Decision Making under Uncertainty in Tool Condition Monitoring

Tool Condition Monitoring is a field which sees a significant divergence between the published research and current industrial practice. It has been proposed that the principal reason for this divergence is the inability of laboratory-developed systems to display sufficient robustness in coping with the diverse range of stochastic influences that are found in industrial deployment – ranging from operator influence to variations in the properties of tools and workpieces used. Furthermore, several studies have shown that a majority of tool condition monitoring systems purchased by industrial concerns are switched off. Principal among the reasons given is the number of ‘false alarms’ resulting in costly system downtime. This work considers the current trends in tool condition monitoring research, in parallel with the requirements of industry, and argues that probabilistic rather than binary decision making systems are required in the next generation of tool condition monitoring systems, if the poor industrial acceptance of the previous generation systems is to be countered. Examples are considered from industry and used to illustrate a suggested approach to both generating and assessing such predictions. These examples explore the range of decision making criteria under which such systems are required to operate – from minimization of downtime to the protection of costly capital equipment or customer resources.