Deterministic Machine Learning for Sensor Alarm Root Causes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional regression-type machine learning models, particularly stochastic models like neural networks and support vector machines, struggle to trace output alarms back to signal deviations that caused the alarms due to their probabilistic nature, making it difficult to identify root causes of anomalies, especially in safety-critical applications.
Innovation Solution
Employ a deterministic machine learning model, such as a multi-variate state estimation technique (MSET) combined with a sequential probability ratio test (SPRT) model, to generate an abductive model that reverses the deterministic ML model, identifying root causes by analyzing the influence of input sources on output results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stochastic machine learning models (neural networks, support vector machines) are used for regression-type predictions, then prediction capability is improved, but the ability to trace output alarms back to root cause signal deviations deteriorates due to probabilistic methods
Solution Approach 1:
The patent inverts the conventional approach by using a deterministic ML model trained on normal operation data to predict expected values, then comparing actual values against these predictions. This reversal enables precise identification of which input signals deviated from normal ranges, allowing root cause tracing while maintaining prediction accuracy through the deterministic nature of the model
2Adaptability or versatility
If stochastic optimization of weights is applied in neural networks and support vector machines, then model training flexibility is improved, but rigorous propagation-of-uncertainty analysis for root cause identification becomes impossible
Solution Approach 1:
The patent replaces the stochastic optimization mechanism (random weight initialization and probabilistic training) with a deterministic approach where weights are fixed based on normal operation data, and predictions are made by comparing actual values against these fixed expectations. This substitution maintains training flexibility through the learning phase but ensures analysis rigor during operation through deterministic, repeatable predictions
3Measurement precision
If human experts manually identify root causes in systems with many monitored assets, then analysis thoroughness is improved, but response time and productivity deteriorate due to the large number of permutations to investigate
Solution Approach 1:
The patent introduces a deterministic ML model as an intermediary between the complex system of monitored assets and human experts. The model automatically compares actual sensor values against predicted normal values, identifies deviations, and traces them back to specific input signals, thereby performing the tedious permutation analysis that would otherwise require manual human investigation of numerous asset combinations
4Ease of manufacture
If black box approaches are used to estimate relationships between input and output uncertainties, then implementation simplicity is improved, but safety-critical application reliability deteriorates due to inability to prove absence of false or missed alarms
Solution Approach 1:
The patent replaces the black box stochastic approach with a deterministic model that produces repeatable, explainable predictions. By training on normal operation data and comparing actual values against fixed expectations, the model provides clear evidence of deviations and their sources, enabling safety-critical applications to verify the absence of false or missed alarms through deterministic reasoning rather than probabilistic estimates
Data Source
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.