AI-based automatic visual packaging control
Article By : Manuel Haß and Chris Montague
Here is how AI can be used to quickly and effectively identify and eliminate complex packaging quality problems in production in real time.
Product packaging has a wide variety of characteristics. Different shapes, colours and materials can thus be the cause of different defect patterns. Data Spree from Berlin shows how artificial intelligence (AI) can be used to identify and eliminate complex packaging quality problems in production in real time quickly and effectively.
ADLINK Technology and Data Spree combine industry-ready hardware and state of the art Vision AI software. For this use case ADLINK has a range of hardware and cutting-edge technology. Using our NVIDIA Jetson Neon series based cameras, allow machine learning to be done in camera at the edge.
Detecting defects at an early stage in ongoing production and logistics and eliminating these at short notice is often a very demanding task in the packaging sector. Reliable automation of visual inspection is a crucial factor for ensuring consistently high quality.
Challenges of visual inspection
Classic image processing systems and traditional algorithms for the visual inspection of packaging defects are often very inflexible and costly to implement. In this case, defect detection must be developed manually by experts, which requires a lot of know-how and time. At the same time, the multitude of possible defect patterns (tears, missing pieces, dents, scratches, geometric deviations, missing contents, printing errors) can only be implemented with a lot of effort or not at all. All this leads to high costs and to the fact that quality requirements for automation often cannot be met.
Solve complex quality problems efficiently with Vision AI
Artificial Intelligence (AI) can be used to reliably detect a wide range of individual error patterns and anomalies. With Data Spree’s Deep Learning DS software, Vision AI software logic can be implemented efficiently and easily in the background. Continuous monitoring of production data, automatic classification of defects and time series analyses are also implemented in a user-friendly manner with Deep Learning DS in production operations.
Figure 1: Deep Learning DS – Data Management and Statistical Analysis of Blister Packaging Surfaces for Automatic Visual Quality Control
To implement AI-based defect detection, images of the packaging are first taken from the production process. Now, certain defects to be detected can be tagged to train the AI. This tagging of the data is called annotation. However, if one wants to detect general defects and deviations, the AI, without defect tagging, can also identify deviations from the norm via anomaly detection after training.
In this process, the AI iteratively trains the detection and localization of deviations or anomalies from the good state or also special error patterns that one would like to identify as a user. Here, the AI functions similarly to the human brain and learns to recognize, assign and localize defects based on the image data – without the need to manually pre-define specific packaging features. With Deep Learning DS, you can quickly and easily perform this learning process yourself. Data Spree also offers the complete process up to productive integration into the system as a service.
This method thus allows the most diverse and complex quality assurance tasks to be implemented quickly and easily – and without a single line of programming code.
Figure 2: Visual quality assurance of blister packs, on the left the detection of minimal damage in the surface, on the right the exact localization of the defects in a heat map display
Automation processes can thus be implemented efficiently and robustly. A ready-to-use prototype can be created in just a few hours and expanded into a productive solution within a very short time. The fast AI models of Data Spree additionally ensure real-time capability in high-frequency production and logistics operations. Another benefit is the flexibility of the learning system. If packaging, packaging properties or products change due to production or logistics changes, the AI can simply be “fed” with new images and retrained. This allows for a quick and effective response to changes in production or logistics without having to start from scratch or purchase and implement a new solution.
Via Deep Learning DS, the data from ongoing production operations and the detected errors can be stored, managed and statistically evaluated in the long term. In this way, the highest quality requirements can be continuously realized in the combination of data management and AI training.
Quick and easy implementation
The trained AI model can be individually integrated into any customer application through the open ONNX standard format. Data Spree’s own execution environment Inference DS also offers a simple graphical user interface in which the AI model can be quickly executed on the respective hardware, such as a smart camera or industrial PC, using a drag-and-drop principle. This saves integration time – and above all costs.
ADLINK and Data Spree Partnership
ADLINK Edge software eco-system has built-in connectors for Data Spree’s Deep Learning DS. This provides both companies with the ability to extend the capabilities of any Vision AI solution to ADLINK’s extensive portfolio of hardware and software in a scalable manner without changing the underlining platform of both systems. Using ADLINK extensive IoT platforms, we can integrate with companies at Edge providing filtered data to Enterprise and cloud.
This article was originally published on EEWeb.
As co-founder of the Berlin-based start-up Data Spree, Manuel Haß implements the shared vision of the automation of the future: making deep learning accessible to everyone in order to automate cognitive processes. After studying computer science at the TU Berlin and stations at ABB, Manuel worked on autonomous vehicles at the DCAITI in Berlin before founding Data Spree.
Chris Montague is Head of Edge Solution Sales EMEA at ADLINK. An IoT professional with over 22 years experience in the hardware, software and IT solutions market, prior to ADLINK he worked for an IT consultancy, advising customers and delivering services from pre-sales to delivery for projects across multiple verticals. He had a Bachelor’s degree in Computer Science from Northumbria University and started his IT career writing code to optimise and streamline databases for large public sector clients.
