3D Printing Quality Prediction via In-Process Scanning and Machine Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In three-dimensional printing, predicting production quality is challenging due to inefficiencies in scheduling and rework processes, leading to idle equipment and bottlenecks, as existing methods lack effective prediction of part failures before completion, resulting in increased costs and reduced productivity.

Innovation Solution

A system using machine learning functions, such as multilayer perceptrons, to predict the likelihood of part failure by scanning regions of interest during printing, calculating input metrics, and training on mechanical testing data to determine the probability of meeting quality specifications, allowing for optimized scheduling and reduced rework.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If machine learning prediction is implemented to predict part quality, then productivity is improved by reducing idle time and optimizing scheduling, but device complexity increases due to the addition of scanning systems and machine learning infrastructure

Engineering Contradiction:
Improvemanufacturing productivityVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system performs quality prediction during the printing process itself, scanning regions of interest and analyzing inputs before the part is completed. This preliminary action allows identification of parts likely to fail early in the printing process, enabling proactive scheduling decisions and rework planning before the part leaves the printer, thus improving productivity without requiring complex post-processing inspection systems

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional mechanical quality inspection methods with machine learning-based prediction. Instead of physically measuring or testing parts after printing, the system uses machine learning functions that analyze scanning data and printing parameters to predict quality outcomes, substituting complex mechanical measurement systems with computational analysis

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Manufacturing precision

If mechanical testing is performed on printed parts to determine quality, then manufacturing precision is improved through accurate quality assessment, but loss of time increases due to testing duration and equipment idle time

Engineering Contradiction:
Improvequality assessment accuracyVSAvoidtesting time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The system skips the traditional time-consuming mechanical testing step by using machine learning prediction during the printing process. The machine learning function analyzes scanning data and printing parameters to rapidly predict quality outcomes without requiring physical testing of each part, thus maintaining quality assessment accuracy while dramatically reducing the time lost to testing

Inventive Principle:
Principle #21Skipping (Rushing through)

Solution Approach 2:

The printing system performs quality assessment itself during the printing process through integrated scanning and machine learning analysis, rather than requiring separate testing equipment and processes. The system uses its own operational data (printing parameters, scanning images) to predict quality, making the quality assessment function self-contained and eliminating the need for external testing operations that would cause equipment idle time

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3749505B1Printing production quality prediction
Publication Date: 2022.12.14 HEWLETT PACKARD DEVELOPMENT COMPANY LP
  • EP3749505B1 patent drawingFigure 1
  • EP3749505B1 patent drawingFigure 2
  • EP3749505B1 patent drawingFigure 3

AI summary

A system and method for providing three dimensional printing production quality prediction is described herein. The system may include logic to scan a region of interest associated with a layer of a plurality of parts during printing to obtain an input for the region of interest and compute an input metric associated with the layer based on the input. In response to an initial set of print jobs, at least a portion of the plurality of parts is mechanically tested to determine at least one output. In response to a production print job, a likelihood of a part of the plurality of parts to satisfy a quality specification is predicted by inputting the input metric to at least one machine learning function comprising the at least one output, wherein the at least one machine learning function is trained during the initial set of print jobs.