Real-time intelligent monitoring and diagnostic system for a CNC turret lathe in a production environment using multi-sensing and neural network

This dissertation presents the results of the research investigation on the real-time intelligent monitoring and diagnostic system for a CNC turret lathe in a production environment using multi-sensing and neural network. A comprehensive review highlighting past and recent developments in sensors for untended machining is first presented. The review presents state-of-the-art machining monitoring sensors from the standpoint of the measurand, specification, characteristics, reliability, precision, signal processing, speed, and limitation. The goal being to offer applied researchers in industry and academia a source of reference. The problem of measuring cutting forces on a CNC turret lathe is addressed and a technique is developed to measure cutting tool forces (feed, radial, and tangential) with the turret allowed to index in a production-type environment. A novel system is developed to calibrate the force measurement system yielding sets of equations that establish the relationship between cutting tool forces and the transducer force output for different turret positions. A comprehensive dynamic characterization of the CNC turret lathe is carried out using modal analysis technique. Several direct and cross transfer functions are measured for the extraction of modal parameters of resonance frequencies and damping ratios and subsequently deriving the bandwidths of the turret structure, toolholder, and the lathe's spindle structure. The frequency bandwidth of the turret structure is particularly valuable for proper setting of the filtering and sampling frequencies during cutting force signal acquisition. The ANOVA analysis technique is used to study the effects of changing cutting conditions during rough and finish turning operations in a production environment on the measured cutting forces and other sensory feedback signals. ANOVA indicated a strong influence of cutting conditions on the feedback sensory signals. A backpropagation neural network is used to fuse multi-sensory feedback signals during turning of AISI 1045 (cold rolled) steel under varying cutting conditions in a production environment to predict tool wear and surface roughness. Features that correlated with tool wear and surface degradation along with cutting parameter index that accounted for machining condition variation were selected for training neural network. Effects of interactions between the varying conditions on tool wear and surface roughness are pronounced. Testing of the trained network gave good estimates under varying cutting conditions.