Fault Diagnosis via Residual Analysis and Causal Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault diagnosis systems for processes, equipment, and sensors are limited in their ability to detect and diagnose single and multiple faults efficiently, often requiring explicit equations and extensive manual calculation of parameters, and they do not effectively incorporate causal relationships between faults.
Innovation Solution
A device and method that utilize multiple models based on correlations between process variables, such as mass and energy balances, to calculate residuals and fault likelihoods, allowing for autonomous fault detection and diagnosis without requiring explicit parameter calculation, and incorporating causal relationships to determine the root cause of faults.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple models based on correlations between process variables are used to calculate residuals and fault likelihoods, then fault detection accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the fault detection task into multiple independent models, each handling specific correlations between process variables. Each model calculates residuals for particular variable relationships, allowing parallel processing and modular complexity management while maintaining high detection accuracy through comprehensive variable coverage.
Solution Approach 2:
The multiple models serve universal functions by collectively analyzing different aspects of process variable correlations. Each model can detect various fault types through residual calculation, and the ensemble of models provides comprehensive fault detection capability across the entire process system, making the system adaptable to diverse fault scenarios.
2Manufacturing precision
If explicit equations and extensive manual calculation of parameters are required, then modeling precision is improved, but ease of operation deteriorates
Solution Approach 1:
The system performs self-service by automatically calculating parameters and generating models without requiring extensive manual intervention. The automated parameter calculation and model generation processes maintain high modeling precision while eliminating the operational burden of manual calculations, allowing the system to serve itself in model development and maintenance.
Solution Approach 2:
Manual mechanical calculation processes are replaced with automated computational systems. The system uses algorithmic approaches to perform parameter calculations and model generation, substituting human-operated mechanical computation with automated electronic processing that maintains precision while dramatically improving ease of operation.
3Measurement precision
If causal relationships between faults are incorporated, then diagnosis accuracy is improved, but device complexity increases
Solution Approach 1:
The system adds another dimension to fault analysis by incorporating causal relationships between faults. This dimensional extension transforms the diagnosis from simple fault detection to causal reasoning, enabling the system to trace fault propagation paths and identify root causes through cause-consequence networks without requiring proportional increases in physical device complexity.
Solution Approach 2:
Cause-consequence networks serve as intermediaries between raw residual data and final diagnosis results. These networks mediate the analysis by organizing causal relationships and fault propagation patterns, allowing the system to achieve high diagnosis accuracy through structured reasoning without directly increasing the complexity of core detection mechanisms.
4Loss of information
If multiple models and causal networks are used, then fault diagnosis capability is improved, but loss of time in data processing increases
Solution Approach 1:
The system performs preliminary actions by pre-establishing cause-consequence networks and model relationships before actual fault detection occurs. This advance preparation of diagnostic frameworks and correlation structures enables rapid fault analysis when needed, as the computational heavy lifting of model development and causal relationship mapping is completed in advance rather than during active fault detection.
Solution Approach 2:
The system maintains continuous monitoring and evaluation across multiple models simultaneously, ensuring that useful diagnostic action continues uninterrupted. By processing residuals from multiple models in parallel and continuously updating fault likelihoods, the system preserves comprehensive diagnosis capability while minimizing processing delays through sustained parallel computation rather than sequential analysis.
Data Source
AI summary
Device and method for the detection and/or diagnosis of faults in a process, equipment and sensors used to measure and control variables of a process, either for single faults or multiple faults. The detection and/or diagnosis are performed on the basis of residual calculation between measured values and values calculated by a plurality of parallel linear models, built up from the existing correlations between the measured variables by the process sensors and by different equations that rule the process, such as mass and heat balances, quality relationships, etc. A fault in a model (i.e. the obtainment of a residual anomaly high by comparing the model estimation and the measurement), increases the probability that some of the parameters or faults associated to that model are failing. When all models in which a participating parameter fail or show a high probability of failure, the anomaly rate of said parameters is maximum. Using said rates and a description of cause-consequence relationship between the parameters a fault probability is obtained, where the root cause of the problem is indicated, indicating the consequences produced in the process. The device comprises data storage means; pre-processing means for filtering data; means for generation and storage of multiple behavior models; residual calculation means for calculating the difference between the measured and the predicted values of the variables; analysis means for determining the need to communicate an anomalous situation; communication means for presenting a process status report.