Liquid Composition Change Detection Using PCA Sensor Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to effectively detect sudden and abnormal changes in the physicochemical composition of liquids, and struggle to distinguish between gradual and sudden changes, particularly in water quality monitoring systems, leading to potential contamination and quality degradation.
Innovation Solution
A computer-implemented method using collocated sensors, data partitioning, and principal component analysis (PCA) to identify clusters in sensor measurements, combined with density-based and centroid-based data partitioning algorithms, to detect and categorize abnormal changes in liquid composition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional threshold-based sensor analysis is used, then simple detection of known errors is achieved, but detection of new anomalies and discrimination between gradual and sudden changes is lost
Solution Approach 1:
The patent transforms sensor measurements into a new parameter space using Principal Component Analysis (PCA). This dimensional transformation converts correlated sensor readings into uncorrelated principal components, enabling the system to detect anomalies in the transformed space where sudden changes manifest as distinct patterns. The parameter change from raw sensor data to principal components resolves the contradiction by providing both reliable detection and adaptability to new anomaly types.
Solution Approach 2:
The patent replaces traditional mechanical threshold-based detection with a data-driven computational approach. Instead of using fixed thresholds that require manual calibration and cannot adapt to new conditions, the system employs unsupervised machine learning algorithms (PCA and DBSCAN) that automatically learn normal patterns from data and detect deviations. This substitution enables detection of new anomaly types and discrimination between gradual and sudden changes without requiring pre-programmed knowledge of specific failure modes.
2Measurement precision
If data partitioning with density-based algorithms is applied, then discrimination between gradual and sudden changes is achieved, but computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the measurement process into distinct stages: first transforming data via PCA to reduce dimensionality and extract essential patterns, then applying DBSCAN clustering to segment data points into normal and anomalous groups. This segmentation approach enables precise discrimination between gradual and sudden changes while managing computational complexity through progressive data processing rather than analyzing all raw data simultaneously.
Solution Approach 2:
The patent introduces principal components as an intermediary between raw sensor measurements and final anomaly detection. The PCA transformation creates a simplified intermediate representation that captures the essential variance in the data with fewer dimensions. This intermediary structure makes the subsequent DBSCAN clustering computationally tractable while preserving the information needed to discriminate between gradual and sudden changes, thus resolving the contradiction between precision and complexity.
Data Source
AI summary
A field of measuring and monitoring liquids is provided. More specifically, the improved detection of a change in physicochemical composition of a liquid is provided, based on a set of measurements from sensors collocated in the liquid, and applying data partitioning to the set of measurements. If at least two measurement clusters are identified, an abnormal change in the composition of the liquid is detected.