A typical example of how the vision systems can benefit, with rapid pay back, is illustrated by a Shelton installation at this installation, where the company in question manufactures a range of fabrics with high performance PTFE membranes. The company have installed two webSPECTOR ®Plus systems consisting of three planes of cameras each looking across the full width of the fabric. Each unit is installed at the end of a production line where it is inspecting a vast range of surface types and finishes on laminated fabrics for all types of defect such as holes, slubs, and streaks. The size of defect to be detected is application dependent, but 0.5 mm is not unreasonable.

The detected faults are used to operate a tagging system, marking the faults for processes further down the production line. Data regarding fault position, type and severity is also interfaced with the factory data logging system. The completed bulk rolls are still sent to the inspection area for separating into individual rolls, but inspectors can now fast forward to the tagged faults and carry out any cutting or mending without inspecting tens of metres of perfect fabric. As a result, throughput is greatly increased.

Since the inspection takes place online, in real time the data is also used by the process operators to provide feedback of the process state. If large quantities of faults are being produced then remedial action can be taken or the process shut down, potentially cutting down on the waste produced by the plant.

The system has three planes of cameras each looking across the full width of the fabric but at different combinations of lighting and viewing angle. The multiple planes are required because different fault types become visible under different conditions. This approach is simply a parallel application of the lighting and viewing techniques applied by the manual inspectors one after the other.

The webTrainer module, which trains the system sensitivity settings for all products, works in an unsupervised fashion. It does this by assuming most of the material passing through the system is good product and then trains itself to expect that. It is also possible to input current manual inspection knowledge into the auto training module, so that it can be more or less sensitive depending on customer wishes.

For each located fault the webSPECTOR ®Plus system records multiple features such as position, size and severity along with an image of the fault. This information can be used by subsequent processes in the form of a roll map showing the roll and the position of the defects contained within it.

To prove the reliability of their vision system platforms Shelton developed the WebCorder module. This is part of a concept designed to ensure the inspection system is operating as expected and is especially useful where quality is critical and there is little tolerance on faults. The entire web can be recorded at production speeds to disk. It can then be replayed as though the material were being run through the system again.

During validation of the system’s abilities for each application, the webCorder is used to compare manual and automatic inspection results. If the manual inspectors see a fault not seen by the system, the webCorder can be moved to this position and the inspection system’s settings adjusted to ensure the fault is detected. These settings can then be incorporated into the auto-training module to ensure subsequent products are also inspected properly. It is more common that the inspection system sees genuine faults not picked up by inspectors. The webCorder is also useful in ensuring that false alarms are not picked up, where the system sees something that is not considered a real fault. By recording the material, it is possible to re-try the same product without having to re-run it through the machine. Physically re-running material often imparts more faults and can cause material to be scrapped unnecessarily.