Sensor Correlation Analysis for Multivariate Time-Series Error Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Monitoring and analyzing multivariate time-series data in equipment processes is time-consuming and expensive, making it difficult to diagnose errors effectively due to varying temporal characteristics of the data.
Innovation Solution
A deep learning-based analysis system that uses a detection device to create multivariate time-series data from sensors and an analysis device with processors to learn correlation degrees between sensors, convert data into image format, and calculate error scores using learning models to detect abnormal tendencies and output error data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multivariate time-series data is monitored and analyzed in detail, then error diagnosis accuracy is improved, but analysis time and cost increase significantly
Solution Approach 1:
The patent segments the analysis process into two distinct models: a first learning model that processes multivariate time-series data to capture temporal correlations, and a second learning model that processes extracted features to identify error patterns. This segmentation allows each model to specialize in specific aspects of analysis, improving overall accuracy while reducing the computational burden on a single model.
Solution Approach 2:
The patent extracts key features and correlations from the complex multivariate time-series data using the first learning model, then feeds only these extracted features to the second learning model. This extraction process removes redundant information and reduces the data volume that needs to be analyzed in detail, thereby reducing analysis time while preserving essential error-diagnostic information.
2Reliability
If multivariate time-series data with varying temporal characteristics is analyzed, then comprehensive error detection is improved, but difficulty of analysis increases
Solution Approach 1:
The patent changes the parameters of data representation by transforming multivariate time-series data with varying temporal characteristics into a standardized feature space through the first learning model. This parameter transformation allows the second learning model to analyze diverse data types using consistent parameters, reducing analysis difficulty while maintaining comprehensive error detection capability.
3Measurement precision
If deep learning models process raw multivariate time-series data directly, then analysis accuracy is improved, but computational complexity and cost increase
Solution Approach 1:
The patent performs preliminary processing of multivariate time-series data using the first learning model to extract correlations and features before the main analysis is conducted by the second learning model. This preliminary action prepares the data in advance, reducing the computational complexity of the subsequent analysis while preserving the information needed for accurate error diagnosis.
Data Source
AI summary
A deep learning-based analysis system includes a detection device configured to create multivariate time-series data through a plurality of sensors of an equipment process, and an analysis device including at least one processor, wherein, when receiving the multivariate time-series data created through the plurality of sensors from the detection device, the processor of the analysis device is configured to obtain a correlation degree between a plurality of sensors based on a first learning model using the received multivariate time-series data as input, and calculate an error score for each sensor based on a second learning model using time-series data for each sensor extracted from the received multivariate time-series data as input.