Industrial Plant State Correction via Explainable Decision Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing industrial plant monitoring and control systems lack effective methods for real-time feedback and correction of processing operations to ensure compliance with quality standards and prevent component failures.
Innovation Solution
A processing system that utilizes machine learning to estimate the current state of an industrial plant, trains a linear classifier to determine the required adjustments, and employs XAI to provide understandable corrections based on monitored variables, ensuring compliance with quality standards and reducing the risk of component failure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If machine learning models are used to estimate plant state and generate control recommendations, then manufacturing precision and reliability are improved, but device complexity increases
Solution Approach 1:
The patent introduces an explainable AI intermediary layer that translates complex machine learning model outputs into understandable natural language explanations. This mediator bridges the gap between the complex ML system and human operators, maintaining high manufacturing precision while reducing the perceived complexity for users through interpretable feedback about quality deviations and control recommendations.
Solution Approach 2:
The patent replaces traditional mechanical control systems with machine learning-based predictive models that analyze sensor data and generate control recommendations. This substitution enables more precise quality monitoring and prediction of equipment failures, improving manufacturing precision while the system manages complexity through automated data processing and intelligent algorithms.
2Reliability
If real-time monitoring and control adjustments are implemented, then reliability is improved, but loss of time for processing increases
Solution Approach 1:
The patent implements preliminary action by using machine learning models to predict equipment failures and quality deviations before they occur. The system analyzes current sensor data and forecasts future states, enabling preventive control adjustments to be made in advance. This approach improves reliability by preventing failures while minimizing processing time losses through proactive rather than reactive interventions.
Solution Approach 2:
The patent establishes continuous feedback loops where sensor data is constantly monitored, analyzed by ML models, and used to generate real-time control recommendations. This feedback mechanism enables the system to detect deviations from optimal operation and automatically suggest corrective actions, improving reliability through continuous monitoring while maintaining efficient processing through automated real-time responses.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Solutions are described for monitoring and controlling the state of operation of an industrial plant (1a). For this purpose, a processing system (40a, 60) obtains, for a current operating condition (CA) of the industrial plant (1a), respective values (300) of a plurality of operating variables (p) of the industrial plant (1a) and estimates (1402), by means of a first classifier, a current state class (AC) of the industrial plant (1a) for the values (300) of the current operating condition (CA). In the case where (1406) the current state class (AC) does not correspond to a requested state class (TC), the processing system (40a, 60) obtains a dataset (400; 202) that comprises, for each operating condition of a plurality of operating conditions that can be implemented by the industrial plant (1a), respective values of the operating variables (p) and a respective state class (v). Next, the processing system (40a, 60) generates (1408) a training dataset (406) as a function of the values (300) of the current operating condition (CA) and of the operating conditions that can be implemented by the industrial plant (1a), and trains (1410) a second classifier configured to estimating the state class (v) of the industrial plant (1a) as a function of the values of the operating variables (p) using the training dataset (406). In particular, the second classifier is a linear classifier, and the processing system (40a, 60) determines a separation plane (502) of the linear classifier that separates the current state class (AC) from the requested state class (TC), and uses (1412) the separation plane (502) to determine the values (600, 600') of the operating variables (p) for an operating condition that has the requested state class (TC). Finally, the processing system (40a, 60) displays (1418) data (MP; 600, 600') that identify the values (600, 600') of the operating condition that has the requested state class (TC) on a screen and/or controls operation of the industrial plant (1a) as a function of the values (600, 600') of the operating condition that has the requested state class (TC).