Sensor Frozen-Period Detection via Adaptive Run-Length Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting frozen periods in sensors rely on predetermined thresholds that are dependent on sampling periods and do not adapt to changes in sensor data, making them ineffective in dynamic environments.
Innovation Solution
A method that computes run-lengths for sensor datasets, clusters them based on run frequency, and identifies frozen periods within the cluster with lower run frequency, using K-means clustering and a minimum discount length to determine anomalous sequences without a preset threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a predetermined threshold is used for detecting frozen periods, then the detection method is simple, but the method becomes ineffective when sensor data changes or sampling periods vary
Solution Approach 1:
The patent applies dynamics by transitioning from a static predetermined threshold to a dynamic adaptive threshold that automatically adjusts based on the sensor data characteristics. The system computes the mean and standard deviation of the sensor data and uses these statistical parameters to dynamically determine the threshold for detecting frozen periods, enabling the detection method to adapt to varying sensor data and sampling periods.
Solution Approach 2:
The patent implements parameter changes by modifying the threshold parameter from a fixed predetermined value to a dynamically computed value based on statistical parameters (mean and standard deviation) of the sensor data. This allows the detection criterion to change according to the actual data characteristics, resolving the contradiction between simplicity and adaptability.
2Ease of operation
If a fixed threshold is set for anomaly detection, then the detection rule is clear and easy to implement, but it cannot accurately identify frozen periods when sensor data distribution changes
Solution Approach 1:
The patent applies feedback by using the statistical properties (mean and standard deviation) of the sensor data to continuously adjust the detection threshold. The system computes these statistical parameters from the actual data and feeds them back into the threshold calculation, creating a closed-loop system that maintains both operational clarity and measurement precision by adapting to data distribution changes.
3Measurement precision
If run-length encoding is applied to convert sensor data into symbols, then anomaly detection can be performed with almost no loss of accuracy, but the method still requires a predetermined threshold that limits its effectiveness
Solution Approach 1:
The patent combines run-length encoding with dynamic threshold computation by changing the threshold parameter from predetermined to data-driven. The system encodes sensor data into run-length symbols and then uses statistical parameters of the original data to dynamically set the threshold for detecting abnormal run lengths, thereby maintaining both high detection accuracy and adaptability to different sensor configurations.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A method 300of detecting at least one frozen period in at least one sensor dataset associated with at least one sensor 222A-222C in a technical system 220A-220C. The method comprises receiving the at least one sensor dataset in time series, computing run-lengths for the at least one sensor dataset, wherein each of the run-lengths is length of consecutive repetitions of a sensor value in the at least one sensor dataset. The method includes clustering the run-lengths into one of two clusters based on a run frequency, wherein the run frequency is a number of times the run-lengths are repeated in the at least one sensor dataset. Further, the method includes identifying a cluster from the two clusters with lower run frequency and detecting the at least one frozen period in the at least one sensor dataset based on the identified cluster.