When shoppers browse supermarket shelves there is nothing more likely to put them off buying products as poorly presented merchandise. It’s a fact that is not lost on Spice Hunter, a leading manufacturer of gourmet spice, which produces around 45,000 jars a day, ranging from nutmeg to chilli powder.
The company’s commitment to producing a high-quality product is matched by its desire to improve the aesthetic quality and presentation of its distinctive square-shaped jars.
To this end the company, which in the past has relied upon manual inspection to detect packaging flaws, has invested in a machine vision system to meet its packaging quality goals. ‘How our products appear is extremely important to us, and this year we have done a lot of work to figure out how to give our jars the best, most consistent shelf appearance,’ says Tim Ewen, quality manager at Spice Hunter.
‘More specifically, we needed a way to ensure that the labels are on straight, the caps are properly applied, and the freshness seal on the cap is present and intact. We had to get as close as possible to zero defects, and quickly realised during our research that machine vision technology was the way forward,’ says Ewen.The task of selecting a vision product was handed to Pete Priebe, a Spice Hunter manufacturing engineer responsible for implementing and maintaining the plant’s automation equipment. According to Priebe, price was the first consideration.
‘We had never used machine vision in the plant before, and wanted to bring in the technology at a relatively low price. So we looked at products in the $5,000 (£3,500) range, and that’s when we discovered the In-Sight vision sensor from Cognex. It seemed to offer the right tools for the job – and at the right price.’
Priebe contacted the company to get an application evaluation under way. According to Christina Fein, the Cognex applications engineer assigned to the evaluation, the vision sensor was not only less expensive but also seemed ideal for the application. ‘When I first spoke to Spice Hunter the firm intuitively felt that since it wanted to do multiple inspections, it needed a vision system with multiple cameras. I explained that the company could perform the label, cap and freshness seal inspections all at once with In-Sight, which is a single-camera product,’ says Fein.
She says the jars presented a number of challenges for the vision sensor to overcome in order to inspect reliably. ‘At fairly high speeds the sensor had to deal with labels that were glossy, glass bottles and jars, and caps that were shiny. Spice Hunter also indicated that the ambient lighting in the factory can vary considerably because of the windows. For the evaluation I received a variety of jar samples that Spice Hunter might actually see during production – not just good and bad jars, but marginal as well since marginal parts are where most vision systems fail.’
Having proved capable of performing all three inspections during evaluation, a vision sensor was purchased and installed on the production line. Having a basic understanding of conventional spreadsheets, Priebe was able to use In-Sight’s vision spreadsheet environment to set up the application. This involved selecting vision tools and parameters from drop-down menus using a video game-like handheld control pad. The spreadsheet then automatically generates tool results into worksheet cells, which can easily be linked together to set up the label, cap and freshness seal inspections. The spreadsheet is transparent, so that Priebe is able to see a reference image of a spice jar, and the vision tools being applied to the image, without having to switch between separate screens.
On the production line the vision sensor’s digital camera is mounted perpendicularly to the jars at a distance of approximately 18in. The camera connects to an industrial-hardened vision processor which links directly to a standard VGA monitor. Because the vision processor is a standalone unit, it does not require the use of a PC during set-up or run time.
As the process begins, empty glass jars are first moved from boxes on to a turntable, which rotates them on to a feeding conveyor. The jars move through various stations where they are filled, banded, capped and labelled. Once they leave the labeller, the jars move towards the inspection point spaced 2.5in apart.
To achieve the highest image contrast, a bright white board is placed directly behind the inspection point, and basic front lighting is used to illuminate each jar as it comes into view. When a jar arrives at the inspection point, a sensor triggers the camera to capture an image of the moving jar and the image is instantly transmitted to the In-Sight processor.
The image is then processed and analysed using a variety of vision software tools. When setting up the application Priebe needed to pick two features common to all jars as reference points from which all measurements would be made. The features that seemed to work best were the leading edge of the jar and the cap, as those are in the same place no matter what type of jar is being run.
If a jar fails any of the inspections, the vision sensor sends a reject signal to a PLC, which then triggers a pneumatic gate to kick the jar off the line into a reject bin. These jars are washed and re-queued back into the line. Jars that pass proceed down the line to a bundling station, where they are packaged up and shipped.
In addition to performing the inspections at required line speeds, the vision sensor helps improve process control. If several label or cap problems in a row are detected, for example, this may alert line operators that the labeller or capping machine need to be adjusted. Operators can then stop the line and fix the problem before additional defects occur. The sensor also records pass/fail data for each jar inspected, including data on the types of defects found, providing quality engineers with more information about the process.
Ultimately, Priebe anticipates that the vision sensor will be able to link interactively with other equipment when problems occur. ‘If the vision sensor sees a trend in labels that are drifting too far to the right or left, we want it to be able to tell the labeller to move things in the other direction automatically without any operator intervention,’ he explains.