With their numerous machines and integrated sensors, production environments such as production lines deliver masses of data every day that are suitable for much more than controlling the current production process. Much of this data is not even actively analyzed today due to missing or incomplete data models. In order to be prepared and to avoid losing data, too much data tends to be stored. Data lakes are a partial solution to this, but they are also associated with costs.

Discussions with data analysts revealed that they do not always receive all the data they need. One of the reasons for this is the growing volume of data and the associated increase in network capacity utilization. However, industrial companies want to evaluate this data in real time. Public cloud solutions initially offered themselves as a quick solution - you simply send all the data to the cloud for data analysis. Depending on the use case and data volume, this approach works quite well. However, other use cases with growing data volumes require a combined edge/cloud-based infrastructure in terms of data security, latency and real-time processing.

Analytical data reduction and evaluation, ADA for short, as a solution

All of IBM's cloud solutions are based on Red Hat Openshift, can run on all common cloud environments and can be managed centrally.

© IBM

The solution: "Analytical data reduction and evaluation (ADA)" directly on the production line in a standards-based edge device, without having to forego all the benefits of a cloud environment. This means that the company evaluates the data at the point of origin in the edge device and decides directly on site where the data should be sent. This can be done using a machine learning algorithm or the company's own container-based applications.

New "Ready to Go" cloud box solutions can do exactly this and more in conjunction with modern container technologies - users regain more control over their data, data volumes are reduced and security is increased.

Once the machine learning model has been created on the basis of current production data, it starts working directly at the edge. In conjunction with modern integration bus technologies, only the relevant data is forwarded to the central data analysis platform, where it is evaluated with other company data. If desired, the unused "raw data" or parts of it can be stored or buffered decentrally. In addition to the greatly reduced amount of data, this architecture is also a major security benefit, as security-relevant data can be encrypted at the point of origin or forwarded or deleted accordingly.

Current ADA solutions based on container technology are flexible by design and can be adapted as required. The number, size or location of the cloud instances is irrelevant; centralized management is almost always possible. This also works with a combination of one or more public or private cloud environments.

The planning decides

As with everything, planning is very important. If certain rules are taken to heart when developing the solutions that will later run in Docker containers, there are virtually no limits to flexibility. The containers run on almost all current cloud infrastructures (on-prem, off-prem, public or private). Multi-cloud management, cloud automation and monitoring solutions round off the picture of such a multi-cloud environment as simple and efficient to manage. The edge components, based on standard cloud technologies, are also integrated here.

Even though there is great awareness of the importance of production process data, many companies are reluctant to take the step of using the data for further analysis. One reason for this is a lack of expertise in container management, such as Kubernetes. Basic knowledge of Docker is already available in many cases and is being built up further. These companies have started to develop solutions based on Docker. As development progressed, challenges also arose:

• If the number of Docker containers increases, for example, companies have realized that they need Docker management like Kubernetes.
• These solutions, which were initially developed purely on Docker, work, but are only scalable to a very limited extent or at a high cost.
• As the number of applications increases, an app catalog is also required so that the ready-made solutions and updates can be installed and used via push or pull.

Creating, distributing and operating such infrastructures yourself from individual open source components can still work in a manageable environment. With a larger number of applications, edge systems and IoT devices, this quickly becomes a mammoth task.

Suitable infrastructure for rapid deployment

A secure, fully functional cloud should be small. This is the basis for bringing more computing power directly to the machines or to the store floor. The current, ready-to-use cloud-in-a-box solutions bring a fully functional cloud with the necessary computing power to the end device, enabling machine learning models to be executed on site. The results flow directly into the further development and optimization of the models - creating a cycle. These standardized cloud boxes have been specially developed for entry into AI-supported manufacturing solutions. This enables companies to expand and develop their ecosystem in their core business.

The core element is a cloud platform based on open source frameworks such as Docker and Kubernetes with an app store. The container-based applications and updates can be compiled and distributed via this company-owned app store. In order to exploit all the advantages of a private cloud environment, the creation of machine learning models and the database used for this must also be considered. This part of the machine learning lifecycle is also part of the cloud-in-a-box solution in the production environment. The respective provision model differs here on the market. Ideally, companies that have their data analyzed in real time on site should not have to send it to a provider or a public cloud every time the machine learning model is adapted, supplemented or changed.

If the whole thing is implemented with a powerful industrial PC for edge implementation, this goes far beyond the previous pure infrastructure solutions, where initial analyses are carried out on normal edge devices. These "Ready to Go" cloud boxes can be connected behind several existing edge devices and read the data either from these or directly from the machine. The data from several end devices, machines or production lines can be merged in this way, evaluated and even stored there for several days in order to subsequently make it available for analyses and machine learning applications. How the architecture is used in the respective environment is adapted and implemented depending on the requirements.