Machine Learning Data Generation via Physics Simulation

Solution Overview

Problem

Acquiring sufficient machine learning data for physical operations requires actual operation of machines under various conditions, which is labor-intensive and time-consuming.

Innovation Solution

A machine learning data generation device that uses physical simulation to generate virtual time series information based on parameter values, identifying and associating parameter values with labels to create new machine learning data without actual machine operation, employing a generative adversarial network (GAN) to generate indistinguishable virtual time series information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If actual machine operation is used to acquire machine learning data under various conditions, then data quality and authenticity are improved, but labor cost and time consumption increase tremendously

Engineering Contradiction:

Improvedata qualityVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates virtual copies of machine operation data through physics simulation. Instead of collecting actual operation data from real machines under various conditions, the system generates synthetic time series data that replicates the physical behavior and characteristics of actual machine operations, thereby obtaining sufficient training data without time-consuming field collection

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary physics simulation to generate labeled training data before actual machine learning model training. By pre-simulating various operation conditions and fault states with known ground truth labels, the system prepares high-quality training datasets in advance, eliminating the need for time-consuming data collection during the learning process

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If actual machine operation is used to acquire machine learning data under various conditions, then data quality and authenticity are improved, but labor cost and time consumption increase tremendously

Engineering Contradiction:

Improvedata qualityVSAvoiddata acquisition efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent creates virtual copies of machine operation data through physics simulation. Instead of collecting actual operation data from real machines under various conditions, the system generates synthetic time series data that replicates the physical behavior and characteristics of actual machine operations, thereby obtaining sufficient training data without time-consuming field collection

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces the mechanical data collection process (physical machine operation, sensor installation, field testing) with a computational physics simulation system. The simulation model substitutes actual machine operation to generate training data, eliminating the need for physical deployment and manual data collection while maintaining data authenticity through physics-based modeling

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If physical simulation is used to generate virtual time series information, then labor cost and time consumption are reduced, but the challenge of identifying accurate parameter values representing internal machine state arises

Engineering Contradiction:

Improvedata generation efficiencyVSAvoidparameter identification accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent employs feedback mechanisms in the physics simulation model where the simulated output time series information is compared with expected physical behavior and actual operational patterns. The simulation model adjusts its internal parameters based on this feedback to ensure that generated virtual data accurately reflects real machine states, thereby maintaining parameter identification accuracy while using efficient simulation-based data generation

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20230134186A1Machine learning data generation device, machine learning model generation method, and storage medium

Publication Date: 2023.05.04 YASKAWA DENKI KK

AI summary

Provided a machine learning data generation device including: at least one processor; and at least one memory device that stores a plurality of instructions which, when executed by the at least one processor, causes the at least one processor to execute: acquiring , in association with a predetermined label actual time series information; executing physical simulation of generating a plurality of pieces of virtual time series information; identifying parameter values based on the plurality of pieces of virtual time series information and the actual time series information, and to associate the identified parameter values with the label; generating a new parameter value and the label based on the identified parameter values; generating virtual time series information corresponding to a new internal state by executing physical simulation through use of the new parameter value; and generating new machine learning data.