Signal Synthesizer Data Pump for ML Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing synthetic time-series signals lack the complex stochastic structure of real signals, making them ineffective for training machine-learning models to detect anomalies in prognostic-surveillance applications, due to copyright and data access issues with real data.
Innovation Solution
A system generates synthetic time-series signals by combining multiple sinusoidal signals with varying periodicities, adding time-varying random noise and amplitude values, and incorporating industry-specific parameters through a roll-of-the-die process to mimic real signal characteristics, allowing for realistic stochastic structure simulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real time-series sensor data is used for training ML models, then the training effectiveness and anomaly detection accuracy are improved, but copyright ownership issues, exorbitant fees, and data access restrictions prevent commercial use
Solution Approach 1:
The patent creates synthetic copies of real sensor data that replicate the stochastic structure and statistical properties of actual time-series signals. These synthetic data copies serve as substitutes for real data, enabling commercial ML training without copyright issues or data access restrictions while maintaining training effectiveness
Solution Approach 2:
The patent transforms real data characteristics into configurable parameters that define synthetic signal generation. By changing parameters such as mean, standard deviation, and autocorrelation coefficients, the system generates synthetic data with controlled statistical properties that match real sensor behavior without using actual proprietary data
2Ease of manufacture
If commonly used synthetic signals are generated through rudimentary techniques, then data accessibility and ease of generation are improved, but the complex stochastic structure of real time-series signals is lost, making them ineffective for training ML models
Solution Approach 1:
The patent implements dynamic signal generation where statistical parameters such as mean and standard deviation vary over time according to autocorrelation models. This creates synthetic signals with realistic temporal dependencies and stochastic behavior that mimic real sensor data dynamics, rather than using static or purely random values
Solution Approach 2:
The patent incorporates periodic components into synthetic signal generation through autocorrelation functions that create repeating statistical patterns. This produces signals with realistic cyclical behavior and temporal structure that resemble actual sensor measurements while maintaining synthetic nature
3Measurement precision
If synthetic signals with realistic stochastic structure are generated, then ML model training effectiveness is improved, but the complexity of signal generation processes and parameter modeling increases
Solution Approach 1:
The patent breaks down the complex task of generating realistic synthetic signals into separate modular components: autocorrelation modeling, parameter estimation, time-varying mean generation, and time-varying standard deviation generation. Each module handles a specific aspect of signal construction, making the overall complex process more manageable and implementable
Data Source
AI summary
The disclosed system produces synthetic signals for testing machine-learning systems. During operation, the system generates a set of N composite sinusoidal signals, wherein each of the N composite sinusoidal signals is a combination of multiple constituent sinusoidal signals with different periodicities. Next, the system adds time-varying random noise values to each of the N composite sinusoidal signals, wherein a standard deviation of the time-varying random noise values varies over successive time periods. The system also multiplies each of the N composite sinusoidal signals by time-varying amplitude values, wherein the time-varying amplitude values vary over successive time periods. Finally, the system adds time-varying mean values to each of the N composite sinusoidal signals, wherein the time-varying mean values vary over successive time periods. The time-varying random noise values, amplitude values and mean values can be selected through a roll-of-the-die process from a library of values, which are learned from industry-specific signals.