A New EWMA Control Chart for Outlier Detection in Nonlinear Profiles of Automobile Engines

ABSTRACT To enhance the effect of outlier detection in the nonlinear profile of automobile engines, this paper presents the time exponentially weighted moving average (TEWMA) control chart. Although the traditional exponentially weighted moving average (EWMA) method has shown good performance in process monitoring, it fails to handle data with irregular time intervals, which limits its application in complex dynamic data. By adaptive weight adjustment, TEWMA can accurately reflect the real dynamic of vibration data. Firstly, the weighted kernel entropy independent component analysis (WKEICA) is used to reduce data dimensionality. Then, the dimensionally reduced data is input into the isolation forest (IF) model to calculate the outlier scores. Next, the outlier scores are imported into the TEWMA control chart, and the values of each feature are dynamically adjusted based on these outlier scores. Finally, a dual threshold adjustment strategy combining traditional and dynamic control limits is introduced to reduce the false positive rate (FPR) and ensure the stability of process monitoring. The company's automobile engine data set is used for verification, and the vibration data is extracted for outlier detection, and the detection accuracy reaches 96.42%. Compared with the EWMA and cumulative sum (CUSUM) control charts, the new method has increased the outlier detection accuracy by 10.79% and 31.42%, respectively. This paper provides a new method and perspective for outlier detection in monitoring the nonlinear profile of automobile engines.