CERN Accelerating science

 
A TS with time lag effect $\mathbf{x}^1_{t - 1} \rightarrow \mathbf{x}^2$ and instantaneous effect $\mathbf{x}^1_t \rightarrow \mathbf{x}^3_t$~\cite{peters2017elements}.
Schematic of the CMS experiment~\cite{focardi2012status}.
The frontend electronics of the HE data acquisition chain, including the SiPMs, the frontend readout cards, and the optical link connected to the back-end electronics~\cite{strobbe2017upgrade}. Each readout card contains twelve QIE11 for charge integration, an Igloo2 FPGA for data serialization and encoding, and a VTTx optical transmitter.
\textsc{AnomalyCD}: our anomaly CD framework: use-case of the HCAL monitoring. The approach builds a GCM and a BN on the binary anomaly-flag data generated from several systems using trained and online-AD tools.
Temporal anomaly CD approach diagram. The approach infers causal interaction among monitoring sensor variables from binary anomaly data.
The active mask of the LHC operation status from August to December of 2022. The active $mask=1$ refers to the LHC during its normal operation run of collision experiment or idle, whereas the $mask=0$ corresponds to the LHC under other non-physics operation states, e.g., technical stop and maintenance development.
Sensor TS reading data from all four RMs of the RBX-HEP07. The HEP07\_i denotes the $i^{\text{th}}$ RM of the RBX.
Online temporal AD on RBX-HEP07 SCH sensors. (Left to right) sensor signal, signal trend estimation, $\Lambda_\iota$ of $\textsc{TrendDriftDetection}$, $\Lambda_\theta$ of $\textsc{MovingSDOutlierDetection}$, and $\Lambda_\eta$ of $\textsc{SpectralOutlierDetection}$.
Online temporal AD on the RBX-HEP07-RM-1 sensors.
Number of detected anomaly flags from all RMs of RBX-HEP07. The humidity sensors have a higher count due to drifting trends.
Anomaly binary flag data from our proposed online AD approach on RBX-HEP07 sensors: a) the raw anomaly data with approximately $400K$ samples and the sparse regions are annotated, and b) sparse compressed data through our sparse handling algorithm with $l_m=10$ and reducing the sample size to approximately $900$.
Anomaly binary flag data from our proposed online AD approach on RBX-HEP07 sensors: a) the raw anomaly data with approximately $400K$ samples and the sparse regions are annotated, and b) sparse compressed data through our sparse handling algorithm with $l_m=10$ and reducing the sample size to approximately $900$.
Temporal GCM of HEP07-RM using time-lag $t=0, \dots, 5$.
Temporal GCM-DAG network of HEP07-RM after edges pruning.
Causal graph of EasyVista's monitoring system during normal operation.
The generated TS anomaly-flag data using our online-AD on the EasyVista sensors.
The generated TS anomaly-flag data using our online-AD on the EasyVista sensors.
Performance ranking for pairwise comparisons using Nemenyi: a) without sparse data handling, and b) with sparse data handling. The $\overline{\text{CD}}$ is the \textit{critical difference distance}, and the horizontal bars denote mean rank differences smaller than the value of the $\overline{\text{CD}}$.
Performance ranking for pairwise comparisons using Nemenyi: a) without sparse data handling, and b) with sparse data handling. The $\overline{\text{CD}}$ is the \textit{critical difference distance}, and the horizontal bars denote mean rank differences smaller than the value of the $\overline{\text{CD}}$.
The estimated TS GCM using \textsc{AnomalyCD} for the EasyVista system from binary anomaly data: a) \textsc{AnomalyCD}, and b) \textsc{AnomalyCD}-Directed.
The estimated TS GCM using \textsc{AnomalyCD} for the EasyVista system from binary anomaly data: a) \textsc{AnomalyCD}, and b) \textsc{AnomalyCD}-Directed.