AI Sensor Prediction Model Retraining for Reference Sensor Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Managing artificial intelligence models trained to predict sensor measurements in industrial environments is challenging due to varying sensor data formats, deployment environments, and computing constraints, leading to difficulties in generalizing AI models across multiple IoT sensors.
Innovation Solution
A computer-implemented method for managing AI models involves obtaining measurements from a reference sensor, comparing predictions with actual measurements using a distance metric, selecting a new reference sensor if the distance exceeds a threshold, retraining the AI model with new data, and evaluating the retrained model before deployment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI models are trained for each individual sensor, then prediction accuracy is improved, but storage resources and computing power requirements increase significantly
Solution Approach 1:
The patent merges multiple individual sensor models into a single generalized AI model that can process data from multiple sensors of the same type. This consolidation reduces storage requirements while maintaining prediction accuracy through centralized model management and automated retraining mechanisms.
Solution Approach 2:
The patent creates a universal AI model that serves multiple sensors simultaneously. This generalized model is designed to handle data from any sensor within a group, making it multi-functional and eliminating the need for separate dedicated models for each sensor.
2Quantity of substance
If AI models are generalized to multiple sensors, then storage resources and computing power are saved, but model performance and adaptability deteriorate
Solution Approach 1:
The patent implements dynamic model management through automated monitoring of prediction accuracy and conditional retraining. When performance degradation is detected (when distance exceeds threshold), the system automatically retrains the generalized model with new data, ensuring continuous adaptation to changing sensor environments and maintaining reliable performance.
Solution Approach 2:
The patent incorporates feedback mechanisms where model predictions are continuously evaluated against actual sensor measurements. The distance metric provides feedback on model performance, triggering automated retraining when performance falls below acceptable thresholds, thus maintaining model reliability in generalized deployments.
3Adaptability or versatility
If manual management of AI models is used, then model customization is improved, but operational complexity and difficulty increase
Solution Approach 1:
The patent implements self-service automation where the system automatically monitors model performance, selects appropriate training data, retrains models when needed, and deploys updated versions without manual intervention. This eliminates the operational burden of manual model management while preserving customization capabilities through automated adaptation to specific sensor groups.
4Adaptability or versatility
If sensor data varies in format, resolution, and frequency, then sensor versatility is improved, but model generalization difficulty increases
Solution Approach 1:
The patent handles varying sensor parameters (format, resolution, frequency) by dynamically adjusting model training parameters and data preprocessing settings. The system adapts to different sensor characteristics through automated parameter optimization during the retraining process, maintaining generalization capability across diverse sensor types without increasing overall system complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The method comprises the following steps, implemented by a computer system, of: - comparing (E4) predictions of measurements of a reference sensor of a group of data sensors placed in a real environment for a given time period, produced by an artificial intelligence model associated with said group of sensors, with the measurements obtained for said reference sensor, using a first distance metric;- when (E5) the obtained distance is greater than a first given threshold, selecting (E6) a new reference sensor for said group from among other sensors of the group, the new selected reference sensor being associated with a smaller distance between the predictions of the artificial intelligence model and the measurements collected for the given time period, - retraining (E7) said artificial intelligence model, - evaluating (E8) said retrained artificial intelligence model using an evaluation set comprising at least a part of said obtained measurements, for which the obtained distance is greater than said first given threshold; and - in case of successful evaluation, making available (E9) said retrained artificial intelligence model for deployment in the environment.;