Design and Implementation of an Anomaly-Detection Model Based on an LSTM Autoencoder for Early Gas Leakage Detection

Purpose: This study proposes a gas-absorbing device with high reliability and availability. To detect anomalies in the device, we applied an anomaly-detection algorithm based on artificial intelligence.BRMethods: We used an LSTM-based autoencoder model to train normal data, and the model was processed to detect anomalies when the mean average error (MAE) deviated more than three times the standard deviation of its mean. To handle false alarms, we introduced a method that assigns increment and decrement scores for all anomaly and normal points and diagnoses an anomaly point when the accumulated score exceeds a designated threshold.BRResults: The model could diagnose anomalies approximately four months before failure, and it could correctly diagnose a return-to-normal after repair.BRConclusion: This algorithm is suitable for systems that are employed in environments that must operate continuously every day.