Sensor Failure Prediction Using Discrete State Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current railway sensor systems, such as those used in the European Train Control System (ETCS), are prone to failures that are difficult to predict, leading to unplanned maintenance, delays, and increased costs, as they rely on post-failure diagnostic systems rather than proactive failure prediction.
Innovation Solution
A method using a machine learning system trained with supervised learning algorithms to analyze discrete conditional information from sensors over time, predicting failures by evaluating signal strength and state changes, allowing for preemptive maintenance and minimizing downtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If post-failure diagnostic systems are used to monitor sensor status, then the system can detect when a sensor has failed, but the system cannot predict failures before they occur, leading to unplanned maintenance and downtime
Solution Approach 1:
The patent applies preliminary action by training a machine learning model with historical sensor data to predict failures before they occur. The system analyzes patterns in discrete conditional information over time and generates early warnings, enabling maintenance to be scheduled in advance rather than responding to actual failures. This transforms the reactive diagnostic system into a proactive prediction system.
Solution Approach 2:
The patent segments the continuous sensor data into discrete conditional information at specific time points. By dividing the monitoring approach into discrete evaluation points and using these segmented data points as input for the machine learning model, the system can identify failure patterns without being overwhelmed by continuous data streams.
2Measurement precision
If complex radar sensor systems are deployed for speed measurements, then measurement accuracy is improved, but the complexity of the system increases and replacement costs rise when failures occur
Solution Approach 1:
The patent implements feedback by continuously monitoring discrete conditional information from sensors and feeding this data into a machine learning model. The model processes this feedback loop of information over time, learning from patterns in the data to predict when sensors will fail, thereby managing complex systems through intelligent feedback mechanisms rather than increasing hardware complexity.
3Reliability
If sensors are monitored continuously to detect failures, then failure detection capability is improved, but the cost of maintenance and system complexity increase
Solution Approach 1:
The patent applies partial action by evaluating sensor data at selected discrete time points rather than continuously monitoring every moment. The system determines discrete conditional information at specific intervals and uses these sampled points as input for the machine learning model, achieving effective failure prediction with reduced computational and operational overhead compared to continuous monitoring.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for predicting a failure (F) of at least one sensor (10, 12), especially, for measuring at least one parameter (P) of a mobile unit (14). To provide a reliable prediction and thus a safely operating system the method comprises at least the following steps: Providing as input data (D) for a machine learning system (16) discrete conditional information (I) of the at least one sensor (10, 12) at several time points (t1, t2), wherein these time points (t1, t2) are such time points at which the discrete conditional information (I) of the at least one sensor (10, 12) has changed and evaluating the input data (D) by using the machine learning system (16) and thus predicting the failure (F) of the at least one sensor (10, 12) by using the machine learning system (16).