Development tools speed machine vision design

Edited by the Engineeringtalk editorial team Jan 10, 2005

The Shelton WebSpector makes uses software development tools to solve customised machine vision requirements.

Industry now benefits from automated computer based inspection or "machine vision" as it is known.

Many vision inspection systems require extensive software development to perform bespoke inspection tasks and have long setup or training times.

To overcome the need for extensive software development timescales, the Shelton WebSpector makes use of a great deal of software development tools for rapid development of clients' customised requirements.

This flexibility was recently demonstrated when a client asked for an exclusion subroutine to be developed to exclude identification details printed on material that would otherwise be shown as a fault.

This is an issue that many manufacturers will face when they print product and company identification onto their materials.

Imagine you have an identification strip with several lanes of printing running down stretchable material.

The range of texts and font sizes can be considerable, even within the same identification strip.

The WebSpector has to isolate each portion of text, and inspect around it, without detecting the text itself, and distinguishing between faults that look very similar to the text.

Some faults can be long faint streaks that may cross the text diagonally or run along side it.

The WebSpector has to be programmed to recognise text from real faults, otherwise inspection would not be possible on such materials.

Shelton's range of WebSpector vision systems have overcome the need for extended setup times.

The WebSpector 's powerful automatic training and web-recording systems, a patent protected feature, developed onto the PC based image processors massively enhance the inspection accuracy and setup efficiency.

Some machine vision applications, such as high quality finished or technical textiles, can involve a large product range with varying textures, colours and densities.

The faults can be varied and quite difficult to spot, even for human operators.

This combination of subtle faults and wide variety of products means that training the system could be a difficult task.

The fault subtleties mean that a large number of nontrivial image process algorithms need to be deployed - up to 50 on a textile inspection model.

Each of these algorithms needs precise setup, referencing complex sensitivity parameters, which can include: expected quality, yield considerations and customer expectations.

The power of the automatic training and web-recording systems has been demonstrated at WL Gore UK and Associates where it is being used for training the system to inspect several thousands of product styles.

WL Gore UK and Associates uses the vision system to inspect high quality PTFE laminate products as they come off the production line.

Any faults are tagged automatically, thus enabling inspectors to move quickly from tag to tag without spending time looking at perfect material.

At each tag the inspectors assess the significance of each fault and carry out any cutting or mending operations.

The vision system automatically trains itself adjusting its settings for new styles that it has never seen before, without manual supervision, incorporating known quality requirements into its decisions.

A database stores details of each product, as well as its trained or "validity" status.

When a new material is first presented to the system it records a sample, trains itself, and then inspects it properly.

The webCorder, a component part of the automated training system, allows the replaying of entire work-orders where the material viewed from all camera planes has been recorded to disc.

By comparing inspection results from the recordings and manual inspection data the system performance can be accurately verified.

Once the replaying gives the required level of inspection, the settings can be incorporated into the auto-training module, which applies them to similar fabrics.

Stuart Speake of WL Gore said: "The software is user-friendly and the fault tracking and statistics generator are useful because they enable us to control quality very tightly".

"The error mapping allows us to look at and analyse exactly what is happening in our production process".

"It takes out the subjectivity and deskills the process".