A numeric/symbolic approach to machine tool supervision

摘要

This work describes the design and evaluation of a sensor-based computer-assisted supervision system, which detects and identifies totally broken, chipped or worn cutters.;A new real-time algorithm (revolution-oriented residual processing algorithm--RORPA) has been developed, which extracts features from x-axis and y-axis displacement sensors attached to the headstock of milling machines. In the raw sensory channel, the algorithm can separate the damage feature from the run-out effect and other reference signals. The RORPA enhances the features of tool breakage, even in the presence of noise, transient and spindle run-out effects.;In addition, a knowledge-based supervision model is developed and applied to enable robust, accurate automated decisions. The model hierarchically integrates the RORPA in a knowledge-based processing environment where rules and objects co-exist. The numeric-symbolic supervision model, which incorporates physical models and empirical knowledge, consists of four tasks: expectation generation, operation monitoring, situation assessment and action execution. The model was developed not only to address the tool breakage problem, but also to include other, multi-sensor, information as will be required for machine tool supervision.;The two aspects of the supervision model have been implemented in the Texas Instruments Explorer I, a Lisp workstation. In the symbolic environment the object-oriented representations implementing the expectation mechanism and the operation monitoring are written in Flavors, an object-oriented programming facility. The situation assessment uses a rule-based, forward-chaining scheme. The RORPA has been implemented in a multiprocessor board, which consists of four TMS32020 digital signal processors interfaced to the NuBus of the Explorer.;The evaluation of the developed supervision system was performed using a variety of real data cases (194 cases). The overall results indicate that all 80 damaged cases, which included totally broken, chipped or worn cutters, were recognized without a missed-detection. On the other hand, there were only 5 false alarm cases out of the 114 sets of test data from undamaged cutters.