Steel bars, used as raw materials for automotive engine and transmission components, require high strength, toughness, and surface quality. Surface defects are particularly dangerous, because they can cause fractures due to stress concentration under ...
Steel bars, used as raw materials for automotive engine and transmission components, require high strength, toughness, and surface quality. Surface defects are particularly dangerous, because they can cause fractures due to stress concentration under repeated loads while driving, which can lead to accidents. Poor shape control during rolling process is a primary cause of these defects, and the process is challenging to optimize due to the dynamic production environment, where changes over time complicate the rolling process and make it difficult to adjust manually.
This study proposes a data mining approach to optimize cross-sectional area and uniformity during rolling, even in a changing production environment. In the first step, we use historical production data and analysis of variance to identify variables that indirectly represent the production environment. We then apply an unsupervised Gaussian mixture model to cluster production data with similar environments. In the second step, shape quality is separated into two response variables: cross-sectional area and uniformity and these are converted into overall preference scores between 0 and 1 using desirability function. After that, we find the optimal process parameters for each cluster that maximize preference scores using the Patient Rule Induction Method. By applying the proposed method to the steel bar rough rolling process, its effectiveness was verified. The products produced within the optimal parameter ranges showed significant improvement, achieving shape quality within the top 63.1% to 99% of all products.