Event Analysis Apparatus Using Bayesian Networks for Alarm Cause Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing event analysis apparatuses in industrial plants primarily perform simple statistical processes, making it difficult for users to understand the cause-and-effect relationships between events such as alarms and operator procedures, which hinders efficient prediction and estimation of event causes and future occurrences.
Innovation Solution
An event analysis apparatus that collects and converts event logs into an event matrix, using Bayesian networks to calculate conditional probabilities and determine cause-and-effect relationships between device events, allowing for the construction of a directed graph structure to represent probabilistic relationships and predict event occurrences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simple statistical process is performed on event logs, then calculation complexity is reduced, but cause-and-effect relationship analysis capability deteriorates
Solution Approach 1:
The patent transforms event log data from simple occurrence counts to probabilistic parameters including conditional probabilities P(Ej|Ei) and joint probabilities P(Ei∩Ej). This parameter transformation enables Bayesian network construction while maintaining computational feasibility by using standardized probability calculations rather than complex causal reasoning.
Solution Approach 2:
The patent introduces Bayesian networks as an intermediary framework between raw event logs and cause-and-effect analysis. The Bayesian network serves as a mediator that structurally organizes event relationships through nodes and edges, enabling intuitive visualization and analysis of causal relationships without requiring complex direct computation from raw logs.
2Measurement precision
If Bayesian network construction with variable time width is implemented, then event relationship analysis precision is improved, but computational complexity increases
Solution Approach 1:
The patent implements dynamic time width adjustment in the Bayesian network construction process. The time width parameter can be varied to optimize the analysis of different event scenarios, allowing the system to adapt the temporal resolution to match the specific characteristics of event sequences being analyzed, thereby improving precision without fixed computational overhead.
Solution Approach 2:
The patent segments the event log data into discrete time-based blocks or windows for processing. By dividing the continuous event stream into manageable segments with configurable time widths, the system can analyze event relationships at different temporal granularities, improving analysis precision while maintaining computational tractability through localized processing.
Data Source
AI summary
An event analysis apparatus configured to analyze events including alarms generated in a plurality of devices and operations targeting the devices is provided. The event analysis apparatus includes an event log collection unit configured to collect an event log including an occurrence date and time of the event, a device identifier (ID) of the device in which the event occurs, and an event type ID of the event, an event log storage unit configured to convert the event log into an event matrix representing presence and absence of occurrence of each device event obtained by coupling the device ID and the event type ID in time series and save the event matrix; and an event analysis unit configured to calculate a conditional probability between the device events to construct a Bayesian network by dividing the event matrix into blocks, each of which has a predetermined reference time width and determining the presence and absence of the occurrence of each of the device events in each of the blocks, and decide a device event as a cause of a device event of an analysis target or a device event to be generated later using the constructed Bayesian network.