Computational Grid for Industrial Process Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for characterizing industrial processes struggle with reliability and stability, particularly in continuous processes, due to the lack of a reliable reference state and variability in conditions, leading to uncertain results and exclusion of potential solutions.
Innovation Solution
A method utilizing a computational grid to measure and process large volumes of data from industrial processes, identifying production windows through unsupervised calculation and convergence analysis, ensuring all potential configurations are considered without a priori conditions, and allowing for scalability and flexibility in processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If reference state comparison methods are used to diagnose industrial processes, then diagnosis can be made using expert rules, but the reliability is uncertain because reliable reference state data is not always available and the reference state may vary widely depending on conditions of use
Solution Approach 1:
The system performs preliminary unsupervised learning to automatically identify production windows and parameter combinations before diagnosis is needed. This pre-characterization of the process eliminates the need for manual reference state establishment, ensuring reliable diagnosis without requiring pre-defined reference conditions.
Solution Approach 2:
The system uses unsupervised learning algorithms that enable the process characterization to be self-performing without external reference data. The algorithm automatically discovers production windows and parameter relationships from process data itself, making the system self-sufficient and eliminating dependency on external reference states.
2Productivity
If clustering algorithms require parameter combinations to follow a determined path with successive eliminations, then convergence towards a solution is achieved, but potential solutions may be excluded and the approach is not suitable for continuous processes requiring high stability and uniformity
Solution Approach 1:
The system employs a dynamic, stochastic search approach that allows parameter combinations to explore the solution space flexibly rather than following a rigid predetermined path. This dynamic exploration ensures all potential solutions are considered while still achieving convergence through iterative refinement of production windows.
Solution Approach 2:
The system systematically varies and explores different parameter combinations within defined ranges to identify optimal production windows. By changing parameters stochastically and evaluating their impact on process outcomes, the system ensures comprehensive coverage of the solution space without excluding potential solutions.
3Reliability
If a computational grid is used to process large volumes of industrial process data, then reliable and stable identification of production windows is achieved, but the device complexity increases
Solution Approach 1:
The computational task is segmented and distributed across multiple processing units in a computational grid. Each unit independently processes portions of the large-volume industrial process data, enabling reliable identification of production windows through parallel computation while managing system complexity through modular architecture.
Data Source
AI summary
A device for diagnosing an evolutive industrial process comprise a plurality of technical steps for the production of a given industrial compound, for which a plurality of state or characterization technical data is available, the device including at least one microprocessor and one memory, an interface module, a counter module for verifying whether a calculation end criterion has been reached, a management module for managing operations and data exchanges between the different modules, a computational grid comprising a plurality of production window calculation modules in order to determine appropriate production windows, a clustering module for distributing parameter combinations between several clusters in order to perform convergence calculations, and a convergence module for verifying whether an expected convergence rate between clusters has been reached.