Machine Abnormality Detection Using Latent Sensor Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection systems face challenges in accurately detecting machine abnormalities with limited sensor installation due to physical restrictions and cost constraints, as they struggle to differentiate between relevant and irrelevant sensor data variations.
Innovation Solution
A computer-based abnormality detection system that uses a machine learning model with an encoding unit to generate latent expressions for estimating data from one sensor, allowing for robust estimation of data from another sensor, even when not installed, by separating variations into different latent expressions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all necessary sensors are attached to the target machine, then measurement precision and reliability are improved, but device complexity and cost increase
Solution Approach 1:
The patent creates a virtual copy of the sensor data through latent expression modeling. Instead of installing physical sensors on the target machine, the system learns the relationship between sensors on a reference machine and creates corresponding latent expressions that replicate the behavior of uninstalled sensors. This allows the system to estimate sensor readings that would otherwise require physical hardware installation.
Solution Approach 2:
The patent introduces latent expressions as an intermediary between physical sensor data and the estimation of uninstalled sensor data. The encoding unit transforms physical sensor readings into latent expressions, which then serve as intermediaries to reconstruct and estimate data from sensors that are not physically installed on the target machine.
2Device complexity
If regression models are used to estimate sensor data, then device complexity is reduced, but measurement precision deteriorates due to inclusion of unrelated variations
Solution Approach 1:
The patent segments the sensor data into distinct latent expressions through the encoding unit. Instead of treating all sensor variations as a single regression problem, the system divides the input data into multiple latent dimensions, each capturing specific patterns of variation. This segmentation allows the decoding unit to selectively reconstruct only the relevant variations needed for accurate sensor data estimation.
Solution Approach 2:
The patent transforms the sensor data from the original measurement space into a latent expression space through encoding. This dimensionality transformation allows the system to represent complex sensor relationships in a compressed latent space, where unrelated variations are separated from relevant signals. The decoding unit then maps these latent expressions back to estimate sensor data with higher precision than traditional regression methods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided are an abnormality detection system and an abnormality detection method capable of performing more stable abnormality detection. An abnormality detection system that detects an abnormality of the target machine by a computer includes a communication unit configured to acquire first data from a first sensor attached to the target machine and second data from a second sensor attached to the target machine, an arithmetic unit, and a memory unit. The arithmetic unit includes an encoding unit trained to generate latent expressions including a predetermined latent expression that estimates the second data on the basis of the first data, a decoding unit trained to restore the first data from the latent expressions, and an abnormality detection unit configured to detect the abnormality of the target machine on the basis of a restoration error between the first data and the first data restored by the decoding unit.