Sensor Data Healing via Machine Learning Intermediary

The patent introduces a machine learning model as an intermediary component that processes raw sensor data to correct measurement errors and fill data gaps. The ML model acts as a mediator between the limited embedded sensors and the required high-quality measurements, using learned patterns from training data to compensate for sensor inaccuracies without requiring larger or more complex hardware

The patent creates a virtual copy of high-quality sensor data through machine learning predictions. Instead of relying on physical sensor replicas, the system generates synthetic sensor readings that mimic what ideal sensors would produce, based on patterns learned from comprehensive training datasets that include both embedded sensor data and reference measurements from laboratory-grade equipment

The patent extracts the essential quality characteristics of laboratory-grade sensor measurements and embeds them within portable devices through machine learning models. Rather than physically extracting and transporting laboratory equipment, the system captures the measurement quality patterns during training and reproduces them using compact embedded sensors and ML algorithms, enabling high-quality measurements in portable form factors

The patent transforms the measurement system from relying on physical hardware parameters (sensor size, processing power) to relying on data-driven parameters (training data quality, model architecture, inference algorithms). This parameter transformation enables portable devices to achieve laboratory-grade measurement quality by changing from hardware-centric to software-centric measurement approaches