Axle Counter Signal Evaluation Using Single-Stage Deep Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional axle counters suffer from inaccurate signal processing due to complex signals, interference, and varying speeds, leading to false positives and negatives, which hinder efficient rail transport and automation.

Innovation Solution

A single-stage deep learning-based algorithm, such as YOLO or SSD, is applied to evaluate axle counter data by simultaneously detecting and classifying signals in a one-dimensional time history, using techniques like convolutional neural networks and anchor regions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional axle counters use simple threshold-based signal processing, then the system is fast and simple to implement, but counting accuracy deteriorates due to false positives and false negatives from complex signals and interference

Engineering Contradiction:
Improvecounting accuracyVSAvoidsignal processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces conventional mechanical/threshold-based signal processing with a deep learning-based neural network system. The neural network automatically learns complex signal patterns from training data, enabling accurate distinction between wheel signals, magnetic track brake signals, and noise without requiring manual threshold configuration or complex rule-based processing systems.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the signal processing approach by changing from fixed threshold parameters to adaptive neural network parameters. The neural network dynamically adjusts its internal parameters during training to optimize signal classification, allowing it to adapt to varying signal characteristics from different train types, speeds, and conditions without requiring reconfiguration of processing rules.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If axle counters use conservative counting to ensure safety, then safety and reliability are improved, but productivity deteriorates due to false positives causing track sections to be incorrectly marked as occupied

Engineering Contradiction:
ImprovesafetyVSAvoidtrain throughput
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent implements feedback mechanisms where the neural network continuously learns from labeled training data to improve its classification accuracy. The system provides feedback on signal identification accuracy, allowing it to refine its decision-making process and reduce false positives while maintaining safety standards, thereby enabling more accurate track occupancy detection that improves train throughput.

Inventive Principle:
Principle #23Feedback

3Measurement precision

If axle counters process signals in real-time using simple methods, then processing speed is maintained, but measurement precision deteriorates due to inability to handle complex signals from varying speeds and interference

Engineering Contradiction:
Improvesignal detection precisionVSAvoidsignal processing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-training the neural network on extensive labeled data representing various signal conditions before actual operation. This pre-training enables the system to quickly and accurately classify new signals during real-time operation without requiring complex processing, as the neural network has already learned the patterns and can make rapid decisions based on its trained knowledge.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4653287A1One-step evaluation of axle counter time curve
Publication Date: 2025.11.26 SIEMENS MOBILITY GMBH
  • EP4653287A1 patent drawingFigure 1~3
  • EP4653287A1 patent drawingFigure 4~5
  • EP4653287A1 patent drawingFigure 6~7

AI summary

A method for evaluating a data set (110) is described, comprising: i) recording a one-dimensional time series as a data set (110) using an axis counter (100); and ii) evaluating the data set (110) by identifying signals in the one-dimensional time series and classifying the signals in the one-dimensional time series in one step using a one-stage algorithm (150) based on deep learning.