Fraunhofer IIS/EAS
Sensor platform for AI testing in SMEs
KNIME is free software for analyzing large amounts of data. The KNIME2Edge platform is designed to enable companies to get started with AI technology without having to invest a large amount of resources and specialist staff at the outset.
In the industrial sector, data has been collected for a long time in order to obtain information about the condition of a system, for example. Until now, data has been evaluated using traditional algorithms. However, AI-based methods are also increasingly being used and introduced in this area. The technologies are moving ever closer to the actual application. Ten years ago, the Google Group and Stanford University still had to use 16,000 processors to teach an artificial intelligence system to distinguish between the image of a cat and that of a human, but modern optimization methods and increasingly powerful computing technology mean that AI algorithms can now run on simple 32-bit microcontrollers. This makes it possible for sensor values to be transferred to a microcontroller very close to the measuring point, in the edge, where they can be evaluated. This makes it possible to use several sensors efficiently, as the amount of data that has to be sent to a central computer can be significantly reduced by processing on site.
The use of such cost-effective computing technology makes AI technologies increasingly attractive for small and medium-sized companies. However, they often lack the experience or simply the access to suitable AI solutions for their industrial sector. Especially for the development of new AI-based approaches for industries that have had little contact with this technology to date, it is often time-consuming to design a new AI system and adapt it to the respective requirements of the application environment(Fig. 1).
Various areas of expertise are required for this development process, as hardware developers and programmers are needed in addition to specialists in the use of artificial intelligence and data analysis in order to implement every aspect of the application system. All these steps require different professions and a considerable amount of time. In order to reduce this development effort and offer an AI system that is also attractive for small and medium-sized enterprises, Fraunhofer IIS/EAS has developed the KNIME2Edge (Konstanz Information Miner) sensor platform for AI testing in SMEs.
This sensor platform is designed to enable interested companies to get started with AI technology without having to invest a lot of initial resources and specialist personnel. The user of the KNIME2Edge solution should only have basic knowledge of data processing (Fig. 2).
Since an initial exploration for suitable solutions often requires many iterations, during which hardware and software requirements can change again, a two-stage methodology was established for KNIME2Edge: In the first step, the user should be able to try out different approaches. The flexibility required for this is achieved through the modular design of the KNIME2Edge system in terms of software and hardware. Once a promising solution has been found, it can either be implemented locally on the sensor platform or - in a second step - in hardware that has been specially adapted or developed for this solution. The advantage of KNIME2Edge is that there is no interruption to the methodology between exploration and hardware-side testing. All steps can be processed within an overall system.
How does KNIME2Edge work?
The platform is based on proven microcontroller and sensor technology and combines this with the free KNIME software, which allows AI algorithms to be created and transferred to the microcontroller simply by dragging and dropping without any programming knowledge. The platform is designed to invite users to experiment with AI solutions without having to integrate them into the existing industrial network. It is designed as a 'plug & play' system, which means that no additional driver installation or user-side configurations are required.
On the hardware side, the sensor platform consists of a base module and various sensor module types, which can be connected to the base module in different combinations and quantities as required using plug & play via a standardized interface. On the sensor side, the platform can therefore be configured according to the respective application criteria. Apart from establishing the electrical connection by plugging the sensors into the sensor platform, no further configuration work is required. This flexible concept also allows additional sensors with other physical measuring principles to be added if required. A wide range of sensors relevant to production can be used. This sensor data is also available after a short adaptation time.
However, the platform can also be used for specialized applications, for example in medical technology. It is also possible to use several base modules. The task of the base module is to establish a connection to the KNIME software on the user's computer, via which the measurement data is then sent to the KNIME application or a developed solution algorithm is transferred to the base module. The connection is established via a USB, Ethernet or Bluetooth interface (Fig. 3). The platform can also be used on the move in battery mode.
On the software side, KNIME is characterized by a code-free interface: The program sequences are put together 'drag & drop' using so-called function nodes (Fig. 4). Each node encapsulates a function required for the program sequence, for example reading a file. The semantic link is created by connecting these nodes using connecting lines. In-depth knowledge of a programming language is not required.
In addition to the KNIME nodes available as standard, for example for a fast Fourier transformation (FFT), Fraunhofer IIS/EAS has developed further nodes that can be used to establish a connection to the sensor platform, to record measurement data and to transfer the algorithms to the base module. These nodes are also integrated into the program sequence using drag & drop.
Processing in KNIME follows the ETL process (Extract, Transform, Load). At the end of the processing chain, target nodes can store the data in files or databases and display it in graphical diagrams. The software can be used to store the data long-term, convert it into traditional data formats, such as CSV or XLSX, or store it in databases. Recorded sensor data can be used within KNIME to develop and test filter or AI algorithms, for example.
Once the algorithms have been developed on the recorded data, the software is transferred directly to the base module and executed. The same system is then used to test the functionality of the algorithm and the overall system in the application. The extended use of the platform often generates other data points where the algorithm still reacts inappropriately. Further optimization can be started with this new data; there is therefore no change in the methodology between testing, optimization and application.
Why KNIME2Edge?
The advantage of the solution is that all steps of input or data acquisition, processing and output are implemented within a free software environment. No additional external tools are required. The variety of nodes already available also offers the option of subjecting the collected data to conventional filtering or smoothing before it is actually processed using AI algorithms. This makes it possible to reduce the complexity or optimize AI solutions.
The platform can be used independently via USB or Bluetooth as well as via the Ethernet interface within an industrial network.
By applying the edge concept with its short transmission paths, the KNIME2Edge platform offers the advantage of short response times and a reduction in the volume of data in the industrial network. In comparison, established AI solutions often forward the recorded sensor data to an external cloud service for further processing, which is often more energy-intensive and costly. In addition, local collection and processing of application data can contribute to improved security and confidentiality of internal data.
The KNIME2Edge system is designed to be user-friendly and is intended to lower the entry barrier to the new world of sensor data acquisition and analysis using AI algorithms. This gives users who have previously only worked with traditional solutions for process monitoring and analysis the opportunity to experiment with new technologies in an uncomplicated way.
What fields of application are there?
The platform is deliberately designed to be flexible in order to give the user as much freedom as possible when selecting the sensors for the application scenario. The main purpose of the system is to find the physical measuring principles that will later be required and to select the corresponding sensors. For example, the user can use the modular system to determine whether pressure is a variable that can be used to gain further insights into the condition of their system. It is also possible to use different pressure sensors to select the one with the best characteristics. However, the result of the investigations may also be that certain sensors provide no further additional information about the condition of a system.
The author: Ronny Schwabe is an expert in system integration at Fraunhofer IIS/EAS in Dresden.
© Fraunhofer IIS/EASThe simple graphical input of the algorithms allows alternatives to be explored quickly. This means that the best alternative for the algorithm can be determined within a short time and tested directly on the system by loading it onto the base module and running it. This provides the user with information about the required sensor types and suitable AI applications. Based on the solutions found, they can ultimately make well-founded assessments of possible AI application scenarios and decisions on the purchase of more complex or more expensive systems for a specific application.
You might also be interested in
Modular AR service platform for industrial production
A cloud-based service platform with open industry interfaces is being developed as part of the "secureAR" project. An AR assistance system plays a central role in this.
read more...All-in-one platform for wireless sensor applications
The Fraunhofer Institute for Reliability and Microintegration IZM has developed a platform that can be modularly adapted to wireless sensor applications. The aim is to save time and money.
read more...AI-based component control for faster changeover times
Fraunhofer IGD's Marquis software makes it possible to inspect components on the move and in real time.
read more...Framework for data and AI lifecycle
Fraunhofer IKS is developing an open, interoperable and technology-neutral framework that supports and optimizes the data and AI lifecycle. This will significantly increase value creation through AI.
read more...Intelligent monitoring using airborne sound monitoring
Structure-borne noise signals from corresponding system components are often evaluated for condition-based machine monitoring. A practical test compares the alternative approach of airborne noise monitoring in combination with AI algorithms with...
read more...Energy self-sufficient IoT communication in 5G networks
In the Internet of Things, sensors act as core components for data acquisition and communication. How can they be connected as wirelessly as possible and supplied autonomously in order to make installations cost-effective?
read more...With a focus on edge & cloud computing
The e-paper of issue 03/23
Sustainability is at the top of the embedded community's agenda. Prof. Dr. Peter Fromm from Darmstadt University of Applied Sciences explains in an interview why embedded technologies are becoming the key to a sustainable future - in cars, in...
read more...Intelligent monitoring using airborne sound monitoring
Structure-borne noise signals from corresponding system components are often evaluated for condition-based machine monitoring. A practical test compares the alternative approach of airborne noise monitoring in combination with AI algorithms with...
read more...The IoT-Bus - what's behind it?
Fraunhofer IIS presented the IoT-Bus with integrated security concept for the first time at embedded world 2017. Project manager Andreas Oeder explains the background.
read more...