Abstract

Glass bottle forming control: on-line measurement of gob position and shape using computer vision

Bruno David - 4 november 1997.

Nowadays, entirely automated machines produce 150 to 200 glass bottles per minute. Any change in bottle shape, machine parts or technical failure requires human intervention. So, the decrease in machine pauses makes it necessary to quantify parameters that until today were only qualitatively estimated by the operators, and to formalize the considerable practical know-how gained by the glass-workers.

The forming of a glass bottle involves a gob of molten glass to be fed into a blank mould before blowing. A gob is defined as a cylindrical piece of molten glass extruded and then cut from the furnace feeder (its temperature at extrusion reaches 1200°C). An assembly of chutes - the delivery system - along which the gob slides under the effect of its own weight enables this transfer.
During its travel from the feeder to the mould, the gob undergoes some deformation (for instance we have measured that its length is roughly multiplied by 2 during the transfer) which can affect the penetration into the mould, called the loading, and can lead to poor quality bottles. For instance a gob arriving off-centre in the mould can cause a thinner or thicker bottle wall than specified.
The great velocity of the gob at the moment it enters the mould (~7 m/s) makes it invisible to the naked eye and the first step towards the control of the loading requires a sensing system able to catch an image of the gob.

In this context, we have equipped a production line with a computer vision sensing system to measure, in real-time, the shape of the gob and its position with respect to the mould axis. The system is based on linear vision, a technology which is well suited to high speed inspection of moving objects : it is composed of a personal computer system, a line scan camera (with a 2048 element linear array) mounted on a computer controlled pan-tilt unit and an acquisition board performing line acquisition, real-time processing and data compression.

Image acquisition :
To initiate the acquisition when a gob passes in front of the camera, we do not have any external signal at our disposal. So we use a template match circuit provided by the acquisition board to detect, in real-time, changes in gray-scale values and this automatically triggers line acquisition. To speed up data transfer between the board and the host computer memory and to reduce storage requirements, we use real-time binary compression also provided by the acquisition board.

Shape measurement :
When the binary image of the gob has been transfered into the host computer memory, we process it to determine some morphological parameters such as area, perimeter, principal axis,...
In order to capture the essence of the structural shape of the gob, we compute its skeleton and we approximate it by a specified number of polygonal segments by means of a polygonal approximation.

Position measurement :
To measure the position of the gob with respect to the mould axis, we use an off-line calibration procedure to find the position of the mould relative to the linear vision system placed a few meters away. The calibration procedure uses a calibration plate implanted in the mould which requires production to be stopped beforehand.
In order to reduce human intervention and to speed up the calibration procedure, we have developed an auto-calibration procedure. The camera is mounted on a computer controlled pan-tilt unit and it positions itself relative to the calibration plate using a visual servoing procedure.

Data processing :
For each gob, the image is displayed on the screen of the PC and all the data related to position and shape information is overlaid at the same time as it is stored in a data base.

This work has been carried out within a more general study of the gob transfer through the delivery system, based on a simulation of gob deformation. The simulation has been validated through a comparison with production line measurements performed with the vision system presented above.
This vision system plays a key role in the control of the loading and more generally in the optimization of the forming process.