Real-time structural damage detection by convolutional neural networks
Certain embodiments may generally relate to structural damage detection. An embodiment may be directed to method for identifying a presence and a location of structural damage. Such method may include training a convolutional neural network (CNN) for a joint of a structure, sending instructions to a modal shaker to induce an input to the structure, receiving, as a result of the induced input, a raw acceleration signal at the joint, computing, based on the trained CNN and the raw acceleration signal, an index value of the joint, and identifying, according to the index value, a presence of a location of structural damage of the structure. In a further embodiment, the index value represents a likelihood of damage at the joint.
1. A method for identifying a presence and a location of structural damage, comprising:
training a one-dimensional convolutional neural network (CNN) applied a joint of a structure;
sending instructions to a modal shaker to induce an input to the structure;
receiving, as a result of the induced input, a raw acceleration signal at the joint;
computing, based on the trained one-dimensional CNN and the raw acceleration signal, an index value of the joint; and
identifying, according to the index value, a presence of a location of structural damage of the structure,
wherein the index value represents a likelihood of damage at the joint, and
wherein the computing of the index value comprises
dividing the acceleration signal to a number of frames that each include a total number of n s samples;
normalizing the frames between −1 to 1;
feeding the normalized frames measured at the joint to the one-dimensional CNN; and
determining a probability of damage (PoD) at the joint by dividing a number of frames classified s damaged by a total number of frames processed by the one-dimensional CNN.
2. The method of claim 1 , wherein a high PoD value within a range of about 0.8 to about 1.0 provides an indication that the joint is likely to be damaged, and a low PoD value within a range of about 0.0 to about 0.5 provides an indication that the joint is likely to be undamaged.
3. The method of claim 1 , wherein the acceleration signal is measured by an accelerometer that is disposed at the location of the joint.
4. The method of claim 1 ,
wherein the training of the one-dimensional CNN comprises conducting a plurality of experiments to generate a training data set for training the one-dimensional CNN,
wherein each of the plurality of experiments comprises measuring acceleration signals at an undamaged joint as a result of an application of a random shaker excitation at the undamaged joint, and
wherein each of the plurality of experiments comprises introducing damage at the undamaged joint to create a damaged joint, and measuring acceleration signals at the damaged joint as a result of the application of the random shaker excitation at the damaged joint.
5. An apparatus, comprising:
at least one memory comprising computer program code; and
at least one processor;
wherein the at least one memory and the computer program code are configured, with the at least one processor, to cause the apparatus at least to:
train a one-dimensional convolutional neural network (CNN) applied a joint of a structure;
send instructions to a modal shaker to induce an input to the structure;
receive, as a result of the induced input, a raw acceleration signal at the joint;
compute, based on the trained one-dimensional CNN and the raw acceleration signal, an index value of the joint; and
identify, according to the index value, a presence of a location of structural damage of the structure,
wherein the index value represents a likelihood of damage at the joint, and
wherein the apparatus is further caused to, in the computation of the index value,
divide the acceleration signal to a number of frames that each include a total number of n s samples;
normalize the frames between −1 to 1;
feed the normalized frames measured at the joint to the one-dimensional CNN; and
determine a probability of damage (PoD) at the joint by dividing a number of frames classified s damaged by a total number of frames processed by the one-dimensional CNN.
6. The apparatus of claim 5 , wherein a high PoD value within a range of about 0.8 to about 1.0 provides an indication that the joint is likely to be damaged, and a low PoD value within a range of about 0.0 to about 0.5 provides an indication that the joint is likely to be undamaged.
7. The apparatus of claim 5 , wherein the acceleration signal is measured by an accelerometer that is disposed at the location of the joint.
8. The apparatus of claim 5 , wherein the at least one memory and the computer program code are further configured, with the at least one processor, to cause the apparatus at least to, in the training of the one-dimensional CNN:
conduct a plurality of experiments to generate a training data set for training the one-dimensional CNN,
wherein each of the plurality of experiments includes measuring acceleration signals at an undamaged joint as a result of an application of a random shaker excitation at the undamaged joint, and
wherein each of the plurality of experiments includes introducing damage at the undamaged joint to create a damaged joint, and measuring acceleration signals at the damaged joint as a result of the application of the random shaker excitation at the damaged joint.
9. A computer program, embodied on a non-transitory computer readable medium, the computer program, when executed by a processor, causes the processor to:
train a one-dimensional convolutional neural network (CNN) applied a joint of a structure;
send instructions to a modal shaker to induce an input to the structure;
receive, as a result of the induced input, a raw acceleration signal at the joint;
compute, based on the trained one-dimensional CNN and the raw acceleration signal, an index value of the joint;
identify, according to the index value, a presence of a location of structural damage of the structure,
wherein the index value represents a likelihood of damage at the joint, and
wherein the processor is further caused to, in the computation of the index value,
divide the acceleration signal to a number of frames that each include a total number of n s samples;
normalize the frames between −1 to 1;
feed the normalized frames measured at the joint to the one-dimensional CNN; and
determine a probability of damage (PoD) at the joint by dividing a number of frames classified s damaged by a total number of frames processed by the one-dimensional CNN.
10. The computer program of claim 9 , wherein a high PoD value within a range of about 0.8 to about 1.0 provides an indication that the joint is likely to be damaged, and a low PoD value within a range of about 0.0 to about 0.5 provides an indication that the joint is likely to be undamaged.
11. The computer program of claim 9 , wherein the acceleration signal is measured by an accelerometer that is disposed at the location of the joint.
12. The computer program of claim 9 , wherein the computer program, when executed by the processor, further causes the processor to, in the training of the one-dimensional CNN:
conduct a plurality of experiments to generate a training data set for training the one-dimensional CNN,
wherein each of the plurality of experiments includes measuring acceleration signals at an undamaged joint as a result of an application of a random shaker excitation at the undamaged joint, and
wherein each of the plurality of experiments includes introducing damage at the undamaged joint to create a damaged joint, and measuring acceleration signals at the damaged joint as a result of the application of the random shaker excitation at the damaged joint.