Failure Prediction Model Reactivation via Duration Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting equipment and process failures in factory operations are inadequate, as they only predict failure occurrence and not duration, leading to increased false positives and operational challenges, especially in process failures where adjustments are needed during ongoing failures.
Innovation Solution
Implementing two prediction models: one for predicting the occurrence of process failures and another for estimating the duration of such failures, allowing for real-time adjustments and reactivation of the prediction model based on the expected end time of the failure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the failure prediction model restarts in the early stage of a failure, then the prediction can detect successive failures, but the false positive rate increases due to similar sensor behavior
Solution Approach 1:
The system performs preliminary action by predicting failure duration before the actual failure occurs. The duration prediction model estimates how long the failure will last, allowing the system to wait until the failure actually occurs before restarting the failure prediction model. This prevents false positives while still enabling detection of successive failures.
Solution Approach 2:
The system uses feedback from the duration prediction model to control when to restart the failure prediction model. The feedback mechanism ensures that the failure prediction model only restarts after the current failure has been confirmed and its duration predicted, using this information to timing the restart optimally.
2Loss of information
If the failure prediction model waits until the failure is confirmed before restarting, then false positives are reduced, but the system cannot predict successive failures that occur during the failure period
Solution Approach 1:
The system performs preliminary action by predicting failure duration before the actual failure occurs. This allows the system to plan when to restart the failure prediction model optimally - after the current failure is confirmed but before successive failures can occur, enabling both accurate failure detection and successive failure prediction.
Solution Approach 2:
The system dynamically adjusts its behavior based on real-time conditions. The duration prediction model continuously monitors the situation and provides updated duration estimates, allowing the system to dynamically determine the optimal restart time for the failure prediction model, balancing false positive reduction with successive failure detection capability.
3Device complexity
If only failure occurrence is predicted without duration estimation, then the prediction model is simpler, but factory operators cannot make timely operational adjustments
Solution Approach 1:
The system segments the prediction functionality into two separate models: a failure prediction model that detects when a failure will occur, and a duration prediction model that estimates how long the failure will last. This segmentation allows each model to focus on a specific aspect, improving overall operational usefulness without creating an overly complex unified model.
Solution Approach 2:
The combined system provides multi-functionality by simultaneously delivering failure occurrence predictions and duration estimates. This universal approach gives factory operators both the timing information needed to prepare for failures and the duration information needed to plan operational adjustments, making the system highly useful for various operational scenarios.
Data Source
AI summary
Aspects of the present disclosure are directed to systems and methods for determining execution of failure prediction models and duration prediction models for a sensor system. Systems and methods can involve receiving streaming data from one or more sensors and for a failure prediction model processing the streaming data indicating a predicted failure with a probability higher than a threshold, obtaining a duration of the predicted failure from a duration prediction model configured to predict durations of detected failures based on the streaming data; deactivating the failure prediction model when the predicted failure occurs; and determining a time to reactivate the failure prediction model based on the obtained duration of the predicted failure.