Remaining Useful Life Estimation Using Neural RUL Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the remaining useful life (RUL) of worn parts, such as regression-based and survival analysis-based approaches, are either data-inefficient or rely on strong assumptions that fail to accurately model the complex degradation dynamics of worn parts, leading to inaccurate RUL predictions.
Innovation Solution
A neural network-based approach that utilizes a historical dataset including condition monitoring data and censor type to train a neural network, allowing it to learn the relationship between RUL distribution and condition monitoring data, thereby providing accurate RUL estimation by considering both censored and uncensored data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If regression-based methods are used for RUL estimation, then the estimation process is simple, but the prediction accuracy is low due to data inefficiency
Solution Approach 1:
The patent transforms the RUL estimation problem from direct regression to distribution parameter estimation. Instead of predicting RUL directly, the neural network estimates parameters (shape parameter and scale parameter) of the RUL distribution, which captures the underlying degradation dynamics more effectively and improves prediction accuracy while maintaining model simplicity
Solution Approach 2:
The patent replaces traditional statistical survival analysis methods with a neural network-based approach. The neural network automatically learns the relationship between condition monitoring data and RUL distribution parameters, substituting manual feature engineering and complex statistical modeling with an adaptive learning system that achieves higher accuracy
2Reliability
If traditional survival analysis-based approaches are used, then the method can handle censored data, but the strong assumptions fail to accurately model complex degradation dynamics
Solution Approach 1:
The patent introduces dynamic adaptability by using a neural network that learns the RUL distribution parameters directly from data without relying on fixed parametric assumptions. The model adapts to complex degradation patterns by automatically adjusting the shape and scale parameters based on observed condition monitoring data, capturing non-linear degradation dynamics that traditional static models miss
Solution Approach 2:
The patent changes from estimating a single RUL value to estimating the full distribution parameters (shape and scale) of RUL. This parameter transformation allows the model to represent uncertainty and variability in degradation processes, providing a more comprehensive and accurate characterization of component reliability while naturally handling censored data
3Ease of operation
If existing RUL estimation methods are used, then the computational process is straightforward, but the features learned are not sufficient for accurate prediction
Solution Approach 1:
The patent replaces manual feature engineering and traditional statistical feature extraction with a neural network that automatically learns relevant features from raw condition monitoring data. The network autonomously identifies degradation patterns and transforms them into meaningful RUL distribution parameters, eliminating the need for domain expertise in feature selection while improving prediction accuracy
Data Source
AI summary
Various embodiments of the teachings herein include a method for estimating a remaining useful life of a worn part. An example includes: collecting a historical dataset of the worn part, wherein the dataset comprises a plurality of tuples, each tuple including: condition monitoring data and remaining useful life at a time the condition monitoring data is observed; and training a neural network with the historical dataset by using the condition monitoring data as input and generating a parameter of distribution of the remaining useful life as output of the neural network.