IPN

As part of a multi-year project, we were able to successfully support the company IFE Automatic Door Systems in the digitization of its products. The project focused on developing a condition monitoring system, performing complex analysis of fault conditions, and implementing AI-based predictive maintenance for IFE products. With the help of our visualization software "IPN Visualytics", the technical experts at IFE Automatic Door Systems can quickly and intuitively analyze the collected data and develop their own solutions for future monitoring of the components.

voestalpine Stahl GmbH, the best-known operator of an integrated steel mill in Austria, has developed a pilot system for maintenance prediction in the area of cooling systems for continuous casting plants in cooperation with IPN. voestalpine Stahl GmbH acted as domain expert for continuous casting and maintenance as well as data supplier. Thanks to the close cooperation, a pilot system for maintenance forecasting with a forecast horizon of more than 28 days was successfully implemented.

Transmitting stations are technical facilities that are distributed across the country and must ensure high availability rates. In the event of a failure, maintenance must be carried out promptly, which is why these stations are technically monitored very closely. Due to the large number of warnings, errors and fault messages, service staff are sent to the stations that cause the most messages or are already in fault mode. Thanks to AI-based analysis of time and order series, it was possible to identify many patterns where quick action could have prevented a malfunction. In addition, it was possible to carry out targeted prioritization of maintenance jobs, which both reduces travel costs and increases availability.

For certain recurring faults in the ongoing operation of fire engines, even their developers were unable to detect any causes. IPN was tasked with identifying the causes of these disruptions using a data-based approach. By combining traditional statistical methods, machine learning and the customer's technical expertise, it was possible to specifically identify the problems and causes. Based on these results, technical and software adjustments could be made to completely eliminate the malfunctions. These measures have increased both Rosenbauer's availability rate and customer satisfaction.

IPN supported KB SfS GmbH in planning and implementing the AI/ML cloud infrastructure and in introducing an automated data processing pipeline. The result is an infrastructure that is suitable for advanced analytics and enables KB SfS GmbH to develop advanced data-driven services for its products. In parallel, IPN assisted in the development of predictive maintenance algorithms for existing and new products of the company, such as brakes, air supply, doors, etc.

IPN received an order from Bayer AG to evaluate the production processes of a chemical reactor online. The goal was to classify the batch just produced into classes such as "production within the normal range" and "production deviating from the normal range" at an early stage. For this purpose, the relevant operating parameters were identified and a quality forecast was made during the production process. The "IPN Advisor" was developed to analyze the operating parameters during production and to give suggestions for control in order to keep the quality high.

RESEARCH PROJECT: Anonymized sensor data from a physical model was approximated using machine learning methods to create virtual sensors without having to install the actual measurement points. A "Support Vector Machine" achieved a coefficient of determination of 99.9%, which means that the physical model and the machine learning model produced almost identical results. These virtual sensors can be used by technicians and researchers for interpretation as well as for predicting failure probabilities and other applications.

For the company "Metallveredelung Huber" (MVH) it is of great importance to continuously identify bottlenecks and increase production efficiency. By combining the company's production, machine and business data in our DataCube, we have created the basis for detailed insights into production and its processes. Based on this data, plant malfunctions are sent in real time to the smartwatches of the maintenance staff in order to significantly reduce downtime. Employees on the machines have visibility into their shift productivity at all times, resulting in increased overall productivity. This is clearly reflected in the weekly and monthly reports that are generated from the DataCube for the management. The current project will enable automated detection of leaks in the process air system and is currently in validation.

In 2019, Plansee conducted a pilot project on predictive maintenance (Industry 4.0) together with IPN. The objective was to evaluate the economic potential of predictive maintenance using the Cold Isostatic Press (CIP) used at Plansee. Fault detection and prediction options for the intensifier and hydraulics of the CIP were analyzed. The results: Predictive maintenance makes both technical and economic sense. It can predict CIP failures due to intensifier and/or hydraulic problems with high probability. Predictive maintenance enables proactive maintenance, increases plant availability and delivery reliability. Analysis of machine data enables the establishment of an OEE metrics system for performance evaluation and root cause analysis.

RESEARCH PROJECT: As part of a joint research project with the research company Salzburg Research and other industrial customers, a model was developed that predicts the print quality of a 3D printer. The main objective of the model is to identify quality-reducing factors of the nozzle and thereby ensure high-quality printing. The forecasts are automatically transmitted to the connected maintenance management program via the Apache Kafka message broker