Decentralized DMDNN Training for Distributed Production Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In distributed industrial production environments, the training of deep neural network models is inefficient due to high computational requirements and poor generalization performance across varying production conditions, making conventional deep learning methods ineffective for decentralized data management.
Innovation Solution
A method for constructing and training a Decentralized Migration Diagram Neural Network (DMDNN) model involves determining decentralized migration diagram learning requirements, distributing management nodes, constructing network calculation nodes, and performing weighted fusion of network weight parameters across distributed management nodes to create a master DMDNN model suited for diverse data sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If deep neural network models are trained with massive scattered data in distributed production management servers, then the model can utilize more data for training, but the training time becomes excessively long and computational requirements increase
Solution Approach 1:
The patent segments the centralized training process into distributed training across multiple production management servers. Each server trains a local DMDNN model using its own data, and then models are aggregated through weighted fusion to create a global model. This segmentation allows parallel training execution, significantly reducing training time while utilizing data from all distributed servers.
Solution Approach 2:
The patent introduces a new dimension of model aggregation through weighted fusion of network weight parameters from multiple distributed models. Instead of traditional centralized training in a single dimension, the system operates across multiple dimensions by combining local models from different servers, achieving both extensive data utilization and efficient training through parallel processing.
2Measurement precision
If deep learning models are trained with production data in specific production conditions, then the model achieves high accuracy for those specific conditions, but the model exhibits poor generalization performance for other production conditions
Solution Approach 1:
The patent creates a universal DMDNN model through weighted fusion of multiple local models trained on different production conditions. The global model inherits capabilities from all local models, making it universally applicable across various production conditions while maintaining the specialized knowledge each local model learned from its specific data distribution.
Solution Approach 2:
The patent merges multiple local DMDNN models trained on different production conditions into a single global model through weighted fusion of network weight parameters. This combining process integrates the specialized knowledge from each local model, creating a comprehensive model that generalizes well across all production conditions while preserving the high accuracy characteristics of individual local models.
3Quantity of substance
If data are gathered in a central management server for training, then the model can access all production data, but privacy protection and data security concerns prevent direct gathering of distributed manufacturing data
Solution Approach 1:
The patent introduces local DMDNN models as intermediaries that enable data utilization without direct data transfer. Each production management server trains a local model using its own data, and only the model parameters (not the raw data) are transmitted for aggregation. This intermediary approach allows the system to access knowledge from all production data while maintaining data security and privacy protection at the source.
Data Source
AI summary
A method for constructing and training a Decentralized Migration Diagram Neutral Network (DMDNN) model for a production process, including: determining a production task input management node, distributed management nodes and a granularity of each of the distributed management nodes, and constructing a production system network; constructing network calculation nodes on each of the distributed management nodes; constructing and training a DMDNN model; and applying the trained DMDNN model in management and control of the production process.