ML-Driven Instrumentation with Human-in-the-Loop Measurement Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current instrument systems for imaging and spectroscopy are inefficient due to manual selection of measurement locations, leading to time-consuming data collection and potential sample damage, especially when dealing with biological materials, and existing analysis methods do not adequately address the disparities in acquisition times and spatial density of information.

Innovation Solution

A method using machine learning to iteratively select measurement locations based on predicted uncertainties and acquisition functions, training models to determine relationships between image patches and physical characteristics, and retraining with new data to optimize measurement acquisition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual selection of measurement locations is used, then operator intuition can identify regions of interest, but data collection becomes time-consuming and requires significant human involvement

Engineering Contradiction:

Improveidentification of regions of interestVSAvoiddata collection time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system enables automated selection of measurement locations through machine learning models that autonomously identify regions of interest and determine optimal measurement points, eliminating the need for continuous human operator intervention while maintaining accurate identification of scientifically significant areas

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical human operator decision-making process with computational algorithms and machine learning models that analyze structural images and automatically determine measurement locations, substituting human intuition with automated intelligent systems

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If uniform sampling grid mode is used for spectroscopic measurements, then systematic data collection is achieved, but acquisition time increases and spatial density of information is limited

Engineering Contradiction:

Improvesystematic data collectionVSAvoiddata acquisition efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system applies different measurement densities to different regions of the sample based on their scientific importance, concentrating measurements in regions of interest identified by the machine learning model while reducing or eliminating measurements in less significant areas, thereby achieving efficient resource allocation

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The machine learning model performs preliminary analysis of structural images before spectroscopic measurements to pre-identify regions of interest and determine optimal measurement locations, allowing the system to avoid unnecessary measurements and focus resources on scientifically valuable areas

Inventive Principle:

Principle #10Preliminary action

3Loss of information

If repeated measurements are performed to collect large volumes of data, then comprehensive parameter space coverage is achieved, but sample damage occurs especially with biological materials

Engineering Contradiction:

Improvedata completenessVSAvoidsample damage

Core Design Contradiction:

Loss of informationVSObject-affected harmful factors

Solution Approach 1:

The system performs measurements only at the necessary minimum number of locations to achieve scientific objectives, using machine learning to identify and focus on critical regions of interest rather than uniformly sampling the entire parameter space, thereby obtaining sufficient information with reduced total measurements

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system uses iterative machine learning models that learn from acquired data and update their predictions of regions of interest, allowing adaptive refinement of measurement strategies based on feedback from previous measurements to maximize information gain while minimizing total measurements and sample exposure

Inventive Principle:

Principle #23Feedback

4Measurement precision

If spectroscopic measurements are performed at multiple locations, then spatial distribution of physical properties is mapped, but acquisition time disparities between structural and spectroscopic measurements limit spatial density

Engineering Contradiction:

Improvespatial distribution mappingVSAvoidacquisition time disparity

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system applies high spatial density of spectroscopic measurements only in identified regions of interest where scientific insights are most valuable, rather than uniformly distributing measurements across the entire sample, thereby achieving effective spatial mapping with reduced total acquisition time

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS12253540B2Machine learning-driven operation of instrumentation with human in the loop

Publication Date: 2025.03.18 UT BATTELLE LLC

AI summary

Systems are provided for machine learning-driven operation of instrumentation with human in the loop. The systems use a model with learnt model parameters to define points for physical-characteristic measurements once the model is trained. The systems use active learning, which considers selection, reinforcement and/or adjustment inputs from the instrumentation's user, to enable describing a relationship between local features of sample-surface structure shown in image patches and determined representations of physical-characteristic measurements.