System and method for automated detection, classification, and remediation of defects using ultrasound testing
A system and method perform automated detection, classification, and remediation of defects in a structure using ultrasound testing. The system includes an autoencoder is trained and configured to generate a de-noised UT scan image from a noisy UT scan image of a structure, a support vector machine configured to detect a defect in the structure, a convolutional neural network configured to classify the defect, and a remediation subsystem configured to remediate the defect. The method implements the system.
1 . A system, comprising:
an input device configured to receive a noisy ultrasonic test (UT) scan image of a structure, wherein the noisy UT scan image includes noise and a plurality of features, and wherein the noisy UT scan image has a first size; and
a processor configured by code executing therein to implement:
a UT scan de-noiser configured to generate a de-noised UT scan image from the noisy UT scan image, the UT scan de-noiser including:
an encoder having a first layer of artificial neurons to compress the noisy UT scan image to an encoded representation having a second size at least two orders of magnitude smaller than the first size;
a bottleneck having a second layer of artificial neurons, responsive to the encoded representation, to pass only the plurality of features as an output; and
a decoder having a third layer of artificial neurons to decompress the output of the bottleneck and to generate the de-noised UT scan image from the decompressed output;
a defect detector configured to automatically determine, without user input, whether the structure has a defect or does not have a defect from the de-noised UT scan image;
a defect classifier configured, responsive to the automatic classification of the structure as having a defect, to classify the defect by type; and
a remediation subsystem configured to remediate the classified defect.
2 . The system of claim 1 , wherein the UT scan de-noiser includes an artificial neural network.
3 . The system of claim 2 , wherein the artificial neural network is an autoencoder.
4 . The system of claim 2 , wherein the artificial neural network is a convolutional neural network (CNN).
5 . The system of claim 2 , wherein the artificial neural network is trained by an inputted training UT scan image.
6 . The system of claim 5 , wherein the trained artificial neural network de-noises the noisy UT scan image to generate the de-noised UT scan image.
7 . The system of claim 1 , wherein the defect detector includes a support vector machine (SVM).
8 . The system of claim 1 , wherein the defect classifier includes an artificial neural network.
9 . The system of claim 8 , wherein the artificial neural network includes a convolutional neural network (CNN).
10 . The system of claim 1 , wherein the defect is selected from the group consisting of: a strain in the structure, a crack in the structure, and a corrosion of the structure.
11 . The system of claim 10 , wherein the remediation subsystem remediates the defect by replacing a portion of the structure having the defect.
12 . The system of claim 10 , wherein the remediation subsystem remediates the defect by placing a covering sleeve over the defect.
13 . A system, comprising:
an autoencoder configured to generate a de-noised UT scan image from a noisy UT scan image of a structure, wherein the noisy UT scan image includes noise and a plurality of features, and wherein the noisy UT scan image has a first size, the autoencoder including:
an encoder having a first layer of artificial neurons to compress the noisy UT scan image to an encoded representation having a second size at least two orders of magnitude smaller than the first size;
a bottleneck having a second layer of artificial neurons, responsive to the encoded representation, to pass only the plurality of features as an output; and
a decoder having a third layer of artificial neurons to decompress the output of the bottleneck and to generate the de-noised UT scan image from the decompressed output;
a support vector machine (SVM) configured to automatically determine, without user input, whether the structure has a defect or does not have a defect from the de-noised UT scan image;
a convolutional neural network (CNN) configured, responsive to the automatic classification of the structure as having a defect, to classify the defect by type; and
a remediation subsystem configured to remediate the classified defect.
14 . The system of claim 13 , wherein the autoencoder is trained by an inputted training UT scan image.
15 . The system of claim 14 , wherein the trained autoencoder de-noises the noisy UT scan image to generate the de-noised UT scan image.
16 . The system of claim 13 , wherein the defect is selected from the group consisting of: a strain in the structure, a crack in the structure, and a corrosion of the structure.
17 . The system of claim 16 , wherein the remediation subsystem remediates the defect by replacing a portion of the structure having the defect.
18 . The system of claim 16 , wherein the remediation subsystem remediates the defect by placing a covering sleeve over the defect.
19 . A method, comprising:
training an autoencoder using a training UT scan image;
applying an input noisy ultrasonic test (UT) scan image of a structure to the trained autoencoder, wherein the noisy UT scan image includes noise and a plurality of features, and wherein the noisy UT scan image has a first size;
compressing the noisy UT scan image using an encoder having a first layer of artificial neurons to generate an encoded representation having a second size at least two orders of magnitude smaller than the first size;
responsive to the encoded representation, passing only the plurality of features as an output using a bottleneck having a second layer of artificial neurons;
decompressing the output of the bottleneck to generate a de-noised UT scan image using a decoder having a third layer of artificial neurons;
applying the de-noised UT scan image to a support vector machine (SVM);
automatically determining by the SVM, without user input, whether the structure has a defect or does not have a defect from the de-noised UT scan image;
responsive to the automatic classification of the structure as having a defect, classifying the defect as to a type of defect using a convolutional neural network (CNN); and
remediating the classified defect using a remediation subsystem.
20 . The method of claim 19 , wherein the defect is selected from the group consisting of: a strain in the structure, a crack in the structure, and a corrosion of the structure.