Bridge impact detection and classification systems and methods
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.
1. A method for classifying a response signal of acceleration data of a structure, the method comprising:
obtaining a plurality of signal features of a response signal, the response signal comprising acceleration data of a structure, wherein the plurality of signal features are selected from the group consisting of a response length feature, a spectral energy feature, a number of peaks feature, a dominant frequency feature, a maximum response feature, a center of mass feature, a slope feature, an average peak power feature, and a response symmetry feature;
inputting a first set of signal features of the plurality of signal figures into a first artificial neural network;
classifying, using the first artificial neural network, the response signal as an impact event or a non-impact event;
inputting a second set of signal features of the plurality of signal figures of the response signal into a second artificial neural network, the second set of signal features being different from the first set of signal features; and
classifying, using the second artificial neural network, the response signal as an impact event or a non-impact event.
2. The method of claim 1 , further comprising training the first artificial neural network and the second artificial neural network using a set of training events, wherein the first artificial neural network is trained using a first subset of signal features of the set of training events, the first subset of signal features being the same features as the first set of signal features, and the second artificial neural network is trained using a second subset of signal features of the set of training events, the second subset of signal features being the same features as the second set of signal features.
3. The method of claim 1 , further comprising weighing the classification of the first artificial neural network and the classification of the second artificial neural network to determine a final classification of the response signal as an impact event or a non-impact event.
4. The method of claim 1 , wherein the first set of signal features is input into and classified using a first cluster of neural networks including the first artificial neural network, and the second set of signal features is input into and classified using a second cluster of neural networks including the second artificial neural network.
5. A method for classifying a response signal of acceleration data of a structure, the method comprising:
obtaining at least one signal feature of a response signal, the response signal comprising acceleration data of a structure;
inputting the at least one signal feature into an artificial neural network;
classifying, using the artificial neural network, the response signal as an impact event or a non-impact event;
inputting the at least one signal feature into at least one additional artificial neural network; and
classifying, using each of the at least one additional artificial neural network, the response signal as an impact event or a non-impact event.
6. The method of claim 5 , further comprising weighing the classifications of the artificial neural network and the at least one additional artificial neural network to determine a final classification of the response signal as an impact event or a non-impact event.
7. The method of claim 5 , wherein the at least one additional artificial neural network is a plurality of additional artificial neural networks, each of the plurality of additional artificial neural networks classifying the response signal as an impact event or a non-impact event.
8. The method of claim 5 , wherein the response signal includes a frequency domain and a time domain, the frequency domain being a Fourier transform of the time domain, the time domain including a plurality of peaks of the acceleration data, the frequency domain including a plurality of peaks of frequency magnitudes, and the at least one signal feature is selected from the group consisting of a response length feature, a spectral energy feature, a number of peaks feature, a dominant frequency feature, a maximum response feature, a center of mass feature, a slope feature, an average peak power feature, and a response symmetry feature, and wherein
the response length feature is a length of the response signal in the time domain,
the spectral energy feature is a sum of the squared fast Fourier transform coefficients in the frequency domain,
the number of peaks feature is a number of peaks within the frequency domain,
the dominant frequency feature is a largest magnitude of the plurality of peaks in the frequency domain,
the maximum response feature is a largest magnitude of the plurality of peaks in the time domain,
the center of mass feature is a summation of the products of the magnitudes of the response signal in the time domain with its corresponding time divided by a summation of the magnitudes of the response signal in the time domain,
the slope feature is a maximum slope between the magnitude of a first peak and the magnitudes of the subsequent peaks in the time domain,
the average peak power feature is an average magnitude of the plurality of peaks in the frequency domain, and
the response symmetry feature is a ratio of an average magnitude of the plurality of peaks in a first half of the length of the time domain to an average magnitude of the plurality of peaks in a second half of the length of the time domain.
9. The method of claim 8 , wherein the at least one signal feature selected from the group includes at least one of the number of peaks feature, the spectral energy feature, or the center of mass feature.
10. The method of claim 8 , wherein the at least one signal feature is a plurality of signal features.
11. A method for training an artificial neural network to distinguish between an impact event and a non-impact event, the method comprising:
obtaining signal data for a plurality of event signals, the plurality of event signals including a plurality of training events and a plurality of validation events, the event signals comprising acceleration data, the signal data for each event signal comprising an event signature of either an impact signature or a non-impact signature, the signal data for each event signal further comprising at least one signal feature selected from the group consisting of 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, and a response symmetry feature; and
training an artificial neural network by:
inputting the signal data of the training events of the plurality of event signals into a neural network;
obtaining a predicted training signature from the artificial neural network for each of the training events;
comparing the predicted training signatures with the event signatures to determine an error level; and
adjusting iteratively a plurality of internal weights of the artificial neural network until the error level is below a selected threshold.
12. The method of claim 11 , further comprising validating the artificial neural network by:
inputting the signal data of the validation events of the plurality of event signals into the artificial neural network;
obtaining a predicted validation signature from the artificial neural network for each of the validation events; and
comparing the predicted validation signatures with the event signatures to determine a performance level.
13. The method of claim 12 , wherein the at least one signal feature is a plurality of signal features.
14. The method of claim 12 , wherein the artificial neural network is a first neural network, and further comprising training and validating a cluster of neural networks using the at least one signal feature, each of the artificial neural networks having a different initial internal weight, the cluster including the first neural network.
15. The method of claim 12 , wherein the artificial neural network is a first neural network, and further comprising training and validating a plurality of neural networks including the first neural network, wherein each of the plurality of neural networks is trained and validated using different subsets of signal features.
16. A system for classifying a response signal of acceleration data of a structure, the system comprising:
at least one processor configured to analyze acceleration data of a structure and classify the acceleration data, the at least one processor configured to perform the steps of:
receiving at least one signal feature of a response signal, the response signal comprising the acceleration data of the structure, the response signal including a frequency domain and a time domain, the frequency domain being a Fourier transform of the time domain, the time domain including a plurality of peaks of the acceleration data, the frequency domain including a plurality of peaks of frequency magnitudes;
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,
wherein the at least one signal feature is selected from the group consisting of a response length feature, a spectral energy feature, a number of peaks feature, a dominant frequency feature, a maximum response feature, a center of mass feature, a slope feature, an average peak power feature, and a response symmetry feature, and wherein
the response length feature is a length of the response signal in the time domain,
the spectral energy feature is a sum of the squared fast Fourier transform coefficients in the frequency domain,
the number of peaks feature is a number of peaks within the frequency domain,
the dominant frequency feature is a largest magnitude of the plurality of peaks in the frequency domain,
the maximum response feature is a largest magnitude of the plurality of peaks in the time domain,
the center of mass feature is a summation of the products of the magnitudes of the response signal in the time domain with its corresponding time divided by a summation of the magnitudes of the response signal in the time domain,
the slope feature is a maximum slope between the magnitude of a first peak and the magnitudes of the subsequent peaks in the time domain,
the average peak power feature is an average magnitude of the plurality of peaks in the frequency domain, and
the response symmetry feature is a ratio of an average magnitude of the plurality of peaks in a first half of the length of the time domain to an average magnitude of the plurality of peaks in a second half of the length of the time domain.
17. The system of claim 16 , wherein the at least one signal feature is a plurality of signal features.
18. The system of claim 16 , wherein the at least one signal feature selected from the group includes the number of peaks feature.
19. The system of claim 16 , wherein the at least one signal feature selected from the group includes the spectral energy feature.
20. The system of claim 16 , wherein the at least one signal feature selected from the group includes the center of mass feature.
21. The system of claim 16 , wherein the at least one signal feature selected from the group includes the response length feature.
22. The system of claim 16 , wherein the at least one signal feature selected from the group includes the dominant frequency feature.
23. The system of claim 16 , wherein the at least one signal feature selected from the group includes the maximum response feature.
24. The system of claim 16 , wherein the at least one signal feature selected from the group includes the slope feature.
25. The system of claim 16 , wherein the at least one signal feature selected from the group includes the average peak power feature.
26. The system of claim 16 , wherein the at least one signal feature selected from the group includes the response symmetry feature.
27. A system for classifying a response signal of acceleration data of a structure, the system comprising:
at least one processor configured to analyze acceleration data of a structure and classify the acceleration data, the at least one processor configured to perform the steps of:
receiving a plurality of signal features of a response signal, the response signal comprising the acceleration data of the structure;
inputting a first set of signal features of the plurality of signal figures into a first artificial neural network;
classifying, using the artificial neural network, the response signal as an impact event or a non-impact event;
inputting a second set of signal features of the plurality of signal figures of the response signal into a second artificial neural network, the second set of signal features being different from the first set of signal features; and
classifying, using the second artificial neural network, the response signal as an impact event or a non-impact event.
28. The system of claim 27 , wherein the at least one processor is further configured to train the first artificial neural network by:
obtaining signal data for a plurality of event signals, the plurality of event signals including a plurality of training events and a plurality of validation events, the signal data for each event signal comprising an event signature of either an impact signature or a non-impact signature;
inputting the signal data of the training events of the plurality of event signals into the first artificial neural network;
obtaining a predicted training signature from the first artificial neural network for each of the training events;
comparing the predicted training signatures with the event signatures to determine an error level; and
adjusting iteratively a plurality of internal weights of the first artificial neural network until the error level is below a selected threshold.