Towards Deep Transfer Learning in Industrial Internet of Things

Area of Science:

• Computer Science
• Artificial Intelligence
• Industrial Internet of Things (IIoT)

Background:

• Machine learning is crucial for data analysis in Internet of Things (IoT) systems.
• Training general models on diverse data enhances flexibility but faces computational limits on IoT devices.
• Transfer learning offers a solution for adapting trained models to specific applications.

Purpose of the Study:

• To propose a general framework for applying transfer learning in industrial Internet of Things (IIoT) systems.
• To categorize transfer learning applications in IIoT into four distinct scenarios.
• To design tailored workflows for each scenario to optimize transfer learning implementation.

Main Methods:

• Developed a general framework for transfer learning in IIoT.
• Categorized IIoT transfer learning into centralized/distributed and large/small dataset scenarios.
• Designed specific workflows for each scenario.
• Applied the transfer learning technique to IIoT component recognition using the VGG-16 model and T-Less datasets.

Main Results:

• Experimental results validated the efficacy of the proposed transfer learning framework across different scenarios.
• The approach significantly reduced training time compared to classical Convolutional Neural Network (CNN) methods.
• Higher accuracy was achieved in IIoT component recognition tasks using transfer learning.

Conclusions:

• The proposed transfer learning framework is effective for IIoT systems, addressing computational limitations.
• Tailored workflows enhance the performance of transfer learning in diverse IIoT application scenarios.
• Transfer learning offers a promising method for efficient and accurate machine learning in industrial settings.