Edge Sensor Data Transmission Reduction via Simulated ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Legacy machines in production lines lack sensor analytics, relying on manual maintenance due to challenges in scalability and the need for extensive training data for machine learning models, which is time-consuming and costly, and are affected by noise and signal distortions in sensor data.
Innovation Solution
A system involving an edge computer and a server that processes and classifies sensor data, using simulated failure mode features to train a classifier, reducing data transmission when classification accuracy meets a threshold, and employing feature extraction techniques like spectral kurtosis and band-limitation to minimize noise and resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning techniques are used to automatically extract features from sensor data, then manual maintenance effort is reduced and analytics automation is improved, but extensive training data is required which increases time and cost
Solution Approach 1:
The patent creates simulated sensor data that copies the characteristics of real sensor data from physical models. This simulated data serves as a substitute for actual training data, allowing ML models to be trained without requiring extensive real-world data collection. The simulated failure modes replicate real failure patterns while avoiding the time-consuming process of collecting actual failure data from legacy machines.
2Loss of time
If simulated data is used to train machine learning models, then actual data collection time and cost are reduced, but the accuracy of failure mode identification may be compromised
Solution Approach 1:
The patent introduces physical models as an intermediary between real machine behavior and simulated sensor data. These physical models capture the essential dynamics and failure modes of legacy machines, generating simulated data that maintains fidelity to real-world patterns. This intermediary layer ensures that simulated training data accurately represents actual failure modes while avoiding the need for extensive real data collection.
3Measurement precision
If continuous data transmission from edge computer to server is maintained for high accuracy classification, then failure mode identification accuracy is improved, but data transmission costs and resource usage increase
Solution Approach 1:
The patent implements a threshold-based transmission strategy where the edge computer only transmits data to the server when classification confidence falls below a certain threshold. When the classifier is confident (above threshold), processing occurs locally without transmission. This partial transmission approach maintains high accuracy for uncertain cases while significantly reducing overall data transmission and associated energy costs.
Data Source
AI summary
Systems and methods described herein are directed to minimizing the resource requirements for edge and network while keeping the accuracy of machine learning classifier by utilizing simulated test data. Once sufficient measured test data is collected by the server, the server instructs the edge computer to reduce the transmission of data received from the corresponding sensors.