摘要
Abstract
Test data from large-scale equipment are characterized by explicit engineering thresholds,progressively skewed distribution due to aging,and the demand for incremental processing.To address these characteristics,an incremental boxplot anomaly detection method that integrates engineering thresh-olds with dynamic whiskers is proposed.Firstly,a hierarchical binning system is constructed based on engi-neering thresholds and the concept of health degree,enabling a fine-grained division of"fault-subhealth-health"states.Secondly,an incremental quartile estimation algorithm based on histograms and the Cumula-tive Distribution Function(CDF)is designed to realize real-time processing of incremental data.Thirdly,the Quartile Skewness Measure(QSM)is adopted to dynamically perceive changes in data distribution pat-terns,adaptively adjusting the whisker range of the boxplot to improve detection robustness under skewed data.Finally,a multi-dimensional diagnostic rules are established to classify anomalies into 4 categories(e.g.,hardware faults,potential risks)and corresponding disposal strategies are provided.Experiments are conducted on voltage data of lithium batteries from a certain type of aircraft.The results show that the pro-posed method can accurately identify over-threshold faults,sub-health trends,and weak outliers,and support real-time incremental updates,making it suitable for full life-cycle health monitoring of large-scale equipment.关键词
异常检测/箱形图/工程阈值/动态触须/偏态数据Key words
outlier detection/boxplot/engineering thresholds/dynamic whiskers/skewed data分类
信息技术与安全科学
