Bridge Impact Detection Using Neural Network Signal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting impacts on bridges, such as those from vehicles exceeding clearance, often fail to distinguish between impact events and normal operating loads, leading to potential structural damage going unnoticed due to lack of timely inspection.
Innovation Solution
The use of artificial neural networks to process accelerometer signals and classify response signals as either impact or non-impact events by extracting features like response length, number of peaks, spectral energy, and dominant frequency, allowing for more accurate differentiation between vehicle strikes and regular bridge usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If accelerometer threshold systems are used to detect impacts, then significant bridge responses can be detected, but vehicle impacts cannot be distinguished from normal operating loads such as trains
Solution Approach 1:
The patent segments the acceleration signal into multiple frequency components using Fast Fourier Transform (FFT), analyzing different frequency bands separately. This allows the system to identify characteristic frequency signatures that distinguish impact events from normal train operations, resolving the inability to differentiate between the two types of events.
Solution Approach 2:
The system dynamically adjusts detection parameters and uses real-time signal processing to adapt to varying bridge conditions and traffic patterns. The neural network continuously learns from incoming data, adjusting its classification thresholds and parameters to maintain high precision in distinguishing impacts from normal loads under changing operational conditions.
2Ease of manufacture
If manual reporting and inspection methods are used, then infrastructure costs are reduced, but impact events may go unnoticed for hours or days
Solution Approach 1:
The system implements self-service through automated monitoring and classification algorithms that continuously analyze acceleration data without human intervention. The neural network automatically detects, classifies, and flags impact events, eliminating the need for manual review of camera footage or constant engineer monitoring, thus reducing both time loss and operational costs.
Solution Approach 2:
The patent replaces manual mechanical inspection methods with automated electronic sensor-based monitoring and computational analysis. Accelerometers continuously monitor bridge vibrations, and computational algorithms automatically process the data, substituting human labor with automated systems that provide continuous monitoring without additional time loss or significant cost increase.
3Measurement precision
If high threshold values are used for impact detection, then false alarms from normal loads are reduced, but lighter impacts such as scraping vehicles go unnoticed
Solution Approach 1:
The system transitions from analyzing only the time domain (single dimension) to analyzing both time and frequency domains (multiple dimensions). By applying FFT to decompose the signal into frequency components, the system can detect subtle impact patterns that occur in specific frequency bands, enabling detection of lighter impacts like scraping vehicles while maintaining low false alarm rates through multi-dimensional pattern recognition.
Data Source
AI summary
A method for classifying a response signal of acceleration data of a structure includes obtaining at least one signal feature of a response signal, inputting the at least one signal feature into an artificial neural network, and classifying, using the artificial neural network, the response signal as an impact event or a non-impact event. One or more signal features may be used, including a response length feature, a number of peaks feature, a spectral energy feature, a dominant frequency feature, a maximum response feature, a center of mass feature, a slope feature, an average peak power feature, a response symmetry feature, or combinations thereof. One or more artificial neural networks may be used. The artificial neural networks may be trained using different combinations of signal features.