Supporting process and quality engineers by automatic diagnosis of cause-and-effect relationships between process variables and quality deficiencies using data mining technologies (AutoDiag)

The through-process detection of cause-and-effect relationships by investigation of process and quality data with data mining techniques has been proved to be a powerful possibility to decrease quality deficiencies. Nevertheless this method is not used area-wide in the companies because of its complexity, the necessary specific knowledge which only few people in the company have and the missing adaptation of the tools to the specific problems of the steel production. These are the reasons to develop, implement and test robust, practicable and easy-to-use solutions which are specialised to steel quality problems. A generic common framework based on a well-known software tool (RapidMiner) was developed and implemented. Each industrial partner has developed individual interfaces to databases at the one hand and to the user interface on the other hand. For the data mining solutions different approaches were investigated: • brute force • individual adapted • smart components For each approach an investigation of an actual problem was performed to be able to show the usability of the developed solution and to analyse the requirements for the transferability. After a training of the personnel the systems were rolled out for daily usage. The experience of the target users was analysed and used for improvements. During the project it could be shown that the developed system can be used in an industrial environment. The increasing number of users, the number of investigations made by them as well as the request to integrate additional data sources show that the system is fully accepted by the users.