A HYBRID APPROACH FOR ROBUST DIAGNOSTICS OF CUTTING TOOLS

A new multisensor based hybrid technique has been developed for robust diagnosis of cutting tools. The technique combines the concepts of pa ttern classification and real-time knowledge based systems (RTKBS) and draws upon their strengths; learning facility in the case of pattern classification and a higher level of reasoning in the case of RTKBS. I t eliminates some of their major drawbacks: false alarms or delayed/la ck of diagnosis in case of pattern classification and tedious knowledg e base generation in case of RTKBS. It utilizes a dynamic distance cla ssifier, developed upon a new separability criterion and a new definit ion of robust diagnosis for achieving these benefits. The promise of t his technique has been proven concretely through an on-line diagnosis of drill wear. Its suitability for practical implementation is substan tiated by the use of practical, inexpensive, machine-mounted sensors a nd low-cost delivery systems.