Functional Neural Network for Vibration Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vibration data analysis methods, particularly in predictive maintenance, face challenges in effectively capturing non-stationary patterns due to fixed temporal-spectral resolution, leading to information loss in high-frequency components or transient events, which is inadequate for real-world applications.
Innovation Solution
The application of a functional neural network (FNN) that processes 2D time-frequency representations of vibration data, treating each frequency component as a functional covariate, allowing for multiresolution analysis and automatic feature extraction, thereby improving predictive maintenance by capturing dynamic changes in vibration patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If fixed temporal-spectral resolution is used in vibration data analysis, then processing simplicity is maintained, but information loss occurs in high-frequency components or transient events
Solution Approach 1:
The patent applies dynamic resolution by making the temporal window size adaptive rather than fixed. The window size automatically adjusts based on the frequency content being analyzed, with smaller windows for high-frequency components to capture transient events and larger windows for low-frequency components to maintain spectral resolution. This dynamic adaptation resolves the contradiction by allowing the system to optimize resolution locally for different frequency bands rather than using a single fixed resolution globally.
Solution Approach 2:
The patent implements local quality by applying different temporal resolutions to different frequency components. Instead of uniform processing, the system uses variable window sizes tailored to specific frequency ranges, providing high temporal resolution where needed (high-frequency transients) and high spectral resolution where needed (low-frequency components). This localized optimization eliminates information loss in critical frequency regions while maintaining overall processing effectiveness.
2Measurement precision
If feature engineering is performed manually to identify important features, then domain knowledge can be applied, but time and labor are consumed
Solution Approach 1:
The patent replaces manual feature engineering with automated machine learning algorithms. Instead of relying on domain experts to manually select and engineer features, the system uses algorithms to automatically identify, extract, and prioritize relevant features from vibration data. This substitution maintains feature selection accuracy through data-driven insights while eliminating the significant time and labor investment required for manual feature engineering processes.
Solution Approach 2:
The patent enables the system to perform feature engineering autonomously without human intervention. The machine learning models automatically learn which features are most predictive of equipment conditions and failures, performing the feature selection task that would otherwise require expert domain knowledge and manual effort. This self-service capability resolves the contradiction by making the system independent of human time investment while maintaining or improving feature selection quality.
3Measurement precision
If high sampling rate is used to capture transient events, then measurement precision is improved, but data processing complexity increases
Solution Approach 1:
The patent extracts only the most relevant information from high-rate sampling data using automated feature extraction algorithms. Instead of processing entire high-volume datasets, the system identifies and extracts key transient events and critical features, discarding redundant information. This extraction approach maintains high measurement precision for transient detection while significantly reducing the computational burden of processing complete high sampling rate datasets.
Solution Approach 2:
The patent applies partial processing by focusing computational resources on detecting and analyzing only the most critical transient events rather than processing all data points uniformly. The system uses adaptive thresholding and event detection to identify only those portions of high-rate data that contain meaningful transient information, performing detailed analysis only where needed. This partial action approach maintains detection accuracy for critical events while reducing overall processing complexity and computational load.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The FNN effectively extracts relevant features from vibration data, enhancing predictive maintenance by providing accurate predictions of system health and failure probabilities, even in complex, non-stationary environments, thus improving machine performance and reducing downtime.
Implementation Method 1
producing a two-dimensional 2D time-and-frequency input data set by applying a wavelet transform to the measured data
Data Source
AI summary
An apparatus for predicting a characteristic of a system is provided. The apparatus may include a memory and at least one processor coupled to the memory. The at least one processor may be configured to perform a method including measuring, at a high sample rate, data relating to an operation of the system over a first time period. The method may further include producing a two-dimensional (2D) time-and-frequency input data set by applying a wavelet transform to the measured data. The method may additionally include generating a set of one or more values associated with one or more system characteristics by processing the 2D time-and-frequency input data set using a functional neural network (FNN).