Using the Weidmüller Industrial AutoML software to evaluate data helps, for example, to improve the quality of products, optimize process sequences or continuously monitor the condition of a machine.

© Weidmüller

Weidmüller takes a different approach here: the 'Industrial AutoML' software relies on the knowledge of the domain experts. This means that the data analysis - which would otherwise have to be carried out by a data scientist from an external partner or from within the company - is provided by the tool, which is 'merely' fed with the expert's application know-how. In other words, the knowledge about the machine and its application flows directly into the modeling. This symbiosis of technical expertise on the machine and data science know-how from the tool delivers results quickly and easily - without the need for extensive training or the purchase of additional know-how.

The software tool automatically runs through the steps required to create ML models for a wide range of ML algorithms. It then automatically determines which ML models most robustly recognize relevant machine states or process criteria.

Navigation

The software tool guides the user through the model development process for the simple development of ML models. ML automation primarily concerns the following steps: Data import, data enrichment, automated model creation and model deployment. To this end, the software consists of the 'ModelBuilder', which is used to create the models, and the 'ModelRuntime' for deployment and use of the models. The software not only helps users to prepare the training data, but also translates and archives their complex application knowledge into a reliable ML application. The expert focuses on their knowledge of machine and process behavior and links this with the machine learning steps running in the background. In this way, the user also benefits from the latest developments in the machine learning environment, which are continuously incorporated into the tool. To further simplify the use of the tools for users, the navigation of the ModelBuilder in particular has been improved.

Execute ML models

The Industrial AutoML tool automates the steps required to create machine learning models for a wide range of ML algorithms.

© Weidmüller

The models can be executed directly on the machine 'on-prem' or in the cloud. In the AutoML ModelRuntime, i.e. the runtime environment, users configure, operate and evaluate their own ML models. The ModelRuntime is based on Docker containers, in which the essential elements for executing the ML models are linked together: Definition of the machines, connection of the data sources, models to be executed and configuration of the model execution. The ML models created are imported and assigned to the respective machine. In this way, the model results can be incorporated into the further operational process, for example to trigger messages about the respective machine status.

Various sources of information

The amount of data to be analyzed is constantly increasing, but it is not always available in the same consistency. As the data usually comes from different information sources - be it sensor data such as temperature or pressure, motor speeds or current curves - it can differ significantly in terms of the time intervals at which it is generated. Temperature data, for example, is recorded slowly, perhaps only once an hour - in contrast to pressure measurement, which is continuous and therefore generates much more data in a short space of time. As a result, there is a different amount of data per information source and the machine learning tool has to distinguish between these different data sources.

The Industrial AutoML tool takes into account the fact that the data is read at different intervals and compensates for these differences without the need for an external tool. The software consolidates the data, regardless of whether it is available at long intervals - as in the case of temperature measurement - or continuously - as in the case of pressure measurement. The models take into account data with different sampling rates (update rates) and use this knowledge to create individual models. Different sampling rates for the input data are not a problem for model training or for the subsequent execution of the models; the values are considered in correlation with the cycle time.

Optimized data requirements

The author: Silke Lödige is a trade press officer at Weidmüller Interface.

© Picture: Weidmüller

When evaluating data in the cloud, for example, it is relevant how much data is transferred to the cloud, as each transfer costs money. With the Industrial AutoML tool and the integrated cycle analysis, the volume can be optimized. Users can control their cloud costs themselves by sending data to the cloud at different update rates. At the same time, they can determine how often the models generate results in order to save on data transmission and storage. For example, in the case of temperature observation, a model based on low update rates should be selected, as it is known that the temperature values only change slowly. This reduces the cloud costs.