Probabilistic Model Reconstructs High-Frequency Wear Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current industrial sensors capture data at low frequencies, which is insufficient for accurately estimating wear in mechanical components like turbine variable stator vanes, as high-frequency sampling over long periods is costly and often misrepresents events, especially in random or environmentally dependent signals.
Innovation Solution
Probabilistic models are used to reconstruct and augment high-frequency cumulative data from low-frequency data samples, combining system behavior knowledge to correct signal magnitudes and optimize information availability without requiring additional sensors or resources, allowing for accurate wear prediction in mechanical components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high frequency sampling is used to accurately capture wear events, then measurement precision is improved, but cost increases prohibitively
Solution Approach 1:
The patent creates a virtual copy of high-frequency data by synthesizing it from low-frequency samples using probabilistic models. Instead of physically capturing high-frequency data with expensive high-rate sensors, the system generates synthetic high-frequency representations that preserve wear event information, thereby achieving accurate wear measurement without the prohibitive cost of actual high-frequency sampling hardware
Solution Approach 2:
The patent replaces the physical mechanical sampling system (high-frequency sensors and data acquisition hardware) with a computational model. The probabilistic model substitutes the physical measurement process, using mathematical relationships and system behavior knowledge to reconstruct what high-frequency data would show, eliminating the need for expensive high-rate sampling infrastructure
2Measurement precision
If high frequency sampling is used to capture wear events, then measurement precision is improved, but loss of time increases due to data storage demands
Solution Approach 1:
The system creates a compressed virtual representation of high-frequency data through probabilistic modeling. By synthesizing high-frequency characteristics from low-frequency samples, the model avoids the need to store and process actual high-volume high-frequency data streams, significantly reducing data processing time and storage requirements while maintaining wear event detection accuracy
3Quantity of substance
If low frequency sampling is used to reduce costs, then loss of substance is reduced, but measurement precision deteriorates due to missed events
Solution Approach 1:
The patent transforms the sampling frequency parameter from a fixed physical constraint to a flexible computational variable. By changing how data is interpreted rather than how it's collected, the system applies probabilistic models that adjust the effective sampling rate through synthesis, allowing low-frequency physical samples to yield high-frequency analytical results, thereby maintaining measurement precision while working with reduced data volumes
4Device complexity
If conventional sensors are used to minimize device complexity, then device complexity is reduced, but measurement precision is insufficient for high frequency wear events
Solution Approach 1:
The patent introduces a computational intermediary (probabilistic model) between the simple conventional sensors and the wear analysis process. This intermediary layer processes low-frequency sensor data to reconstruct high-frequency wear event information, allowing conventional simple sensors to achieve the measurement precision of complex high-frequency systems through the mediating mathematical model
Data Source
AI summary
A method for augmenting low frequency rate data includes obtaining mechanical system operational data at a first and a second frequency rate, the first rate larger than the second rate, computing time intervals contained in the data, extracting respective samples of first and second frequency data, calculating individual cumulative values, resampling the first frequency rate data to simulate data at the second frequency rate, computing a multiplication factor, applying probabilistic model to the resampled first frequency data using the multiplication factor and rules, algorithms, and/or formulations, combining the resampled first frequency rate data with the second frequency rate data, analyzing the synthesized data set to determine if the mechanical system is viable to operate, and if the determination indicates an approaching end to the operational viability providing an indication to a control processor of the mechanical system. A system and non-transitory computer-readable medium are disclosed.