Physics-AI Data Synthesis and Validation for Machinery Fault Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Physics-based models generate high-quality sensor data but are time-consuming, while generative learning is fast but prone to errors, necessitating a hybrid approach for efficient and accurate data generation and validation.
Innovation Solution
A hybrid system combining Finite Element Analysis (FEA) with generative models like Variational Autoencoders (VAEs) for data synthesis, followed by rule engine and AI-based validation to ensure data quality, using statistical features and classifiers for robust fault diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based models are used to generate sensor data, then data quality is improved, but processing time increases
Solution Approach 1:
The system segments the data generation process into two distinct stages: (1) physics-based simulation to generate a limited set of high-quality training data, and (2) generative adversarial network to rapidly produce large volumes of synthetic data. This segmentation allows each method to operate in its optimal performance zone, resolving the contradiction between quality and speed.
Solution Approach 2:
The generative adversarial network acts as an intermediary between the physics-based model and the final data generation task. The physics-based model generates a small dataset that trains the GAN, which then serves as a fast generator for large-scale data production, mediating between the high-quality but slow physics simulations and the need for rapid data generation.
2Productivity
If generative learning is used to generate data, then processing speed is improved, but data accuracy deteriorates
Solution Approach 1:
The system performs preliminary action by using physics-based simulations to generate a high-quality training dataset before deploying the generative adversarial network. This pre-trained foundation ensures that the subsequent fast data generation process is guided by accurate physical principles, preventing accuracy deterioration.
Solution Approach 2:
The generative adversarial network incorporates feedback mechanisms where the discriminator component continuously evaluates generated data against the physics-based training set, providing corrective feedback to the generator. This feedback loop ensures that rapidly generated data maintains fidelity to physical reality, resolving the accuracy-speed contradiction.
3Reliability
If physics-based simulations are used, then data reliability is improved, but computational resources are consumed
Solution Approach 1:
The system applies partial action by using physics-based simulations only for the essential training phase rather than for all data generation. A limited but sufficient set of physics-based simulations (partial action) creates the foundation for the GAN, which then handles the bulk of data generation with minimal computational resources, resolving the reliability-resource contradiction.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
In sensor data analytics, physics-based models generate high quality data. However, these models consume lot of time as they rely on physical simulations. On the other hand, generative learning takes much less time to generate data, and may be prone to error. Present disclosure provides system and method for generation of synthetic machine data for healthy and abnormal condition using hybrid of physics based and generative model-based approach. Finite Element Analysis (FEA) is used for simulating healthy and faulty parts in machinery with set of parameters and precondition(s). Small output data from FEA is fed into a generative model for generating synthesized data by learning data distribution knowledge and representing into latent space. Rule engine is built using statistical features wherein realistic bounds serve as faulty data indicators. Synthesized data which does not satisfies features bounds are discarded. Further, AI-based validation framework is used to analyze quality of synthesized data.