Predictive Anomaly Detection in Electronic Devices Using ML Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current process monitoring methods fail to predict anomalies in electronic devices, such as overheating or increased processing capacity, in a timely manner, leading to potential failures and unavailability, especially as the volume of metric data increases, and are often specific to the system or process being monitored.
Innovation Solution
A method that involves a learning step to analyze historical data, determine statistical indicators, and establish rules correlating these indicators with anomalies, followed by a monitoring step that periodically checks new data against these rules to predict future anomalies, using techniques like decision trees and neural networks to anticipate issues before they occur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If process monitoring methods analyze data representative of events occurring in the course of a process, then anomalies can be detected, but the alert is generated only after the anomaly has occurred and has been detected, making it too late to take effective measures
Solution Approach 1:
The patent applies preliminary action by training machine learning models on historical metric data and event logs to establish predictive patterns before anomalies occur. The system continuously learns from past data to predict future anomalies, enabling preventive measures to be taken before the actual anomaly happens, thus resolving the timing issue between detection and prevention.
Solution Approach 2:
The patent monitors multiple metric indicators simultaneously and uses machine learning to identify which combinations of metrics are most predictive of anomalies. By focusing on the most relevant metrics rather than all available data, the system achieves effective prediction while managing computational resources efficiently.
2Measurement precision
If the number of metric data generated by a process increases, then more comprehensive monitoring is achieved, but it becomes harder to predict the imminent occurrence of an anomaly, particularly for a person
Solution Approach 1:
The patent replaces manual analysis of metric data with machine learning algorithms. The system automatically processes large volumes of metric data and event logs, identifying complex patterns and correlations that would be impossible for human operators to detect manually. This substitution of mechanical/human analysis with automated ML systems resolves the contradiction between comprehensive monitoring and prediction difficulty.
Solution Approach 2:
The patent creates predictive models that replicate the relationship between metric patterns and anomalies based on historical data. These models serve as virtual copies of the system's behavior, allowing prediction of future anomalies without requiring direct human analysis of the actual metric data, thus managing complexity while maintaining comprehensive monitoring.
3Reliability
If known process monitoring methods are used, then anomaly detection is achieved, but the methods are almost always specific to the process studied and/or to the system used, and therefore cannot be generalized
Solution Approach 1:
The patent implements a universal machine learning framework that can be applied to different processes and systems by training on their specific historical data. The same core architecture and methodology work across diverse applications (IT systems, manufacturing, healthcare, etc.), achieving both specialized accuracy for each system and generalizability across domains through the flexible ML approach.
Data Source
AI summary
A device (DS) monitors a process using at least one electronic device (EE1-EE4) in operation and generating first data of a metric. This device (DS) comprises:learning means (MA) configured to analyse automatically second data which are representative of events that have occurred in the course of the process, in order to determine anomalies of a chosen type, and then automatically determine an indicator representative of this metric, then a correlation between these determined anomalies and this indicator, and then at least one rule defining this correlation, andmonitoring means (MS1) configured to analyse newly generated first data periodically, and group by group, by checking whether at least one value of the indicator determined on the basis of the aforesaid data satisfies this determined rule, in order to predict the occurrence of the anomaly in a future group of first data when this at least one value satisfies this rule.