Automating Refresco's can picking lines with deep learning
- Productivity of packing line significantly improved
- Enhanced reliability and 24/7 operation
- Customer implemented an additional 30-40 SKUs
- Ability to analyse repetitive faults and conduct root cause analysis
Eliminating human error from pallet inspection
The challenge
Bytronic Vision Intelligence was approached by a leading beverage manufacturing company to automate a manual inspection in its can packing lines.
The primary objective was to efficiently layer 12 packs of Monster Energy cans into 4×6 and 4×4 layers to be palletised.
The project faced specific challenges, including occasional twisted packs and packs being missed, resulting in incorrectly laid pallets. In this event it takes 15-20 minutes for an operator to clear out the line and remove the incorrect layer from the pallet.
To address these problems, Refresco first attempted a remote inspection system, through a CCTV camera and an operator watching from a distance. However, operators would occasionally miss an error, so Refresco approached Bytronic to install an innovative vision system.
The solution
Design and technical decisions
During development, Bytronic made several design and technical decisions to ensure the success of the project.
To capture a comprehensive view of the packs, the camera was mounted 3.5 metres above the packing line.
Strobing lighting is unpleasant for operators and also causes a risk of epilepsy, therefore flood lighting was used. A liquid lens was selected to facilitate easy focusing and provide flexibility for future applications. Integration of the vision system was achieved through EtherNET/IP communication to the line PLC.
The Cognex In-Sight 2800 camera was chosen due to its 1.6MP sensor, liquid lens, and, most importantly, the easy to use “Classify” tool driven by edge learning.
Edge learning is a form of artificial intelligence where the processing takes place on-device where the data originates. Due to this, the In-Sight 2800 offers a cost effective solution that is easy and quick to set up.
Implementation steps
The project started with an on-site proof of concept, where Bytronic rigged a camera and light on the packing machine to capture and process images.
Subsequently, a comprehensive desktop exercise was conducted at Innovation Campus to determine the exact technical specifications required.
Throughout the development process, collaboration between Bytronic and the customer played a crucial role, ensuring design approval and obtaining necessary access to the production line for the proof of concept.
Adding an additional 30-40 SKUs and improving productivity
The result
The implementation of the vision system yielded remarkable success.
Bytronic successfully implemented the automation for the initial SKUs, and, thanks to training from Bytronic, there is potential to implement an additional 30-40 SKUs.
The implementation of the vision system had a substantial impact on the can packing line. With enhanced reliability and the benefit of deep learning, the productivity of the line has significantly improved.
24/7 operation and the ability to analyse repetitive faults and conduct root cause analysis has empowered the customer to implement continuous process improvements and explore further preventative measures.
Based on the success of this application, it is evident that automation tasks utilising the In-Sight 2800 vision system hold vast potential across various industries. Bytronic’s engineers emphasised the significant potential for leveraging the vision system for other effortless and repetitive automation tasks.