MVTEC MERLIC
The customer produces toast from different types of grain, like whole meal and white flour, fully automatically. The packaging of the toast consists of different variants, e.g., transparent or with different colors, and varies in size. The packaging process becomes more complex as the imprint may change within a short period.
Incorrectly packed toast packages lead to disruptions in the production process. The sporadic control checks by the employees are not effective enough to detect packaging errors to a sufficient extent at an early stage. The wrongly packaged products can then not be gripped by a robot arm during further processing, for example. This leads to an uncontrolled - and undesired - production stop.
The errors that occur are:
To improve process stability, production was to be modified and automated by using industrial image processing. The products are to be differentiated into "bad products" and "good products". "Good products" are all toast packages that are not defective and are classified as "buyable" by the customer.
The system should also work for different product sizes, packaging, and toast types without having to adapt the algorithm. "Bad products" should then finally be ejected via a separate pneumatic system.
A machine vision automated quality inspection was set up with the following components:
By means of hardware triggers, images are taken by two cameras from the top and from a transverse side of the toast package. The two image sources can be managed and configured easily and conveniently using the Image Source Manager integrated
in MERLIC 5.
Afterwards, the captured images are read in, labeled, trained, and evaluated in the free software MVTec Deep Learning Tool.
The tool "Classify Image" in MERLIC subsequently evaluates the quality of the toast packages and - if necessary - rejects them. Not only is a distinction made between "bad products" and "good products", but the specific defect class is also determined, based on the above-mentioned defects. This makes it possible to show in detail the type and frequency of errors that occur.
This classification was performed using the new Deep Learning feature in MERLIC 5. Due to the flexibility of MERLIC 5, the company was able to implement an IO map through a self-developed communicator plugin. In addition, MERLIC's front-end can quickly and clearly show users a visual representation of the toast package and highlight relevant image areas via heatmap.
MOREOVER: The implementation of the machine vision solution enabled the documentation and control of production errors. This allows quality management to take targeted measures to prevent production errors.
Published on: October 27, 2021
