Scheme evaluation method of coal gangue sorting robot system with time-varying multi-scenario based on deep learning

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Data Science

Background:

• The effective configuration of coal gangue sorting robot systems (CGSRS) is challenged by time-varying raw coal flow (TVRCF) and complex field conditions.
• Existing multi-task allocation strategies for CGSRS face limitations in handling dynamic scenarios and high time complexity.

Purpose of the Study:

• To propose a novel deep learning-based scheme evaluation method for CGSRS.
• To address the challenges posed by time-varying, multi-scenario TVRCF in optimizing CGSRS performance.
• To enhance the accuracy and stability of CGSRS multi-task allocation strategies.

Main Methods:

• Dataset generation for TVRCF across multiple scenes and conditions, considering belt speed and coal flow variations.
• Development of a CGSRS scheme evaluation model using DenseNet, trained on RGB sample sets derived from gangue queue labels.
• Implementation of the scheme evaluation method for predicting random gangue queues and assessing solution stability.

Main Results:

• The proposed deep learning model demonstrates accurate and stable solutions for CGSRS scheme evaluation.
• Significant reduction in time complexity, with highly stable performance observed.
• The method proves superior to traditional multi-task allocation strategies, unaffected by data variations.

Conclusions:

• The deep learning-based CGSRS scheme evaluation method offers a robust solution for optimizing robot system configurations.
• This approach significantly enhances efficiency and stability in dynamic mining environments.
• This research marks the first application of deep learning to multi-task allocation problems within CGSRS.