Inline Fluid State Detection Using Neural Networks and Density Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In-line monitoring of fluid quality is challenging due to the difficulty in detecting immediate changes and assessing potential threats, especially in complex systems, as traditional methods require sampling and pretreatment, leading to low measurement frequency and accuracy issues.
Innovation Solution
A method using a neural network that creates a reference dataset from normal fluid states, projects in-line measurements into a reduced vector space, and employs a kernel density estimator to evaluate deviations from this reference, enabling rapid and adaptive monitoring of fluid quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If in-line measuring devices are used to directly detect changes in fluid quality, then measurement speed and frequency are improved, but the ability to reliably assess potential threats and detect contamination is worsened
Solution Approach 1:
The patent combines multiple in-line measuring devices that record different measured variables (electrical conductivity, temperature, pH value, etc.) into a unified monitoring system. By merging these measurements and evaluating them collectively through statistical methods, the system achieves both rapid detection capability and reliable threat assessment, resolving the contradiction between measurement speed and assessment reliability.
Solution Approach 2:
The patent transitions from evaluating single measured variables to multivariate analysis by examining multiple measured variables simultaneously across different dimensions. This dimensional expansion allows the system to detect contamination patterns that would be invisible in single-parameter monitoring, thereby improving threat assessment reliability while maintaining the speed advantages of in-line measurement.
2Reliability
If multiple measuring devices are used to record several in-line measured values for different measured variables, then the ability to assess fluid quality condition is improved, but the complexity of rapid evaluation despite large amount of information is worsened
Solution Approach 1:
The patent replaces complex manual evaluation of multiple measured variables with automated statistical methods and computer-based analysis. By substituting mechanical/manual assessment with electronic data processing and statistical algorithms, the system maintains high assessment accuracy while dramatically reducing evaluation complexity and enabling rapid automated decision-making.
Solution Approach 2:
The patent transforms multiple raw measured variables into statistical parameters and indicators that simplify complex data interpretation. By changing the form of data representation from raw measurements to statistical metrics (mean values, standard deviations, trend indicators), the system reduces evaluation complexity while preserving the ability to accurately assess fluid quality conditions.
3Measurement precision
If non-inline measuring devices with sampling and pretreatment are used, then detection limits and accuracy are improved, but measurement time and measurement frequency are worsened
Solution Approach 1:
The patent extracts and eliminates the time-consuming sampling and pretreatment steps from the measurement process by implementing direct in-line measurement. By removing these intermediate steps and measuring fluid quality parameters directly in the process stream, the system achieves rapid continuous monitoring while maintaining detection capability through sophisticated multivariate analysis of the streamlined measurement process.
Data Source
Figure 1
Figure 2
AI summary
Method for the automated in-line detection of deviations of an actual state of a fluid (1) from a reference state (RZ) of the fluid (1), wherein measured values (2) which are captured substantially at the same time are evaluated in a combined manner with respect to at least three measurement variables (MV1,MV2,MV3;...), wherein the measurement variables (MV1,MV2,MV3;...) are different measurement quantities (3) of the fluid (1) and/or a measurement quantity (3) of the fluid (1) which is measured at different measuring points (4), comprising at least the following method steps of: - creating a reference data set, wherein reference measured values (21) are mapped to a reference vector (rtj) of the vector space (VR) by means of a neural network: - in-line measurement, wherein measured values at a time ti and measured values at all times (t1,...,ti-1) preceding the time ti are mapped to a measurement vector xti of the vector space by means of a neural network; - comparing the measurement vector xti with the n reference vectors rtj by means of a kernel density estimator p h (xti) of a predefinable window width (h) (formula AA), wherein (formula BB) is a probability density function (PDF); creating an assessment for the time ti with respect to a deviation of the actual state from the reference state (RZ) on the basis of the value from the kernel density estimator p h (xti).