| 摘要 |
<p>A system for predicting the life of a cutting tool in an automated metal cutting machine, comprises a plurality of sensors for measuring physical data (601) relating to a cutting operation of said machine, and a computer connected to these sensors. The computer comprises a feature extractor (603) for transforming the physical data into feature values, and a tool life predictor (605) for arriving at a prediction of the life of the cutting tool. The tool life predictor has an influence diagram having an input level of feature nodes, intermediate nodes connected to the feature nodes and to a machining class node, and an output node connected to the intermediate nodes. The tool life predicator further comprises a first module for computing averages of feature values for each sensor, a second module responsive to the feature averages for making a classification of the feature average as indicative of a long life or a short life, and a third module connected to the influence diagram for combining the classifications to arrive at a classification and a conditional probability at each intermediate node and a classification at the output node which is the predicted life of the cutting tool. Furthermore, a method to enable a real-time expert system to learn is provided which comprises the steps of creating a relational structure corresponding to the influences that a first and a second variable have on a third variable, obtaining data for these variables over some duration of time, determining a tolerance zone around a sample which divides classes and using a dynamic distance classifier, which is responsive to the tolerance zone, to determine decision boundaries for the first and second variables and probabilities associated with the decision boundaries. The probabilities indicate the confidence with which a decision boundary classifies values of the first and second variables with respect to the value of the third variable. <IMAGE></p> |
