AI Model Transfer Function for Cross-Environment Device Operation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models trained for specific industrial devices are difficult to transfer between different environments due to variations in installation, operating conditions, and adjacent installations, leading to different data distributions and reduced accuracy when applied to similar devices.
Innovation Solution
A method involving detecting operating parameters using sensors in different environments, performing Fourier transforms, and generating a transfer function to enable the reuse of AI models across identically constructed devices without retraining, allowing for accurate model transfer and scaling in industrial applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an AI model is trained for a specific device in a specific environment, then the model achieves high accuracy for that device, but the model cannot be transferred to other devices in different environments without retraining
Solution Approach 1:
The patent applies parameter changes by transforming the sensor signals from different operating environments into the frequency domain using Fourier transforms. This transformation changes the representation parameters of the data, allowing the model to learn environment-invariant features. The transfer function is derived by comparing frequency-domain representations across environments, enabling the model to adapt to different operating conditions without full retraining.
Solution Approach 2:
The patent introduces a transfer function as an intermediary element that bridges different operating environments. This transfer function, derived from the ratio of Fourier transforms of sensor signals from different environments, acts as a mediator that allows the AI model to translate predictions from one environment to another. The intermediary transfer function captures the environmental differences and enables model transferability across diverse operating conditions.
2Measurement precision
If an AI model is retrained for each new device installation, then the model maintains high accuracy, but the complexity and time required for deployment increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-training the AI model on sensor data from a reference device in a controlled environment. The model learns general fault detection patterns that are transferable to other identically constructed devices. The transfer function is pre-computed from Fourier transforms of reference data, enabling rapid adaptation to new installations without requiring time-consuming retraining processes.
Solution Approach 2:
The patent uses copying by creating a transfer function that replicates the environmental characteristics from the reference device to the target device. Instead of retraining the model from scratch for each new installation, the system copies the pre-trained model and applies the transfer function to adapt it to the new environment, significantly reducing deployment time while maintaining accuracy.
3Measurement precision
If an AI model is trained on data from one operating environment, then the model performs well in that environment, but the model accuracy decreases when applied to devices in different operating environments
Solution Approach 1:
The patent transforms sensor signals from the time domain to the frequency domain using Fourier transforms. This parameter change in signal representation reveals underlying patterns that are invariant across different operating environments. By training the model on frequency-domain features rather than raw time-domain signals, the model learns to recognize faults based on spectral characteristics that remain consistent across different installations and operating conditions.
Solution Approach 2:
The transfer function serves as an intermediary that compensates for environmental differences between training and deployment scenarios. By multiplying the model's output by the transfer function (which captures the ratio of environmental characteristics), the system adjusts predictions to account for differences in operating environments, thereby maintaining reliability across diverse conditions.
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
Enables the accurate transfer and reuse of AI models across different environments, reducing the need for retraining and facilitating the scaling of AI models in industrial settings, improving prediction accuracy and operational efficiency.
Implementation Method 1
performing a first Fourier transform for the first sensor signal and a second Fourier transform for the second sensor signal via a computing apparatus
Data Source
AI summary
A computer-implemented method for operating a technical device with a model based on artificial intelligence including detecting a first operating parameter of a reference apparatus in a first operating environment and a first permissible operating mode, detecting a second operating parameter of the reference apparatus in a second operating environment and a second permissible operating mode, performing a first Fourier transform for the first sensor signal and a second Fourier transform for the second sensor signal, determining a quotient from the first and second Fourier transforms as a transfer function, generating and training a first model based on artificial intelligence for the reference apparatus in the first operating environment, generating a second model based on artificial intelligence for the technical device in the second operating environment using the first model and applying the transfer function, and operating the technical device with the second model.