Process Plant Component Monitoring With Graph-Based Sensor Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large and complex process plants, training machine learning algorithms for condition monitoring is complicated by overlapping and noisy sensor data, requiring significant manual effort and domain-specific knowledge, which is often not feasible.
Innovation Solution
Generate a graph from the digital flow diagram of the process plant, specifying relevant input nodes for the target node based on the system's structure, using metrics or rules to select sensor data sets for training, thereby automating the selection of relevant data for improved algorithm performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual selection of sensor datasets is performed to improve training quality, then the quality of training is improved, but the workload and time required increase significantly
Solution Approach 1:
The system performs self-service by automatically selecting relevant sensor datasets through graph analysis of the flow diagram. The method autonomously identifies input nodes and their associated sensors without requiring manual domain expertise, allowing the system to serve itself in the data selection process while maintaining high training quality.
Solution Approach 2:
The invention changes the parameter of data selection from manual expert-based selection to automated algorithm-based selection. By transforming the selection criterion from domain knowledge to graph structure analysis, the system achieves both automation and consistent quality in training data selection.
2Quantity of substance
If all sensor data is used for training to ensure comprehensive coverage, then data completeness is improved, but noise and irrelevant information increase
Solution Approach 1:
The method extracts only the relevant sensor datasets needed for training by analyzing the flow diagram structure. It identifies and extracts data from input nodes that have direct functional relationships with the target component, while excluding irrelevant sensors and noisy data from other parts of the system.
Solution Approach 2:
The system segments the overall sensor network into relevant and irrelevant portions based on the flow diagram structure. By dividing the sensor dataset into component-specific subsets, it maintains data completeness for the target component while eliminating noise from unrelated sensors.
3Measurement precision
If domain-specific knowledge is required for data selection to ensure accuracy, then selection accuracy is improved, but the complexity and resource requirements increase
Solution Approach 1:
The flow diagram serves as an intermediary that bridges domain knowledge and automated selection. Instead of requiring experts to directly select sensors, the method uses the pre-existing flow diagram (which encodes domain knowledge) as a mediator to automatically identify relevant data sources through graph analysis.
Solution Approach 2:
The invention replaces the mechanical process of expert-based manual selection with an automated computational system. By substituting human domain expertise with algorithmic analysis of the flow diagram structure, it reduces resource requirements while maintaining selection accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method and a corresponding device for monitoring the state of a component of a process plant comprising a plurality of process-related interconnected components. The state of a component is determined based on process variables from at least one sensor relevant to the component, and corresponding data sets are recorded and stored for each sensor. According to the invention, a graph is generated from a digital flow diagram of the plant, which contains the structure of the plant with its components and their functions as additional information and their functional relationships. In this graph, the components of the plant are represented as nodes, and the functional relationships between the components are represented as lines of action according to a flow direction defined in the flow diagram, based on the additional information and relationships specified in the flow diagram.In the graph, a node is selected as the target node. The target node corresponds to a component whose state is to be monitored using a machine learning (ML) algorithm. Based on a rule and/or metric, input nodes of the previously selected target node are then identified, and sensor data sets from the input nodes that are relevant for monitoring the component being monitored, as well as the data sets from the relevant sensors of the input nodes, are selected to train the ML algorithm of the target node. Using this trained ML algorithm, the state determination of the component represented by the previously selected target node is significantly improved.