Local Sensor Analytics Feedback Tuning for Detection Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing analytics models embedded in IoT devices are slow to update and lack scalability, leading to imprecision and inefficiency in recognizing errors, particularly in critical applications like maintenance diagnostics and safety monitoring, due to their reliance on pre-computed models and aggregate data, which fail to adapt to changing conditions.

Innovation Solution

A method and system for continuously updating and improving local analytics by selecting models and parameters based on the likelihood of hits, misses, false alarms, and correct rejections, using a processor and database to evaluate sensor data and adjust criteria dynamically, incorporating feedback loops for self-correction and adaptation across devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If pre-computed models and aggregate data are used for local analytics, then device complexity is reduced and ease of operation is improved, but measurement precision and reliability deteriorate due to imprecision from assumptions and lack of adaptability

Engineering Contradiction:
Improveease of operationVSAvoidmeasurement precision
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system dynamically updates model parameters and selects models based on current operational conditions rather than using static pre-computed models. Local analytics devices continuously adapt their analytical models to changing conditions, transforming the system from static to dynamic to improve measurement precision while maintaining ease of operation through automated adaptation.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system implements feedback loops where outcomes from local analytics are continuously monitored and used to update model parameters and select improved models. This feedback mechanism allows the system to learn from actual performance data, correcting imprecision from assumptions by continuously refining models based on real-world outcomes.

Inventive Principle:
Principle #23Feedback

2Device complexity

If pre-computed models based on first principles or big data aggregation are used, then device complexity is reduced, but reliability deteriorates due to errors that cannot be recognized or self-corrected

Engineering Contradiction:
Improvedevice complexityVSAvoidreliability
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

Local analytics devices perform self-updating of model parameters and self-selection of models based on their own operational outcomes. The system is self-correcting, automatically identifying and fixing errors without external intervention, thereby improving reliability while maintaining low device complexity through autonomous operation.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system uses feedback from actual analytics outcomes to continuously validate and correct model performance. By monitoring hits, misses, false alarms, and correct rejections, the system automatically adjusts parameters and selects models that improve reliability, eliminating the need for complex external validation systems.

Inventive Principle:
Principle #23Feedback

3Manufacturing precision

If models are updated only in new product generations or tailored for large installations, then manufacturing precision is maintained through controlled updates, but productivity deteriorates due to insufficient scalability for updating and adapting analytical modules

Engineering Contradiction:
Improvemanufacturing precisionVSAvoidproductivity
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The system creates a universal framework for model updating that can be applied across all local analytics devices regardless of scale or application. The same automated model selection and parameter update mechanisms work for both individual devices and large installations, eliminating the need for separate update processes and improving productivity while maintaining manufacturing precision through consistent methodologies.

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

Solution Approach 2:

The system pre-establishes a database of models and outcomes and implements automated selection algorithms that can immediately begin updating analytical modules upon deployment. This preliminary preparation eliminates the need for time-consuming manual tailoring for each installation, dramatically improving productivity while maintaining precision through the pre-vetted model database.

Inventive Principle:
Principle #10Preliminary action

4Measurement precision

If signal detection theory optimization is applied without experimental control, then measurement precision of sensor sensitivity can be improved, but reliability of response criteria optimization deteriorates due to inability to isolate and optimize response criteria independently

Engineering Contradiction:
Improvemeasurement precisionVSAvoidreliability
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The system segments the optimization process into two independent components: sensor sensitivity optimization and response criteria optimization. By separating these previously confounded elements, the system can independently optimize each aspect using outcome data, improving the reliability of response criteria while maintaining the measurement precision of sensor sensitivity through separate adjustment mechanisms.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3539028B1Systems and methods for supervision of local analytics
Publication Date: 2022.12.28 3M INNOVATIVE PROPERTIES CO
  • EP3539028B1 patent drawingFigure 1
  • EP3539028B1 patent drawingFigure 2
  • EP3539028B1 patent drawingFigure 3

AI summary

Systems and methods for dynamically optimizing models used for sensor data analytics. An action is taken based on an analytics determination by systematically varying parameters of the analytical model using actions taken based on the analytics to determine the relative frequencies of hits, misses, false alarms, and correct rejections for particular model parameters. The model parameters for local analytics are selected based upon on signal detection theory analysis and the value or cost of each hit, miss, false alarm, or correct rejection.