ML-Based Anomaly Detection for Package Conveyor Diversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing package conveyor systems in fulfillment centers face challenges in efficiently detecting and diverting anomalous products, such as those with defective seals or damaged packaging, from shipping lanes as they travel along the conveyor.
Innovation Solution
A method and system that utilize machine learning models trained on digital images of products to identify anomalous products. The system includes image capture devices to capture product images, computing devices to process these images using trained machine learning models, and package sorters to divert identified anomalous products from shipping lanes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used to identify anomalous products, then detection accuracy is improved, but system complexity increases
Solution Approach 1:
The patent replaces traditional mechanical inspection systems with machine learning models that process digital images to identify anomalous products. The machine learning model analyzes visual characteristics from captured images to detect anomalies such as defective seals or damaged packaging, substituting physical inspection mechanisms with intelligent image processing systems.
Solution Approach 2:
The patent introduces an intermediary processing layer between the image capture device and the package sorter. The machine learning model serves as this intermediary, receiving digital images, analyzing them for anomalies, and providing classification results that trigger diversion actions. This intermediary layer manages the complexity by centralizing the decision-making logic.
2Loss of time
If real-time image processing is performed, then diversion timing is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-training machine learning models on extensive datasets of anomalous and non-anomalous products before actual operation. This pre-training phase prepares the model to make rapid predictions during real-time conveyor operation, reducing the processing time required during actual anomaly detection without compromising accuracy.
Solution Approach 2:
The patent optimizes processing parameters including image resolution, processing frequency, and model inference settings to achieve the fastest possible anomaly detection. By adjusting these parameters, the system balances processing speed with detection accuracy, enabling real-time decisions on product diversion.
3Productivity
If multiple products are inspected simultaneously, then productivity is improved, but measurement precision deteriorates
Solution Approach 1:
The patent segments the inspection process by capturing images of individual products at specific intervals along the conveyor, rather than attempting to inspect multiple products simultaneously in a single operation. The image capture device takes discrete images of products as they pass through the inspection zone, allowing the machine learning model to analyze each product individually with high precision while maintaining high throughput through continuous inspection.
Data Source
AI summary
A method of automatically diverting products from a shipping lane on a package conveyor system using a package sorter. One or more computing devices communicatively coupled to a network receives a plurality of digital images of products, generates an anomalous data set and a non-anomalous data set therefrom, and trains a machine learning model using the anomalous data set and the non-anomalous data set. Digital image are received of a target product traveling on the conveyor system. Prior to the target product reaching the package sorter, the trained machine learning model determines that the target product is an anomalous product. A command signal is delivered to the package sorter to cause the package sorter to divert the anomalous product from the shipping lane.