Maintenance Support System Using Site-Level Model Reuse
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional installation maintenance support systems require relearning and significant computer resources for each installation, leading to inefficiencies and reduced credibility of the certainty factor for maintenance work content identification across installations with different specifications.
Innovation Solution
A maintenance support system that includes an element unit inference model acquisition unit to reuse learned models across installations with similar configuration elements, a maintenance work inference unit to estimate maintenance work content based on feature quantities from sensor data, and a maintenance work learning unit to iteratively evaluate and update the maintenance work inference model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If individual learning is performed for each installation, then the maintenance work content can be identified for that specific installation, but computer resources increase and learning time extends significantly
Solution Approach 1:
The patent segments the installation into multiple sites and performs learning independently for each site rather than for the entire installation. This allows learning results to be reused across installations that share common sites, reducing redundant learning while maintaining identification accuracy for each specific site's maintenance work content.
Solution Approach 2:
The patent creates a universal learning result for each site that can be applied across multiple installations. Once a site's maintenance patterns are learned, the results are reused for all installations containing that site, making the learning process multi-functional and eliminating redundant learning across similar installations.
2Reliability
If individual learning is performed for each installation, then the maintenance work content can be identified accurately, but computer resources for learning have to be prepared each time
Solution Approach 1:
The patent discards the approach of performing complete individual learning for each installation and recovers only the necessary site-specific learning results. By reusing learning results from common sites across different installations, the system reduces computer resource consumption while maintaining the certainty factor for maintenance work content identification.
3Productivity
If learning is performed for each installation separately, then the maintenance work content can be identified, but the results cannot be utilized in other installations
Solution Approach 1:
The patent creates universal learning results at the site level that can be applied across multiple installations. Each site's learning result becomes a reusable asset that improves both maintenance identification efficiency for that site and cross-installation applicability, as the same learning results serve multiple installations containing that site.
Solution Approach 2:
By segmenting the installation into independent sites, the patent enables learning results to be transferred and reused across installations. This segmentation allows the system to balance productivity for specific sites with adaptability across different installations, as each site's maintenance patterns can be independently learned and universally applied.
4Reliability
If the certainty factor is individually calculated for each installation, then the maintenance work content can be presented with confidence, but the decrease in population parameter of data lowers credibility
Solution Approach 1:
The patent merges data from multiple installations that share common sites to calculate the certainty factor. By combining population parameters across installations with identical or similar sites, the system increases the data population used for certainty factor calculation, thereby improving its credibility while still providing installation-specific maintenance guidance.
Data Source
AI summary
A maintenance support system (500) supports maintenance work to be carried out by a maintenance worker for an installation (30). An element unit inference model acquisition unit (107) sets an element unit inference model (108) for estimation of a maintenance work content in units of installation configuration elements as initial values of a maintenance work inference model (105) for estimation of the maintenance work content for the installation (30). A maintenance work inference unit (109) estimates the maintenance work content to be presented to the maintenance worker by applying the maintenance work inference model (105) to a feature quantity of an abnormality. A maintenance work learning unit (104) iteratively evaluates whether the maintenance work content has been effective or not for cancellation of the abnormality, learns an effective maintenance work content concerning the feature quantity of the abnormality, and updates the maintenance work inference model (105).