Track Circuit Anomaly Detection Using SVM Signal Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting anomalous behavior in track circuit devices, such as Grade Crossing Predictors, are manual and time-consuming, relying on expert feedback to identify abnormal impedance and raw data signals, which is inefficient.
Innovation Solution
A machine learning-based approach using a support vector machine (SVM) algorithm that extracts features like the number of signal values greater than a threshold and standard deviation from labeled time series data to automatically detect anomalies in track circuit signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual visual inspection of IPIZ and IMAG signals is used for anomaly detection, then detection accuracy based on expert feedback is achieved, but time consumption and operational efficiency deteriorate significantly
Solution Approach 1:
The patent replaces the manual mechanical inspection process with an automated machine learning system. The SVM algorithm automatically processes IPIZ and IMAG signals to detect anomalies, substituting human expert visual inspection with computational analysis that maintains detection accuracy while eliminating time consumption associated with manual review
Solution Approach 2:
The system enables self-service anomaly detection by training the machine learning model to independently identify anomalies without requiring expert intervention. The SVM algorithm autonomously analyzes signal patterns and classifies anomalies based on learned features from training data, making the detection process self-sufficient
2Reliability
If manual expert inspection is used for each signal figure, then detailed anomaly identification is achieved, but operational complexity and resource requirements increase
Solution Approach 1:
The patent segments the anomaly detection task into distinct automated components: feature extraction from signals, SVM model training, and anomaly classification. This segmentation transforms the complex manual inspection process into manageable automated stages, reducing operational complexity while maintaining reliability through systematic processing
Solution Approach 2:
The machine learning model serves as an intermediary between raw signal data and anomaly identification. The SVM algorithm mediates the analysis by automatically extracting features and making classifications, eliminating the need for direct human expert involvement while preserving detection reliability
3Productivity
If automated machine learning methods are implemented for anomaly detection, then time efficiency and productivity are improved, but system complexity and implementation difficulty increase
Solution Approach 1:
The patent applies preliminary action by pre-training the SVM model with labeled training data before deployment. This preliminary training phase establishes the anomaly detection capabilities in advance, allowing the system to achieve high productivity during operation without requiring complex real-time decision-making logic
Solution Approach 2:
The system manages complexity through parameter changes by adjusting SVM hyperparameters and feature extraction parameters during training. This allows the model to adapt to different signal characteristics and anomaly types, maintaining high productivity across varying operational conditions without requiring fundamentally different system architectures
Data Source
AI summary
Examples of techniques for anomaly detection in a grade crossing prediction system are disclosed. Aspects include receiving a training data set comprising a plurality of labelled time series of signal values from a track circuit in a grade crossing predictor system, removing one or more non-unique values from each labelled time series of signal values in the plurality of labelled time series of signal values, extracting a plurality of features from the plurality of labelled time series of signal values, the plurality of features comprising: a number of signal values for each labeled time series of signal values in the plurality of labelled time series of signal values that are larger than a first threshold and a standard deviation for each labelled time series of signal values in the plurality of labelled time series of signal values, and training a machine learning algorithm utilizing the plurality of features.