Josef Ressel Centre for Time-Series Failure Prediction and Prevention
Machine failures in industrial process chains can have serious financial consequences. Conducting maintenance at the optimal time, together with detecting and predicting critical production settings, helps prevent such failures. The Josef Ressel Centre at the Institute of Software Design and Security of FH JOANNEUM Kapfenberg is developing a data-based model to predict and subsequently prevent malfunctions and failures of machines and testing equipment.
The Josef Ressel Centre for Time-Series Failure Prediction and Prevention is supported by the Christian Doppler Research Association (CDG) and the Austrian Federal Ministry of Labour and Economy (BMAW). AVL DiTEST and voestalpine Tubulars are its corporate partners.
Optimal maintenance scheduling and unforeseen events
The question of when and how often industrial machinery should be serviced presents a dilemma: carrying out maintenance too frequently and unnecessarily takes time and generates additional costs, while the probability of a malfunction rises the longer a functional check of the process-relevant machines is delayed. In addition, even with perfectly maintained machines, an unlucky combination of machine condition and production parameter settings can lead to critical situations arising at any time. In the worst-case scenario, this could lead to a complete breakdown of the industrial process chain which is very time-consuming and cost-intensive to remedy.
To solve this dilemma, the research team at the Josef Ressel Centre is developing models based on time series data. These are sequential measurements taken by sensors indicating the condition of the various machine components. These time series data are used to predict failures which can then be avoided by prompt servicing – a concept known as predictive maintenance.
However, unfavourable production parameter settings can also lead to unexpected failures. Production depends on a large number of parameters which, when optimally aligned, ensure a smooth production process and good product quality. But an unlucky combination of individual parameters can cause deviations in process control, which in turn can lead to plant downtime and, in the worst case, to tool breakages and system damage. The influencing parameters are many and varied, including insufficient forming temperatures, borderline lubricant concentrations, or incorrect rolling mill settings.
Time-series failure prediction eliminates two potential risks: production failures due to technical problems, and the unintentional manufacture of substandard products due to defective machinery and testing equipment.
Using machine learning and artificial intelligence for data analysis
A particular challenge faced by this project is that often the time series data on machine conditions measured by the sensors are not linked to past failures and their technical causes. Consequently, a three-phase research plan was designed as a means of gaining the best possible insights. In phase 1, machine learning is used to establish a connection between the data and the technical causes. In phase 2, failure prediction models are developed, and the aim of phase 3 is to use artificial intelligence and traditional statistics to derive interpretable explanations for the phase 2 results. The phases do not run in a linear sequence, however, but are cyclically interlinked due to their interdependencies. This maximizes efficiency and knowledge generation.
The goal of the Josef Ressel Centre is to develop a generalized model for failure prediction and prevention that is ready for industrial application when the project ends in 2028. The research will create the basis for fewer downtimes and failures, as well as for greater safety, efficiency, and cost and time savings in industrial process chains.
Funding providers
In Josef Ressel Centres, high level application-oriented research is carried out by outstanding scientists in cooperation with innovative companies. The Christian Doppler Research Association is regarded internationally as a best practice example for encouraging such cooperation. Josef Ressel Centres are jointly funded by the BMAW and the participating companies.
Corporate partners
Industrial sponsors AVL DiTEST and voestalpine Tubulars are providing funding for the research project. The two corporate partners provide huge volumes of data for the research at the Josef Ressel Centre. AVL DiTEST focuses attention on a portable measuring device that records the emissions of diesel vehicles and is used by testing organisations. voestalpine Tubulars addresses the complex process of manufacturing seamless steel tubes.
