A New Fault Classification Model For Prognosis And Diagnosis In CNC Machine

Process safety and reliability are critically important for large-scale and highly integrated modern industry. Faults can cause long-term disturbances, influence normal operation, and may even lead to large economical loss and ecological crisis. This thesis presents a new fault classification model and an integrated approach to fault diagnosis which involves the combination of ideas of Neuro-fuzzy Networks (NF), Dynamic Bayesian Networks (DBN) and Particle Filtering (PF) algorithm on single platform. In the new model we categorize faults in two aspects, namely first and second degree faults. First degree faults are instantaneous in nature and second degree faults are evolutional and appear as a developing phenomenon which start from an initial stage, graduates through development stage and finally ends at a mature stage. These categories of fault have a lifetime which is inversely proportional a machine tool life according to modified version of Taylor’s equation.For fault diagnosis, our framework consists of two phases:the first focusing on fault prognosis which is done online and the second dwelling on fault diagnosis which depends on both off-line and on-line modules. On the first phase a neuro-fuzzy predictor is used take a decision. The decision determines whether to embark Conditional Based Maintenance (CMB) or fault diagnosis depending on the magnitude of a fault. The second phase only comes into action when an evolving fault goes beyond a critical threshold limit called CBM limit. And if the CBM limit is exceeded a command will be issued for fault diagnosis to commence. During this phase DBN and PF techniques are used as intelligent fault diagnostic systems to determine the magnitude, time and location of the fault. The feasibility of the approach has been tested in a simulation environment using CNC machine as a case study and the results are studied and analyzed.