Deterministic ML Root Cause Analysis for Fault Signal Traceability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Regression-type machine learning models, particularly stochastic models like neural networks and support vector machines, struggle to trace output alarms back to signal deviations causing them, making it difficult to identify root causes of anomalies, especially in safety-critical applications due to their probabilistic nature and inability to perform rigorous propagation-of-uncertainty analyses.
Innovation Solution
Employ a deterministic machine learning model, such as a multivariate state estimation technique (MSET) combined with a sequential probability ratio test (SPRT), and generate an abductive model to reverse-engineer the model and identify the root cause of operational results by analyzing the influence of input data sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If stochastic machine learning models (neural networks, support vector machines) are used for regression-type predictions, then the model can handle complex non-linear relationships and provide probabilistic predictions, but the model cannot perform rigorous propagation-of-uncertainty analyses to trace output alarms back to input signal deviations
Solution Approach 1:
The patent inverts the conventional approach by using a deterministic model that can be mathematically reversed. Instead of trying to trace through the probabilistic stochastic model, the invention creates an inverse deterministic model that takes output alarms and systematically works backward to identify the specific input signal deviations that caused them, enabling rigorous propagation-of-uncertainty analyses
2Ease of operation
If black box approach is used to estimate relationships between input and output uncertainties, then the analysis is simpler to perform, but the results cannot be proven through analytical propagation-of-uncertainty analyses and may produce false alarms or missed alarms
Solution Approach 1:
The patent introduces a deterministic model as an intermediary between the input signals and output predictions. This intermediary model has the property that it can be mathematically inverted, serving as a bridge that allows rigorous analytical propagation-of-uncertainty analyses while maintaining the ability to handle complex non-linear relationships
3Productivity
If stochastic optimization of weights is used in neural networks and support vector machines, then the model can be trained on large datasets with multiple epochs, but the same input signals produce slightly different output signals each time, making root cause analysis impossible
Solution Approach 1:
The patent changes the fundamental parameter of the model from probabilistic to deterministic. By using a deterministic model that produces identical output signals for the same input signals every time, the invention enables consistent root cause analysis while maintaining training efficiency through deterministic optimization methods
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques for identifying a root cause of an operational result of a deterministic machine learning model are disclosed. A system applies a deterministic machine learning model to a set of data to generate an operational result, such as a prediction of a "fault" or "no-fault" in the system. The set of data includes signals from multiple different data sources, such as sensors. The system applies an abductive model, generated based on the deterministic machine learning model, to the operational result. The abductive model identifies a particular set of data sources that is associated with the root cause of the operational result. The system generates a human-understandable explanation for the operational result based on the identified root cause.