Vision system
for the inspection of cans

In the fish processing industry, even the smallest shortcoming can mean returns, complaints and image damage. Sound familiar? If so, you are not alone – most producers today are struggling with labour shortages and increasing quality demands.

For these reasons, among others, we supported one Polish producer of tinned food, salads and fish preserves in the implementation of a vision-based can seal control system.

A system that automates quality control of cans and detects defects invisible to the naked eye – with an accuracy of one-tenth of a millimetre.

Find out how we solved his problem and how we can help your business.

Tooth problem: errors detected by the can quality control system

The canned fish producer sought to increase the proportion of quality packaging among the products reaching consumers. Among the key critical canning defects were:

• teeth – dynamic, point-like deformation of the seam;
• sagging – incomplete curvature of the seam;
• dents – mechanical damage to the seam.

These defects can lead to leaking cans, resulting in spoiled contents, soiled surroundings and even loss of credibility with the end customer. The system must detect these regardless of the colour of the seam – white or coloured.

For this task, we used a modular quality control system based on HYDRA technology. The video nest monitoring the quality of the cans included five field cameras using infrared. Four, placed symmetrically at 90o intervals, map the sides of the cans and the fifth, placed above the tape, their lids.

We also installed dedicated lighting in the video socket. All cameras and lighting have been synchronised to within 1µs. The whole thing is triggered by an occupancy sensor and classification is supervised by a computer equipped with deep learning (DL) algorithms.

The above hardware configuration allows quality inspection at a maximum speed of up to 300 cans per second.

Solution used

The Scanway system supports the production line as follows:

Key features:

• water droplet blowing system;
• a non-destructive mechanical rejecter with a buffer into which only cans defined as defective (NOK) go;
• recording of detection recipes for different product references;
• generation of cyclical quality reports;
• Transparent HMI for adjusting detection sensitivity;
• archiving of NOK images with metadata.

Operating principle

The cans, go first under a blow-off system, which removes water droplets. An occupancy sensor then activates the cameras and lighting, and image data is sent to an industrial computer.

A computer, equipped with a DL algorithm, analyses the images and identifies any anomalies. Cans defined as NOK are routed through the rejector to a buffer and the data on them – along with the images – is archived. At the end of the production cycle, a quality report is generated.

Implementation challenge: water droplets

During the testing phase, we encountered a challenge: lingering water droplets that caused the cans to be classified as NOK, resulting in false positives. The problem, therefore, was not the ‘false NOKs’ themselves, but precisely the presence of water that caused them.

We dealt with this effectively by implementing a blowing system that removed drips before inspection – and by tuning the machine vision algorithm.

Effects of implementing vision-based quality control

Compared to the previous manual control, the system has proven to be more effective. Thanks to its implementation:

• We have objectified the assessment of can quality and increased its precision;
• we have significantly reduced the number of defective cans reaching the market;
• we have increased the automation of the production line;
• we have delegated manual control staff to other tasks.