed to detect failure of a microdrill with a 0.39 mm diameter. The proposed method measures the thrust force and encodes the data by calculating four consecutive averages and standard deviations during each drilling operation, which takes four to seven seconds. The Restricted Coulomb Energy (RCE) type neural networks are used for classification of the signals after encoding. The training period of RCE neural networks is very short compared to backpropogation neural networks and can indicate if their previous training is satisfactory to identify the presented signals. The proposed approach was tested by using microdrills with a 0.39 mm diameter. After training the neural networks on the experimental data, less than 10% error was observed on the test data during identification of normal and poor operating conditions (just before breakage, breakage, and drilling with a broken tool).
