This paper introduces a new method for tool condition monitoring in transfer machining stations. The new method is developed based on a combination of wavelet transform, signal reconstruction, and the probability of threshold crossing. It consists of two parts: training and decision making. Training is aimed at determining the alarm threshold and it is done in six steps: (1) Calculate the wavelet packet transform of the sensor signals (spindle motor current) obtained from normal tool conditions. (2) Select feature wavelet packets that represent the principal components of the signals. (3) Reconstruct the signals from the feature wavelet packets (this removes the unwanted noises). (4) Calculate the statistics of the reconstructed signals. (5) Calculate the alarm thresholds based on the statistics of the reconstructed signals, and (6) Calculate the probability of the threshold crossing (the number of threshold crossing conforms a Poison distribution). The decision making is done in two steps: (1) Check the threshold crossing, and (2) Calculate the number of threshold crossing to determine whether an alarm shall be given. As demonstrated using a practical example from a drilling transfer station, the new method is effective with a success rate over 90 percent. Also, it is fast (the monitoring decision can be done in milliseconds) and cost-effective (the implementation cost shall be less than
展开▼