Investigation On Intelligent Acoustic Nondestructive Evaluation System For Typical Complex Engineering Structures

Complex engineering structures are made of two or more different materials through a variety of processing methods.The prestressed structure of bridge and the wind turbine blade are two typical complex engineering structures.The prestressed structure of bridge is made of steel strand with high tensile strength and concrete with high compressive strength,so that the span of the bridge can be larger and larger.The web structure of wind turbine blade is made of carbon fiber and resin composite,which can reduce its self-weight,and improve the strength of the blade,so that the length of the wind turbine blades can be more than 100 meters.With the development of modern engineering technology,in order to improve the bearing capacity of various components,composite structures are widely used and their sizes are getting larger and larger.The manufacturing process of such large-size composite components is generally complex with the characteristics of simultaneous forming of materials and components.Various factors are likely to cause manufacturing quality problems of components,such as uncompacted grouting in pipes of prestressed structure,adhesion and degumming defects in web structure of wind turbine blades.Composite component is the key component of a structure,which relates to the quality and safety of the whole project.Therefore,how to improve the production quality of large size composite component is of great significance,and the detection technology of large size complex engineering structure has an important role in its processing and application.Compared with acoustic nondestructive evaluation technology of the traditional non-composite components and small size samples,the acoustic nondestructive evaluation technology of large size composite components has put forward new requirements.The first is that the detection efficiency should be higher,the second is that the evaluation accuracy of composite structure should be higher,and the third is that the evaluation system should be intelligent.The main contents and results of this paper are as follows:(1)A wireless system based on second pulse synchronization of GPS was developed.In order to meet the demand of high efficiency and high precision of long-distance acoustic detection in prestressed bridge structure,wireless transmission mode was adopted to realize the separation of acoustic emission and receiving,and GPS second pulse was used to develop the hardware system of high precision synchronous acquisition.In terms of upper computer,Lab VIEW was used to realize the application program with strong usability,flexibility,reusability,scalability,high efficiency and high degree of automation,and an automatic sound extraction algorithm was designed.The model tests and field engineering applications showed that the system can be used for the transmission method and the reflection method in the field of complex bridge detection,and the instrument can achieve nanosecond synchronization,with strong anti-interference ability and intelligent evaluation,which improve the detection efficiency.(2)A high-performance and real-time synchronous positioning acquisition acoustic transceiver system was developed.In order to meet the requirements of fast C scan at large area of wind power blades,FPGA was used as the main controller,nanosecond pulse transmission has been realized.Combined with the multi-threading of producer-consumer architecture with Lab VIEW software,the processing speed was greatly improved and pulse repetition frequency above 1 KHz has been achieved.To approach the problem of distinguishing the overlapped web echo,an improved matching tracking algorithm was proposed to decompose the overlapped signals.To adapt to the characteristics of attenuation in amplitude and frequency gradient of ultrasonic echo,the performance of MP algorithm to separate weak echo signal was improved by adding frequency modulation factor.(3)A kind of defect information processing method based on Hilbert-Huang transform(HHT)algorithm was developed.On account of the characteristics of anisotropy and irregular shape of the prestressed grouting pipe structural,impact-echo signals are often severely disturbed by noise and modal aliasing problems arise.And based on this,the improved HHT algorithm was proposed.This algorithm can effectively distinguish different sizes of grouting defects by processes of preprocessing echo data through band-pass filter,complementary ensemble empirical mode decomposition(CEEMD),the screening of intrinsic mode function(IMF)and HHT transformation.Simulation models,experiment models and field application tests verified the validity and practicability of the algorithm.(4)A method of automatic defect identification in wind turbine blade based on deep convolution neural network(DCNN)was proposed.The low-level signals acquired by sensors were directly input into the deep learning network structure to extract features that were important for defect classification.Based on the real ultrasonic detection data of different types of defects in wind turbine blades,two kinds of deep learning models of two-dimensional wavelet packet transform CNN(WPT-CNN)and one-dimensional single-point time-domain CNN were designed.The results showed that putting the two-dimensional WPT coefficients into CNN can provide more feature information,which effectively reduced the dependence of the model on network depth.While taking the time-domain waveform of single-point as the input of deep CNN needs to increase the number of network layers to improve the performance.(5)An automatic acoustic detection method for grouting defects in prestressed pipes based on multi-points array DCNN model was proposed.The input of the model is to organize multiple signals in a certain region to form an array in order to provide more stable and centralized regional feature information and spatial structured information for training and prediction of DCNN model.The feasibility and validity of DCNN architecture in defect detection of prestressed pipelines were verified by simulation and actual prestressed models,and the results showed that the multi-points array CNN model has higher defect classification and prediction ability than the single-point signal based CNN architecture.