Sectors

Pharmacy

Blister control, package pick and place or sealing controls are some of the pharmaceutical processes where we commonly find the most optimal solution using machine vision systems.

vision-online-APLICACIONES-FARMACIA-control-de-blisters-y-comprimidos-en-3D

Inspection of liquid holding capsules

Due to the growing demand for capsules containing liquid drugs in order to facilitate their ingestion, there is an increasing need for a machine vision system that offers a solution to inspect and guarantee the quality of the product.

Due to the growing demand for capsules containing liquid drugs in order to facilitate their ingestion, there is an increasing need for a machine vision system that offers a solution to inspect and guarantee the quality of the product.

Image of a set of samples

Before finding the solution for the implementation of the system it is necessary to analyze the samples and decide which are the defects that should be considered and with what precision, in this case:

- Wrong reference. (1)

- Empty capsule. (2)

Identification of defects

Once the defects to be detected have been defined, a machine vision system is designed and implemented to carry out the analysis of these capsules. It has different components, including the acquisition device and the lighting, essential elements to acquire the image that must be analyzed later.

Image of the software analysis

By means of the analysis and control carried out by the system, it is possible to reduce the number of defective products that circulate in the station, consequently reducing the responsibility that falls on the operator and providing a very high and constant reliability to the inspection.

vision-online-APLICACIONES-FARMACIA-inspeccion-360-control-de-tapones

360º bottle inspection

Vision system that offers a solution that allows checking the presence, the correct positioning of the seal, as well as the inspection of the correct closure of the cap.

Machine vision system that offers a solution that allows to check the presence, the correct positioning of the seal, as well as the inspection of the correct closing of the cap.

Inspection of 360 bottles by machine vision.

To carry out the inspection of the entire bottle, several matrix cameras are required to be able to analyze in detail the 360º that the seal must cover. Together with the selected lighting, the system can detect any malformation, absence or misplacement of the seal.

By means of the customized graphic environment created, the inspection can be viewed and the reason for the anomaly detected in the product can be detected, thus guaranteeing excellent quality in the final product, all this helping to improve the company's image.

vision-online-APLICACIONES-FARMACIA-control-de-capsulas-mediante-IR-e-hyperspectral-277x300

IR and Hyperspectral Capsules Inspection

Farmaceutical-application-VisionOnline

Machine Vision Reads Dot-Print Text

Summary: Dot-print text verification is being done by an automated machine vision system on medical packaging workflow in a pharmaceutical company.

In a production line of a large and international company that manufactures pharmaceutical products as sterile dressings, there’s necessary the final products encasing into boxes which are printed with an ink-jet device. Furthermore, such boxes are marked with lot code and expiry date and as quality parameter, company has to determine (in one of the last production stages) if those codes are defective, illegible or don’t belong to current production recipe. In order to cover this requirement, pharmaceutical company along with Vision Online S.L. created an automatic vision-based system to release human operator of performing this task. Doing so, the results are an improvement in the quality company statements, a decrease in the code checking time and, an avoidance of human-made mistakes. The challenge in system development was considerably huge due to the variation in dot-print characters quality (given by the ink-jet device) was immense high either. Before means that on some occasions, character’s dots were as close between them as they seem to be made by a unique stroke. In other situations, types were so close between them that they touched in more than one point. And even, in some prints, the size of the types (height and width) changed along to the print movement. In conclusion, this point can be summarized in the fact that the variations on dot-print characters are as high as they (characters) seem to be different on their typography along the prints.

The vision-based system proposed by Vision Online S.L. is composed by a monochromatic camera, an industrial computer and a C# WPF customized app supported by Halcon Libraries. Thus, the system can identify dot-print text on medical boxes and compare it with the database’s info. The system commands the final product’s expulsion from the production line, if faulty.

As was said before, the C# WPF application works under Halcon libraries (compiled as dynamic [.dll]). Thanks to that, the inner machine vision algorithm is developed on Halcon and is supported by a pretrained OCR classifier based on convolutional neural networks. The usage of classifiers based on CNN was mandatory thanks to all variations said before. Specifically, the stages and explanation how algorithm is composed are described next.

Farmaceutical-application1-VisionOnlineFarmaceutical-application2-VisionOnlineFarmaceutical-application3-VisionOnlineFarmaceutical-application4-VisionOnline

  • First of all, the program trains a 2D matching pattern to position dynamically the OCR’s ROI.
  • When pattern (sandglass) is found, the system locates a region for the search of characters (which corresponds to the OCR).
  • After that, it’s used a high accuracy segmentation to separate individual characters and to apply the OCR pre-trained function.
  • For this case, some variables like character’s width, height and stroke (character line stroke) are considered as constant.
  • Due to an expectation of only numeric characters, the pretrained OCR classifier based on CNN is restricted to just numbers searching and to “Universal_NoRej.occ” library usage.
  • Finally, coincidence percentages higher than 97% were achieved as well as the identification of defects.

Copyright 2019 © All rights reserved