Data mining improves production quality

Research News /

Why is it that one machine component wears out so quickly when others last for ages? The PRODAMI tool provides a targeted diagnosis of faults that occur in production plant, enabling component and other failures to be detected and corrected in advance.

The robot arm picks up the car door and moves it across to the body of the vehicle, where it is automatically attached to the hinges under automatic, central software control. This complex manipulation adds an element of risk to the underlying production process. The slightest programming inaccuracy can lead to production errors – such as repeated quality problems affecting a certain part of the vehicle. Such problems are not always detected immediately. “Quality problems due to the unreliability of manufacturing processes are often detected too late for their origin to be diagnosed and localized, preventing corrective or optimization measures from being introduced at the appropriate time. This is equally true at the product development and factory planning stage as during production planning on a corporate level,” says Dr. Helge Björn Kuntze, department head at the Fraunhofer Institute for Information and Data Processing IITB in Karlsruhe.

He and his team of researchers have developed the PRODAMI software that enables such errors to be detected rapidly and eliminated. “Employees can use our software to monitor, plan and control manufacturing systems in real time, in any production department,” says Kuntze. It is based on data mining tools, which can search through huge volumes of data in a minimum of time and identify characteristic information patterns out of a multitude of production and sensor data. The patterns provide a quick and easy way of detecting errors. Until now, the task of interpreting the flood of data delivered by sensors, numerical and process control systems was extremely time-consuming – and the data were therefore seldom utilized. The IITB is partnered in the PRODAMI project by the Fraunhofer Institutes for Production Systems and Design Technology IPK in Berlin and for Industrial Mathematics ITWM in Kaiserlautern. Data mining methods are already being employed successfully in marketing applications, financial planning, and genetic research. But it is rare to find them in a production environment, firstly because they haven’t until now provided real-time data analysis, and secondly because the data sources are too varied. PRODAMI, by contrast, can be easily integrated in different data and communication structures and analyzes the data in real time. To ensure that the PRODAMI tools can be integrated smoothly in online production processes, machine learning methods are also employed. The researchers will be presenting PRODAMI in a joint Simulation exhibit at the Hannover Messe from April 20 to 24 (Hall 17, Stand D60).