Air Conditioning Supervised Learning for Refrigerant Leak Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing refrigerant amount determination devices fail to accurately detect small leaks due to their inability to precisely assess the impact on air conditioning performance, leading to uncertain maintenance timing.
Innovation Solution
A learning device that generates inference models using supervised learning to identify normal ranges of feature amounts from operation data, allowing early detection of abnormalities in air conditioning systems by comparing actual data against trained models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If refrigerant amount indicator value is calculated from operation data, then refrigerant leakage can be detected, but accuracy of detecting small leaks and determining maintenance timing is insufficient
Solution Approach 1:
The patent segments the detection process into multiple independent feature amounts (first feature amount from first data group, second feature amount from second data group). Each feature amount is calculated separately and then integrated, allowing the system to detect different aspects of refrigerant leakage independently and combine them for comprehensive accuracy.
Solution Approach 2:
The patent introduces a learning device as an intermediary component that generates inference models. This learning device processes operation data and creates models that infer normal ranges of feature amounts, serving as a mediator between raw data and final leakage detection, thereby improving measurement precision while preserving performance impact information.
2Reliability
If refrigerant amount indicator is used to detect leakage, then leakage detection is enabled, but timing for maintenance cannot be precisely determined
Solution Approach 1:
The patent performs preliminary actions by calculating multiple feature amounts and generating inference models before actual leakage detection. The learning device pre-processes operation data to establish normal ranges, so when leakage occurs, the system can immediately compare actual values against pre-established benchmarks, enabling precise maintenance timing determination.
Solution Approach 2:
The patent changes parameters by using multiple different data groups (first data group, second data group) to calculate different feature amounts. This multi-parameter approach allows the system to detect leakage while simultaneously assessing its impact on air conditioning performance, thereby determining maintenance timing based on both detection and performance degradation.
3Measurement precision
If small refrigerant leakage occurs (10%), then refrigerant amount indicator shows leakage, but air conditioning performance is not significantly decreased
Solution Approach 1:
The patent applies local quality by focusing on specific feature amounts derived from different data groups. Instead of treating all data uniformly, it calculates a first feature amount from a first data group and a second feature amount from a second data group, allowing localized analysis of different operational aspects to distinguish significant from insignificant leakage.
Solution Approach 2:
The patent changes parameters by introducing multiple feature amounts with different characteristics. By comparing the first feature amount (indicating leakage) with the second feature amount (reflecting performance impact), the system can differentiate between small leaks that don't affect performance and significant leaks that do, reducing false alarms.
Data Source
AI summary
Operation data of an air conditioning apparatus includes a first data group and a second data group that is not the same as the first data group. A learning device includes: a first calculation unit configured to calculate a first feature amount from the first data group of the air conditioning apparatus during a learning period; and a learning unit configured to generate a first inference model that infers a first normal range of the first feature amount from a second data group by performing supervised learning using the second data group with the first feature amount being set as truth data, the first feature amount being obtained by calculation by the first calculation unit.