Exhibitors & Products
Events & Speakers

In this interview, Philipp Zehnder explains what the StreamPipes solution can offer and why mid-sized companies will be able to develop AI applications in less than an hour in the near future.

What is StreamPipes?

StreamPipes is designed to help make IoT data analyses in the production and logistics environment accessible to users who are not very familiar with the relevant technology. Our goal with StreamPipes is to provide a very easy way of evaluating continuous data streams in the IoT environment in the form of an open source solution that can run completely without being connected to an external cloud provider. Machines or other data sources can be connected within minutes. In a graphic editor, we then provide an extendable kit of algorithms and functions that allow the data to be evaluated independently.

What applications do you have in mind?

We support applications ranging from automated data cleansing and condition monitoring ̶ which detects when threshold levels have been exceeded, for example ̶ to machine learning techniques such as the evaluation of camera data or predictive maintenance.

You advertise lower development costs – how does that work?

An integrated alerting mechanism enables incidents to be detected, for example, and the right people to be notified. In the area of predictive maintenance, it helps to detect faults at an early stage in order to prevent unexpected downtime. The emphasis is always on specialist users being able to model these rules themselves in a very flexible way via the graphical user interface. StreamPipes primarily relies on reusable algorithms – these can be used to link an analysis component, such as anomaly detection, with a variety of data sources. As a result, the actual development work only needs to take place once, but the algorithm can be connected to other components any number of times – this reduces the software development costs, which is a key deciding factor for smaller companies in particular.

Who will its users be within the company?

Our target group consists of specialist users such as production experts. Today, connecting and continuously evaluating machine data still requires a great deal of specialist IT knowledge. This knowledge is often a very limited resource, however, especially in small and medium-sized manufacturing companies. Our solution allows users to first find out what kind of data their machines pro-duce and also to understand the importance of this data. Using this basic knowledge, rules or algorithms based on AI methods can then be employed to obtain added value from the data.

When will your application enter the market?

Over the last few years we’ve developed many of the innovations that StreamPipes offers today in the course of various research projects at German and EU level as part of our work at the Research Center for Information Technology in Karlsruhe. In early 2018, we then made the software available as open source in order to provide as many users as possible with an easy introduction to IoT data analyses. The tool can be downloaded and installed directly from our website (https://streampipes.org).

Which protocols does StreamPipes support?

The current version contains interfaces to the most common IoT protocols such as OPC UA and MQTT. What’s more, users can quickly connect files themselves. The current version includes more than 80 preconfigured algorithms and data sinks to help users get started quickly. We work closely with industry partners on the further development of StreamPipes and offer workshops and pilot testing as part of proof-of-value projects, which quickly demonstrate the added value of the solution. In addition, we continuously incorporate the latest findings from research, which we develop in partnership with companies, into the solution.

What are you planning for the near future?

The next step will be to enable specialist users to apply artificial intelligence techniques themselves in order to further simplify and automate the data analysis – our aim is to cut the time between a machine first being connected and the first usable AI prediction to less than an hour.