Modified GRU for Predicting Information Handling Resource Failures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information handling systems face challenges in predicting component failures, particularly due to irregular telemetry data collection frequencies and missing data fields, which existing methods like LSTM struggle to handle effectively, especially when scaling to large datasets.
Innovation Solution
Implementing a deep recurrent neural network with a modified gated recurrent unit (GRU) that imputes missing values and trains a pattern recognition engine to predict failure status using merged telemetry and failure statistics, enabling efficient prediction of information handling resource failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If discrete cosine transformation is used for data imputation, then missing values can be filled, but the method is not scalable to large datasets and requires many elements within telemetry data
Solution Approach 1:
The patent replaces the discrete cosine transformation (mathematical signal processing method) with a deep recurrent neural network with modified gated recurrent unit (machine learning model). This substitution enables the system to handle large datasets more efficiently while maintaining imputation accuracy, as the neural network can learn patterns from data without requiring many elements like cosine transform does.
Solution Approach 2:
The patent modifies the gated recurrent unit by adding a decay rate parameter that exponentially decays based on time interval. This parameter change allows the model to adapt to irregular sampling frequencies and improve scalability by efficiently handling varying data densities without requiring uniform data distribution.
2Reliability
If LSTM approach is used for failure prediction, then pattern recognition can be performed, but it requires separate imputation and training steps increasing complexity
Solution Approach 1:
The patent merges the data imputation function and the training function into a single integrated deep recurrent neural network model. The modified gated recurrent unit simultaneously performs missing value imputation and learns failure patterns from the same computational pass, eliminating the need for separate imputation and training steps required by traditional LSTM approaches.
Solution Approach 2:
The modified gated recurrent unit is designed to perform multiple functions: it imputes missing values, handles irregular sampling frequencies, and performs failure pattern recognition all within a single model architecture. This multi-functional design simplifies the overall system compared to separate specialized components.
3Adaptability or versatility
If telemetry data is collected at irregular frequencies, then real-world conditions are captured, but existing methods struggle to handle the inconsistent time intervals
Solution Approach 1:
The patent introduces a dynamic decay rate mechanism in the modified gated recurrent unit that adapts to varying time intervals between data collections. The decay rate exponentially decays based on the time interval, allowing the model to dynamically adjust its behavior for irregular sampling frequencies while maintaining prediction accuracy, unlike static approaches that assume uniform sampling.
Data Source
AI summary
A method may include receiving telemetry data associated with one or more information handling resources, receiving failure statistics associated with the one or more information handling resources, merging the telemetry data and the failure statistics to create training data, and implementing a gated recurrent unit to: (i) impute missing values from the training data and (ii) train a pattern recognition engine configured to predict a failure status of an information handling resource from operational data associated with the information handling resource.