Alarm Flood Pattern Mining for Root-Cause Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Building automation systems (BAS) generate high volumes of alarms due to univariate and multivariate nuisance alarms, leading to 'alarm floods' that overwhelm operators and hinder maintenance efforts, with existing solutions failing to effectively manage multivariate alarms caused by complex interactions and sequences.
Innovation Solution
An alarm flood management system utilizing machine learning and pattern mining methods on ANSI/ASHRAE standard 135-compliant event logs to predict and suppress alarm floods, providing root-cause and strategy information to operators, comprising an extracting device, aligning device, clustering device, pattern mining device, controller, and segmenting device to process and analyze alarm data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional alarm management techniques (delay timers, deadbands) are applied to univariate nuisance alarms, then chattering alarms are reduced, but multivariate alarm floods caused by complex interactions and sequences remain unmanaged
Solution Approach 1:
The system segments alarm floods into individual alarm sequences and identifies root cause alarms within each sequence. By breaking down complex multivariate alarm floods into manageable segments and analyzing their temporal relationships, the system can identify and suppress the root cause alarm that triggers the cascade, effectively managing both univariate and multivariate alarm scenarios.
Solution Approach 2:
The system introduces an intermediary alarm flood management system between the building automation system and operators. This intermediary analyzes alarm sequences, identifies patterns, determines root cause alarms, and suppresses appropriate alarms before they reach operators, enabling effective management of complex multivariate alarm floods that traditional direct management techniques cannot handle.
2Reliability
If all alarms generated by BAS are presented to operators, then complete system monitoring is maintained, but operators become overwhelmed and alarm effectiveness decreases
Solution Approach 1:
The system extracts and identifies the root cause alarm from alarm flood sequences using pattern recognition and temporal relationship analysis. By taking out only the essential root cause alarm and suppressing the consequential cascade alarms, the system maintains complete system monitoring capability while presenting a manageable number of alarms to operators, directly addressing the contradiction between monitoring completeness and operator workload.
Solution Approach 2:
The system implements feedback by continuously analyzing alarm sequences, learning from historical patterns, and dynamically adjusting alarm suppression decisions. The feedback mechanism ensures that complete system monitoring is maintained while intelligently filtering alarms based on learned patterns, preventing operator overwhelm while preserving reliability.
3Productivity
If machine learning and pattern mining methods are implemented to predict and suppress alarm floods, then alarm rates are reduced, but system complexity and computational requirements increase
Solution Approach 1:
The system performs preliminary action by pre-processing alarm sequences, clustering similar patterns, and building prediction models in advance. By preparing alarm flood patterns and root cause identification rules beforehand, the system reduces real-time computational complexity while maintaining high alarm processing efficiency, effectively balancing productivity improvement against system complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides an alarm flood management system (1) including an extracting device (2), an aligning device (3), a clustering device (4), a pattern mining device (5), a controller (6), and a segmenting device (9). The extracting device (2) extracts alarm log data into alarm flood sequences. The aligning device (3) aligns the alarm flood sequences for obtaining the relationship between the alarm flood sequences. The clustering device (4) clusters the alarm flood sequences according to the relationship between the alarm flood sequences for obtaining groups of alarm flood sequences. The pattern mining device (5) performs pattern mining on the groups of alarm flood sequences for obtaining patterns of the alarm flood sequences. The controller (6) utilizes the patterns of the alarm flood sequences for providing a root-cause/strategy. The controller (6) suppresses the incoming alarm data by the predicted incoming alarm data and updates the root-cause/strategy according to the predicted incoming alarm data.