Anomaly Detection with Artificial Intelligence: Post-mortem analysis of LHC ion beam losses during high-energy beam dumps

The analysis of beam losses and detection of anomalies during particle collision experiments in the Large Hadron Collider is of paramount importance to ensure the integrity and safety of the machine’s equipment. This study addresses the research gap by focusing on the post-mortem detection of anomalies specifically for ion-ion collisions. This area has received less attention in comparison to proton-proton collisions. The aim of this work is to classify beam dumps and detect anomalies. It also compares the ion dump classification performance of two models: one trained on proton data and another trained on ion data. The methodology is driven by comprehensive data evaluation and uses statistical methods to identify correlations between beam losses and relevant variables. Based on the statistical analysis, linear and polynomial regression models are developed and classification thresholds are derived from these models. To further improve the classification accuracy, three polynomial regression models are tested. Polynomials of different orders are automatically selected using the Bayesian information criterion, root mean square error and R-2-adjusted criteria. The results show that the linear ion model achieves an accuracy of 93.75% for the classification of ion dumps, while the proton model achieves 79.17% for the same task. This discrepancy indicates significant differences between proton and ion dumps. However, there are no differences in the classification of known anomalies such as asynchronous dump tests and 10 Hz dumps. In conclusion, this study highlights the need for tailored anomaly detection systems specifically designed for ion dumps. The results emphasize the importance of analyzing and understanding beam anomalies in ion-ion collisions and provide insight into the classification performance of different models. The implementation of dedicated anomaly detection mechanisms for ion dumps will assist experts in identifying anomalous behavior. This will help experts to improve the overall operational safety and efficiency of the Large Hadron Collider.