Looking back on IoT retrofitting
Author: EIS Release Date: Feb 16, 2023
A collaboration between Würth Elektronik, FEGA & Schmitt and IAV demonstrates how industrial manufacturing equipment can be upgraded and digitised.
The bias of progress dictates that the latest technology is believed to be superior. Sometimes, however, it is better to use tried and tested, current, or even legacy technologies, but there are challenges with older equipment, including efficiency, inconsistent quality, expensive maintenance and manual labour.
Considerations when replacing older manufacturing equipment include the fact that it can represent significant capital investments and years of planning. A retrofitting approach is more cost-effective.
Production environments can be greatly improved with the introduction of intelligent automation. Existing systems can create additional value through increased automation without writing off the older machines or necessitating a major investment. Automated machines must accurately record and analyse operating data for modernisation to succeed. Furthermore, it is essential that any conclusions drawn from these evaluations can be extended to other manufacturing sites.
Retrofitting is the process of updating or adding new features to existing equipment using the IoT in a non-invasive way. Transforming a machine into its digital twin means production can be examined and optimised in greater detail and better efficacy.
Adding connectivity
Using an open source concept, Würth Elektronik, FEGA & Schmitt and IAV have realised a proof of concept for industrial cutter monitoring (Figure 1). The concept was by FEGA & Schmitt, Würth Elektronik supplied connectivity and sensing components and, together with IAV, provided cloud infrastructure (Figure 2). IAV also offered data analysis and full system integration services.
The goal was to develop an easy-to-install product for FEGA customers, to monitor industrial cutting machines and detect use based on current measurements, as well as to detect possible problems with the cutting tools before they occurred.
Sometimes, a particular combination of tool movements can cause tools to break. By identifying this set of movements, a failure prediction can be made. Consequently, there will be significantly less production downtime. A current measurement, on the other hand, provides the ability to determine machine utilisation and simplifies the planning process.
A strict requirement for the installation was not to interfere with the customer’s infrastructure nor to cause any process downtime.
Customers receive comprehensive system availability information from the finished product. Predictive maintenance is provided through the use of sensors and AI-supported data evaluation.
Prototyping boards
FeatherWings are a set of stackable prototyping boards with different functionality. Würth Elektronik created a range of open source, compatible FeatherWing development boards, with sensor wings, WE Pro-Ware wireless connectivity, Wi-Fi and various power supplies. There is a GitHub repository for all open source boards, including schematics, bill of materials, software and cloud connectivity descriptions for Azure and AWS.
Sensor FeatherWings (Figure 2) were used to create the initial data points. As the acceleration is closely connected to the movement of the cutter hand, the use of an acceleration sensor is a good starting point for monitoring movements.
The development board with four sensors is compatible with the Adafruit Feather form-factor and Sparkfun’s QWIIC-connect, which provides a standard I2C interface that is also compatible with STEMMA QT and Grove/Gravity.
Reduced network traffic
Gateway/cloud connectivity can be established in two distinct methods. Using an industrial Raspberry Pi with LTE connectivity, vast amounts of data are sent to the cloud for spectral analysis throughout the model generation phase. After the model is created, connectivity is switched to an Adrastea-I LTE-M/NB-IoT module, which is claimed to “greatly” reduce network traffic and, consequently, costs. Both methods have been tested in cloud-connected production environments.
The node is connected to the cloud via a gateway using a proprietary Thyone-I Wireless 2.4GHz radio module. Security should not be underestimated, therefore the gateway to cloud connectivity uses the TLS protocol, and the node employs a similar approach with the secure element (ATECC608A-TNGTLS from Microchip Technologies) on one side and the cloud key vault on the other. The whole connection is protected and encrypted between all the communication participants, nodes, gateways and the cloud.
Vibration measurement
To select the appropriate accelerometer, a clear understanding of the application and its measurement tasks are essential. In this case, a three-axis acceleration MEMS sensor, the WSEN-ITDS, was used to detect cutter arm movements.
Current measurements must be non-invasive because the devices being monitored cannot be interfered with and should be easily applicable to similar machines. Here, the Wago split-core current transformer 855-4101/400-001 and the SparkFun ACS723 Hall-effect sensor breakout were used (Figure 3). The advantage of using a Hall-effect sensor is that the circuit being sensed and the circuit reading the sensor are electrically isolated for the circuit to operate at higher DC or AC voltages than the main board.
For proof of concept, two versions of the connectivity solutions were used. Version one, used in the initial data collecting stage, was an industrial IoT Raspberry Pi-compatible gateway. A Linux-based system was used to generate C-code and optimise the data collection and transfer, as a vast amount of data is necessary to validate machine behaviour.
For the cloud, a dashboard was created for real-time monitoring of the data using Node-Red and Grafana. The timestream data was analysed to identify trends and patterns with machine learning. Similar process patterns are automatically recognised and labelled (Figure 4). The remaining patterns are marked as unknown. This data serves as the basis for process statistics that can be used for process monitoring, quality assurance and predictive maintenance.
Real-world test
Many challenges were encountered during the real-life testing, for example, loss of data as a result of the distance and various radio sources in the manufacturing hall, constant movement of the stackable boards and power supply (or lack thereof).
Acceleration sensors were mounted on the cutter arm without any nearby power sources, supported by a LiPo battery. Despite the low standby current consumption, constant data transmission during the initial stage drained the battery. Vast amounts of information were transmitted daily, resulting in an empty battery every two to three days. The solution was to use a solar panel to charge the battery using an open source Adafruit board.
The second problem was the location of the sensors and radio modules. The sensor has to be situated on the tool handle, which is a moving part. On the machine, all moving parts are protected by metal housings, which act as a Faraday cage. Despite being small and efficient, the integrated antenna was of no use. Instead, an external antenna was attached to the outside of the housing.
The current sensing part was a composition of split-core current transformers and Hall-effect sensors for each phase, combining two sensors calibrated by Würth Elektronik.
Fast prototyping
Making a proof of concept with open-source components can dramatically reduce prototyping time. Combining pre-existing boards with standard pinning and sensors with standard connectors eases test and experimentation with the setup.
Using two stages in the proof-of-concept prototyping allows for the creation of an effective model in the first stage, which can then be implemented in the second. The second stage will deploy local models on the microcontroller and send out only the bare minimum amount of data. The necessary data will be sent to the cloud using the Adrastea-I cellular module.