AI Object Recognition and Digital Twin Sorting Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated sorting and palletizing systems face inefficiencies in object recognition, estimation, and handling due to limitations in accurately identifying diverse object properties and selecting appropriate sorting algorithms, leading to potential damage and errors.
Innovation Solution
A computer-implemented method utilizing a trained Convolutional Neural Network (CNN) for object recognition and estimation, combined with a trained Artificial Intelligence (AI) algorithm for sorting algorithm selection, and verified through a digital twin model to ensure accurate and efficient object handling and sorting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional automated sorting systems are used, then sorting operations can be performed, but accuracy in object recognition and handling is insufficient
Solution Approach 1:
The patent replaces traditional mechanical sorting systems with an AI-based digital twin model that uses convolutional neural networks for object recognition and reinforcement learning for sorting algorithm selection. This substitution of mechanical recognition systems with intelligent algorithms significantly improves object recognition accuracy while maintaining handling reliability through virtual simulation and validation.
Solution Approach 2:
The patent creates a digital twin (virtual copy) of the physical sorting system that allows for accurate object recognition and algorithm validation without affecting real objects. This copying approach enables precise measurement and recognition in the virtual environment, which can then be transferred to the physical system, improving both accuracy and reliability.
2Reliability
If sorting algorithms are tested with real objects, then algorithm effectiveness can be verified, but objects may be damaged during testing
Solution Approach 1:
The patent uses a digital twin (virtual copy) of the sorting system to test and verify sorting algorithms. By performing all algorithm validations in the virtual environment with virtual objects, the system achieves reliable algorithm verification without exposing real objects to any damage risk. The virtual copy perfectly replicates physical conditions while eliminating harmful effects.
Solution Approach 2:
The patent performs preliminary testing and validation of sorting algorithms in the digital twin environment before deploying them to the physical system. This preliminary action in the virtual space allows complete algorithm verification without any risk to real objects, ensuring reliability before actual implementation.
3Measurement precision
If manual sorting methods are used, then handling accuracy can be maintained, but productivity is reduced
Solution Approach 1:
The patent replaces manual sorting operations with an automated system driven by convolutional neural networks for object recognition and reinforcement learning algorithms for sorting decisions. This substitution maintains high handling accuracy through AI-based perception while dramatically increasing productivity through automated execution at industrial speeds.
Solution Approach 2:
The system uses AI algorithms that automatically recognize objects, select appropriate sorting algorithms, and control execution without human intervention. This self-service capability maintains the accuracy benefits of manual handling while achieving the productivity of automated systems.
4Adaptability or versatility
If diverse sorting algorithms are implemented, then adaptability to different objects is improved, but system complexity increases
Solution Approach 1:
The patent implements a digital twin that contains the complete library of diverse sorting algorithms in a virtual environment. This copying approach allows the system to maintain high adaptability by having access to multiple sorting strategies without increasing physical system complexity. The virtual space handles the complexity of algorithm diversity while the physical system remains simple and focused on execution.
Solution Approach 2:
The system uses dynamic algorithm selection through reinforcement learning, where the appropriate sorting algorithm is chosen based on real-time object characteristics and conditions. This dynamic approach provides adaptability to diverse objects without requiring all algorithms to be simultaneously active in the physical system, thereby managing complexity effectively.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
It is provided a computer-implemented method, in particular for object recognition, estimation and handling, the method comprising: classifying and/or identifying at least one object (202, 204, 206) using a trained Convolutional Neural Network, CNN, (300), particularly with transfer learning capabilities, the object (202, 204, 206) having object properties associated with the object (202, 204, 206) and being observable through a sensor device, the trained CNN (300) outputting a classification result associated to the object (202, 204, 206) and indicative of estimated object properties, wherein the object properties are at least one of a shape, a labelling, a weight, a dimension, a material of the object, (sub-)components, a composition and a content; identifying and/or selecting at least one sorting algorithm (304) using at least one trained Artificial Intelligence, AI, algorithm (302), preferably being in conjunction with the CNN (300), based on the classification result of the at least one object (202, 204, 206); and verifying and/or validating the identified and/or selected at least one sorting algorithm (304, 308) using a digital twin model (306) of a sorting apparatus (200).