Installation-Aware Device State Estimation Using Collation Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing diagnostic technologies face challenges in accurately diagnosing devices with different installation conditions due to differences in installation environments, leading to decreased diagnostic accuracy and redundant processing when generating models for each device.
Innovation Solution
A data processing apparatus that generates collation data based on design and installation conditions, allowing for accurate state estimation by collating sensor data with simulated data for various installation scenarios without redundant model generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a model is generated from normal data for a certain device (device M), then the model can be used for diagnosis, but diagnostic accuracy decreases when applied to another device (device N) with different installation conditions
Solution Approach 1:
The patent segments the diagnostic model into two independent parts: (1) a general model structure learned from normal data of device M, and (2) installation condition-specific parameters or adjustments tailored to device N's specific installation environment. This segmentation allows the general diagnostic framework to be reused while adapting to specific installation conditions, thereby maintaining diagnostic accuracy across different devices without regenerating entire models.
2Measurement precision
If models are generated for each device to account for installation condition differences, then diagnostic accuracy is maintained, but redundant processing occurs in data acquisition and model generation
Solution Approach 1:
The patent performs preliminary action by pre-learning the general diagnostic model structure from normal data during an offline training phase. This pre-learned structure serves as a reusable template that can be quickly adapted to different installation conditions without requiring complete model regeneration. The preliminary extraction of common diagnostic patterns eliminates redundant processing when diagnosing multiple devices with different installation conditions.
3Reliability
If normal data is acquired for each device to generate device-specific models, then installation condition differences are accounted for, but data acquisition and model generation become redundant
Solution Approach 1:
The patent merges the data acquisition and model generation processes by utilizing normally-operating data from the target device itself, combined with the pre-learned general model structure. Instead of requiring separate normal data acquisition campaigns for each device, the system merges the universally applicable diagnostic patterns (learned once) with device-specific operational data, thereby reducing data acquisition time while maintaining diagnosis reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A collation data generation unit (103) generates collation data to be used for estimating a state of a device in accordance with an installation condition which is a condition to be employed when the device is installed. A sensor data acquisition unit (104) acquires sensor data of a diagnostic target device (200) which is an installed device. A state estimation unit (105) collates the sensor data with the collation data, and estimates a state of the diagnostic target device (200).