Laser Cutting Parameter Refinement with Bayesian Re-Optimization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for calculating process parameters in laser cutting are time-consuming and costly, especially when dealing with slightly different material compositions or thicknesses, and require extensive training for AI-based approaches, making them inefficient for optimizing parameters across varying settings.

Innovation Solution

A computer-implemented method that determines material and machine properties to optimize process parameters using a re-optimization algorithm based on preconfigured initial parameters, which expands the parameter domain through an acceptability classifier and Bayesian optimization, reducing the need for extensive experimentation and personal assessment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If AI-based approaches (neural networks) are used to determine optimized process parameters, then the automation and speed of parameter calculation is improved, but the resource intensity and cost of training and testing the model increases significantly

Engineering Contradiction:

Improvespeed of parameter calculationVSAvoidresource intensity of training

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent introduces a surrogate model as an intermediary between the complex neural network and the actual parameter optimization task. This surrogate model captures the essential relationships from limited training data, enabling fast predictions without requiring full neural network training. The surrogate model acts as a mediator that translates scarce training resources into practical optimization capability.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent performs preliminary action by training the neural network model once on comprehensive data to create a surrogate model, then uses this pre-trained surrogate for rapid parameter calculation in subsequent applications. This preliminary training phase eliminates the need for repeated resource-intensive training when solving new optimization problems.

Inventive Principle:

Principle #10Preliminary action

2Reliability

If extensive training data is collected to improve the reliability of AI-based parameter determination, then the accuracy and robustness of the model is improved, but the time and cost required for data collection and model training increases

Engineering Contradiction:

Improvereliability of parameter determinationVSAvoidtime for data collection and training

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial action by using a limited subset of training data to create a surrogate model that captures the essential relationships. Instead of requiring exhaustive training data coverage, the surrogate model learns from representative samples, achieving sufficient reliability for practical application without the time cost of comprehensive data collection.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The patent creates a simplified copy (surrogate model) of the complex neural network relationships. This surrogate model replicates the essential input-output relationships from the trained neural network but can be evaluated rapidly without requiring the original training data or computational resources, thus maintaining reliability while reducing time requirements.

Inventive Principle:

Principle #26Copying

3Adaptability or versatility

If the parameter domain is expanded to cover all possible material compositions and thicknesses, then the versatility and applicability of the optimization tool is improved, but the complexity of the model and the resources required for training increase

Engineering Contradiction:

Improveapplicability to different materialsVSAvoidcomplexity of optimization model

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the parameter domain into manageable regions or clusters based on material properties and processing conditions. The surrogate model is trained to handle these segmented regions, allowing the system to achieve high versatility across different materials while keeping the model complexity manageable through localized optimization strategies.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates a universal surrogate model that can handle multiple material types and processing conditions through a single unified framework. This universal model replaces the need for separate complex models for each material type, achieving versatility while reducing overall system complexity through model consolidation.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11927927B2Computer implemented method of and optimisation tool for refinement of laser cutting processing parameters by means of an optimization tool

Publication Date: 2024.03.12 BYSTRONIC LASER AG

AI summary

The present disclosure relates to a method of calculating process parameters. which are optimized for processing a workpiece with specific material properties by means of a laser machine, comprising the method steps of: determining material properties for which the process parameters should be optimized; determining preconfigured initial process parameters; executing a re-optimization algorithm until a target objective function is minimized or maximized for calculating optimized material-specific process parameters by accessing a storage with a statistical model, wherein the statistical model is based on Bayesian optimization using Gaussian Processes as priors.