Uncertainty Quantification of Data-driven Quality Prediction Model For Realizing the Active Sampling Inspection of Mechanical Properties in Steel Production

• DOI: 10.1007/s44196-024-00451-6
• Publication Date: 2024-04-02T12:02:00Z
• Authors: Yong Song; Feifei Li; Zheng Wang; Baozhong Zhang; Borui Zhang
• Keywords: Active sampling; 0209 industrial biotechnology; Quality inspection; Electronic computers. Computer science; Mechanical performance prediction; QA75.5-76.95; 02 engineering and technology; Uncertainty quantification

Abstract Pre-production quality defect inspection is a crucial step in industrial manufacturing, and many traditional strategies suffer from inefficiency issues. This especially true for tasks such as mechanical performance testing of steel products, which involve time-consuming processes like offline sampling, specimen preparation, testing. The volume significantly impacts the production cycle, inventory, yield, labor costs. Constructing data-driven model predicting product implementing proactive sampling based on prediction results an appealing solution. However, uncertainty models poses challenging problem that needs to be addressed. paper proposes active approach products quantification predictive performance. objective reduce both frequency omission rate site. First, ensemble improved lower upper bound estimation established interval specific value quantitatively estimated using probability distributions. Then, failure built size distribution. By determining appropriate threshold, trade-off between accuracy detection (recall rate) balanced, enabling establishment strategy. Finally, this functionality integrated into manufacturing execution system factory, realizing sampling. proposed validated real datasets. When threshold set 30%, recall samples are 75% 100%, respectively. Meanwhile, only 5.33%, while controlling risk omission. represents 50% reduction compared rules commonly used actual production. overall efficiency improved, costs reduced.