Over the past decades, various methods have been developed to analysis and monitor the dynamic metal processes, specially the extensively used cold rolling process. However, some limitation still exists for the traditional data analysis tools to be implemented well for these processes. For example, the performance of many of the traditional data analysis approaches cannot be guaranteed when the distribution assumption is violated. Meanwhile, it is still lack of systematic method to make good use of the huge condition parameters. In this article we develop a viable on-line anomaly incipient detection technique towards the cold rolling process of steel sheets. Based on the condition-based SPC, the proposed approach can monitor the multi condition parameters as well as the corresponding output characteristic in a real-time manner simultaneously and efficiently. It provides a framework for statistical process monitoring development under such dynamic manufacturing environment in order to improve the detecting Sensitivity and Specificity. The real data practical application verifies that this proposed approach can have an excellent performance without the normal distribution assumption, thus it has great potential to be employed in a large application area.