Fault Detection Training Using RMS Error Threshold Expansion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional fault detection systems in process plants face challenges in accurately identifying abnormal operating conditions due to complex systems and unintuitive initial parameter inputs, leading to potential sequence stalls and equipment damage.
Innovation Solution
An improved training technique for fault detection systems using root-mean-squared (RMS) error threshold values and a maximum system matrix size to iteratively expand the system matrix with on-line process data, defining normal operating conditions and reducing operator confusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fault detection systems use complex system models and multiple parameters, then measurement precision may improve, but device complexity increases significantly
Solution Approach 1:
The patent extracts and isolates the critical fault detection function from the complex process control system. By using a separate MSET module that operates independently from the main control system, the invention achieves accurate fault detection without adding complexity to the core control architecture. The MSET is trained offline on normal operating data and then deployed as a standalone fault detection component.
Solution Approach 2:
The fault detection system is segmented into distinct functional components: an offline training phase that builds the MSET model, and an online detection phase that uses the trained model. This segmentation allows the complex model building to occur separately from the real-time control operations, maintaining system simplicity during critical operational phases while achieving high detection accuracy.
2Measurement precision
If traditional systems require multiple unintuitive parameter inputs for training, then measurement precision may improve, but ease of operation deteriorates
Solution Approach 1:
The MSET training system performs self-service by automatically selecting and weighting the most significant process variables based on their correlation with the target variable. The system calculates correlation coefficients and automatically determines the optimal set of input variables without requiring operator intervention or intuitive knowledge of which parameters to select. This eliminates the need for operators to understand complex parameter selection while maintaining high detection accuracy.
Solution Approach 2:
The patent transforms the training process from requiring multiple manual parameter inputs to using a single automated correlation-based selection method. By changing from a manual parameter-specification approach to an automated correlation analysis approach, the system achieves both high precision and ease of operation. The training data requirements are simplified from needing multiple tuned parameters to simply needing historical process data.
3Reliability
If fault detection systems use comprehensive monitoring of all process variables, then reliability may improve, but loss of information increases due to false alerts
Solution Approach 1:
The patent applies local quality by focusing monitoring resources on the most critical relationships between process variables. Instead of uniformly monitoring all variable combinations, the MSET identifies and monitors only the specific variable relationships that are most indicative of faults. This localized approach to monitoring improves reliability by concentrating on meaningful patterns while reducing false alerts from irrelevant variable combinations.
Solution Approach 2:
The system uses partial action by monitoring only the essential variable relationships needed for fault detection rather than all possible relationships. The MSET selectively tracks specific correlations between variables that are most relevant to detecting abnormal conditions, achieving reliable fault detection without the excessive monitoring that would generate false alerts from less relevant data.
Data Source
AI summary
A real-time control system includes a fault detection training technique to implement a data-driven fault detection function that provides an operator with information that enables a higher level of situational awareness of the current and likely future operating conditions of the process plant. The fault detection training technique enables an operator to recognize when a process plant component is behaving abnormally to potentially take action, in a current time step, to alleviate the underlying cause of the problem, thus reducing the likelihood of or preventing a stall of the process control system or a failure of the process plant component.