Package Quality Inspection Using Adaptive ML Retraining

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Manual quality control of packages produced in roll-fed packaging machines is tedious and wasteful, requiring large quantities to be assessed, and lacks efficiency in detecting defects and ensuring food safety.

Innovation Solution

A method utilizing machine learning, specifically a deep learning model, to continuously monitor and evaluate package quality by capturing image data, determining quality measurements, and adjusting the packaging machine settings, with the ability to re-train the model when performance metrics deviate, ensuring autonomous operation and early detection of defects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual quality control is used to evaluate batches of finished products, then quality assessment can be performed, but it is tedious and leads to waste of product

Engineering Contradiction:
Improvequality assessmentVSAvoidproduct waste
Core Design Contradiction:
ReliabilityVSLoss of substance

Solution Approach 1:

The patent replaces manual mechanical quality control with an automated optical inspection system using image data capturing devices and machine learning models. The system captures images of packages during production and uses trained neural networks to automatically detect defects in sealing, folding, and packaging material, eliminating the need for manual opening and inspection of packages.

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

Solution Approach 2:

The system performs quality inspection during the packaging process itself, before the packages are completed and sent to storage or distribution. By inspecting packages in real-time on the production line, the system can identify and flag defective packages immediately, preventing further processing of faulty products and reducing overall waste.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If manual quality control is used to ensure package quality, then quality can be assessed, but large quantities of packages have to be assessed

Engineering Contradiction:
Improvepackage qualityVSAvoidinspection efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent replaces manual inspection with automated optical imaging and machine learning analysis. Multiple image capturing devices scan packages continuously during production, and trained neural network models automatically analyze the images to detect defects, enabling inspection of every package rather than relying on manual sampling of large quantities.

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

Solution Approach 2:

The inspection system operates continuously during package production, with image capturing devices and machine learning models analyzing packages in real-time as they move along the production line. This continuous automated inspection maintains high productivity while ensuring every package is assessed without the need for batch-by-batch manual evaluation.

Inventive Principle:
Principle #20Continuity of useful action

3Reliability

If machine learning model is used for autonomous quality control, then confidence in package quality is enhanced, but the model needs continuous re-training to maintain performance

Engineering Contradiction:
Improvequality confidenceVSAvoidmodel maintenance
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system incorporates feedback mechanisms where performance metrics of the machine learning model are continuously monitored. When the model's performance falls outside acceptable intervals, the system automatically triggers re-training mode, collects additional image data as training data, and re-trains the model to maintain optimal performance. This closed-loop feedback ensures high reliability while managing complexity through automation.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The machine learning model maintenance is partially automated through self-service mechanisms. The system automatically detects when re-training is needed based on performance metrics, collects relevant image data, and initiates re-training without requiring constant manual intervention. This reduces the operational complexity of model maintenance while maintaining high quality assurance standards.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4390590A1A computer implemented method for secure transmission of a data message from a source node to a target node
Publication Date: 2024.06.26 TETRA LAVAL HOLDINGS & FINANCE SA
  • EP4390590A1 patent drawingFigure 1
  • EP4390590A1 patent drawingFigure 2
  • EP4390590A1 patent drawingFigure 3

AI summary

The present invention relates to a method (100) for quality control of packages produced in a roll-fed packaging machine. The method (100) comprising a series of recurring steps, said series of recurring steps comprising: obtaining (S102), from an image data capturing device, image data depicting at least a portion of a package; in response to the roll-fed packaging machine being in an autonomous mode: determining (S106) a quality measurement of the package by inputting the image data into a machine learning model, such as a neural network; determining (S108) an action to be performed, based on the determined quality measurement; communicating (S110) the action to a control system of the roll-fed packaging machine; determining (S112) a performance metric of the machine learning model based on the quality measurement of the package and quality measurements of previously assessed packages; and assigning (S114) the mode of the roll-fed packaging machine to a re-training mode if the performance metric is outside a performance metric interval; in response to the roll-fed packaging machine being in the re-training mode: collecting (S116) the image data as training data for the machine learning model, re-training (S118) the machine learning model using the training data, and assigning (S120) the mode of the roll-fed packaging machine to the autonomous mode.