PCA-Based Maintenance Prediction for Measurement Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting maintenance on measurement devices with multiple sub-elements are inadequate, as they fail to provide timely maintenance recommendations and often require extensive analysis, especially when the number of sub-elements is high, and do not effectively identify which sub-elements require attention.
Innovation Solution
The proposed solution involves collecting time series data from measurement sub-elements, applying principal components analysis to reduce the data dimensionality, and performing change-point detection on the resulting reduced matrix signal to identify changes in distribution, which are interpreted as potential malfunctions, thereby scheduling maintenance and pinpointing faulty sub-elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional change-point detection methods are applied to measurement devices with a large number of sub-elements, then the device health monitoring becomes possible, but the computational complexity and analysis time increase significantly
Solution Approach 1:
The patent segments the large set of measurement sub-elements into a smaller number of principal components through dimensionality reduction. Instead of analyzing all individual sub-elements separately, the method groups them into representative principal components that capture the essential variations in device health, thereby reducing computational complexity while maintaining monitoring effectiveness.
Solution Approach 2:
The patent introduces principal components as intermediary representations between the raw measurement sub-elements and the change-point detection process. These principal components serve as mediators that summarize the state of multiple sub-elements into fewer variables, making the subsequent change-point detection computationally feasible while still reflecting the overall device health status.
2Measurement precision
If all measurement sub-elements are analyzed simultaneously to detect device deterioration, then the detection accuracy improves, but the time required for maintenance prediction increases
Solution Approach 1:
The patent segments the analysis process by first reducing the dimensionality of the measurement data into principal components, then performing change-point detection on these reduced components. This segmentation allows the method to maintain detection accuracy by preserving the essential information in fewer variables, while significantly reducing the time required for maintenance prediction compared to analyzing all original sub-elements simultaneously.
3Productivity
If the number of measurement sub-elements is reduced through dimensionality reduction, then the computational efficiency improves, but the ability to identify specific faulty sub-elements deteriorates
Solution Approach 1:
The patent implements feedback by using the principal component analysis results to guide the identification of specific faulty sub-elements. After performing change-point detection on the reduced principal components, the method uses the loading information from the PCA to trace back which original sub-elements contribute most to the detected changes, thereby recovering the sub-element identification capability that would otherwise be lost in the dimensionality reduction process.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
An apparatus for predicting maintenance comprises means for: - Collecting a plurality of time series of measurements from a corresponding plurality of measurement sub-elements of the measurement device, - Obtaining a matrix measurement signal comprising said time series of measurements, - Applying a principal components analysis to the matrix measurement signal in order to obtain principal components of the matrix measurement signal, wherein the number of principal components is lower than the number of time series of measurements, - Obtaining a reduced matrix measurement signal as a function of the matrix measurement signal and at least one of the principal components, wherein the reduced matrix measurement signal comprises principal time series of values along a first dimension, and principal values vectors along a second dimension, - In response to detecting a change point in the distribution of the values of the principal time series, emitting a maintenance recommendation notification.