Fault Diagnosis Model Partial Relearning for Rare Error Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fault diagnosis systems face challenges in achieving high accuracy when receiving rare combinations of signals from devices, as relying solely on past results may not effectively identify solutions in such cases.
Innovation Solution
A system that includes a storage unit for a machine learning-based model to estimate error handling for network devices, an obtainment unit for diagnosis results, a comparison unit to assess multiple candidates for handling, and an execution unit for relearning based on partial operation information to adjust the diagnostic model, especially when hardware and machine learning results differ.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fault diagnosis is performed by referring to past combination cases, then common error patterns can be identified efficiently, but rare error combinations cannot be diagnosed with high accuracy
Solution Approach 1:
The system performs preliminary machine learning to construct a diagnostic model in advance, enabling it to handle rare error combinations that have not yet appeared in past cases. The model is built by learning from operation information and handling histories, preparing the system to diagnose errors before they actually occur in practice.
Solution Approach 2:
The system implements a feedback mechanism where diagnosis results are compared with model estimates, and when discrepancies are found, the model undergoes relearning. This closed-loop feedback allows the system to continuously improve its diagnostic accuracy by incorporating new information from actual device diagnoses, particularly for rare error patterns.
2Reliability
If a machine learning model is constructed using all operation information, then comprehensive error patterns can be learned, but the model becomes complex and difficult to update
Solution Approach 1:
The diagnostic model is divided into multiple layers, each performing different estimation functions. This segmentation allows the system to update only specific layers when new error patterns are encountered, rather than retraining the entire model, thereby reducing complexity while maintaining comprehensive diagnostic capability.
Solution Approach 2:
Instead of performing complete relearning of the entire model when new error patterns are detected, the system performs partial relearning only on the necessary layers or components. This partial action reduces the computational burden and model update complexity while still improving diagnostic accuracy for the specific error patterns encountered.
Data Source
AI summary
In connection with an error event detected in an image forming apparatus, a fault diagnosis service server diagnoses the error event and obtains a diagnosis result for handling to be performed on the error event detected in the image forming apparatus. In a case where the diagnosis result indicates a plurality of candidates for handling to be performed on the error event, the fault diagnosis service server compares the plurality of candidates for handling to a result estimated by a diagnosis model with respect to the error event. In a case where the plurality of candidates for handling is different from the result estimated by the diagnosis model, the fault diagnosis service obtains partial operation information from a history of the image forming apparatus and executes machine relearning based on the obtained partial operation information with respect to a layer in a latter stage of a plurality of layers included in the diagnosis model, each layer performing a different estimation.