Intelligent automated inspection, representing the uncertainty of the real world

In many industrial process control situations the need to identify and classify product defects is key to enabling process improvements. Whilst there are many defect detection systems on the market, there are few commercial products which also provide satisfactory classification. At present it is difficult to represent the inherent uncertainty which is found in many industrial classification problems. Theoretical techniques do exist but how they should be applied is not necessarily intuitive. It is suggested there are two reasons why classification techniques are difficult to apply. First, most classification problems assume that a description of what is to be classified already exists. In many applications this is not the case. Second, the working environment of many `real world´ applications is continually open to change. The knowledge acquisition task is not a one off process since the information will vary over time. This research attempts to bridge the gap between theoretical techniques for managing uncertainty and real world applications. The difficulty in applying the theoretical techniques to real world problems shall be the focus of the following discussion