Production data is a key requirement for the use of machine learning in manufacturing. However, easy access to this data is often lacking. A lean IoT solution enables structured and secure collection of machine data – without deep interventions in existing systems and with a clear focus on later use by artificial intelligence.
The IoT expert and co-developer of this solution is Perinet GmbH, which implements an end-to-end infrastructure from the field level to IT (including the cloud) based on standards. This allows for direct, smart, and secure integration of sensors and actuators with the cloud, the internet, and company IT.
The challenge: data access and integration
One of the most important tasks in industrial companies is the continuous optimization of production processes. This is the driving force behind digitalization, as modern digital technologies can continuously extract data from the individual machines of an industrial operation. This data is then analyzed – not only using simple statistical methods but also with the help of artificial intelligence. However, in practice, this is not as straightforward as it sounds. Many companies encounter limitations.
- There is no unified data basis, as the existing machines are heterogeneous. They come from various manufacturers and have partly analog controls, but some already feature digital interfaces.
- Accessing relevant machine data often requires custom programming and deep IT integration. This leads to high costs and long project durations.
In many cases, there are also security concerns, since the OT (Operational Technology) of industrial systems is typically isolated from external networks. However, digital technologies generally require a connection to an external network.
Requirements for Machine Learning in Manufacturing
This is why industrial companies are looking for solutions that allow them to securely collect all types of production data without major modifications to their existing infrastructure. In recent years, the focus has increasingly shifted to establishing the foundations for the use of machine learning in manufacturing.
Machine learning is a form of artificial intelligence that independently identifies patterns and correlations based on large volumes of data and derives decisions or predictions from them. In manufacturing, machine learning uses various methods, such as:
- Classification models for error detection
- Regression analyses to predict energy consumption
- Clustering methods for pattern recognition in production
An ideal solution integrates existing machines with minimal effort, provides data in a structured format, and takes IT security into account. Perinet and Zentinel MDS have developed a specially adapted solution for their customers.
Challenges at a Glance
- Heterogeneous machine landscape with no standardized interfaces
- High effort for individual data connection and IT integration
- Security concerns when connecting OT systems
- Lack of infrastructure for structured data collection
The solution: integrating machine learning into manufacturing
The solution enables continuous data collection with minimal network and infrastructure changes. It eliminates the need for complex software installations and incurs no additional licensing costs. Network integration is simple, as existing systems do not need to be replaced or modified. Communication is fully encrypted using mTLS, ensuring that the solution meets the security requirements of industrial applications.
Components for data collection
The system combines Single Pair Ethernet (SPE), an IoT adapter, an edge server, as well as a SQL database and a dashboard.
The ZentNode adapter is an intelligent networking solution for data collection to support machine learning in manufacturing. It is suitable for recording production metrics (OEE, unit counts, scrap, downtimes), energy consumption, and analog process values. It processes the data locally and transmits it securely via Single Pair Ethernet (SPE) to an IT system, an edge server, or an MQTT broker.
The ZentEdge server is a robust edge computer that collects, stores, and processes machine data from ZentNode. The data is stored in the SQL database of ZentEdge and made accessible to your existing systems via standard IT protocols. ZentEdge serves as a bridge between operational technology (OT) and information technology (IT), using Single Pair Ethernet (SPE).
Single Pair Ethernet (SPE) handles communication between the individual components. It enables data transmission at 100 Mbps and power supply through a single cable, significantly reducing installation effort. SPE eliminates the need for complex gateways and allows the integration of industrial devices into IT networks using standard protocols such as HTTPS and MQTT, or via a REST API. All data transmission is additionally secured using mTLS certificates.
Data as a driver of optimization
Structured data collection for machine learning in manufacturing enables companies to optimize numerous downstream processes. This data forms the foundation for machine learning and can be applied in many use cases:
Predicting failures. If, for example, machine vibrations or temperature patterns indicate wear, maintenance can intervene in good time before downtime occurs.
Identifying quality issues. Machine learning enables earlier and more targeted detection and analysis of defects and process deviations, helping to uncover correlations between process parameters and product quality.
Controlling material flow. A detailed data base helps detect bottlenecks in material flow and take countermeasures. This increases delivery reliability and reduces production downtimes.
Improving planning processes. The gained transparency helps identify constraints earlier and adjust detailed planning based on current operational data.
Overall, machine learning provides production managers with a solid foundation for informed decision-making. It allows them to respond to changes in near real-time, increasing their flexibility in daily operations. Additionally, it creates a valuable pool of data that can be used for simulations or long-term trend analysis.
The result: a foundation for data-driven innovation
The solution enables companies to establish data-based processes during ongoing operations. It offers an entry point into the use of AI without requiring major upfront investments or extensive IT resources. The insights gained can be applied across different areas of production.
The benefits are clear: more efficient workflows, reduced costs, and a higher level of production maturity. At the same time, this creates the basis for realizing further data-driven innovations. These include digital twins, automated process control and integration with ERP systems such as SAP S/4HANA.
Based on this foundation, additional features can be integrated in the future, such as self-learning process optimization or connections to external data sources like supply chain information or customer requirements. This makes the system a sustainable solution for medium- and long-term digitalization strategies.
Results at a glance
- Seamless data collection without interfering with machinery
- High security through encrypted communication
- Structured data ready for machine learning
- Flexible expandability for further digital applications
