Wearable Sensor Time-Series Normalization for Drift-Resistant Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for processing sensor data from wearable sensors lack efficient techniques to accurately predict and reconstruct temporal characteristics, especially when dealing with multivariate time series, leading to inconsistencies and sensor drift issues.

Innovation Solution

A computer-implemented method and device that processes sensor data by determining features and values from time series, using an artificial neural network to normalize and predict or reconstruct values, incorporating meta-information such as location and demographic data to enhance accuracy and reduce sensor drift.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional sensor data processing methods are used, then the processing is simpler, but the accuracy of temporal characteristics evaluation deteriorates and sensor drift issues occur

Engineering Contradiction:
Improveaccuracy of temporal characteristics evaluationVSAvoidcomplexity of processing method
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by transforming the sensor data through normalization techniques that adjust statistical parameters (mean and standard deviation) of the time series data. This allows the system to maintain high measurement precision in evaluating temporal characteristics while managing complexity through standardized transformation operations rather than complex architectural designs.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional mechanical or rule-based sensor drift compensation methods with a data-driven approach using artificial neural networks. The neural network learns temporal patterns and predicts future sensor values, substituting complex physical calibration mechanisms with intelligent algorithms that automatically adapt to sensor drift without requiring additional hardware complexity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If sensor drift compensation techniques are applied, then the reliability of sensor data improves, but the processing time and computational resources increase

Engineering Contradiction:
Improvereliability of sensor dataVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent implements preliminary action by pre-processing the sensor data through normalization and feature extraction before main analysis. The artificial neural network is trained in advance on historical sensor data to learn temporal patterns and drift characteristics. This preliminary preparation enables faster real-time processing while maintaining high reliability, as the heavy computational work is performed beforehand rather than during critical measurement periods.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system applies self-service by using the sensor data itself to compensate for its own drift through the neural network's predictive capabilities. The network learns from the sensor's historical behavior patterns and automatically corrects for drift without requiring external calibration references or additional measurement devices, thereby improving reliability without proportionally increasing processing time.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If multivariate time series are processed without normalization, then the processing is faster, but the accuracy deteriorates due to statistical differences and sensor drift

Engineering Contradiction:
Improveaccuracy of prediction and reconstructionVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies parameter changes through normalization transformations that standardize the statistical parameters (mean, standard deviation) of multivariate time series data. This enables accurate prediction and reconstruction across different sensors and conditions while maintaining processing efficiency through vectorized operations and optimized neural network computations that handle normalized data more effectively.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The normalization approach provides universality by creating a standardized processing framework that works across multiple sensor types and experimental conditions. The same normalized processing pipeline can be applied to different multivariate time series without requiring condition-specific adjustments, thereby maintaining both high accuracy and processing efficiency across diverse applications.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4696231A1Device and computer-implemented method for processing sensor data captured with sensor elements in particular worn on the body
Publication Date: 2026.02.18 ROBERT BOSCH GMBH
  • EP4696231A1 patent drawingFigure 1~2
  • EP4696231A1 patent drawingFigure 3
  • EP4696231A1 patent drawingFigure 4

AI summary

Device and computer-implemented method for processing sensor data acquired by sensor elements (102), in particular by an accelerometer, a gyroscope, a velocity sensor, a magnetometer, a pressure sensor, a blood pressure sensor, a VO2 sensor, or a pulse sensor, wherein a time series of the sensor data of the sensor element (102) is provided for each sensor element (102), wherein a feature associated with the time series is determined for each time series depending on the time series, wherein several features associated with the time series are determined for each time series depending on a part of the time series, wherein a value associated with the time series is determined for each time series depending on the features associated with the time series, and wherein a time series value associated with the time series is determined for each time series depending on the values ​​associated with the time series.wherein, for each time series, a time series value assigned to the time series is determined depending on a subset of the values ​​assigned to the time series, and wherein, for at least one time series, at least one predicted value assigned to the at least one time series is determined depending on the time series values ​​assigned to the at least one time series.