Deep Learning Material Classifier for Recycling Adaptability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current material sorting systems rely heavily on rule-based methods that require manual parameterization by experts, leading to inefficiencies and labor-intensive adaptations to changes in material flows, limiting their ability to accurately and efficiently classify and sort materials in recycling processes.

Innovation Solution

A deep learning-based classification system that learns relevant features from data and optimizes performance metrics, reducing the need for human expertise and enabling flexible adaptation to new material classifications using supervised, semi-supervised, and unsupervised training methods with dual energy X-ray data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If rule-based methods are used for material classification, then the system provides interpretable sorting decisions, but the system requires manual parameterization by experts and cannot adapt efficiently to new material flows

Engineering Contradiction:
Improveadaptability to new material flowsVSAvoidmanual parameterization effort
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The deep learning model performs self-parameterization by automatically learning optimal classification parameters and features from training data, eliminating the need for manual expert parameterization. The system adapts to new material flows by retraining on new data without requiring expert intervention to adjust parameters.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system transforms fixed manual parameters into learnable parameters that are automatically optimized through training. The deep learning model dynamically adjusts its internal parameters (weights and biases) based on training data, enabling automatic adaptation to different material flows without manual reparameterization.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If deep learning models are used for material classification, then adaptability to new material flows improves, but the need for training data and model training time increases

Engineering Contradiction:
Improveflexibility in classifying new material objectsVSAvoidmodel training time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary training with diverse material data before deployment to build a robust pre-trained model. This preliminary action reduces the need for extensive retraining when encountering new material flows, as the model already possesses general classification capabilities that can be fine-tuned with minimal additional training.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual expert parameterization is used, then the system provides accurate sorting decisions, but the availability of experts is limited and adaptation is slow

Engineering Contradiction:
Improvesorting decision accuracyVSAvoidadaptation speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system replaces the mechanical process of manual expert parameterization with an automated deep learning-based parameter optimization process. The model automatically learns classification parameters from data, substituting human expert labor with computational processes that can rapidly adapt to new material flows while maintaining high sorting accuracy.

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

4Loss of information

If rule-based systems are deployed, then the system structure is simple and understandable, but the system provides hardly any information about the actual sorting decision

Engineering Contradiction:
Improveinformation about sorting decisionVSAvoidsystem complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The system introduces intermediate feature representations and attention mechanisms that bridge the gap between complex deep learning models and interpretable decisions. These intermediaries provide insights into which features and regions contribute most to sorting decisions, delivering actionable information to recyclers without requiring simplification of the underlying complex model architecture.

Inventive Principle:
Principle #24Intermediary (Mediator)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

The deep learning system achieves robust and accurate material classification with reduced manual effort, improving sorting efficiency and adaptability to changes in material flows, enhancing the recycling process by providing traceable and high-quality sorting decisions.

Implementation Method 1

deep learning-based sorting on dual energy X-ray transmission data

Methodology Applied
Scientific EffectX-ray transmission: X-Ray

Implementation Method 2

a respective pair of sensor data, wherein the respective sensor data comprises dual energy X-ray data

Methodology Applied
Scientific EffectDual energy X-ray: X-Ray

Data Source

PatentEP4194107A1Apparatus and method for classifying material objects
Publication Date: 2023.06.14 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP4194107A1 patent drawingFigure 1
  • EP4194107A1 patent drawingFigure 2
  • EP4194107A1 patent drawingFigure 3

AI summary

The application concerns an apparatus and a method for classifying material objects. The apparatus comprises a deep learning model. The apparatus is configured to, in an initialization phase, subject the deep learning model to supervised learning based on a training data obtained from, for each of a training set of material objects, a pairs of sensor data obtained by a measurement of the respective material object and label information associating the respective material object with a target classification. Additionally, the apparatus is configured to, using the deep learning model, classify a predetermined material object based on sensor data obtained by a measurement of the predetermined material object.