Data is the lifeblood of intelligent and networked production. The big challenge for companies is to generate added value from the data. This added value can be process optimization, but it can also be the foundation for a new business model.

Weidmüller has made it its mission to support its customers on the path from data acquisition to the business model. To this end, the Detmold-based company has defined four stages: Data acquisition, data pre-processing, data communication and data analysis and business logic. Weidmüller is building and expanding a suitable product portfolio along these four points.

Data acquisition

The company has developed the u-sense sensor series for data acquisition in the field. As one of the first representatives of the series, 'u-sense vibration' monitors motors, gears and pumps. It records mechanical parameters as well as temperature and other electrical variables. The sensor module in an IP66/IP67 housing is connected directly to the rotating device to be monitored by means of an adapter plate or a hole and detects vibrations in accordance with ISO 10816. The measuring range covers 2 to 16 g at a sampling rate of 1 kHz. Wiring is not required. If there are deviations from the standard behavior, the sensor module reports this to a higher-level unit via Bluetooth Low Energy 5.0. Data preparation, i.e. preprocessing, already takes place in the sensor module.

An AA battery, which can be replaced, serves as the power supply. The sensor operates at ambient temperatures from -20 to +85 °C and has FCC, cULusx, ATEX, IECEX (Zone 2 / 22) certificates.

Vibration detection is only the first step. Weidmüller plans to further expand the functions of the sensor module. For example, the sensors are to be equipped with a microphone in order to detect changing noises in the environment and thus be able to draw conclusions about the functionality and service life of a motor.

Data pre-processing and data communication

Data provides important information in the production environment and usually finds its way into an ERP system without detours via gateways. Pre-selection is crucial to ensure that this is not flooded with irrelevant data. The data pre-processing itself takes place at different levels depending on the effort and performance requirements: directly in the sensor module, in the gateway or in the IPC. However, the data must also be visualized. This is where Human Machine Interfaces (HMI) come into play, enabling simple visualization and operation on any end device.

With 'u-create Procon-Web', an HMI/SCADA software, dynamic user interfaces can be parameterized and configured. It is based on open and manufacturer-independent web platforms such as HTML5, CSS3 or JavaScript. The system configuration takes place independently of the operating system in the browser. This means that not only can different end devices be used - from the machine controller to an industrial tablet to a smartphone - but the software also requires no installation on the client.

The software moves away from a device-specific display and supports a role and rights system in order to offer the displayed data in a device- and role-specific manner. This means that the groups of people accessing the data, such as plant operators, production managers, quality managers or maintenance staff, are shown precisely the information that corresponds to their tasks and the nature of the operating devices.

Data analysis and business logic

The 'u-sense vibration' sensor module for monitoring the condition of rotating devices can be mounted on the cooling fins or directly on a rotating component.

© Weidmüller

In order to generate added value from data, it must be analyzed and interpreted. Artificial intelligence or machine learning plays a central role in this context. As not every company has a data scientist on hand, Weidmüller has developed an industrial automated machine learning tool. This enables domain experts, i.e. an application specialist, to develop ML solutions independently.

Automated machine learning is used in many areas, from anomaly detection and classification to prediction. However, in order to detect anomalies and make predictions from them, for example for predictive maintenance, data must be collected and correlated. Process-relevant data from machines or systems is usually available in sufficient quantities. In order to extract added value from this data, it is analyzed using machine learning methods and corresponding models are developed.

The software essentially provides two modules. With the 'Model Builder', the domain expert can generate ML solutions for anomaly detection, classification and error prediction. To do this, he detects and classifies deviations from normal behavior in data sets, which are then used for model building. The data set enriched with the application knowledge is the input variable for the subsequent automatic generation of the ML models. The user selects the most suitable model according to certain criteria such as model quality, execution time or preferred parameters. The selected model can be exported and transferred to the execution environment. In the second module, the AutoML software, the models are executed on the machine - on premise or cloud-based - in the runtime environment.