Welding Defect Prediction Using Surface Valley Geometry and AI

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing non-contact inspection methods for manufactured objects, such as ultrasonic flaw detection and X-ray CT, face challenges in accurately detecting defects within complex objects and are limited by equipment size and cost.

Innovation Solution

A learning device that generates an estimation model to predict defect sizes in additively manufactured objects by learning from welding conditions, dimensions of narrow portions, and positional relations, allowing for accurate defect size prediction and reduction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If ultrasonic flaw detection is used to inspect manufactured objects, then defect detection capability is improved, but the probe cannot be applied until after surface cutting and reflected waves cannot be detected smoothly for complex objects

Engineering Contradiction:

Improvedefect detection capabilityVSAvoidprobe application difficulty

Core Design Contradiction:

Measurement precisionVSEase of operation

Solution Approach 1:

The invention performs preliminary measurement of surface shape and narrow portion dimensions before welding to predict defect occurrence. By acquiring surface shape data and calculating narrow portion dimensions (bottom width, opening width, valley depth) in advance, the system can identify high-risk areas for unwelded defects before they occur, enabling preventive control rather than post-inspection

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention replaces the mechanical probe-based ultrasonic detection system with an optical measurement system (laser sensor or camera) that non-contactly measures surface shape. This substitution eliminates the need for physical probe contact and surface cutting, allowing measurement of complex manufactured objects in their original state

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If X-ray CT apparatus is used for non-contact inspection, then defect detection capability is improved, but the apparatus size limits object inspection and the apparatus itself is expensive

Engineering Contradiction:

Improvedefect detection capabilityVSAvoidapparatus size and cost

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The invention extracts only the essential measurement information (surface shape and narrow portion dimensions) needed for defect prediction, rather than using comprehensive but expensive X-ray CT scanning. By focusing on specific geometric features (bottom width, opening width, valley depth of narrow portions), the system achieves defect prediction capability with simpler, more affordable equipment

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The invention uses inexpensive optical measurement devices (laser sensors or standard cameras) instead of expensive X-ray CT apparatus. These simpler measurement tools, while having limited penetration capability, are sufficient for surface and near-surface defect prediction when combined with welding condition data and machine learning models

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

3Measurement precision

If machine learning model is trained with comprehensive welding parameters and surface shape data, then defect prediction accuracy is improved, but data acquisition complexity and processing time increase

Engineering Contradiction:

Improvedefect prediction accuracyVSAvoiddata acquisition and processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The invention focuses machine learning attention on specific local features (narrow portion dimensions: bottom width, opening width, valley depth) rather than processing all surface data uniformly. By identifying and emphasizing the most critical geometric parameters that correlate with defect occurrence, the model achieves high accuracy with reduced data processing requirements

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system performs preliminary calculation of narrow portion dimensions from surface shape data before feeding into the machine learning model. This preprocessing step extracts key features (bottom width, opening width, valley depth) in advance, reducing the complexity of data input to the learning device and improving processing efficiency

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250033133A1Learning device, defect determination apparatus, learning method, defect determination method, welding control device, and welding device

Publication Date: 2025.01.30 KOBE STEEL LTD

AI summary

A learning device includes: a data acquisition unit configured to acquire information on a welding condition when beads are deposited, a dimension related to a narrow portion forming a valley portion in a surface shape of an additively manufactured object before the beads are deposited, a positional relation between the narrow portion and a target position of the bead, and a defect size of an unwelded defect; and a learning unit configured to generate the estimation model by learning a relation between the welding condition, the dimension related to the narrow portion and the positional relation, and the defect size. The dimension related to the narrow portion includes at least one of a bottom width, an opening width representing an interval between top portions on both sides, both sides, and a valley depth from the top portion to a bottom of the valley portion.