TWO-STEP METHODOLOGY FOR STATISTICAL ANOMALY DETECTION AND PREDICTION USING XGBOOST REGRESSION IN BLOWER MOTOR VIBRATION TIME SERIES DATA

초록

Analyzing the vibration of blower motors in industrial sites to detect and predict anomalies is important for increasing operational efficiency and improving predictive maintenance. Existing methods risk malfunctions due to over-sensitivity, and deep learning approaches, in particular, require large datasets for training and are difficult to maintain. This research is divided into two phases: diagnostic analysis and predictive analysis. The first phase, diagnostic analysis, utilizes the Pruned Exact Linear Time (PELT) algorithm and Statistical Process Control (SPC) techniques to identify vibration data points with abrupt pattern changes and outside the normal operating range. The second step, predictive analysis, utilizes the Extreme Gradient Boosting (XGBoost) Regression algorithm to identify patterns in the vibration data to predict the occurrence and timing of potential failures. The algorithm used in the study provides computational efficiency and high prediction accuracy, which can compensate for the shortcomings of existing methods. The study also presents a methodology that can effectively detect and predict anomalies in blower motors and similar mechanical equipment in industrial environments. © INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING.

키워드