Root-Cause Indicators for Interpreting Industrial ML Anomalies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for anomaly detection in industrial systems are complex 'black-box' models that are difficult to interpret, making it challenging to understand the physical root-causes of anomalies, necessitating labor-intensive manual investigations and lacking human-understandable explanations.
Innovation Solution
A method and apparatus that utilize explainable AI to determine root-cause indicators by training a root-cause model on feature values derived from machine learning model outputs, allowing for automated and efficient identification of physical causes of anomalies using domain expert labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex machine learning models are used for anomaly detection, then detection accuracy is improved, but interpretability deteriorates
Solution Approach 1:
The patent introduces an intermediary explanation model that translates complex black-box model outputs into human-understandable root cause indicators. This mediator layer processes the opaque predictions and generates interpretable explanations about physical system states, thereby maintaining detection accuracy while recovering interpretability.
Solution Approach 2:
The system segments the anomaly detection process into distinct components: the black-box detection model, an explanation generation module, and a root cause classification system. This segmentation allows each component to specialize - the first in accurate detection, the others in interpretability - resolving the contradiction between accuracy and understandability.
2Loss of information
If manual root cause investigation is performed, then accurate physical understanding is achieved, but time consumption increases
Solution Approach 1:
The system enables self-service by automatically generating root cause explanations without requiring manual expert investigation. The explanation model autonomously processes anomaly detections and produces human-understandable root cause indicators, eliminating the need for time-consuming manual analysis while maintaining accurate physical understanding.
Solution Approach 2:
The system performs preliminary action by pre-computing and storing explanation models during training phases. When anomalies occur in production, these pre-prepared explanation mechanisms can immediately generate root cause indicators without requiring real-time manual intervention, significantly reducing response time.
3Loss of information
If domain expert knowledge is integrated into the model, then explanation quality improves, but model complexity increases
Solution Approach 1:
The patent incorporates domain expert knowledge by transforming it into structured parameters and features that the explanation model can process. Rather than directly embedding complex expert systems, the knowledge is converted into measurable parameters that guide the explanation generation, improving quality while controlling complexity through parameterization.
Data Source
AI summary
A computer-implemented method is provided for automatically determining a root-cause indicator of a technical system concerning an output of a machine learning model trained to analyze the technical system based on sensor data of different sensors measuring parameters of the technical system, including receiving measured sensor data of the technical system and an output of the machine learning model for the received measured sensor data as input, determining an explanation information for the output of the machine learning model, training a root-cause model based on root-cause labels assigned to the output of the machine learning model to map the set of feature values of the explanation information onto a root-causes indicator, determining one of the root-cause labels by inputting a new set of feature value determined for an output of the machine learning method, and outputting the determined root-cause label as root-cause indicator to a user interface.