Machine Part Life Expectancy Prediction via Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting the life expectancy of machine parts are inaccurate in operating environments, as they rely on data from design-phase testing, which does not reflect actual operational conditions, and require significant service data that is often unavailable, leading to unexpected equipment stoppages and suboptimal part replacement timing.
Innovation Solution
A method using a neural network to create a life expectancy model based on service life data, including maintenance operation counts, to predict when parts should be replaced, incorporating pseudo-explanatory variables and interpolation techniques to enhance accuracy with limited data, and verifying the model against observed service life data to ensure reliable predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional reliability engineering methods using Weibull distribution are used to predict failure probability, then the prediction can be obtained based on design-phase testing data, but the prediction accuracy deteriorates because the testing data does not reflect actual operational conditions
Solution Approach 1:
The patent transforms the prediction approach by changing from using design-phase testing parameters to using operational service data parameters. The system collects actual operational data including maintenance operation counts and service life information from real-world usage, then uses neural networks to learn patterns from this operational data rather than relying on controlled testing conditions.
Solution Approach 2:
The patent replaces the traditional mechanical reliability engineering approach (Weibull distribution based on physical testing) with an information-processing approach using neural networks and machine learning. This substitution allows the system to process and learn from operational data patterns that cannot be captured through conventional testing methods.
2Adaptability or versatility
If parametric models are built to approximate probability density function based on failure data characteristics, then the model can be adapted to failure data, but the requirement for significant service data worsens the applicability because such data is often unavailable
Solution Approach 1:
The patent applies partial action by using only the essential features of service data (maintenance operation counts and service life) rather than requiring comprehensive failure data. The neural network is trained on partial operational information that is readily available, eliminating the need for extensive service data collection while still achieving accurate predictions.
Solution Approach 2:
The patent creates a virtual model of part degradation by copying and learning from patterns in operational data. The neural network builds a digital representation of how parts degrade based on maintenance operations, allowing predictions without requiring physical failure data or extensive service records.
3Ease of manufacture
If design-phase testing data is used for failure prediction, then the prediction method is simple to implement, but the prediction accuracy deteriorates due to deviation from actual operational conditions
Solution Approach 1:
The patent enables the system to self-improve by automatically collecting operational data and using it to train and refine the neural network model. The system serves itself by continuously learning from actual usage patterns, eliminating the need for manual model adjustments while improving accuracy over time.
Data Source
AI summary
Methods, computer program products, and systems are presented. The methods include, for instance: obtaining service life data of a part including a maintenance operation count of the part. The life expectancy of the part is formulated as a function of a life span of the part and the maintenance operation count at a point in time based on the obtained service life data and data interpolated therefrom. A life expectancy model is built based on the function and a plurality of life expectancies are predicted by applying simulated inputs to the life expectancy model. The life expectancies are produced after verification to indicate when to replace the part.