CPC G01S 7/41 (2013.01) [B25J 5/007 (2013.01); B25J 15/0019 (2013.01); G01M 3/38 (2013.01); G01N 23/203 (2013.01); G01N 29/069 (2013.01); G01S 13/885 (2013.01); G05D 1/0212 (2013.01); G06N 3/08 (2013.01); F16L 2101/30 (2013.01); G01N 2223/628 (2013.01); G01N 2223/646 (2013.01); G01N 2291/0289 (2013.01)] | 4 Claims |
1. A multi-scale inspection and intelligent diagnosis system for tunnel structural defects, comprising:
a traveling section;
a supporting section, disposed on the traveling section, and comprising a rotatable telescopic platform, wherein two mechanical arms working in parallel are disposed on the rotatable telescopic platform, one of the two mechanical arms being a multi-joint snake-shaped mechanical arm, and the other being a load measuring arm with a built-in inspection device;
an inspection section, mounted on the supporting section, and configured to perform multi-scale inspection on surface defects and internal defects in different depth ranges of a same position of a tunnel structure, and transmit inspected defect information to a control section, the inspection section including:
a laser 3D scanner, mounted on the rotatable telescopic platform, and configured to acquire tunnel panorama image information comprising defects of cracks and water leakage on a tunnel lining surface;
a ground penetrating radar and an ultrasonic imaging device which are placed on a top end of the multi-joint snake-shaped mechanical arm, wherein the ground penetrating radar is configured to obtain defects of deeper layers, and the ultrasonic imaging device is configured to obtain depths and widths of the cracks on the tunnel lining surface; and
an X-ray backscattering device, built in the load measuring arm with the built-in inspection device, and configured to perform millimeter-level measurement on an inside of the cracks on the tunnel lining surface; and
the control section, configured to:
construct a deep neural network-based defect diagnosis model;
construct a data set by using historical surface defect and internal defect information and a marked pixel-level defect type, and train the deep neural network-based defect diagnosis model; and
receive defect information in real time and input the received defect information into a trained defect automatic recognition and diagnosis model, and automatically recognize a type, a position, and a contour of a defect.
|