Sensor Monitoring Thresholds for Control Signal Change Anomalies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale infrastructure systems like power plants and manufacturing apparatuses, monitoring numerous sensors to prevent failures and quickly recover from them is challenging due to the complexity and size of the systems, often leading to erroneous anomaly detections, especially when control signals change suddenly.
Innovation Solution
A monitoring apparatus that acquires measurement values and control signals, uses a machine learning model to generate prediction values, and adjusts anomaly detection thresholds based on control signal changes to minimize false positives by setting higher thresholds during periods of control signal changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of sensors is increased to monitor all parts of a large-scale system, then the monitoring coverage is improved, but the complexity of monitoring and the number of people required increases
Solution Approach 1:
The monitoring apparatus integrates multiple functions into a single system: it acquires sensor values from multiple sensors, generates prediction values using machine learning models, detects anomalies by comparing actual and predicted values, and determines system states. This multi-functional integration allows comprehensive monitoring coverage while reducing the need for separate monitoring systems for each sensor.
Solution Approach 2:
The patent introduces prediction values generated by machine learning models as an intermediary element. Instead of directly monitoring all sensor values, the system uses prediction values as a reference to compare against actual sensor readings, enabling anomaly detection without requiring direct human monitoring of every sensor.
2Speed
If anomaly detection is performed continuously on all sensor values, then detection speed is improved, but erroneous detections increase especially during control signal changes
Solution Approach 1:
The system performs preliminary actions by generating prediction values using machine learning models before actual anomaly detection. These prediction values serve as dynamic thresholds that anticipate normal system behavior during control signal changes, allowing the system to distinguish between expected variations and actual anomalies.
Solution Approach 2:
The system uses feedback by comparing actual sensor values against prediction values and using the results to determine system states. This feedback mechanism allows continuous refinement of anomaly detection accuracy while maintaining fast detection speed, as the system learns from the difference between predicted and actual values.
3Ease of operation
If fixed thresholds are used for anomaly detection, then the detection process is simple, but false alarms occur during normal control signal variations
Solution Approach 1:
The patent applies dynamics by replacing fixed thresholds with dynamic prediction values generated by machine learning models. These prediction values adapt to changing system conditions and control signals, maintaining detection process simplicity while significantly reducing false alarms during normal operational variations.
Solution Approach 2:
The system changes the parameter used for threshold comparison from fixed values to dynamic prediction values. This parameter change allows the anomaly detection threshold to automatically adjust based on current system state and control signals, reducing false alarms while keeping the detection logic simple.
Data Source
AI summary
According to one embodiment, a monitoring apparatus includes a processing circuit. The processing circuit is configured to generate second data including a prediction value of a second sensor correlated with a first sensor from first data including a measurement value of the first sensor of which a measurement value changes suddenly in a case where the control signal changes, detect an anomaly of the system or an anomaly of at least one sensor, and make it difficult to detect the anomaly in a case where the determination signal indicates that there is a change in the control signal.