Industrial Process Anomaly Detection Using Scaled Entropy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional entropy techniques in industrial process monitoring require prior knowledge of data scale, making it difficult to compare signal entropies across different time periods, especially when data values drift or change scale, leading to overwhelming data volumes and irrelevant information transmission over networks.
Innovation Solution
A system that uses scaled entropy calculations to detect anomalous operation in industrial processes by apportioning time-series data into sub-intervals, determining minimum and maximum values, scaling entropy based on the range, and performing statistical analysis to identify anomalies without prior knowledge of the signal scale, thereby generating reports for user devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional entropy techniques are used to monitor industrial processes, then anomaly detection capability is provided, but the technique requires prior knowledge of data scale which prevents comparison across different time periods when data drift occurs
Solution Approach 1:
The patent transforms the entropy calculation by introducing a scaling factor that normalizes entropy values based on observed data ranges. This parameter transformation allows entropy comparisons across different time periods even when data scales drift, converting the absolute entropy requirement into a relative scaled entropy metric that adapts to changing data conditions
2Loss of information
If all sensor data is transmitted over the network for analysis, then complete data availability is achieved, but network resources are unnecessarily consumed by irrelevant data
Solution Approach 1:
The patent extracts only the essential anomaly detection information from the full sensor data stream by computing scaled entropy metrics locally at edge devices or gateways. This extraction approach transmits only the computed entropy values and anomaly flags to central systems, filtering out the vast majority of raw sensor data that would otherwise consume network bandwidth while preserving the critical information needed for anomaly detection
Solution Approach 2:
The patent introduces scaled entropy computation as an intermediary processing layer between raw sensor data and central monitoring systems. This intermediary performs local data reduction by transforming high-volume raw measurements into compact entropy metrics, serving as a mediator that preserves anomaly information while dramatically reducing the data volume transmitted over the network
3Measurement precision
If thousands of sensors operate at high measurement frequencies for accurate monitoring, then measurement accuracy is improved, but the volume of data generated becomes overwhelming
Solution Approach 1:
The patent segments the high-frequency sensor data stream into smaller processing windows or batches, computing scaled entropy metrics for each segment rather than processing every individual data point in real-time. This segmentation allows the system to maintain high measurement accuracy by preserving the full-resolution data for entropy calculation while reducing the overall processing burden and data volume that needs to be managed and transmitted
Solution Approach 2:
The patent maintains continuous monitoring capability by continuously computing scaled entropy metrics from the incoming sensor data stream, ensuring that anomaly detection remains active and responsive without requiring the transmission or storage of every raw data point. This continuous transformation of data into compact metrics preserves the useful monitoring function while managing data volume
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Automated detection of anomalous operation of equipment in an industrial process. A reporting architecture utilizes scaled entropy calculations that enable comparing signal entropies across a plurality of time periods without prior knowledge of the scale of the signal. The reporting architecture combines the scaled entropy values with statistical analyses to detect anomalous time periods that represent anomalous operation of equipment in an industrial process. The reporting architecture generates reports of the anomalous operation for transmission to particular user devices via a communications network.