German automation supplier Weidmüller has unveiled a new machine learning (ML) tool, named edgeML, designed to facilitate machine learning directly at the production site without reliance on cloud connectivity. This innovative tool allows for the execution of machine learning algorithms on programmable logic controllers (PLCs) or industrial PCs, effectively bringing the processing power to the edge where the data is generated.
The edgeML tool is also offered as a Docker container, which enables deployment on any industrial controller that supports running such containers. Its manufacturer-independent nature allows it to execute various machine learning models across different platforms. Moreover, the no-code design of the software enables users to implement ML models on controllers without requiring expertise in Python or data science, thereby simplifying the adoption of machine learning technologies in industrial settings.
According to Weidmüller, implementing machine learning directly at the machine presents multiple advantages. By processing and storing data locally, companies can avoid the need for data transfer typically required in cloud-based systems. This approach ensures that sensitive data remains within the company's infrastructure. The immediate detection of discrepancies in production processes leads to quicker troubleshooting, minimising prolonged downtimes and reducing defects in production.
EdgeML also offers significant cost savings, eliminating the necessity for cloud licences and reducing expenses associated with data transmission and storage. Furthermore, it is particularly beneficial for production lines where machines are not connected to the Internet due to security concerns, providing local ML capabilities without compromising safety.
The development process for a machine learning model using edgeML begins with the collection of data from the system, which is then imported into ModelBuilder software. This user-friendly interface allows the creation of ML models based on the collected data, which are subsequently transferred to the edgeML tool. The tool supports the Open Neural Network Exchange (ONNX) format, enabling users to create models outside of ModelBuilder, including those developed in Python, thus allowing flexibility in the implementation of machine learning in familiar environments.
In instances where a model may not perform as expected, users can easily replace it without the need to adjust communication settings, thereby simplifying model lifecycle management (MLOps). To further streamline the development of ML solutions, Weidmüller plans to introduce a calibration feature for created models. This enhancement is currently available in ModelRuntime, allowing a standard model for a particular machine family to serve as a template, which can then be adapted for other machines of the same class. As these models learn from various machines, they can be scaled and reused as necessary.
Looking ahead, Weidmüller has plans to enhance the accessibility of edgeML even further. This includes developing a connector that will enable the tool to function across various fieldbuses and protocols, broadening its applicability across diverse systems. The developments in edgeML signal a significant advancement in the integration of machine learning within industrial automation, promising increased efficiency and cost-effectiveness for businesses in the sector.
Source: Noah Wire Services