A steel component quality evaluation method based on acoustic tomography

Abstract

The present invention provides a steel component quality assessment method based on acoustic tomography, applicable to the field of steel component quality assessment. The method comprises the following steps: fabricating a test model, conducting acoustic tomography testing and obtaining results, processing result files, image preprocessing, model construction, model training, and model application. The proposed steel component quality assessment method based on acoustic tomography offers advantages such as comprehensive testing, rapid analysis, reliable results, and low operator skill requirements. It can be widely applied in the field of steel component quality assessment.

Description

Technical Field

[0001] The present invention relates to the field of steel component quality evaluation, in particular to a steel component quality evaluation method based on acoustic wave tomography. Background Art

[0002] With the rapid development of modern industry, various steel components have been widely used in the fields of construction, bridges, ships, aerospace, etc. The quality and safety of steel components are crucial to the stability and reliability of the entire structure.

[0003] However, due to the influence of the external environment and the fatigue of the material itself, steel components may suffer from damage such as cracks, corrosion, fatigue and wear during use. If these damages are not discovered and handled in time, they may lead to serious consequences or even catastrophic accidents.

[0004] To ensure the quality and safety of steel components, regular quality assessment and damage detection are necessary. While traditional steel component inspection methods, such as visual inspection, ultrasonic testing, and magnetic particle testing, can detect defects to a certain extent, they often suffer from incomplete quality assessments and limited accuracy when used to evaluate steel component quality. These methods are particularly limited in complex structures or in inaccessible areas. Summary of the Invention

[0005] The present invention proposes a steel component quality assessment method based on acoustic tomography. By analyzing the propagation characteristics of sound waves inside the material, the defects and damage inside the steel component can be intuitively displayed in the form of images. Compared with traditional detection methods, acoustic tomography technology has the advantages of comprehensive detection, fast detection speed, high resolution, and low operator skill requirements. It is particularly suitable for quality assessment of large steel components or complex structures, and can solve the problems of existing steel component detection and quality assessment methods such as incompleteness and limited accuracy.

[0006] A steel component quality evaluation method based on acoustic tomography comprises the following steps:

[0007] 1) Make a test model: Classify and code the common quality problems of steel components. Common quality problems of steel components are coded with Q i ={1,2,3......}, i represents the i-th problem, and the corresponding test model Q is made ij , there are multiple test models;

[0008] 2) Performing ultrasonic tomography test and obtaining results: for the test model Q ij The velocity cloud diagram of the test results obtained by ultrasonic tomography is F ijThe detection points used in the acoustic tomography detection are arranged in a straight line.

[0009] 3) Processing of result files: The velocity cloud F ij Define a label, the label is Q i , and finally obtain the labeled speed cloud map dataset G ij ;

[0010] 4) Image preprocessing: ij Preprocessing is performed, and the preprocessing includes image size normalization, image grayscale and histogram equalization. After the preprocessing is completed, the data set G ij Divide into training group and validation group according to a certain ratio;

[0011] 5) Model construction: Build an ALexent neural network recognition model M to perform label prediction. The ALexent neural network recognition model M uses a self-gating activation function, which is expressed as formula (1):

[0012]

[0013] Where β is a fixed hyperparameter with a value of 0.5, σ(x)∈(0,1);

[0014] The ALexent neural network recognition model M is subjected to Droput regularization, and all neurons are eliminated with a probability of 60%;

[0015] 6) Model training: Using the training group to train the model, and after the training is completed, using the validation group to validate and optimize the model, and finally obtain a trained model;

[0016] 7) Model application: After obtaining the ultrasonic tomography test results of the steel component to be tested and performing image preprocessing, the trained model is used to perform label prediction. Based on the label prediction results, the quality of the steel component to be tested is evaluated.

[0017] Furthermore, the ALexent neural network recognition model is divided into 7 layers in total, which are formed by the superposition of convolutional layers, pooling layers, and fully connected layers, and a flattening layer is added at the same time.

[0018] Furthermore, the parameters of the ALexent neural network are set as follows: batch size is 64, momentum is set to 0.95, ω decay rate is set to 0.001, and learning rate is 0.001.

[0019] Furthermore, the training effect of the ALexent neural network recognition model uses the precision rate P as an evaluation indicator. The expression of the precision rate P is shown in formula (2). When the precision rate P ≥ 95%, the model meets the requirements:

[0020]

[0021] Where: TP represents the number of actual positive samples predicted as positive samples, and FP represents the number of actual negative samples predicted as positive samples.

[0022] The beneficial effects of the present invention are:

[0023] The beneficial effects of the present invention are as follows: based on the acoustic tomography technology, the velocity cloud map inside the steel component obtained has better accuracy than the traditional quality inspection method; at the same time, the ALexent neural network is used for label prediction and quality evaluation, which reduces the error and workload of manual inspection, and realizes the unification and automatic intelligence of the quality evaluation standard of the steel component; when using the ALexent neural network for label prediction, Droput regularization and self-gating activation function are used, which improves the generalization ability of the neural network model, reduces the overfitting phenomenon, and ensures the stability and reliability of the model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the steel component quality evaluation method based on acoustic tomography of the present invention. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0026] See also Figure 1 The embodiment of the present invention provides a steel component quality evaluation method based on acoustic tomography, comprising the following steps:

[0027] S101, making a test model; S101 specifically includes:

[0028] Common quality problems of steel components are classified and coded. Quality problems of steel components are coded i Indicates that i = 2, Q1 indicates the problem of surface cracks on steel components, and Q2 indicates the problem of internal bubbles on steel components. 100 corresponding test models are made, denoted as Q ij , i=1,2,j=1~100;

[0029] S102, performing an acoustic tomography test and obtaining a result; S102 specifically includes:

[0030] For the test model, the velocity cloud diagram of the test results obtained by using the acoustic tomography method is F ij , i=1,2,j=1~100;

[0031] S103, processing of result files;

[0032] The processing of the result file includes converting the obtained velocity contour F ij , define the label, the label is 1 or 2, and finally obtain the labeled speed cloud map dataset G ij , i=1,2,j=1~100;

[0033] S104, image preprocessing;

[0034] For the dataset G ij Preprocessing is performed, and the preprocessing measures include image size normalization, image grayscale and histogram equalization. After the preprocessing is completed, the data set G ij According to a certain ratio, the grouping divides the data set G ij Divided into training group and validation group, where the training group accounts for 80% of the total data, with a number of 160, and the validation group accounts for 20% of the total data, with a number of 40;

[0035] S105, build model;

[0036] An ALexent neural network recognition model M is constructed to perform label prediction. The model M uses a self-gating activation function, which is expressed as formula (1):

[0037]

[0038] Where β is a fixed hyperparameter with a value of 0.5, σ(x)∈(0,1);

[0039] The model building further includes applying Droput regularization to the model M and eliminating all neurons with a probability of 60%;

[0040] The ALexent neural network recognition model is divided into 7 layers in total, which are formed by superposition of convolutional layers, pooling layers, fully connected layers, etc., and a flattening layer is added at the same time; the main parameter settings of the ALexent neural network are shown in Table 1.

[0041] Table 1 Main parameter settings of ALexent neural network

[0042] Batchsize 64 momentum 0.95 Decay rate 0.001 Learning rate 0.001

[0043] S106, model training;

[0044] The training effect of the ALexent neural network uses the precision rate P as the evaluation index, and the expression is shown in formula (2). When P ≥ 95%, the model meets the requirements.

[0045]

[0046] Where: TP represents the number of actual positive samples predicted as positive samples, and FP represents the number of actual negative samples predicted as positive samples.

[0047] The model is trained using the training group data. After the training is completed, the model is verified using the validation group. The model is verified on the validation group. It is stipulated that the evaluation index precision rate P is greater than 90%, which is considered to meet the requirements. The prediction work is performed. The precision rate evaluation results are shown in Table 2.

[0048] Table 2 Precision evaluation results

[0049] TP(pcs) 37 TP+FP (pieces) 40 Accuracy (%) 92.5

[0050] As shown in Table 2, the accuracy evaluation result is 92.5%, which meets the requirements, that is, the model is trained well for subsequent work.

[0051] S107, model application;

[0052] The model application includes applying the obtained trained model to the quality evaluation of steel components. The application of the obtained trained model to the quality evaluation of steel components includes obtaining the ultrasonic tomography detection results of the steel components to be tested, performing image preprocessing, and then using the trained model to perform label prediction. According to the label prediction results, the quality of the steel components to be tested is evaluated.

[0053] In the above embodiment, the present invention discloses a steel component quality evaluation method based on acoustic tomography, which includes making a test model, performing acoustic tomography detection and obtaining results, processing result files, image preprocessing, model construction, model training and model application; the steel component quality evaluation method based on acoustic tomography proposed by the present invention has the advantages of comprehensive detection, fast analysis speed, reliable results, and low operator skill requirements, and can be widely used in the field of steel component quality evaluation.

[0054] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A steel component quality evaluation method based on acoustic tomography, characterized in that: The following steps are involved: 1) Make a test model: Classify and code the common quality problems of steel components. Common quality problems of steel components are coded with Q i ={1,2,3......}, i represents the i-th problem, and the corresponding test model Q is made ij , there are multiple test models; 2) Performing ultrasonic tomography test and obtaining results: for the test model Q ij The velocity cloud diagram of the test results obtained by ultrasonic tomography is F ij The detection points used in the acoustic tomography detection are arranged in a straight line. 3) Processing of result files: The velocity cloud F ij Define a label, the label is Q i , and finally obtain the labeled speed cloud map dataset G ij ; 4) Image preprocessing: ij Preprocessing is performed, and the preprocessing includes image size normalization, image grayscale and histogram equalization. After the preprocessing is completed, the data set G ij Divide into training group and validation group according to a certain ratio; 5) Model construction: Build an ALexent neural network recognition model M to perform label prediction. The ALexent neural network recognition model M uses a self-gating activation function, which is expressed as formula (1): Where β is a fixed hyperparameter with a value of 0.5, σ(x)∈(0,1); The ALexent neural network recognition model M is subjected to Droput regularization, and all neurons are eliminated with a probability of 60%; 6) Model training: Using the training group to train the model, and after the training is completed, using the validation group to validate and optimize the model, and finally obtain a trained model; 7) Model application: After obtaining the ultrasonic tomography test results of the steel component to be tested and performing image preprocessing, the trained model is used to perform label prediction. Based on the label prediction results, the quality of the steel component to be tested is evaluated.

2. A steel component quality evaluation method based on acoustic tomography according to claim 1, characterized in that: The ALexent neural network recognition model is divided into 7 layers in total, which are formed by the superposition of convolutional layers, pooling layers, and fully connected layers, and a flattening layer is added at the same time.

3. The steel component quality evaluation method based on acoustic tomography according to claim 1, characterized in that: The parameters of the ALexent neural network are set as follows: Batchsize is 64, momentum is set to 0.95, ω decay rate is set to 0.001, and learning rate is 0.

001.

4. The steel component quality evaluation method based on acoustic tomography according to claim 1, characterized in that: The training effect of the ALexent neural network recognition model uses the precision rate P as an evaluation indicator. The expression of the precision rate P is shown in formula (2). When the precision rate P ≥ 95%, the model meets the requirements: Where: TP represents the number of actual positive samples predicted as positive samples, and FP represents the number of actual negative samples predicted as positive samples.

Images