Automatic gas cylinder re-test line powered by HALCON

For a company in the food industry, Swiss Safety Center developed a fully automatic machine called Autosonic™ mini+ Auto. This machine is capable of retesting gas cylinders and performs a set of inspections in a complete and fully automated process to reduce labor costs and the risk of typos, using machine vision and OCR algorithms:

• Non-compliant gas cylinders due to material defects cannot be sold because of the risk of explosion. Thus, the conformity must be ensured through appropriate tests, usually ultrasonic and visual tests. This quality control for each cylinder is periodically necessary and mandatory by law.
• Each cylinder receives a digital conformity certificate where its generic information is listed (manufacturer, serial number, dates etc.). These data must be retrieved, interpreted and stored. The challenge is that the data is engraved into a metallic and curve surface (cylinder shoulder).

The core of the automated machine is a combination of ultrasonic testing and machine vision. The result of the process is a pallet of ready-to-use cylinders with a newly engraved conformity mark.

A palletizing robot feeds and unloads the line, and a precise hydraulic head engraves new data into a free space on the cylinder shoulder after all tests have been successfully passed. The complete set of collected data from each test is stored in a database and each compliant cylinder automatically receives a new digital certificate of conformity. There are three vision stations, each consisting of a laser profile sensor (LMI Gocator). Two stations are located at the entrance and exit of the line, one station is directly in front of the engraving station. All sensors acquire 2D profile data which they combine in order to output 3D surfaces.

The machine vision software MVTec HALCON is used to process 3x 3D surface images with the following objectives:

• Reading of all engraved data and identification of the relevant data at the entrance of the machine (e.g. manufacturer, serial number, dates, some symbols etc.)
• Finding a suitable free space on the cylinder shoulder where new data can be engraved if all tests are passed.
• Rereading of the engraved data for double checking and to find redundancies including the additionally engraved data at the exit of the machine.

The processed steps with HALCON:

• Acquisition of the surface data from the sensor through the GenICam GenTL interface. They also used LMI SDK to get additional data from the sensor.
• Using of iterative dynamic filtering (Filter for Edges – Derivate Gauss) in order to rectify the original surface.
• Blob analysis and a bit of denoising in order to segment characters, then a "paint_image" to get black data on a white background.
• Shape-based matching for symbols identification in order to read a mix of about 150 alphanumeric characters and symbols. Then they subtracted those symbols from the image.
• Applying OCR classification (MLP) on the remaining alphanumeric characters.
• "Neural Nets for classifying group of related characters" was added at the end in order to apply a certain logic and syntax to the text.

Customer benefits

The Swiss Safety Center converted a manual or semi-automatic process into a completely automatic one, which generates a lot of added value for the customer:

• Minor labor costs
• Increased productivity
• Reduced risk of incorrect data (human reading & typing, data manual merging, etc.)
• All data converge in one platform (easier data management)

Text and images kindly provided by Swiss Safety Center.