Bayesian Risk Inference for Dynamic Asset Survival Curves
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional asset survival rate prediction models assume constant wear rates, leading to inefficiencies and inaccuracies due to their inability to handle dynamic wear rates and real-time data, resulting in unreliable maintenance and replacement schedules.
Innovation Solution
A Bayesian risk inference engine (BRIE) that combines expert-driven models with condition-based alarms to dynamically filter wear rate possibilities, using a random-walk model to accommodate temporally dependent wear rates and incorporate real-time data for accurate survival curve predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional RUL models assume constant wear rates, then the models are simpler to implement, but the accuracy of survival curve predictions deteriorates due to inability to handle dynamic wear rates
Solution Approach 1:
The patent transforms the static constant wear rate assumption into a dynamic wear rate model that evolves over time. The wear rate is modeled as a time-dependent parameter that can change based on operating conditions, allowing the model to adapt to real-world variability while maintaining computational tractability through structured probabilistic formulations.
Solution Approach 2:
The patent changes the wear rate parameter from a constant value to a dynamic variable that can take different values over time. This is achieved by introducing time-varying wear rate parameters that capture the evolving degradation process, enabling more accurate prediction of asset survival while managing complexity through parametric modeling approaches.
2Measurement precision
If conventional RUL models handle real-time data with dynamic wear rates, then prediction accuracy improves, but processing and memory resources are overwhelmed
Solution Approach 1:
The patent segments the continuous real-time data processing into discrete time intervals or batches. By dividing the degradation process into manageable segments, the model can process real-time data incrementally without overwhelming computational resources, while still capturing dynamic wear rate variations for accurate survival predictions.
Solution Approach 2:
The patent applies partial action by selectively processing only the most relevant real-time data points that significantly impact wear rate estimation. Rather than processing all available data equally, the model focuses on critical observations that drive degradation, reducing computational burden while maintaining prediction accuracy.
3Productivity
If conventional models use constant average wear rates, then computational efficiency is maintained, but reliability of maintenance scheduling deteriorates due to overconfidence and brittleness
Solution Approach 1:
The patent introduces feedback mechanisms where real-time sensor data continuously informs and updates the wear rate estimates. This closed-loop approach allows the model to adjust predictions based on actual asset performance, improving maintenance schedule reliability while maintaining computational efficiency through iterative updating rather than complete recalculation.
Solution Approach 2:
The model performs self-updating by automatically adjusting wear rate parameters based on incoming real-time data without requiring external intervention. This self-service capability enables the system to adapt to changing conditions autonomously, improving reliability while avoiding the computational overhead of manual model updates.
Data Source
AI summary
A computing platform is configured to: (a) generate a predicted health distribution of an asset for a failure mode based on a prior health distribution and a wear rate distribution. The computing platform is further configured to (b) update, the predicted health distribution based on (i) an observed state distribution corresponding to an observed state associated with the asset and (ii) a normalized value representative of a probability of the observed state over one or more health values of the asset. The computing platform is further configured to (c) generate a survival curve of the asset based on the predicted health distribution and a set of wear rates; iteratively perform (a)-(c) for each failure mode of the set of failure modes; aggregate the survival curve for each failure mode into an aggregate survival curve; and cause a client device to display a visual representation of the aggregate survival curve.