Streaming Decision Tree Updates for Industrial Automation Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing control systems using decision trees in industrial automation face inefficiencies due to the need for retraining the entire decision tree when anomalous outputs are detected, which can disrupt operations and consume resources.
Innovation Solution
The system modifies a portion of the decision tree associated with anomalous outputs, rather than retraining the entire tree, by adjusting thresholds, splitting nodes, or adding new nodes, allowing for incremental updates and improved efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire decision tree is retrained when anomalous outputs are detected, then the accuracy and reliability of the decision tree is improved, but the processing time and computational resources are significantly increased
Solution Approach 1:
The decision tree is divided into multiple nodes that can be independently identified and modified. When an anomalous output is detected, only the specific node(s) responsible for the anomaly are updated rather than retraining the entire tree, thus maintaining reliability while reducing retraining time and computational resources.
2Adaptability or versatility
If the entire decision tree is retrained when anomalous outputs are detected, then the decision tree adapts to new conditions, but the system downtime and operational disruption increase
Solution Approach 1:
By segmenting the decision tree into independent nodes, the system can perform localized updates to specific nodes without stopping the entire system. This allows the decision tree to adapt to new conditions while minimizing system downtime and maintaining operational efficiency.
Solution Approach 2:
The decision tree is designed to be dynamically modifiable at the node level during operation. Nodes can be added, removed, or modified without retraining the entire tree, enabling continuous adaptation while the system remains operational and productive.
3Reliability
If the entire decision tree is retrained when anomalous outputs are detected, then the decision tree remains accurate, but the computational power and processing requirements increase
Solution Approach 1:
The decision tree is segmented into independent nodes that can be updated individually. When an anomaly is detected, only the affected node(s) are retrained using minimal computational resources, rather than consuming significant processing power to retrain the entire tree, thus maintaining accuracy while reducing energy consumption.
Data Source
AI summary
A non-transitory computer-readable medium comprising computer-executable instructions that, when executed, are configured to cause a processor to perform operations that include receiving process input data associated with one or more automation devices, determining a plurality of operating conditions corresponding to the one or more automation devices; clustering the process input data based on the plurality of operating conditions, receiving a decision tree representative of the plurality of operating conditions and corresponding process input data, determining a splitting criterion for one or more nodes of the decision tree based on the clustered process input data, wherein the splitting criterion is configured to link the process input data to at least two operating conditions based on the decision tree, generating control logic for the one or more automation devices based on the decision tree and splitting criterion, and sending the control logic to the one or more automation devices.