Anomaly Prediction Module for System Profile Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Information handling systems face challenges in predicting and preventing anomalies in processing systems without requiring raw sensor data, additional sensors, or actual system failures, which can lead to unscheduled downtime and reduced productivity.
Innovation Solution
An anomaly prediction module within a management system learns system profiles by gathering sensor data and building models to predict anomalies, allowing for proactive remediation plans to be implemented, using machine learning algorithms to analyze fan, power, thermal, and system load sensor data to identify potential issues before they occur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used requiring raw sensor data and additional sensors, then measurement precision may be improved, but device complexity and cost increase
Solution Approach 1:
The patent uses virtual copies of sensor data through machine learning models instead of physical additional sensors. The anomaly prediction module creates virtual sensor readings by learning from existing sensor data patterns, eliminating the need for additional physical sensors while maintaining detection capability
Solution Approach 2:
The patent replaces the mechanical approach of adding more physical sensors with an information-processing approach using machine learning algorithms. The system substitutes physical measurement infrastructure with computational models that analyze existing sensor data to predict anomalies
2Device complexity
If traditional anomaly detection waiting for actual failures is used, then device complexity remains low, but loss of time and productivity decrease
Solution Approach 1:
The patent implements preliminary action by predicting anomalies before they occur. The anomaly prediction module analyzes current sensor data patterns and forecasts potential failures in advance, allowing the system to take preventive measures before actual failures happen, thus reducing downtime without requiring complex additional hardware
3Reliability
If machine learning-based anomaly prediction is implemented, then reliability and productivity are improved, but use of energy and computational resources increase
Solution Approach 1:
The patent applies partial action by implementing anomaly prediction selectively rather than continuously analyzing all sensor data at full computational capacity. The system uses machine learning models to predict specific anomaly types based on relevant sensor data patterns, consuming computational energy only when prediction is needed rather than continuously processing all data streams
Data Source
AI summary
An information handling system includes a processing system including a first sensor, and a second sensor, and a management system including an anomaly table, a learned model table entry associated with the processing system and including a learned model and a first sensor data history, and a prediction module to implement a prediction algorithm. The management system is configured to: receive first sensor data and second sensor data, determine an estimate of a first value of the first sensor data using a second value of the second sensor data, determine a residual of the first value by a comparison of the estimate to the first value, determine a significance of the residual, where the significance having a significant value is associated with a predicted anomaly, determine that an anomaly table entry has a known anomaly class for the predicted anomaly, and perform a remediation plan to resolve the predicted anomaly.