SOM-Based Process Monitoring for Nonlinear Batch Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the behavior of dynamic process systems, especially nonlinear ones, require complex modeling and are not reliable or feasible for many applications, particularly in industries like pharmaceuticals where batch processes are common.
Innovation Solution
A data-based method using self-organizing maps (SOMs) to predict future process behavior by learning from historical data, eliminating the need for rigorous modeling and allowing for the representation of complex nonlinear dynamics, with automated learning and visualization of predicted trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simulation-based prediction methods are used to accurately predict complex nonlinear dynamic processes, then prediction reliability is improved, but device complexity and modeling effort increase significantly
Solution Approach 1:
The patent uses self-organizing maps to create a simplified data-based copy of the complex nonlinear process behavior. Instead of building rigorous simulation models, the system learns from historical data and creates a virtual representation that captures essential process dynamics, enabling predictions without complex modeling
Solution Approach 2:
The patent replaces traditional mechanical simulation-based prediction approaches with a data-driven neural network approach. Self-organizing maps substitute the need for explicit mathematical process models with learned patterns from historical data, eliminating complex modeling requirements while maintaining prediction capability
2Adaptability or versatility
If data-based identification methods are applied to nonlinear dynamic models without structural specification, then adaptability to complex processes is improved, but device complexity becomes virtually infinite
Solution Approach 1:
The patent changes the fundamental parameter of model representation from explicit mathematical structures to learned data patterns. By using self-organizing maps with adjustable neuron weights and topological arrangements, the system adapts to complex nonlinear processes through parameter learning rather than structural specification
3Ease of operation
If extrapolation methods are used to predict future process behavior, then ease of operation is improved, but prediction accuracy deteriorates due to ignoring earlier past effects
Solution Approach 1:
The patent performs preliminary learning from the entire historical data set before making predictions. The self-organizing map is trained in advance with comprehensive process data, capturing long-term patterns and effects that simple extrapolation would miss, while maintaining operational simplicity during actual prediction
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method and to a correspondingly designed system for predicting the operation of a technical installation in which a process-engineering process having at least one process step runs, wherein datasets characterizing the operation of the installation and containing values of process variables are acquired in a time-dependent manner and stored in a data memory and, in a learning phase, a self-organizing map (SOM) is learned using historical datasets per process step or batch, symptom threshold values and their tolerances are ascertained and stored for each neuron per SOM and all permissible temporal profiles of victor neurons are ascertained and stored per process step or batch for all timestamps. The invention is characterized in that, in an evaluation phase, a time starting from which the operation of the technical installation should be predicted is determined, in that at least one victor neuron profile for the overall process step or batch is then ascertained using current datasets of at least one process step by way of the learned victor neuron profiles, and in that the values, stored in the neurons of this victor neuron profile, of the process variables are displayed on an output unit after the previously determined time in the form of a predicted profile. By predicting the future dynamic process behaviour, the invention allows early corrective interventions in the operation of a technical installation.