Manufacturing companies are still finding it difficult to digitize their machines and systems. There is no standard and the range of different solutions is constantly growing. In the end, it is difficult to make the right decision. The only thing that is beyond question is that the use of proprietary data and the use of new tools and processes are becoming important in order to remain competitive.

But where are the entry problems at the moment? The architectures and technologies to be used are largely known, although many manufacturers have not yet implemented them. For many, the biggest challenge is deciding who will take responsibility for implementation - in other words, the famous OT/IT convergence needs to be overcome:

• While the IT team can implement solutions by making machine data available to cloud systems, and using services available in the cloud. This comes with potential costs, security and IP protection issues. It also requires IT staff and data scientists or cloud engineers or outsourcing of this expertise, which incurs high costs.
• The automation engineer, on the other hand, works on and with the machine, knows all the electronic and mechanical parts as well as all the software that runs the machine. He knows which data is important and meaningful and what needs to be improved. But he does not know how. He is usually not in a position to implement AI to optimize the process.

Many companies are faced with this dilemma. Large companies are trying to solve the problem for themselves by increasing their digitalization budget, generating cloud computing and new IT resources in general. But what remains for SMEs that cannot afford such an expensive strategy?

The architecture of container-based edge stacks.

© Coligo Technologies

For this clientele, Coligo Technologies has developed a plug-and-play software solution that runs on an edge device - the EdgeBox - and is aimed at automation engineers. Most of the functionality is already implemented in the firmware, the EdgeStack, which is based on Linux and Docker containers. Data from the field is collected via OPC UA. The data is then made available in a streaming component and can be accessed by various containers:

• Store: database for local persistence
• Monitor: a local, user-specific dashboard that can be edited directly in the app
• Expose: enables data to be transferred to an MES or cloud system - for hybrid architectures
• Analysis: for local AI applications

By using container technology and microservice architecture, the EdgeStack can be operated not only on an EdgeBox, but also on other hardware and in other configurations. In addition, it can be easily expanded by integrating new microservices, which can be future Coligo functions or customer-specific applications. The platform is individually expandable and scalable.

The front end as an APP

The app allows the user to collect, display and save files and start AI models.

© Coligo Technologies

A standard AI process includes historical data collection, pre-processing, exploratory data analysis (EDA), feature engineering and selection, model selection and training, evaluation and deployment, and finally inference. However, too few automation engineers are familiar with this process. In addition, the lifecycle of AI models does not end here: Monitoring and updating is required as soon as their performance deteriorates. Coligo's goal was to integrate AI into the concept of user-friendliness. The challenge was to provide an AI offering to a user who has no prior knowledge of AI so that they can also draw on their very valuable knowledge - how the machine works, what problems it has, what data is meaningful.

As a solution, the automation technician has access to the front end as an app on a tablet or smartphone. The app connects to the EdgeBox via its own WLAN access point. To work with the app, the user requires no training to collect the necessary data, design a dashboard, select and configure one or more AI models and, if required, provide raw data or results in a cloud. The solution offers four AI models as standard, without programming: anomaly detection, time series prediction, object detection and recognition in video streams, audio classification and recognition in audio streams. Additional models can also be implemented and executed on the platform to solve individual problems. An engineering team is available to identify, train and specifically develop possible models.

The cockpit

The author: Dimitri Philippe is Managing Director of Coligo Technologies.

© Coligo Technologies

A cockpit for managing the EdgeBoxes rounds off the portfolio. It offers user administration with rights, device and fleet management. It also enables complete firmware updates or the updating of a specific container on a specific device or fleet of devices, among other things.