Component Malfunction Prediction Using Conditional Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting malfunctions in industrial assets are complex, computationally inefficient, and unable to predict malfunctions over an extended future time horizon, requiring separate models for each component type and lacking the ability to handle correlated random events.
Innovation Solution
A method using conditional probability distributions to predict malfunctions based on current parameter values, allowing for a general approach applicable to various component types, with a two-step process to determine future parameter distributions and malfunction probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate reliability models are created for each component type, then prediction accuracy for specific components is improved, but device complexity and model creation effort increase significantly
Solution Approach 1:
The patent applies universality by creating a single generic malfunction prediction model that can be applied to any component type. The model uses standardized inputs (parameter measurements, operational conditions) and outputs (malfunction probability, time to malfunction) that work across different component types, eliminating the need to create separate models for each component while maintaining prediction accuracy through the generic nature of the probability distribution functions.
Solution Approach 2:
The patent utilizes parameter changes by representing component states and malfunction probabilities through adjustable probability distribution parameters rather than fixed model structures. The generic model adapts to different component types by modifying these parameters (e.g., mean, variance of parameter distributions) rather than changing the fundamental model architecture, thus reducing complexity while maintaining accuracy.
2Duration of action of moving object
If traditional reliability models are used for extended future time horizon predictions, then prediction capability is improved, but computational efficiency deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the extended future time horizon into smaller discrete time intervals. The generic model computes malfunction probabilities incrementally for each interval rather than attempting to predict the entire extended horizon in a single computationally intensive calculation. This segmented approach reduces computational complexity while maintaining accuracy across extended time periods.
3Reliability
If component-specific reliability models are developed, then prediction accuracy is improved, but ease of manufacture and deployment worsens
Solution Approach 1:
The patent implements universality by designing a single generic model framework that can predict malfunctions for any component type without requiring separate model development. The model takes standardized inputs (parameter measurements, operational conditions) and produces consistent outputs (malfunction probability, time to malfunction) across different component types, eliminating the need for extensive model creation and validation efforts for each specific component.
Solution Approach 2:
The patent uses copying by applying the same generic model structure and methodology across different component types. Rather than creating unique models for each component, the approach copies the proven generic model framework and adapts it through parameter adjustment, significantly reducing model creation effort while maintaining prediction accuracy through the replicated successful approach.
4Ease of operation
If Markov models are used for prediction, then ease of operation is improved, but measurement precision and ability to predict random events deteriorates
Solution Approach 1:
The patent applies parameter changes by moving from fixed Markov model transition probabilities to flexible probability distribution functions with adjustable parameters. This allows the model to capture random events and unprecedented states by adapting the distribution parameters (mean, variance, shape) to match actual component behavior, thereby improving measurement precision while maintaining ease of operation through the generic framework.
Data Source
Figure 1
Figure 2
Figure 3~5
AI summary
A computer-implemented method for predicting a malfunction is provided. The method comprises receiving, by a server, a current value of a parameter of a component (30) measured at the component (30), determining a conditional probability distribution for the parameter of the component (30) for a future point in time given the current value of the parameter based on the current value of the parameter, and determining, at the server, a conditional probability for the malfunction at the future point in time given the current value of the parameter.