Machine vision helps make paper

From wet end to roll take-off, today's papermaking machines can be over 100m long, with paper webs of up to 11m wide running at speeds of 120km/h through the machine. A machine failure under these conditions not only represents a huge loss of production, but can also incur substantial repair costs.

A calender roll, for example - where the web is finished to a smooth, printable surface - can cost up to £120,000. Defects in the paper web, which can harm the calender as it passes through, have also been estimated to cause annual losses of over £800,000 per production line.

Because of such potential losses, the industry has long used condition monitoring techniques to avoid unplanned shutdowns. Vibration monitoring and analysis are typically used but there are failures, such as hot spots on the polymer rolls of the calender section, which have been undetectable until now.

To tackle this particular problem, under the auspices of the European Commission-supported Eutist-IMV initiative, the Finnish company Metso Paper has been working with a series of project partners to develop monitoring systems based on integrated machine vision technologies. These systems will protect the polymer rolls and monitor the moisture gradients of rotating machine elements in the wet end of paper machines.

The calender section of a paper machine comes immediately after the coating section of the machine. Its rôle is to adjust the thickness and surface qualities of the almost dry, coated web by warming and pressing it between rolls. This is in contrast to the earlier press section, where the web is also pressed through rolls but only to remove water with absorbent fabric.

Both calender and press rolls, however, can be harmed by the presence of foreign bodies in the web. These contaminants can include fibres, clay and stones that all occur naturally in paper pulp. Although mainly filtered out, any material getting through to the roll sections can cause damage.

In the calender, the polymer-covered rolls are sensitive to local temperature differences. These 'hot spots' typically arise as a consequence of dirt, coating compound or even a piece of paper attaching themselves to the roll surface. As the foreign body continuously passes through the high pressure 'nip' where two rolls are pressed together, the surface heats up locally at this point, eventually causing damage to the polymer cover, which in turn can damage adjacent rolls. Traditional monitoring methods such as vibrational analysis with acceleration sensors cannot pick up these small spots, particularly if the spot is in the middle of the roll.

The problems that foreign bodies in the web can cause to press rolls are perhaps not quite as catastrophic, but particles can enter the absorbent fabrics used in the press, reducing their operational lifetime and the efficiency of the press section. Once particles have been detected, however, they can be washed away with cleaning showers.

The Eutist-IMV THERM project is developing a low-cost infrared imaging system that has applications in both these problem areas of calender and press rolls. In conjunction with Satakunta Polytechnic Institute, Pori, Finland, Metso Paper has trialed the system at its Järvenpää pilot plant.

Infrared imaging has been used in the past to monitor calender rolls, but all it could do was measure the average temperature of bands on the rolls. Technically, continuous spot-specific imaging has proved difficult because of the high speeds involved. This entailed the use of expensive high-speed infrared cameras, several of which would be needed to monitor a single calender.

The Therm solution on the other hand uses low-cost cameras whose thermal images are converted in an external synchronisation module into a two-dimensional temperature map of the entire roll surface. Further image processing techniques then enhance the automatic software detection of spot-like objects on the surface. The radial position of hot spots can now be detected to a resolution of 30mm.

Averaging is used to obtain a higher signal-to-noise ratio from the low-cost cameras. The spot detection itself uses a residual calculation to discover the spot-like temperature differences in the image. The calculation compares the values of each pixel with the horizontal and vertical mean values at those rows and columns. The greater the difference, the higher will be the intensity of the corresponding pixel in the residual image. According to Markku Kojo, Metso's senior development engineer at Järvenpää, with the new camera system 'we will move from one-dimensional defect detection to the inspection of the whole roll surface two-dimensionally.' He sees the main benefit of the system as extending the run-time of the paper machine and the avoidance of unpredictable, time-consuming repairs.

In the wet end of the paper machine, the same synchronisation technique can also be used with Metso's own moisture measurement system called IQInsight. At the moment its development centres on monitoring the uniformity of the press fabrics by correlating temperature and moisture levels across the web. A high-end infrared camera and advanced spectrum analysis technology is needed to ensure adequate sensitivity, but costs are said to be reasonable as only one camera is required.Metso Paper is also involved in another Eutist-IMV project, called Topcapi, in collaboration with the German software company Parsytec. According to the UK's Eutist-IMV project co-ordinator, Mark Sawyer of Edinburgh University's Parallel Computing Centre (EPCC, one of the five European Eutist centres), this project involves more of a predictive approach to machine performance, rather than the condition monitoring concept behind the Therm project.

'This is a more conventional machine vision application,' he says. 'The goal is to detect foreign particles before they enter the calender section. It is very high-speed monitoring, but the concept has been proved and the next stage is to work with systems integrators to develop feedback mechanisms to enable operators to take action.'

Because of the web speed, such on-line inspection has to be able to react to coating defects within 500msec. And since 400msec are needed for any mechanical intervention, there is only a 100msec window for the detection and classification of defects. The most common of these are colour splashes, some harmless but some potentially damaging. To avoid false alarms, an accurate classification system is needed to distinguish between the two and this is what the Topcapi project is currently focusing on.

Pilot tests of a Parsytec WIS system have shown that all critical coating defects such as splashes, holes and wrinkles can be detected within the 100msec window, but the differences between tolerable and harmful defects need to be studied on each line under real production conditions in order to 'tune' the classifier. The potential for controlling coating defects in real time, however, is estimated to be an annual saving of over £600,000.