Research on Quality Anomaly Control Charts in Manufacturing Processes Based on Isolation Forest

With the continuous advancement of digitalization in the manufacturing industry, substantial amounts of data are generated during the production process. Many data-driven methods aim to quickly and accurately identify anomalies related to product quality or process states within massive highdimensional datasets. This study leverages binary partition trees to segment samples in high-dimensional datasets, demonstrating that anomalous samples require fewer partitions compared to ordinary samples. Using the Monte Carlo method, the average path length for the separation of anomalous samples is analyzed. Based on this analysis, a one-sided anomaly score control chart is developed to monitor anomalies in high-dimensional datasets. Application results demonstrate that the proposed method efficiently identifies anomalous samples in such datasets. These findings provide technical support for quality anomaly detection in multivariate manufacturing processes.