ML Prognostics Signal Clustering for Faster Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing abundance of high-fidelity network-connected sensors leads to a capacity for data accumulation that outpaces processing power in cloud computing, resulting in inefficiencies in machine learning (ML) prognostics, particularly in terms of throughput and latency.
Innovation Solution
A prognostics acceleration system that configures ML prognostic analyses by clustering signals based on correlation, automatically generating configurations that satisfy user constraints for accuracy and completion time, thereby reducing cumulative compute time and enabling parallelization without compromising model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single ML model is used to analyze all sensor signals, then model accuracy is maintained, but compute time increases quadratically and throughput decreases
Solution Approach 1:
The patent divides the set of N sensor signals into K clusters, where each cluster contains a subset of signals. Instead of using one ML model to analyze all N signals (which would require O(N²) compute time), the system uses K separate ML models, each analyzing a smaller subset of signals. This segmentation reduces the quadratic compute time while maintaining accuracy through correlation-based clustering that groups related signals together.
2Productivity
If sensor data is partitioned into smaller subsets for parallel processing, then throughput increases and latency reduces, but model accuracy deteriorates
Solution Approach 1:
The system segments sensor signals into K clusters that can be processed in parallel by K ML models, increasing throughput and reducing latency. The segmentation is optimized through correlation-based clustering to preserve important relationships between signals.
Solution Approach 2:
Before partitioning signals for parallel processing, the system performs preliminary correlation analysis to cluster signals based on their relationships. This preliminary action ensures that signals with strong correlations are grouped together in the same clusters, so that parallel processing does not break important signal relationships. The correlation-based clustering is computed in advance to guide the partitioning strategy.
3Loss of time
If correlation-based clustering is performed to optimize partitioning, then compute time is reduced quadratically, but system complexity increases
Solution Approach 1:
The system performs correlation-based clustering as a preliminary action before deploying the parallel ML model architecture. By computing signal correlations and forming clusters in advance, the system reduces the complexity of the main prognostic processing task. The clustering step is a one-time or periodic operation that enables subsequent efficient parallel processing without requiring complex real-time coordination between models.
Data Source
AI summary
Systems, methods, and other embodiments associated with quadratic acceleration boost of compute performance for ML prognostics are described. In one embodiment, a prognostic acceleration method includes separating time series signals into a plurality of alternative configurations of clusters based on correlations between the time series signals. Machine learning models are trained for individual clusters in the alternative configurations of clusters. One or more of the alternative configurations of clusters is determined to be viable for use in a production environment based on whether the trained machine learning models for the individual clusters satisfy an accuracy threshold and a completion time threshold. Then, one configuration is selected from the alternative configurations of clusters that were determined to be viable configurations. Production machine learning models are deployed into the production environment to detect anomalies in the time series signals based on the selected configuration.