Automated Process-Graph Modeling for Site-Wide Plant Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current industry practices face challenges in generating efficient, automated optimization models for site-wide plant optimization due to the complexity of manufacturing processes, reliance on manual methods, limited AI utilization, and the inability to leverage historical plant sensor data effectively, leading to suboptimal operational strategies and increased computing times.
Innovation Solution
A computer-implemented method that maps a process flow diagram to a directed acyclic graph, learns behavioral relationships using regression functions, and generates an optimization model with an adjacency matrix to provide setpoints for optimal control variables, utilizing machine learning to automate the generation of optimization models for site-wide plant optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual optimization model generation is used, then domain experts can generate accurate models, but the process is time-consuming and requires both domain experts and optimization experts
Solution Approach 1:
The system enables automated generation of optimization models using historical plant sensor data and machine learning algorithms, eliminating the need for manual model generation by domain and optimization experts. The automated system processes sensor data to learn process behaviors and generates optimization models independently, significantly reducing time consumption while maintaining model accuracy through data-driven approaches.
2Adaptability or versatility
If conventional end-to-end learning methods are used, then single process optimization can be achieved, but the methods cannot handle site-wide optimization problems with multi-plant processes
Solution Approach 1:
The system creates a universal optimization framework that can handle both single-process and site-wide multi-plant optimization problems. By using a standardized approach with process flow diagrams mapped to graphs, regression functions for each node, and adjacency matrices for interconnections, the system provides multi-functionality that scales from individual process optimization to enterprise-wide optimization without requiring fundamentally different methodologies.
Solution Approach 2:
The system segments the complex site-wide optimization problem into manageable components: individual process nodes represented as graphs, regression functions learned for each node based on historical data, and adjacency matrices representing inter-process connections. This segmentation allows the system to handle complexity by breaking down the overall optimization into smaller, tractable sub-problems that can be solved and then integrated.
3Loss of information
If historical plant sensor data is used directly for optimization, then data-driven insights can be obtained, but the data comes from different temporal resolutions making end-to-end solutions difficult
Solution Approach 1:
The system performs preliminary processing of historical sensor data by learning regression functions for each process node before generating optimization models. This preliminary action involves training machine learning models on historical data to capture process behaviors and relationships, transforming raw sensor data into meaningful predictive models that can be used in optimization. This preprocessing step handles temporal resolution variations by learning from historical patterns rather than requiring synchronized real-time data.
Data Source
AI summary
A computer implemented method for automatically generating an optimization model for site-wide plant optimization includes mapping a process flow diagram of a plant process to a graph comprising nodes and edges, wherein the nodes represent processes and the edges represent flows between processes. A behavior is learned for each node of the graph based at least on historic data of the plant process. One or more regression functions are modeled for each node to predict an output of each of the processes, wherein the one or more regression functions are modeled based on the learned behavior for each node.