The system will use opto- electronic scanners positioned above the production line, linked to a computer that processes the digitised video images of the cakes and compares them with images of correctly decorated cakes stored in its database. It is capable of learning from a small sample of decorations what any particular type of cake should look like.
The prototype inspection system was developed to check the icing on Mr Kipling's Bakewell slices. These are made at Manor Bakeries in Stoke-on-Trent (part of Ranks Hovis McDougall) on a production line that rolls off more than 50 types of cake. The slices come in the form of continuous strips that are separated after baking and decorating.
Bruce Batchelor of the Machine Vision Research Group at the University of Wales says the system could be used to check the appearance of all these products. He is hoping to install a test system at Manor Bakeries later this year.
The prototype was developed under the government-sponsored Link research programme. The Ministry of Agriculture provided half the funding with the remainder coming from the industrial partners, RHM and Vision Dynamics of Hemel Hempstead.
At the moment, the quality-control inspection of Mr Kipling cakes is done manually. But as the line produces thousands of metres of cake strips each day, this becomes a demanding and tedious task.
'Food manufacturers are well aware that the appearance of food products affects sales,' said Professor Batchelor. 'There is an outstanding need for equipment that can automatically monitor quality during manufacture.'
With today's faster production lines, smaller components and the requirements of zero-defect programmes, computer vision systems are already being used for inspection tasks in such industries as car manufacture, aerospace and electronics.
Computer-controlled vision systems are attractive to the food industry because there need not be any contact with the product, but there have been very few applications. In most manufacturing, product specifications are extremely precise, but some variation is usually allowed in food. This means the computer has to be programmed to accept more than one correct product. 'This allowable variation makes it inherently more difficult to apply computer-controlled vision systems to food manufacture,' Professor Batchelor said.
'A bottle made in a mould should be exactly the same each time - and when it isn't, it gets rejected. But even with strict control of ingredients, cooking time and temperature, cakes will rise differently.'
Indeed, variation is deliberate in cake decoration, because it is thought to make the product look hand-made.
The patterns on Mr Kipling Bakewell slices are produced by a dribble bar - an arm above the production line that moves across and oscillates as it dribbles icing on to the cake. The scanner takes pictures of the surface of the cake and feeds this information to the computer, which detects any unacceptable variations. The system can then either eject cakes or alert staff that there is a problem.
In the longer term, Professor Batchelor says, the inspection system could provide feedback to other systems on a manufacturing line, getting them to alter their behaviour and so correct the fault. For example, if the decoration is wrong, it may be the belt speed needs to be changed. Machine vision systems could also be used to detect foreign bodies, wrong shapes or discoloration. They can measure linear dimensions, volume and three-dimensional shapes.
A second element of the system allows it to be adapted to inspect new cake decorations. Using artificial intelligence techniques, the program interrogates a baker about the design and then suggests which inspection software routine is closest to the new product. 'For example, a new design that uses yellow icing may be very similar to one that uses pink icing,' Professor Batchelor said.