Facility State Change Detection Using Run Length Probability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for monitoring facility states and detecting abnormalities require extensive data learning and parameter adjustment, leading to delayed detection and the need for specialized knowledge, resulting in inefficiencies and inaccuracies in preventive maintenance.
Innovation Solution
A monitoring device that uses a data acquisition unit, state change detection unit, update unit, and information presentation unit to detect changes in facility states based on a preset prior distribution and run length probability distribution, allowing for immediate operation and accurate estimation of change times without the need for extensive learning or parameter adjustment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a model with high accuracy is built by learning from a large amount of data and adjusting parameters, then measurement precision is improved, but loss of time increases due to several months required for system import
Solution Approach 1:
The patent applies preliminary action by pre-defining multiple candidate models with different structures and parameters before actual use. When a state change is detected, the system can immediately switch to a pre-prepared model rather than learning from scratch, thus achieving high accuracy without the time-consuming parameter adjustment process
Solution Approach 2:
The patent changes parameters by selecting different pre-defined models with varying structures and parameters based on the detected state change. This allows the system to adapt to different operational conditions by switching models rather than continuously adjusting parameters, resolving the contradiction between accuracy and import time
2Reliability
If physical quantities are monitored by sensors to detect abnormalities, then reliability is improved, but difficulty of detecting and measuring increases due to requiring highly specialized knowledge
Solution Approach 1:
The patent introduces an intermediary layer (the monitoring device with multiple candidate models) between the sensor data and the abnormality detection decision. This intermediary automatically processes the specialized analysis, allowing users without specialized knowledge to achieve reliable abnormality detection by simply using the system's interface
Solution Approach 2:
The system performs self-service by automatically selecting appropriate models and detecting state changes without requiring user expertise. The monitoring device independently handles the complex analysis of sensor data, freeing users from needing specialized knowledge while maintaining high reliability
3Reliability
If a change in state is detected by monitoring physical quantities, then reliability is improved, but loss of time increases due to time lag between defect occurrence and detection
Solution Approach 1:
The patent applies preliminary action by preparing multiple candidate models that represent different operational states in advance. The system can immediately recognize state changes by comparing current data against these pre-defined models, detecting abnormalities at their earliest stages rather than waiting for significant physical quantity changes, thus reducing the time lag between defect occurrence and detection
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A state change detection unit obtains the data generation probability on the basis of the value of observation data and the value of a parameter of a prior distribution, obtains, on the basis of the data generation probability, a run length probability distribution for when the time-series observation data acquired up to the current point in time is used as a condition, and detects a change in the state of a facility on the basis of the run length probability distribution. Furthermore, an update unit updates the value of the parameter of the prior distribution using the value of the observation data, to generate the prior distribution to be used for calculating the data generation probability at the next point in time. In cases when the current point in time is determined to be a point of change in the state of the facility, the state change detection unit also searches for a change indication point, i.e. a point in time at which an indication of the change in the state of the facility began to appear, on the basis of the run length probability distribution.