Disturbance Estimation With Explainable ML for Water-Intensive Processes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex industrial processes like papermaking and water treatment generate vast amounts of data, making it difficult for existing systems to accurately interpret and address disturbances using machine learning algorithms, leading to unreliable recommendations for improving process performance.
Innovation Solution
A method and arrangement that involves measuring and pre-processing data, estimating disturbances by normalizing and scaling pre-selected variables, and forming recommendations by mapping these estimations to status categories, using a combination of machine learning values and explanation values to provide actionable insights for adjusting setpoints and raw materials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used to analyze vast amounts of process data, then the ability to detect disturbances is improved, but the interpretability and reliability of the results deteriorate
Solution Approach 1:
The patent introduces explanation values (such as SHAP values, LIME method, or DeepLIFT method) as an intermediary between the machine learning algorithm outcomes and the user. These explanation values translate the complex, hard-to-interpret ML outputs into understandable information about how different input variables contribute to disturbance detection, thereby maintaining both high detection accuracy and interpretability
Solution Approach 2:
The patent segments the analysis by applying the ML algorithm separately to different groups of variables (pre-selected groups) rather than analyzing all variables simultaneously. This segmentation makes the complex multivariable analysis more manageable and interpretable while still capturing the essential disturbance patterns
2Reliability
If all process variables are analyzed simultaneously, then comprehensive disturbance detection is achieved, but the complexity of the system increases
Solution Approach 1:
The patent divides the set of all process variables into multiple pre-selected groups, where each group is associated with specific disturbance types or process areas. The ML algorithm is applied separately to each group, which reduces the computational complexity and makes the system more manageable while maintaining comprehensive coverage of all variables through the aggregation of results from different groups
Solution Approach 2:
The patent applies the ML algorithm to pre-selected groups of variables rather than all variables at once, using partial action on subsets of data. This approach reduces system complexity while still achieving reliable disturbance detection by focusing computational resources on the most relevant variable groups for each disturbance type
Data Source
AI summary
The invention provides a method for estimating disturbances and giving recommendations for process performance of a water intensive industrial process. The method takes into account a huge number of process variables.