Sensor Signature Control for Unexpected Industrial Operations
Solution Overview
Problem
Existing control systems for industrial automation processes face inefficiencies in training and retraining model predictive control systems, leading to suboptimal use of computing resources and operational performance.
Innovation Solution
The system receives processed sensor data from multiple sources, identifies signatures indicating unexpected operations, and performs root cause analysis to determine relationships between datasets, allowing for real-time adjustments and model updates to optimize industrial automation processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If model predictive control systems are trained and retrained in real-time or near real-time, then operational performance and optimization are improved, but computing resource usage (energy, processing power, storage) increases
Solution Approach 1:
The patent segments the control system into multiple components: a process model for predictions, a sensor data processing system for feature extraction, and an anomaly detection system. This segmentation allows each component to operate independently with optimized resource allocation, enabling real-time performance without requiring full system retraining continuously.
Solution Approach 2:
The system performs preliminary actions by pre-training the process model offline and pre-processing sensor data to extract relevant features before real-time operation. This preliminary preparation reduces the computational burden during real-time execution, allowing fast anomaly detection without intensive real-time model retraining.
2Reliability
If full model retraining is performed frequently, then model accuracy and adaptability are improved, but time consumption and system downtime increase
Solution Approach 1:
Instead of performing full model retraining, the system applies partial updates by detecting anomalies in sensor data and retraining only specific components of the process model when needed. This partial action maintains model accuracy while significantly reducing retraining time and computational overhead.
Solution Approach 2:
The system implements continuous feedback loops where sensor data is monitored, anomalies are detected, and model updates are triggered only when performance degradation is detected. This feedback mechanism ensures model accuracy is maintained through targeted updates rather than frequent full retraining cycles.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A non-transitory computer-readable medium comprising computer-executable instructions that, when executed, cause a processor to perform operations including receiving a first dataset from a first automation component, the first dataset corresponds to raw data acquired by a first sensor; receiving a second dataset from a second automation component, the second dataset corresponds to raw data acquired by a second sensor; receiving data indicating an expected operation related to operations of an industrial automation system including the first and second automation components; determining a signature based on the first and second datasets and the data indicating the expected operation, wherein the signature indicates an unexpected operation as compared to the expected operation; performing a root cause analysis using the signature to determine a relationship indicating a first set of changes of the first dataset corresponding to a second set of changes in the data indicating the expected operation.
