Robotic-Aided Inspection of Bulky; Problem Ceramics

Many manufactured items lend themselves readily to automated handling and inspection, which is why such techniques are making such speedy inroads into factories worldwide. Equally, there are some inspection tasks that are extremely difficult to carry out, because of the size, weight or shape of the manufactured goods concerned. If the efficiency benefits of robotic handling and inspection are to be made available to all manufacturers, thereís still a substantial amount of improvement needed.

A classic example of such difficult workpieces concerns non-flat ceramic products: toilet bowls, bidets, washbowls, and so on, which are a major item of commerce. Why are they so difficult to integrate into modern manufacturing? They are heavy, fragile, have surfaces that are easily damaged; they are three-dimensional with elaborate curved surfaces, and they have highly reflective, glazed surfaces that offer serious problems to machine vision systems. Product designs change frequently, because of fashion trends in home decoration. The goods also have high quality standards, and tiny flaws must be spotted reliably.

A recent case study of a project in Portugal shows how robotics can be drafted to assist a human--manned inspection line for such goods, and increase throughput quite considerably. The project study is by J. Pires, a mechanical engineer at Coimbra University, working with Sergio Paulo of Roca Comercio e Ceramica SA (Leiria, Portugal). It involved the new ABB S4 C+ robot controller, a distributed system running on several networked PCs in the production area. The ceramic pieces come direct from the plantís high-temperature ovens on pallets moved by forklift equipment.

Manual inspection is carried out on two parallel inspection lines, 15 m long, with rejects being removed. At the far end of the line, the pieces have to be re-palletised before entering the distribution system. There are several constraints on the pallet loads: there are four levels per input pallet and 8 pieces per level, in a special order to keep the pallet in equilibrium, while the output picture is different, with five levels. A good deal of switching of goods is involved and all have to be protected by hard paper interlayers.

The idea of the automation project is to spend no more than 12 seconds per piece, to accept errors in pallet loading at the input side (itís done manually), and to tolerate some non-error dimensional variations (plus/minus 1 cm) in the items. Finally, the system needs to work on a 24-hour basis, and involve little operator parameterisation, while still being Ďteachableí about new designs. The design of the handling equipment turns out to be fairly complicated and in the case study involved one PC per robot, although thatís not strictly necessary.

The system was designed from scratch, using none of the common commercial software packages. ActiveX controls are used to provide a visual interface and the PCs used the familiar PCROB package for this purpose. The system provides simple messages about fault conditions such as missing pieces, incorrect pallets and unknown models. At the back end of the line, barcodes are attached, so the reloading of pallets is relatively simple.

Further improvements are planned, based on operational experience. ABB Robotics Portugal provided some assistance in the work. Itís described in Robotics & Computer-Integrated Manufacturing, Vol. 21 (2005), pp.11-17.


J. Norberto Pires, Departamento de Engenharia Mecanica (CEMUC)

Universidade de Coimbra, Pinhal de Marrocos,

3030 Coimbra, Portugal

Phone: +351-239-790700


E- mail:

To comment on this article, write to us at

To find out more about Technical Insights and our Alerts, subscriptions and research services, access