Process State Simulation Using ML-Inferred Hidden Variables
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing simulation models for industrial processes struggle when not all necessary information is available, leading to inaccurate predictions of process variables and suboptimal control actions.
Innovation Solution
A method using a trained machine learning model to bridge the gap between available process snapshot data and complete process state information, combined with a simulation model to predict process variables and evaluate control actions, leveraging cloud computing for speed and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a simulation model is used to predict process variables, then the prediction accuracy is improved, but the method fails when not all necessary information is available in the process snapshot
Solution Approach 1:
A machine learning model is introduced as an intermediary between the available process snapshot data and the simulation model. This ML model infers missing process state information that the simulation model requires, enabling the simulation to run accurately even when direct measurements are unavailable. The ML model acts as a bridge that transforms partial observations into complete initial state representations.
Solution Approach 2:
The patent replaces direct physical measurement of all process variables with a computational approach. Instead of requiring physical sensors for every process state variable, a machine learning model computationally infers missing information from available measurements, substituting physical measurement infrastructure with intelligent data processing.
2Reliability
If all process state information is directly measured, then the simulation model can accurately predict process variables, but the complexity and cost of the measurement system increases
Solution Approach 1:
The patent extracts only the essential measurements that are actually available from the process, rather than implementing a complete measurement system for all possible variables. The machine learning model then derives the remaining necessary information from these extracted measurements, reducing the measurement infrastructure to only what is strictly necessary.
Solution Approach 2:
The system uses the available measurements themselves to generate the complete initial state information needed by the simulation model. Rather than requiring external measurement infrastructure for all variables, the system self-generates the necessary information by combining available measurements with ML-based inference.
Data Source
Figure 1
Figure 2
AI summary
A computer-implemented method (100) for predicting, based on the state of an industrial process at a first point in time that is described by a process snapshot record (1) with values of a first set of variables, a value (6) of at least one process variable of the industrial process at a second, later point in time, comprising the steps of: • mapping (110), by means of at least one trained machine learning model (2), the process snapshot record (1) to at least one initial state record (3), wherein this initial state record (3) characterizes the state of the process at the first point in time and contains an estimate of at least one hidden variable that is not comprised in the process snapshot record (1); • providing (120) the initial state record (3) to a simulation model (4) of the process; • simulating (130), using the simulation model (4), the further development of the process; • obtaining (140), from the simulation model (4), a final state record (5) that characterizes the state of the process at the second point in time; and • determining (150), based on said final state record (5), the sought value (6) of the process variable at the second point in time.