Method for detecting looseness of steel bracket

Abstract

The invention belongs to the field of engineering construction steel structure detection, and particularly discloses a steel bracket looseness detection method, which comprises the following steps of: arranging a looseness bolt at the joint of a truss of steel structure bracket outside an engineering; an aluminum head force hammer is used for knocking the middle point of a support cross beam, and sound signals of bolts with different loosening degrees are collected through a sound pressure sensor; establishing a one-dimensional convolutional neural network according to the collected sound signal and the bolt loosening condition; a plurality of sound pressure sensors are arranged around the to-be-tested steel structure support; knocking and collecting all sound signals by using the same method, and then carrying out independent component analysis to obtain an independent sample; inputting the independent sample into the trained one-dimensional convolutional neural network to obtain a corresponding result; and judging the loosening degree of the to-be-detected steel structure bracket. The real-time performance and the detection efficiency of looseness detection of the steel support can be improved, operation is easy and convenient, expensive professional equipment is not needed, looseness of the steel support can be rapidly evaluated, implementation can be conducted on the construction site, and the influence on the construction progress is small.

Description

Technical Field

[0001] The invention belongs to the field of engineering construction steel structure detection, and particularly relates to a method for detecting looseness of a steel bracket. Background Art

[0002] Due to factors such as complex construction environments, soil variations, unstable geological conditions, and human intervention, steel supports may become loose during construction. This loosening can lead to uneven settlement or a decrease in overall stability. Therefore, promptly identifying loose steel supports and tightening bolts is crucial for ensuring safe construction and operation.

[0003] Traditional inspection equipment is mainly optical inspection equipment. Inspectors can only perform deformation inspection on the periphery of the steel bracket or rely on visual inspection. These inspection methods have poor real-time performance and low inspection efficiency. They often cannot fully grasp the status of the steel bracket and their safety is not guaranteed. Summary of the Invention

[0004] In order to address the deficiencies of the prior art, the present invention aims to provide a method for detecting loose steel supports, which can effectively solve the problems in the prior art of poor real-time detection of loose steel supports, low efficiency, and being unfavorable for rapid detection and timely tightening, thereby avoiding safety problems caused by loose steel support bolts.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for detecting looseness of a steel support, comprising:

[0007] S1: Build a steel structure support outside the project and use a torque wrench to loosen bolts at the connection of the steel structure support. Use an aluminum hammer to strike the midpoint of the crossbeam of the steel structure support. Use an acoustic pressure sensor to collect sound signals of bolts with different degrees of looseness to establish a sample library.

[0008] S2: Take the sound signals of bolts with different degrees of looseness as input and the corresponding bolt looseness as output to establish a one-dimensional convolutional neural network;

[0009] S3: m sound pressure sensors are evenly arranged within a certain range of the circumscribed circle of the corner points of the steel structure support to be tested;

[0010] S4: Use an aluminum hammer to strike the midpoint of the central beam of the steel bracket to be tested, and synchronously collect all sound signals through the sound pressure sensor. The sampling frequency of the sound pressure sensor is the same as that of step S1, and the sampling data point is n;

[0011] S5: Perform independent component analysis on the sound to obtain m independent samples;

[0012] S6: Input m independent samples into the neural network trained in step S2 in sequence, and obtain m results d i , the looseness degree d of the steel structure support to be tested is equal to m d i The maximum value, that is, d=max(d i ), where i = 1, 2,…, m.

[0013] Furthermore, the looseness of the bolts in step S1 is divided into ten levels from fully tightened to fully loosened, named 0 to 9, where 0 represents fully tightened and 9 represents fully loosened.

[0014] Furthermore, in step S1 , the sampling frequency of the sound pressure sensor is not less than 40 kHz, the sampling duration is 1 s, and the number of sampling times is 3.

[0015] Furthermore, the sound signal collected in step S1 is sliced ​​every 1024 points, and each slice data is subjected to maximum normalization regularization to form a sample library.

[0016] Furthermore, in step S3, the sound pressure sensor is arranged within a 2-meter range of the circumscribed circle of the corner points of the steel structure support to be tested.

[0017] Furthermore, in step S3, the sound pressure sensors cannot be located in the same plane, and m is not less than 8.

[0018] Furthermore, in step S4, the number of sampling data points n is not less than 4096.

[0019] Furthermore, the independent component analysis in step S5 specifically includes:

[0020] S5-1: The data measured by m sound pressure sensors around the steel structure support to be tested are combined into a set of sample data {x i}, {x i} is an n-dimensional vector;

[0021] S5-2: Data preprocessing; the specific calculation formula is:

[0022]

[0023] in, is the data after removing the mean of the signal, E(x i ) is the sound signal {x i}mean;

[0024] S5-3: Data whitening; the specific calculation is:

[0025] Will Composition matrix:

[0026]

[0027] in, is the matrix composed of the data of m sound pressure sensors after removing the signal mean;

[0028] Calculate the variance matrix [C x ]:

[0029]

[0030] [C x ] to perform spectrum analysis:

[0031] [C x ][V]=[V][D];

[0032] Where [D] is the spectral matrix and [V] is the eigenvector matrix.

[0033] Calculate the whitened matrix [Z]:

[0034]

[0035] S5-4: Randomly generate an n×n dimensional weight matrix:

[0036] [W]=[{w1}{w2}...{w n}],

[0037] Each of its elements comes from a Gaussian distribution with mean zero and variance 1;

[0038] S5-5: Calculate the separation matrix:

[0039] To {w i} Perform iterative update calculations in sequence, where i = 1, 2, .., n, then:

[0040]

[0041] Where, is the updated weight vector; E(*) is the sample expectation; g(*) is a nonlinear function, g′(*) is the derivative function of g(*); Perform orthogonal normalization, when Relative to stopping the iterative calculation;

[0042] S5-6: Solve for the independent component matrix [S]:

[0043] [S]=[W][Z];

[0044] S5-7: Form m independent samples {s i}, i=1,2,…,m.

[0045] Furthermore, in step S6, for independent samples, the maximum value of the time-course amplitude is used as the first data, 1024 data points are taken continuously, and then the maximum value normalization regularization process is used as the input of the convolutional neural network.

[0046] The beneficial effects of the present invention are:

[0047] The present invention uses an aluminum-headed hammer to strike a steel support constructed outside a project. A sound pressure sensor is used to collect sound signals corresponding to different loose bolts to establish a sample library. The sound signals of bolts with different degrees of looseness are used as input, and the corresponding bolt looseness levels are used as output. A one-dimensional convolutional neural network is then established. The aluminum-headed hammer is then used to strike the steel support to be tested. Independent component analysis is performed on the collected sound signals. Finally, the independent samples are input into the one-dimensional convolutional neural network to obtain the looseness level of the steel support to be tested. The present invention can improve the real-time performance and efficiency of steel support loosening detection, is easy to operate, does not require expensive specialized equipment, facilitates rapid assessment of steel support looseness, can be implemented on-site, and has minimal impact on construction progress. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the method flow of the present invention;

[0049] Figure 2 This is a schematic diagram of the structure of the detection equipment of the present invention using a steel structure bracket outside the project to sample sound signals;

[0050] Figure 3 It is a structural schematic diagram of the steel support loosening detection equipment of the present invention. DETAILED DESCRIPTION

[0051] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of application of the present invention.

[0052] like Figure 2-3 As shown, the present invention proposes a steel support loosening detection device, which is mainly used to address the problems of difficulty in accurately grasping the loose state of the support, poor real-time detection, and low detection efficiency when using traditional detection methods to detect loose steel structure supports during construction. The steel support loosening detection device includes an aluminum head hammer 2, an acoustic pressure sensor 3, a connecting line 4, a data acquisition device 5, and a computer 6. The acoustic pressure sensors 3 are evenly arranged around a steel structure support 7 and a steel support to be tested 8. The acoustic pressure sensors 3 are connected to the data acquisition device 5 via a connecting line 4, and the data acquisition device 5 is connected to the computer 6 via a connecting line.

[0053] The present invention first builds a steel structure bracket 7 outside the project, uses a torque wrench to arrange bolts 1 with different degrees of looseness at the connection of the steel structure bracket 7, uses an aluminum head hammer 2 to knock the midpoint of the beam of the steel structure bracket 7, and collects sound signals around the bolt 1 through the sound pressure sensor 3. Then, the sound signal is transmitted to the computer 6 through the connecting line 4 and the collector 5. The sound signal is sliced ​​in the computer 6 to form a sample library. The sound signals of the bolts 1 with different degrees of looseness are used as input and the corresponding looseness degree of the bolt 1 is used as output to establish a one-dimensional convolutional neural network.

[0054] Then, multiple sound pressure sensors 3 are arranged around the steel bracket 8 to be tested, and an aluminum head hammer 2 is used to knock the midpoint of the beam of the steel bracket 8 to be tested. The sound pressure sensor 3 collects sound signals around the steel bracket 8 to be tested, and then the sound signals are transmitted to the computer 6 through the connecting line 4 and the collector 5. Independent component analysis is performed in the computer 6 to obtain independent samples, which are then input into the trained one-dimensional convolutional neural network. According to the output result of the neural network, the looseness degree of the steel bracket to be tested is determined.

[0055] Accordingly, the present invention also proposes a method for detecting looseness of a steel support, which specifically comprises the following steps:

[0056] S1: Build a steel structure support outside the project and use a torque wrench to loosen the bolts at the support connection. Use an aluminum-head hammer to strike the midpoint of the support beam. Use a sound pressure sensor to collect sound signals of bolts with different degrees of looseness to establish a sample library.

[0057] Among them, the sampling frequency of the sound pressure sensor is not less than 40kHz, the sampling time is 1s, and the number of sampling times is 3. The collected sound signal is sliced ​​every 1024 points, and each slice data is processed with maximum normalization to form a sample library.

[0058] The present invention divides bolts with different degrees of looseness into ten levels from fully tightened to fully loosened, and names them 0 to 9, where 0 represents fully tightened and 9 represents fully loosened.

[0059] S2: Take the sound signals of bolts with different degrees of looseness as input and the corresponding bolt looseness as output to establish a one-dimensional convolutional neural network.

[0060] S3: Within a 2-meter radius of the circumscribed circle of the corners of the steel support to be tested, evenly arrange m sound pressure sensors. The sensors cannot be located in the same plane, and m is not less than 8. The present invention evenly arranges sound pressure sensors around the steel support to ensure that all areas of the support are tested, avoiding missing any potentially loose parts.

[0061] S4: Use an aluminum head hammer to hit the midpoint of the central beam of the steel bracket to be tested, and synchronously collect all sound signals through the sound pressure sensor. The sampling frequency of the sound pressure sensor is the same as that of step S1, and the sampling data point is n, which is not less than 4096.

[0062] S5: Perform independent component analysis on the sound to obtain m independent samples; specifically, it includes:

[0063] S5-1: The data measured by m sound pressure sensors around the steel structure support to be tested are combined into a set of sample data {x i}, {x i}(i=1,2,…,m) is an n-dimensional vector;

[0064] S5-2: Data preprocessing; the specific calculation formula is:

[0065]

[0066] in, is the data after removing the mean of the signal, E(x i ) is the sound signal {x i}mean;

[0067] S5-3: Data whitening; the specific calculation is:

[0068] Will Composition matrix:

[0069]

[0070] in, is the matrix composed of the data of m sound pressure sensors after removing the signal mean;

[0071] Calculate the variance matrix [C x ]:

[0072]

[0073] [C x ] to perform spectrum analysis:

[0074] [C x ][V]=[V][D];

[0075] Where [D] is the spectral matrix and [V] is the eigenvector matrix.

[0076] Calculate the whitened matrix [Z]:

[0077]

[0078] S5-4: Randomly generate an n×n dimensional weight matrix:

[0079] [W]=[{w1}{w2}...{w n}],

[0080] Each of its elements comes from a Gaussian distribution with mean zero and variance 1;

[0081] S5-5: Calculate the separation matrix:

[0082] To {w i} Perform iterative update calculations in sequence, where i = 1, 2, .., n, then:

[0083]

[0084] Where, is the updated weight vector; E(*) is the sample expectation; g(*) is a nonlinear function, g′(*) is the derivative function of g(*); where g(x) = x 3 , then g′(x)=3x 2 ;right Perform orthogonal normalization, when Relative to stopping the iterative calculation;

[0085] S5-6: Solve for the independent component matrix [S]:

[0086] [S]=[W][Z];

[0087] S5-7: Form m independent samples {s i}, i=1,2,…,m.

[0088] S6: Input m independent samples into the neural network trained in step S2 in sequence, and obtain m results d i , the looseness degree d of the steel structure support to be tested is equal to m d i The maximum value, that is, d=max(d i ), where i = 1, 2,…, m.

[0089] Among them, for independent samples, the maximum value of the time-course amplitude is taken as the first data, 1024 data points are taken continuously, and then the maximum value normalization regularization processing is used as the input of the convolutional neural network.

[0090] The present invention uses an aluminum-headed hammer to strike a steel support constructed outside a project. Sound signals corresponding to different loose bolts are collected using an acoustic pressure sensor to establish a sample library. The sound signals of bolts with different degrees of looseness are used as input, and the corresponding bolt looseness levels are used as output. A one-dimensional convolutional neural network is then established. The aluminum-headed hammer is then used to strike the steel support to be tested. Independent component analysis is performed on the collected sound signals, and the independent samples are input into the one-dimensional convolutional neural network to determine the looseness level of the steel support to be tested. This method can significantly improve the real-time performance and efficiency of steel support loosening detection. It is also simple to operate, does not require expensive specialized equipment, facilitates rapid assessment of steel support looseness, can be implemented on-site, and has minimal impact on construction progress.

[0091] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. A method for detecting looseness of a steel support, characterized by: include: S1: Build a steel structure support outside the project and use a torque wrench to loosen the bolts at the connection of the steel structure support; An aluminum hammer was used to strike the midpoint of a steel structure support beam. The sound pressure sensor was used to collect the sound signals of bolts with different degrees of looseness to establish a sample library. S2: Take the sound signals of bolts with different degrees of looseness as input and the corresponding bolt looseness as output to establish a one-dimensional convolutional neural network; S3: m sound pressure sensors are evenly arranged within a certain range of the circumscribed circle of the corner points of the steel structure support to be tested; S4: Use an aluminum hammer to strike the midpoint of the central beam of the steel bracket to be tested, and synchronously collect all sound signals through the sound pressure sensor. The sampling frequency of the sound pressure sensor is the same as that of step S1, and the sampling data point is n; S5: Perform independent component analysis on the sound to obtain m independent samples; S6: Input m independent samples into the neural network trained in step S2 in sequence, and obtain m results d i , the looseness degree d of the steel structure support to be tested is equal to m d i The maximum value, that is, d=max(d i ), where i = 1, 2,…, m.

2. The method for detecting looseness of a steel support according to claim 1, characterized in that: In step S1 , the looseness of the bolts is divided into ten levels from fully tightened to fully loosened, and is named 0 to 9, where 0 represents fully tightened and 9 represents fully loosened.

3. The method for detecting looseness of a steel support according to claim 1, characterized in that: In step S1 , the sampling frequency of the sound pressure sensor is not less than 40 kHz, the sampling time is 1 s, and the number of sampling times is 3.

4. The method for detecting looseness of a steel support according to claim 1, wherein: The sound signal collected in step S1 is sliced ​​into 1024 points, and each slice data is processed by maximum normalization to form a sample library.

5. The method for detecting looseness of a steel support according to claim 1, characterized in that: In step S3, the sound pressure sensor is arranged within a 2-meter range of the circumscribed circle of the corner points of the steel structure support to be tested.

6. The method for detecting looseness of a steel support according to claim 1, characterized in that: In step S3, the sound pressure sensors cannot be located in the same plane, and m is not less than 8.

7. The method for detecting looseness of a steel support according to claim 1, characterized in that: In step S4, the number of sampling data points n is not less than 4096.

8. The method for detecting looseness of a steel support according to claim 1, characterized in that: The independent component analysis in step S5 specifically includes: S5-1: The data measured by m sound pressure sensors around the steel structure support to be tested are combined into a set of sample data {x i }, {x i } is an n-dimensional vector; S5-2: Data preprocessing; the specific calculation formula is: in, is the data after removing the mean value of the signal, E(x i ) is the sound signal {x i }mean; S5-3: Data whitening; the specific calculation is: Will Composition matrix: in, is the matrix composed of the data of m sound pressure sensors after removing the signal mean; Calculate the variance matrix [C x ]: [C x ] to perform spectrum analysis: [C x ][V]=[V][D]; Where [D] is the spectral matrix and [V] is the eigenvector matrix. Calculate the whitened matrix [Z]: S5-4: Randomly generate an n×n dimensional weight matrix: [W]=[{w1}{w2}...{w n }], Each of its elements comes from a Gaussian distribution with mean zero and variance 1; S5-5: Calculate the separation matrix: To {w i } Perform iterative update calculations in sequence, where i = 1, 2, .., n, then: Where, is the updated weight vector; E(*) is the sample expectation; g(*) is a nonlinear function, g′(*) is the derivative function of g(*); Perform orthogonal normalization, when Relative to stopping the iterative calculation; S5-6: Solve for the independent component matrix [S]: [S]=[W][Z]; S5-7: Form m independent samples {s i }, i=1,2,…,m.

9. The method for detecting looseness of a steel structure support according to claim 1, characterized in that: In step S6, for independent samples, the maximum value of the time-course amplitude is used as the first data, 1024 data points are taken continuously, and then the maximum value normalization regularization process is used as the input of the convolutional neural network.

Images