Article Preview
Top1. Introduction
One of the major concerns of any manufacturer is the accuracy of tool elements. But the desired accuracy is never achieved and this results in various kinds of errors. Sources of errors can be broadly classified into three categories: geometric, thermal and cutting force induced errors. Identification, modeling and estimation of these errors are performed in (Gong, Yuan, & Ni, 2000; Khan & Wuyi, 2010; Li, 2001a, 2001b). As far as CNC, lathe machines etc. are concerned, the main aim is to move a tool (or workpiece) along the X, Y or Z direction relative to a workpiece (or tool) in order to perform a tool-related manufacturing process such as milling, drilling, grinding etc. In our experimental set up, the lathe tool is moved only in X/Y direction assuming that the workpiece will remain in its position. Tool positioning and its verification is quite important in these applications as machining errors, controller dynamics, environmental effects and process effects deviate the tool from the desired path and this results in tool positioning error. Earlier tool positioning and its verification was done using Strain Gauge, Capacitive sensors etc. Compensation of positional error of a CNC machine tool using laser interferometry have been performed in (Barman & Sen, 2010). Various other methods for compensation of errors are discussed in (Fines & Agah, 2008; Jurkovic, Korosec, & Kopac, 2005; Kono, Matsubara, Yamaji, & Fujita, 2008; Leete, 1961; Liang, Li, Yuan, & Ni, 1997; Sze-Wei, Han-Seok, Rahman, & Watt, 2007; Turek, Jedrzejewski, & Modrzycki, 2010).
It is desired to design a machine vision system for lathe tool positioning and verification without embedding sensors. To achieve this, image processing and bio-inspired technique were employed. Motion sensors were initially embedded in order to verify the performance of developed vision system. Sequential images of the tool were captured by a single 1.2 mega pixel monochrome AVT Stingray (F-125B) CCD camera having Navitar lens with 0.25x magnification. For calibrating this camera, a calibration grid (Distortion Target 5 Freq., PN: 64-864) of Max Levy Autograph, Inc was used. Dimensions of grid are 20*20 mm, with dot diameter 0.25 mm and dot spacing 0.5 mm. This calibration grid is NIST (National Institute of Standards and Technology) traceable. Image of grid was captured using CCD camera. Total number of pixels were calculated in the grid and equated to 20 mm to find out dimension of a single pixel. Calibration factor came out to be 0.0224 mm/pixel.
Images before and after the tool movement were termed as reference and tool-moved images respectively. The least movement of the tool was 30 micron as the imaging system was unable to resolve lower motions. Tool movement was calculated using image processing technique from the captured images. It was observed that the distance traversed by the tool from embedded sensors and image processing technique did not match and therefore there was tool positional error in the vision system.
Camera calibration error, imaging setup error, image measurement error and environmental effects etc. are the key factors for this error. In order to develop a robust machine vision system, this error must be optimized so that the output of developed vision system matches with that of the motion controller. This paper proposes a hybrid version of nature inspired meta-heuristic Bat algorithm and CLONALG, a model of Artificial Immune System (AIS), to optimize these errors. It was seen that this hybrid technique is better than standalone ones as it conquers the limitations of individual algorithms without losing their leverage.