Hybrid First-Principles and Inferential Model for Root-Cause Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chemical and petro-chemical plants face challenges in performing root-cause-analysis due to the large number of process variables involved, inconsistent and discrete nature of undesirable events, and the lack of efficient tools for calculating causal correlations, which hinders quick identification of root-cause variables.
Innovation Solution
A hybrid online first principles model combined with an empirical inferential model is used to generate continuous Key Performance Indicators (KPIs) for representing undesirable plant events, allowing for continuous KPI estimation and prediction, thereby facilitating root-cause analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full-scale first-principles dynamic model is used for offline simulation, then model accuracy and fundamental principle representation are improved, but device complexity and computational burden increase significantly
Solution Approach 1:
The patent segments the complex first-principles dynamic model into a simplified steady-state first-principles model component and a data-driven inferential model component. This segmentation allows the system to retain the physical accuracy of first-principles modeling while reducing computational complexity through the steady-state approximation, and further simplifies the dynamic behavior representation through the inferential model trained on historical data.
Solution Approach 2:
The patent changes the operational parameters of the model by transitioning from a dynamic model that requires continuous solving of differential equations to a steady-state model that operates at equilibrium points. This parameter change reduces computational burden while maintaining accuracy through the combination with the inferential model that captures dynamic transitions.
2Device complexity
If a steady-state first-principles model is used instead of a dynamic model, then device complexity is reduced, but the ability to represent dynamic plant behavior deteriorates
Solution Approach 1:
The patent merges the steady-state first-principles model with a data-driven inferential model to create a hybrid modeling approach. The steady-state model provides accurate physical relationships at equilibrium conditions, while the inferential model learns to predict dynamic behavior and transitions between steady states from historical process data, thereby compensating for the limitations of the steady-state approximation.
Solution Approach 2:
The inferential model acts as an intermediary that bridges the gap between the steady-state first-principles model and the actual dynamic plant behavior. It translates the static model outputs into dynamic predictions by learning from historical data how the system evolves between steady states, effectively mediating between the simplified model and complex reality.
3Measurement precision
If thousands of process variables are monitored for root-cause analysis, then measurement precision is improved, but the difficulty of detecting and measuring the root cause increases
Solution Approach 1:
The patent extracts and isolates the root-cause variables from the thousands of monitored process variables by using the hybrid model to calculate event indicators. The system identifies which specific variables have the strongest causal relationship with plant events by analyzing model predictions and historical correlations, thereby extracting only the critical few variables that matter for root-cause analysis.
Solution Approach 2:
The patent performs preliminary analysis by continuously monitoring and storing calculated event indicators for all process variables before actual plant events occur. This preliminary action builds a database of variable-event relationships that enables rapid root-cause identification when events happen, as the system has already pre-processed and organized the causal relationships in advance.
4Productivity
If continuous KPI estimation is implemented using the hybrid model, then productivity is improved, but use of energy and computational resources increases
Solution Approach 1:
The patent implements a dynamic hybrid modeling approach where the system adapts its computational behavior based on process conditions. The steady-state first-principles model is evaluated at current process conditions to determine if the system is near steady state, and the inferential model provides dynamic predictions when transitioning between states, optimizing computational resource usage based on actual process dynamics.
Solution Approach 2:
The patent applies partial action by selectively using the computationally intensive inferential model only when needed for dynamic predictions, rather than continuously running full dynamic simulations. The system uses the simpler steady-state model for routine calculations and resorts to the more complex inferential approach only when dynamic behavior must be captured, thereby reducing overall computational energy consumption.
Data Source
AI summary
Computer-based methods and system perform root-cause analysis on an industrial process. A processor executes a hybrid first-principles and inferential model to generate KPIs for the industrial process using uploaded process data as variables. The processor selects a subset of the KPIs to represent an event occurring in the industrial process, and divides the selected data into time series. The system and methods select time intervals from the time series based on data variability and perform a cross-correlation between the loaded process variables and the selected time intervals, resulting in a cross-correlation score for each loaded process variable. Precursor candidates from the loaded process variables are selected based on the cross-correlation scores, and a strength of correlation score is obtained for each precursor candidate. The methods and system select root-cause variables from the selected precursor candidates based on the strength of correlation scores, and analyze the root-cause of the event.