IMPLEMENTASI EXPONENTIALLY WEIGHTED MOVING AVERAGE (EWMA) PADA DATA TEMPERATUR SENSOR UNTUK DETEKSI ANOMALI PROSES

Abstrak

This study aims to implement the Exponentially Weighted Moving Average (EWMA) method to detect anomalies in sensor temperature data from an electrophoretic painting process. The dataset consists of 720 observations recorded at 30-second intervals. The analysis begins with descriptive statistics to understand the data’s distribution and characteristics. EWMA is then applied using two smoothing parameters, λ = 0.2 and λ = 0.3, to compare their sensitivity in detecting anomalies. Results show that λ = 0.2 detects 218 anomalies (30.3%), while λ = 0.3 detects 141 anomalies (19.6%). A residual analysis is conducted to evaluate the randomness of the data after smoothing, and the Ljung–Box test reveals significant autocorrelation in the residuals (p-value = 0.0000), indicating that the model has not fully removed the time-dependent structure. These findings demonstrate that EWMA is effective for sensor-based quality monitoring, but model refinement is still necessary to reduce false alarms. Further development is recommended by optimizing the λ parameter and integrating EWMA with time-series or machine learning models to ensure robust anomaly detection in real-world industrial environments.