Pile construction quality real-time detection AI model training method, pile construction quality real-time detection method and device

Abstract

The present application relates to the technical field of acoustic wave transmission method pile detection, and particularly relates to a training method of a real-time pile construction quality detection AI model, a real-time pile construction quality detection method and device. The method comprises: obtaining first detection data by a first acoustic wave transmission method pile detection operation during concrete pouring, the first detection data containing acoustic wave parameters of the cast-in-place concrete pile body at different depths; obtaining second detection data by a second acoustic wave transmission method pile detection operation after concrete pouring, the second detection data containing acoustic wave parameters of the solidified concrete pile body at different depths; obtaining concrete pile body quality defect detection result data by analyzing the second detection data; associating the first detection data with the concrete pile body quality defect detection result data to form a training sample pair; and iteratively training a preset neural network using the training sample pair to obtain a real-time pile construction quality detection AI model. Real-time defect early warning is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of acoustic wave transmission method pile detection, and particularly relates to a training method of a pile construction quality real-time detection AI model, a pile construction quality real-time detection method and a pile construction quality real-time detection device. BACKGROUND

[0002] As an important form of foundation structure, the construction quality of a concrete bored pile is crucial to the safety of a project. At present, the construction quality of a pile is mainly detected after the concrete is poured and reaches a certain strength, mainly including acoustic wave transmission method, low-strain reflected wave method, core drilling method, etc. For example, in the acoustic wave transmission method pile detection, according to the existing specification, the concrete strength of the pile to be detected is required to reach at least 70% of the design strength, and is not less than 15 MPa. Since the construction quality of a pile is detected after the concrete is poured and reaches a certain strength, it means that the quality defects of the pile have been formed and consolidated at the time of the construction quality detection of the pile. If serious defects are found, a large amount of time, manpower and material resources need to be spent for remediation, and even the pile foundation may need to be abandoned and rebuilt, resulting in delay of the construction period and substantial increase of the cost. It can be seen that the lag of the construction quality detection of the pile makes the quality control in the concrete pouring process unable to be fed back in real time, and the best intervention opportunity is missed.

[0003] However, with the rapid development of artificial intelligence (AI) technology, especially the powerful ability in data pattern recognition and complex system prediction, a new idea is provided to break through the lag of the construction quality detection of the pile. If the AI technology can be applied to the concrete pouring process, the acoustic wave data of the concrete in the plastic or initial setting state can be intelligently analyzed and predicted, which is expected to provide immediate and operable intervention information for the construction party when the defects are formed in the early stage or are about to be formed. SUMMARY

[0004] The purpose of the present application is to provide a training method of a pile construction quality real-time detection AI model, a pile construction quality real-time detection method and a pile construction quality real-time detection device, which solve the technical problem that the existing acoustic wave transmission method pile detection cannot feed back the construction quality in the concrete pouring process.

[0005] In a first aspect, a training method of a pile construction quality real-time detection AI model is provided, comprising: obtaining first detection data by a first acoustic transmission method pile detection operation during concrete pouring, the first detection data containing acoustic parameters of a cast-in-place concrete pile body at different depths; obtaining second detection data by a second acoustic transmission method pile detection operation after concrete pouring, the second detection data containing acoustic parameters of a solidified concrete pile body at different depths; obtaining concrete pile body quality defect detection result data by analyzing the second detection data, the concrete pile body quality defect detection result data containing defect type and defect depth position information; associating the first detection data with the concrete pile body quality defect detection result data to form a training sample pair; using the training sample pair to iteratively train a preset neural network to obtain a pile construction quality real-time detection AI model capable of generating a concrete pile body quality defect prediction result according to the first detection data.

[0006] As an optimization and / or instantiation of the above-mentioned training method of a pile construction quality real-time detection AI model, further: the first acoustic transmission method pile detection operation is performed by real-time acquisition of acoustic parameters of the cast-in-place concrete layer under the rising concrete pouring surface to form the first detection data.

[0007] As an optimization and / or instantiation of the above-mentioned training method of a pile construction quality real-time detection AI model, further: the acoustic transmission method pile detection device used in the first acoustic transmission method pile detection operation includes an acoustic detector, an acoustic emission probe, an acoustic receiving probe, a probe hanging mechanism, and a control system, the acoustic emission probe and the acoustic receiving probe are respectively connected to the acoustic detector and are respectively arranged in the first acoustic pipe and the second acoustic pipe by the probe hanging mechanism, the first acoustic pipe and the second acoustic pipe are arranged on opposite sides of the pile hole and extend in the vertical direction; during the first acoustic transmission method pile detection operation, the control system controls the height of the acoustic emission probe and the acoustic receiving probe through the probe hanging mechanism to match the rising concrete pouring surface, so that the acoustic emission probe and the acoustic receiving probe are kept in the cast-in-place concrete layer.

[0008] As an optimization and / or instantiation of the above-mentioned training method of a pile construction quality real-time detection AI model, further: the first acoustic transmission method pile detection operation and the second acoustic transmission method pile detection operation use the same acoustic transmission method pile detection device to ensure that the depth reference of the first detection data and the second detection data is consistent.

[0009] As an optimization and / or instantiation of the training method of the foundation pile construction quality real-time detection AI model, further: in the first acoustic wave transmission method pile detection operation process, the control system controls the acoustic wave emission probe and the acoustic wave receiving probe through the probe hanging mechanism to move synchronously from bottom to top at a lifting speed matching the rising speed of the concrete pouring surface.

[0010] As an optimization and / or instantiation of the training method of the foundation pile construction quality real-time detection AI model, further: the types of acoustic wave parameters of the first detection data include any one or several of acoustic time, acoustic velocity, peak amplitude, signal energy, main frequency, spectral bandwidth, attenuation coefficient, signal signal-to-noise ratio, waveform features extracted from original waveform data, and power spectral density features extracted from power spectral density (PSD) data.

[0011] As an optimization and / or instantiation of the training method of the foundation pile construction quality real-time detection AI model, further: the defect types include any one or several of holes, segregation, sediment, mud, reduced diameter, and expanded diameter.

[0012] As an optimization and / or instantiation of the training method of the foundation pile construction quality real-time detection AI model, further: the first detection data is associated with the concrete pile body quality defect detection result data obtained by analyzing the second detection data to form a training sample pair, including: segmenting the concrete pile body quality defect detection result data into multiple concrete pile body quality defect detection result data segments according to the foundation pile depth, each concrete pile body quality defect detection result data segment contains at least one quality state record, wherein: when a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defect-free state, the quality state record contains a defect-free assignment; when a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defective state, the quality state record contains a defect type assignment, a defect bottom depth assignment and a defect top depth assignment, the defect type assignment indicates the defect type, the defect bottom depth assignment indicates the defect bottom depth position, and the defect top depth assignment indicates the defect top depth position; segmenting the first detection data into multiple first detection data segments according to the foundation pile depth, each first detection data segment is stored as a depth-indexed sequence or array data structure; aligning the concrete pile body quality defect detection result data segments with the first detection data segments one by one to ensure the spatial consistency of each concrete pile body quality defect detection result data segment and each first detection data segment in the foundation pile depth direction; from each first detection data segment, extract a uniform dimension numerical feature vector as the input feature of the training sample, and the quality state record contained in the concrete pile body quality defect detection result data segment aligned with the first detection data segment to which the numerical feature vector belongs is taken as the true label of the numerical feature vector. The numerical feature vector and the true label are bound to form a training sample pair.

[0013] As an optimization and / or instantiation of the training method of the foundation pile construction quality real-time detection AI model, further: the preset neural network includes: a learnable preprocessing layer for preprocessing the first detection data, the parameters of the learnable preprocessing layer are learned and optimized together with other parameters of the preset neural network in the training process of the preset neural network; a feature extraction layer for extracting features from the preprocessed first detection data, the feature extraction layer at least includes a one-dimensional convolution layer; a sequence modeling layer for capturing long-distance dependency in the feature sequence, the sequence modeling layer at least includes a long short-term memory network (LSTM) layer or a gated recurrent unit (GRU) layer; and an output layer for generating a concrete pile body quality defect prediction result based on the features and the long-distance dependency, the output layer includes a classification output branch for predicting the defect type and a regression output branch for predicting the defect depth position.

[0014] In a second aspect, a pile construction quality real-time detection method is provided, comprising: receiving first detection data, the first detection data being obtained by a first acoustic transmission method pile detection operation when concrete is poured, and containing acoustic parameters of the cast-in-place concrete pile body at different depths; inputting the first detection data into a pile construction quality real-time detection AI model for processing to obtain a concrete pile body quality defect prediction result; wherein the pile construction quality real-time detection AI model is a pile construction quality real-time detection AI model trained by the training method of the pile construction quality real-time detection AI model of the first aspect.

[0015] In a third aspect, a pile construction quality real-time detection device is provided, comprising a processor coupled with a memory, the memory being used to store a computer program or instructions, and the processor being used to execute the computer program or instructions in the memory, so that the pile construction quality real-time detection device executes the pile construction quality real-time detection method of the second aspect.

[0016] The training method of the pile construction quality real-time detection AI model, the pile construction quality real-time detection method and the pile construction quality real-time detection device provided by the present application effectively solve the hysteresis problem of pile construction quality detection compared with the prior art. By obtaining the first detection data (containing acoustic parameters of the cast-in-place concrete pile body at different depths) when the concrete is poured, and using the pile construction quality real-time detection AI model trained based on the first detection data for prediction to generate a concrete pile body quality defect prediction result, real-time early warning of defects in the early stage of formation or just before formation can be realized, so that the construction party can intervene in time, such as adjusting the construction process, thereby avoiding or significantly reducing the formation of defects, and changing the pile construction quality control from post-repair to pre-prevention and real-time control.

[0017] The present application will be further described below in conjunction with the accompanying drawings and specific embodiments. Additional aspects and advantages of the present application will be partially given in the following description, partially become apparent from the following description, or be understood by practice. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The drawings provided in the accompanying drawings and their descriptions in this specification can be used to explain the present application, but do not constitute an improper limitation on the present application.

[0019] Figure 1 The principle diagram of the acoustic transmission method pile detection device used in the training method of the pile construction quality real-time detection AI model of the embodiment of the present application.

[0020] Figure 2It is a structural schematic view of the pile construction quality real-time detection device of the embodiment of the present application, and the pile construction quality real-time detection AI model is arranged in the pile construction quality real-time detection device.

[0021] Figure 3 It is a structural schematic view of the pile construction quality real-time detection AI model arranged in the pile construction quality real-time detection device of the embodiment of the present application.

[0022] Figure 4 It is the sound time-depth curve, the sound velocity-depth curve, the peak amplitude-depth curve and the PSD-depth curve obtained by analyzing the second detection data used in the training method of the pile construction quality real-time detection AI model of the embodiment of the present application.

[0023] Figure 5 It is the sound time-depth curve, the sound velocity-depth curve, the peak amplitude-depth curve and the PSD-depth curve obtained by analyzing the first detection data used in the training method of the pile construction quality real-time detection AI model of the embodiment of the present application.

[0024] In the figure, it is marked as: 1-sound wave transmission method pile detection device, 11-sound wave detector, 12-sound wave emission probe, 13-sound wave receiving probe, 14-probe hanging mechanism, 15-control system, 16-first sound measuring pipe, 17-second sound measuring pipe, 21-pile hole, 22-cast-in-place concrete layer, 23-defect, 3-pile construction quality real-time detection device, 31-processor, 32-memory, 33-network interface, 34-input device, 35-output device, 41-learnable pre-processing layer, 42-feature extraction layer, 43-sequence modeling layer, 44-output layer. DETAILED DESCRIPTION

[0025] The present application will be described in detail below with reference to the drawings. Those skilled in the art will be able to implement the present application based on these descriptions. Before the present application is described in detail with reference to the drawings, it should be specifically pointed out that:

[0026] The technical solutions and technical features provided in each part including the following description can be combined with each other without conflict. In addition, in the case of possibility, these technical solutions, technical features and related combinations can be given a specific technical subject and protected by a related patent.

[0027] The embodiments of the present application involved in the following description are generally only a part of the embodiments and not all the embodiments, and based on these embodiments, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of patent protection.

[0028] The terms "comprise", "comprising", and any variations thereof, in the Specification and corresponding claims and related parts thereof, are intended to cover not exclusively inclusive. Other related terms and units can be reasonably interpreted based on the related content provided in the specification.

[0029] Figure 1 The principle diagram of the pile detection device for the training method of the pile construction quality real-time detection AI model of the embodiment of the present application is shown in FIG. 1. As shown in the figure, the pile detection device 1 of the acoustic transmission method comprises an acoustic detector 11, an acoustic emission probe 12, an acoustic receiving probe 13, a probe hanging mechanism 14, a control system 15, a first acoustic measuring pipe 16 and a second acoustic measuring pipe 17. Figure 1

[0030] The acoustic detector 11 is the core equipment of the pile detection device 1 of the acoustic transmission method, responsible for generating acoustic signals and receiving and processing acoustic data. The acoustic detector 11 has functions of signal generation, data acquisition, signal processing and data storage.

[0031] The acoustic emission probe 12 is used to emit acoustic signals into the concrete, and the acoustic receiving probe 13 is used to receive acoustic signals after passing through the concrete. The acoustic emission probe 12 and the acoustic receiving probe 13 are respectively connected with the acoustic detector 11 through signal lines to transmit excitation signals and receive signals.

[0032] The first acoustic measuring pipe 16 and the second acoustic measuring pipe 17 are oppositely arranged on both sides of the pile hole 21 and extend in the vertical direction. The acoustic emission probe 12 is arranged in the first acoustic measuring pipe 16 through the probe hanging mechanism 14, and the acoustic receiving probe 13 is arranged in the second acoustic measuring pipe 17 through the probe hanging mechanism 14. The first acoustic measuring pipe 16 and the second acoustic measuring pipe 17 provide a test channel for acoustic wave propagation, ensuring that acoustic waves can be transmitted and propagated in the concrete.

[0033] The probe hanging mechanism 14 comprises a hanging device and a lifting control device, which is used to control the vertical position of the acoustic emission probe 12 and the acoustic receiving probe 13 in the first acoustic measuring pipe 16 and the second acoustic measuring pipe 17. The probe hanging mechanism 14 can accurately control the lifting speed and the stopping position of the probe, ensuring that the probe can move to the specified depth according to the detection needs.

[0034] ​The control system 15 is the intelligent control center of the acoustic wave transmission method pile detection device 1, responsible for coordinating the work of each component. The control system 15 is connected with the acoustic wave detector 11 and the probe hanging mechanism 14, and can automatically control the detection process. It should be noted that the control system 15 is mainly responsible for the hardware control and basic data acquisition of the acoustic wave transmission method pile detection device 1, and the subsequent pile construction quality real-time detection device 3 as a computer device independent of the control system 15 is specially responsible for running the pile construction quality real-time detection AI model, and intelligently analyzing and predicting quality defects from the first detection data obtained from the control system 15.

[0035] The control system 15 receives the depth position information feedback by the lifting control device of the probe hanging mechanism 14 in real time, and synchronously receives the acoustic wave parameter data feedback by the acoustic wave detector 11, and aligns the time stamp and associates the depth mark. Specifically, the control system 15 assigns a corresponding depth mark to the acoustic wave parameter data of each depth position, ensuring that the acoustic wave parameter and the accurate depth position of the probe at the time of collection establish a one-to-one correspondence, thereby forming an acoustic wave parameter data sequence containing depth index.

[0036] During the first acoustic wave transmission method pile detection operation, as the concrete pouring proceeds, the control system 15 controls the height of the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 through the probe hanging mechanism 14 to match the rising concrete pouring surface, so that the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 remain in the cast-in-place concrete layer 22. Specifically, the control system 15 controls the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 through the probe hanging mechanism 14 to move synchronously from bottom to top at a lifting speed matching the rising speed of the concrete pouring surface.

[0037] During this process, the acoustic wave transmitting probe 12 continuously transmits acoustic wave signals to the cast-in-place concrete layer 22, and the acoustic wave receiving probe 13 receives the acoustic wave signals passing through the cast-in-place concrete layer 22. The acoustic wave detector 11 processes and analyzes the received acoustic wave signals in real time to obtain basic acoustic wave parameters, including but not limited to acoustic time, sound speed, peak amplitude, signal energy, main frequency, spectral bandwidth, attenuation coefficient, signal signal-to-noise ratio, etc., and records the original waveform data and power spectral density (PSD) data.

[0038] The first acoustic wave transmission method pile detection operation forms the first detection data by real-time acquisition of the acoustic wave parameters of the cast-in-place concrete layer 22 under the rising concrete pouring surface. The first detection data contains the acoustic wave parameters of the cast-in-place concrete pile at different depths, which reflect the acoustic characteristics of the cast-in-place concrete at different depth positions.

[0039] After the concrete pouring is complete and reaches the specified strength, a second acoustic transmission method pile inspection operation is performed using the same acoustic transmission method pile inspection device 1. At this point, the control system 15 controls the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 to ascend layer by layer in the first and second acoustic detection tubes 16 and 17 at preset depth intervals, performing a comprehensive inspection of the solidified concrete pile body.

[0040] The second acoustic wave transmission pile inspection operation acquires second inspection data, which includes acoustic wave parameters of the cured concrete pile at different depths. Because the first and second acoustic wave transmission pile inspection operations utilize the same acoustic wave transmission pile inspection device 1, the depth reference of the first and second inspection data is consistent, providing a reliable data foundation for subsequent data association and AI model training.

[0041] pass Figure 1 The acoustic wave transmission pile detection device 1 shown can acquire first and second detection data during and after concrete pouring, respectively, providing complete data support for training an AI model for real-time pile construction quality detection. The real-time detection capability and precise control function of this acoustic wave transmission pile detection device 1 enable the acquisition of effective acoustic wave data while the concrete is still in a plastic state, laying the technical foundation for real-time prediction of pile construction quality.

[0042] Figure 2 This is a schematic diagram of the structure of a real-time detection device for pile construction quality according to an embodiment of the present invention, in which a real-time detection AI model for pile construction quality is deployed. Figure 2 As shown, the real-time detection device 3 for pile construction quality provided in this embodiment is used to execute the real-time detection method for pile construction quality, and realizes real-time prediction of concrete pile body quality defects by running the real-time detection AI model for pile construction quality to perform intelligent analysis on the first detection data.

[0043] During the training stage of the AI ​​model for real-time detection of pile construction quality, the real-time detection device 3 for pile construction quality mainly undertakes the functions of training and verifying the AI ​​model for real-time detection of pile construction quality.

[0044] like Figure 2 As shown, the real-time detection device 3 for pile construction quality includes a processor 31 , a memory 32 , a network interface 33 , an input device 34 and an output device 35 .

[0045] The processor 31 serves as the core computing unit of the pile construction quality real-time detection device 3, and is responsible for performing inference calculation of the pile construction quality real-time detection AI model. The processor 31 can include a central processing unit (CPU), a digital signal processor (DSP), a microprocessor, an application specific integrated circuit (ASIC), a microcontroller (MCU), a field programmable gate array (FPGA), or one or more integrated circuits for implementing logical operations. Preferably, the processor 31 can adopt an artificial intelligence (AI) dedicated processing chip to improve processing speed in the implementation provided below. For example, the processor 31 can be a graphics processing unit (GPU), a tensor processing unit (TPU), or other dedicated AI chip.

[0046] The memory 32 is coupled with the processor 31, and is used to store parameters, program codes, and input and output data of the pile construction quality real-time detection AI model. The memory 32 includes an internal memory for storing AI model parameters and intermediate calculation results that are running, and a storage device for persistently storing trained pile construction quality real-time detection AI model files, historical detection data, and analysis results. The memory 32 can include a random access memory (RAM), a dynamic random access memory (DRAM), a static random access memory (SRAM), a flash memory, a solid state drive (SSD), a hard disk drive (HDD), or a combination of two or more of the above.

[0047] The network interface 33 is responsible for the communication connection between the pile construction quality real-time detection device 3 and external devices. The network interface 33 communicates data with the control system 15 of the acoustic wave transmission method pile detection device 1 through wired or wireless networks. The network interface 33 supports multiple communication protocols, such as Transmission Control Protocol (TCP) / Internet Protocol (IP), User Datagram Protocol (UDP), serial communication, etc., to ensure the stability and real-time performance of data transmission.

[0048] The input device 34 is used for the interaction between the operator and the pile construction quality real-time detection device 3. The input device 34 can include a keyboard, a mouse, a touch screen, etc., allowing the operator to set detection parameters, start or stop detection tasks, view detection status, etc. The input device 34 can also be used for manual input or correction of data, and manual verification and labeling of the prediction results of the pile construction quality real-time detection AI model.

[0049] The output device 35 is used to show the detection results and system state information to the operator. The output device 35 can include a display, a printer, an alarm, etc. The display is used to display the prediction results of the pile construction quality real-time detection AI model in real time, including defect type, defect location, confidence, etc. information, as well as the visualization chart of the detection data. The alarm gives an audible and visual alarm when a serious quality defect is detected, reminding the operator to take timely intervention measures.

[0050] The foregoing respectively introduces the hardware structure and function configuration of the acoustic wave transmission method pile detection device 1 and the pile construction quality real-time detection device 3. The acoustic wave transmission method pile detection device 1 and the pile construction quality real-time detection device 3 constitute a complete hardware environment for the pile construction quality real-time detection AI model training method. Among them, the acoustic wave transmission method pile detection device 1 is responsible for collecting the first detection data and the second detection data respectively at the time of concrete pouring and after concrete pouring, providing the original data source for the training of the pile construction quality real-time detection AI model; The pile construction quality real-time detection device 3 undertakes the training calculation and verification test task of the pile construction quality real-time detection AI model, and completes the iterative optimization of the preset neural network through the computing power and storage capacity. After the training of the pile construction quality real-time detection AI model is completed, the pile construction quality real-time detection device 3 will deploy the trained pile construction quality real-time detection AI model, receive the first detection data provided by the acoustic wave transmission method pile detection device 1 in actual engineering application, get the concrete pile body quality defect prediction result, and realize the role conversion from the training platform to the application system.

[0051] The specific implementation steps and technical details of the pile construction quality real-time detection AI model training method based on the above hardware environment will be described in detail below.

[0052] A training method of a pile construction quality real-time detection AI model, by performing acoustic wave transmission method pile detection at the time of concrete pouring and after concrete pouring respectively, obtaining training data and training a preset neural network. Specifically, the following steps are included.

[0053] Step 1: Obtain the first detection data.

[0054] At the time of concrete pouring, the first detection data is obtained through the first acoustic wave transmission method pile detection operation. The first detection data contains the acoustic wave parameters of the cast-in-place concrete pile body at different depths.

[0055] Specifically, the first acoustic wave transmission method pile detection operation acquires the acoustic wave parameters of the cast-in-place concrete layer 22 under the rising concrete pouring surface to form the first detection data.

[0056] During the first pile detection operation by the first acoustic transmission method, the control system 15 controls the height of the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 to match the rising concrete pouring surface through the probe hanging mechanism 14, so that the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 are kept in the cast-in-place concrete layer 22. The control system 15 controls the acoustic wave transmitting probe 12 and the acoustic wave receiving probe 13 to move synchronously from bottom to top at a lifting speed matching the rising speed of the concrete pouring surface through the probe hanging mechanism 14.

[0057] The types of the acoustic wave parameters of the first detection data can include any one or several of the following: acoustic time, acoustic velocity, peak amplitude, signal energy, dominant frequency, spectral bandwidth, attenuation coefficient, signal signal-to-noise ratio, waveform features extracted from the original waveform data, and power spectral density features extracted from the power spectral density (PSD) data.

[0058] Specifically, the acoustic time refers to the propagation time of the acoustic wave from the transmitting probe to the receiving probe, which can reflect the compactness of the medium on the acoustic wave propagation path; the acoustic velocity is the propagation speed of the acoustic wave in the medium, which is usually calculated by dividing the propagation distance by the acoustic time, and is an important indicator for evaluating the quality of concrete; the peak amplitude represents the maximum amplitude value in the received signal, reflecting the degree of attenuation of acoustic energy; the signal energy is the total energy of the entire waveform signal, which is used to evaluate the energy loss of the acoustic wave during propagation; the dominant frequency is the frequency component with the most concentrated energy in the signal spectrum, which can reflect the physical properties of the medium; the spectral bandwidth represents the effective frequency range of the signal spectrum, reflecting the frequency distribution characteristics of the signal; the attenuation coefficient describes the attenuation law of the acoustic wave amplitude with the propagation distance, which is used to quantify the absorption and scattering degree of the medium to the acoustic wave; the signal signal-to-noise ratio is the power ratio of the useful signal to the noise, reflecting the influence of signal quality and detection environment; the waveform features include time domain characteristic parameters such as the rise time, pulse width, and waveform symmetry of the waveform; and the power spectral density features include frequency domain characteristic parameters such as the energy distribution, spectral peak position, and spectral line width in the frequency domain.

[0059] Step 2: Obtain second detection data.

[0060] After the concrete is poured, the second detection data is obtained by the second pile detection operation by the second acoustic transmission method. The second detection data includes acoustic wave parameters of the solidified concrete pile body at different depths.

[0061] The first pile detection operation by the first acoustic transmission method and the second pile detection operation by the second acoustic transmission method use the same acoustic transmission method pile detection device 1 to ensure that the depth reference of the first detection data and the second detection data is consistent.

[0062] In this embodiment, the first detection data and the second detection data are stored in a sequence or array data structure indexed by depth, specifically including an acoustic time sequence, an acoustic velocity sequence, a peak amplitude sequence, and a power spectral density (PSD) sequence. The acoustic time sequence reflects the variation of acoustic wave propagation time with pile depth, and can represent the compactness of concrete at different depths; the acoustic velocity sequence represents the distribution characteristics of acoustic wave propagation velocity along the depth direction, and is an important basis for evaluating the uniformity of concrete quality; the peak amplitude sequence describes the trend of the maximum amplitude of the received signal with depth, reflecting the energy attenuation and scattering of the acoustic wave; the PSD sequence exhibits the distribution pattern of the power spectral density parameter in the depth direction, containing rich frequency domain feature information. This data organization method ensures the efficiency of data processing, provides a structured input format for the pile construction quality real-time detection AI model, and enables the neural network to effectively learn the spatial correlation between acoustic parameters and depth positions, thereby achieving accurate prediction of concrete pile quality defects. At the same time, these sequence data can also be visualized as corresponding curve graphs, which is convenient for engineers to perform manual analysis and interpretation.

[0063] In order to facilitate understanding of the relationship between the data structure and the visual representation, Figure 4 and Figure 5 respectively show the visualization results of the second detection data and the first detection data. Specifically, the values in the acoustic time sequence are arranged according to the depth index, and the acoustic time-depth curve graph can be obtained by taking the depth as the horizontal axis and the acoustic time as the vertical axis; the values in the acoustic velocity sequence are arranged according to the depth index, and the acoustic velocity-depth curve graph can be obtained by taking the depth as the horizontal axis and the acoustic velocity as the vertical axis; the values in the peak amplitude sequence are arranged according to the depth index, and the peak amplitude-depth curve graph can be obtained by taking the depth as the horizontal axis and the peak amplitude as the vertical axis; the values in the power spectral density sequence are arranged according to the depth index, and the PSD-depth curve graph can be obtained by taking the depth as the horizontal axis and the PSD energy as the vertical axis.

[0064] Step 3: analyze the second detection data to obtain the quality defect detection result.

[0065] Through analysis of the second detection data, the concrete pile quality defect detection result data is obtained. The concrete pile quality defect detection result data contains defect type and defect depth position information.

[0066] The defect types can include any one or several of a hole, segregation, sediment, clay, reduced diameter, and expanded diameter. The hole defect refers to a cavity or hole formed inside the concrete pile body, which is mainly caused by reasons such as blockage of the guide pipe during concrete pouring, interruption of concrete supply, too fast lifting of the guide pipe, or insufficient fluidity of the concrete, and can seriously weaken the integrity and bearing capacity of the pile body; the segregation defect refers to a phenomenon that the components of the concrete are separated and unevenly distributed during pouring, which is usually caused by improper concrete mix ratio, too large water-cement ratio, insufficient mixing, or too fast pouring speed, and leads to uneven concrete strength; the sediment defect refers to a loose layer formed by mud, sand, and other impurities deposited at the bottom of the pile, which is mainly caused by incomplete hole cleaning at the bottom of the pile or too long interval between hole cleaning and pouring, and can affect the development of the pile end bearing capacity; the clay defect refers to foreign impurities such as soil and silt mixed in the concrete pile body, which is caused by reasons such as hole wall collapse during hole forming, poor guide pipe sealing, or hole wall instability during concrete pouring, and can form weak parts in the pile body to affect the continuity and strength; the reduced diameter defect refers to a phenomenon that the diameter of a certain section of the pile body is smaller than the designed diameter, which is mainly caused by reasons such as extrusion of the surrounding soil, change of underground water level, deformation of the sleeve, or improper selection of hole forming equipment, and can reduce the cross-sectional area of the pile body and the bearing capacity; the expanded diameter defect refers to a phenomenon that the diameter of a certain section of the pile body is larger than the designed diameter, which is usually caused by reasons such as hole wall collapse in soft soil layer, over-excavation during hole forming, existence of soft interlayer in the stratum, or improper drilling parameters, and increases the amount of concrete and can lead to uneven stress distribution.

[0067] Among them, the hole, segregation, sediment and clay are the most common defect types in foundation pile construction and have the greatest impact on structural safety. The hole defect is usually caused by interruption of concrete pouring, blockage of the guide pipe, or insufficient fluidity of the concrete, and the sound time is prolonged and the amplitude is attenuated when the sound wave passes through the hole area; the segregation defect is usually caused by improper concrete mix ratio or uneven mixing, which is characterized by a significant decrease in sound velocity and abnormal spectral characteristics; the sediment defect is mainly caused by incomplete hole cleaning at the bottom of the pile, which can cause a low-density medium layer in the sound wave propagation path and cause a sudden change in acoustic parameters; the clay defect is usually caused by hole wall collapse or improper hole cleaning during hole forming, and scattering and energy loss occur when the sound wave is transmitted.

[0068] For different defect types, if timely detection can be made, corresponding engineering measures can be taken: when hole defects occur, concrete pouring should be immediately suspended, the state of the duct and the fluidity of the concrete should be checked, and the concrete mix should be adjusted or the duct should be cleaned if necessary; when segregation defects occur, the concrete mix and mixing quality need to be rechecked, and the aggregate gradation and water-cement ratio need to be adjusted; when sediment defects occur, the secondary hole cleaning operation at the pile bottom should be strengthened to ensure that the sediment thickness meets the design requirements; when mud inclusion defects occur, the hole wall stability needs to be evaluated, and wall protection measures or adjustment of the hole forming process need to be taken if necessary. For the defects of shrinkage and expansion, it can be judged whether the deformation degree affects the pile body bearing capacity, and measures such as re-drilling or local reinforcement can be taken if the deformation is serious.

[0069] From the technical characteristics of real-time detection by acoustic transmission method, there are significant differences in the difficulty of real-time detection of different defect types. Due to the large difference in medium density, the acoustic signal will have significant acoustic time extension, amplitude rapid attenuation and frequency spectrum distortion, and these characteristic changes are obvious and easy to identify, so it is the most easily detected defect type. Therefore, the training method of the foundation pile construction quality real-time detection AI model of the embodiment takes hole defects as the main identification target (as shown in the defect 23 in the schematic diagram). Figure 1 The defect 23 in the schematic diagram represents a hole defect.

[0070] Step 4: Forming a training sample pair.

[0071] The first detection data is associated with the concrete pile body quality defect detection result data to form a training sample pair. Specifically, it includes:

[0072] (1) The concrete pile body quality defect detection result data is segmented into multiple concrete pile body quality defect detection result data segments according to the depth of the foundation pile, and each concrete pile body quality defect detection result data segment contains at least one quality state record. When a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defect-free state, the quality state record contains a defect-free assignment; when a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defective state, the quality state record contains a defect type assignment, a defect bottom depth assignment and a defect top depth assignment, the defect type assignment indicates the defect type, the defect bottom depth assignment indicates the defect bottom depth position, and the defect top depth assignment indicates the defect top depth position.

[0073] (2) The first detection data is segmented into multiple first detection data segments according to the depth of the foundation pile, and each first detection data segment is stored as a sequence or array data structure indexed by depth.

[0074] (3) Aligning the concrete pile body quality defect detection result data segments with the first detection data segments one by one to ensure the spatial consistency of each concrete pile body quality defect detection result data segment and each first detection data segment in the depth direction of the pile.

[0075] (4) Extracting a numerical feature vector of uniform dimension from each first detection data segment as the input feature of the training sample, and taking the quality state record contained in the concrete pile body quality defect detection result data segment aligned with the first detection data segment to which the numerical feature vector belongs as the true label of the numerical feature vector, and binding the numerical feature vector and the true label to form a training sample pair.

[0076] Step 5: Training the preset neural network.

[0077] Using the training sample pair, the preset neural network is iteratively trained using the pile construction quality real-time detection device 3 to obtain a pile construction quality real-time detection AI model capable of generating concrete pile body quality defect prediction results based on first detection data.

[0078] The training sample pair forming method of step 4 has the following technical features: first, this method realizes the accurate spatio-temporal association of real-time detection data during construction and standard detection results after construction, and establishes a direct mapping relationship between "real-time acoustic parameters" and "final quality state" by deeply aligning the first detection data and the concrete pile body quality defect detection result data. This data association method can provide reliable true label sources for real-time prediction, effectively solving the problem of difficult label data acquisition in the field of pile construction quality real-time detection.

[0079] Secondly, the segmented training sample construction strategy is adopted to decompose the continuous first detection data and concrete pile body quality defect detection result data into multiple data segments, and each segment independently forms a training sample pair. This method can improve sample utilization, so that the detection data of a single pile can generate multiple training samples, effectively increasing the amount of training data. At the same time, the first detection data segments of different depth segments contain different geological conditions and construction state information, which helps to enhance the generalization ability of the pile construction quality real-time detection AI model, so that the model can adapt to diversified construction environments and geological conditions.

[0080] Thirdly, the designed multi-dimensional quality status record labeling mechanism can accurately describe the quality status of the concrete pile body. When the quality status record indicates a defect-free state, it includes a defect-free value; when it indicates a defective state, it includes defect type assignments, defect top depth assignments, and defect bottom depth assignments, forming a structured labeling system. This labeling mechanism enables the trained AI model for real-time pile construction quality inspection to not only identify the presence of quality defects, but also simultaneously predict the specific defect type and precise defect depth location, realizing the comprehensive functions of defect detection, classification, and positioning. Its output results can directly meet the actual needs of the project.

[0081] Finally, by extracting numerical feature vectors of uniform dimension from each first detection data segment, the problem of fusing heterogeneous data from multiple sources, such as acoustic time series, sound velocity series, peak amplitude series, and power spectral density series, can be effectively solved. This uniform feature extraction method ensures the effective integration of different types of acoustic wave parameters, improves the computational efficiency of the neural network, and retains the key characteristic information of multiple acoustic wave parameters, providing a rich and standardized input data format for the AI ​​model for real-time pile construction quality detection.

[0082] Figure 3 Schematic diagram of the structure of the AI ​​model for real-time detection of pile construction quality deployed in the real-time detection device for pile construction quality according to an embodiment of the present invention. Figure 3 As shown, the preset neural network includes:

[0083] (1) Learnable preprocessing layer 41: This layer preprocesses the sound time sequence, sound velocity sequence, peak amplitude sequence, and PSD sequence. The parameters of this layer are learned and optimized along with the other parameters of the neural network during training. After the input data is processed by the learnable preprocessing layer 41, the output is a fully connected (128-neuron) feature representation. This representation is then passed to the BatchNorm layer for batch normalization and finally activated by the ReLU function.

[0084] (2) Feature extraction layer 42: used to extract features from the pre-processed first detection data. The feature extraction layer 42 includes at least one-dimensional convolution layer.

[0085] like Figure 3 As shown, in this embodiment, the feature extraction layer 42 specifically includes:

[0086] ① Feature extraction layer 1 (14×14): uses a 1D Conv(32, 3×1) convolution operation, followed by a ReLU activation function and a MaxPooling(2×1) pooling operation;

[0087] ② Feature extraction layer 2 (10x10): 1D Conv (64, 3x1) convolution operation is adopted, then ReLU activation function is passed, and then MaxPooling (2x1) pooling operation is performed.

[0088] (3) Sequence modeling layer 43: used to capture long-distance dependencies in feature sequences, and the sequence modeling layer 43 at least contains a long short-term memory (LSTM) layer or a gated recurrent unit (GRU) layer.

[0089] As shown in FIG. 5, in the embodiment, the sequence modeling layer 43 (5x5) contains two LSTM layers, each of which contains 128 neurons, and the output is set to Dropout (0.2) to prevent overfitting. Figure 3

[0090] (4) Output layer 44: used to generate a concrete pile body quality defect prediction result based on features and long-distance dependencies. The output layer 44 contains a classification output branch for predicting a defect type and a regression output branch for predicting a defect depth position.

[0091] In the embodiment, specifically, the output layer 44 contains:

[0092] ① Branch 1: fully connected layer to Softmax, used to predict a defect type;

[0093] ② Branch 2: fully connected layer to Linear, used to predict defect depth position information.

[0094] ​BatchNorm is the abbreviation of Batch Normalization, which is used to stabilize the training process and accelerate network convergence; Conv is the abbreviation of Convolution, 1D Conv represents one-dimensional convolution operation, which is used to extract local features of sequence data; 1D Conv(32, 3x1) represents one-dimensional convolution with 32 feature channels and a kernel size of 3x1; ReLU is the abbreviation of Rectified Linear Unit activation function, which can introduce nonlinearity and solve the problem of gradient disappearance; MaxPooling is the abbreviation of Max Pooling, which is used to reduce the feature dimension and retain important features, and MaxPooling(2x1) represents a pooling window size of 2x1; LSTM is the abbreviation of Long Short-Term Memory, which can effectively process sequence data and capture long-distance dependencies; Dropout is the English name of the random inactivation technique, which prevents overfitting by randomly setting some neuron outputs to zero, and Dropout(0.2) means randomly setting 20% of neuron outputs to zero during training; Softmax is a normalized exponential function, which is used for probability output in multi-classification tasks; Linear represents a linear layer, which is used for numerical output in regression tasks. Figure 3 The size information annotated as (14x14), (10x10), (5x5) represents the spatial dimension size of the output feature map of the layer, reflecting the dimension change process of data between layers of the network.

[0095] Thus, the first detection data is first standardized by the learnable preprocessing layer 41, converting multi-source heterogeneous data such as acoustic time series, acoustic velocity series, peak amplitude series, and PSD series into a unified feature representation, then sequentially passing through two one-dimensional convolution blocks of the feature extraction layer 42 to extract local feature patterns, and then passing through the sequence modeling layer 43 to capture long-distance time series dependencies of acoustic parameters, and finally producing concrete pile quality defect type classification results and defect depth position regression results in the output layer 44.

[0096] The above-mentioned architecture design of the preset neural network fully considers the time series characteristics and multi-parameter feature fusion requirements of the first detection data. Through the learnable preprocessing layer 41, the effective integration of various acoustic parameters is realized, the one-dimensional convolution layer extracts the local pattern features of the acoustic signal, the LSTM layer models the long-distance time series dependencies of the acoustic parameters with depth changes, and the multi-task output design realizes the synchronous prediction of concrete pile quality defect detection, classification, and positioning. This network structure can directly learn the mapping relationship of the concrete pile quality defect prediction results from the first detection data.

[0097] In the training process of the pile construction quality real-time detection AI model, for the double-branch design of the output layer 44, the classification branch adopts the cross-entropy loss function to optimize the defect type prediction accuracy, the regression branch adopts the mean square error loss function to optimize the defect depth position prediction accuracy, and the total loss function is the weighted combination of the two. Through the multi-task learning mechanism, the pile construction quality real-time detection AI model can simultaneously learn two related tasks of defect recognition and position positioning, thereby improving the overall performance and generalization ability of the pile construction quality real-time detection AI model. After training is completed, the pile construction quality real-time detection AI model can output the concrete pile body quality defect prediction result containing the defect type assignment, the defect top depth assignment and the defect bottom depth assignment according to the first detection data collected in the concrete pouring process.

[0098] Figure 4 The acoustic time-depth curve, the acoustic velocity-depth curve, the peak amplitude-depth curve and the PSD-depth curve obtained by analyzing the second detection data used in the training method of the pile construction quality real-time detection AI model of the embodiment of the application. Figure 5 The acoustic time-depth curve, the acoustic velocity-depth curve, the peak amplitude-depth curve and the PSD-depth curve obtained by analyzing the first detection data used in the training method of the pile construction quality real-time detection AI model of the embodiment of the application. Figure 4 and Figure 5 The detection data shown is derived from the detection results of the same pile, wherein Figure 4 corresponding to the second acoustic wave transmission method pile detection operation result after the concrete pouring is completed and reaches the specified strength, Figure 5 corresponding to the first acoustic wave transmission method pile detection operation result in the concrete pouring process, both are completely consistent in depth reference and spatial position.

[0099] From Figure 4 It can be observed that in the depth range of about 12-14 meters, all four curves show obvious abnormalities: the sound time curve shows that the propagation time is significantly prolonged, the sound velocity curve shows that the velocity is sharply decreased, the peak amplitude curve shows that the amplitude is greatly attenuated, and the PSD energy curve also shows corresponding energy loss. The consistency of these characteristic changes indicates that there is obvious concrete pile body quality defect in this depth section, and the defect type assignment is "hole", the defect top depth assignment is 12.0 meters, and the defect bottom depth assignment is 14.0 meters, which is confirmed by field verification.

[0100] By comparison Figure 4 and Figure 5It can be found that the acoustic time curve, the sound speed curve, the peak amplitude curve and the PSD energy curve corresponding to the first detection data also present similar abnormal characteristic patterns in the same 12-14 meter depth range: acoustic time prolongation, sound speed reduction, amplitude attenuation and PSD energy reduction, but these abnormal characteristics are relatively weak and fuzzy relative to the visualization results of the second detection data, because the cast-in-place concrete layer 22 is still in a plastic state. This corresponding relationship theoretically verifies the learnable feature correlation between the first detection data and the second detection data, proves the theoretical feasibility of the "construction period real-time detection data-post-construction standard detection result" correlation training method proposed by the present application, and provides a data basis and technical basis for the formation of the training sample pair.

[0101] The above describes the relevant content of the present application. Those skilled in the art will be able to implement the present application based on these descriptions. Based on the above content of the present description, all other embodiments obtained by those skilled in the art without creative labor shall fall within the scope of the present application.

Claims

1. A training method for an AI model for real-time detection of pile foundation construction quality, characterized by: The method comprises the following steps: During concrete pouring, first detection data is obtained through first acoustic transmission method pile detection operation, and the first detection data comprises sound wave parameters of the cast-in-place concrete pile body at different depths; After concrete pouring, second detection data is obtained through second acoustic transmission method pile detection operation, and the second detection data comprises sound wave parameters of the solidified concrete pile body at different depths; Through analysis of the second detection data, concrete pile body quality defect detection result data is obtained, and the concrete pile body quality defect detection result data comprises defect type and defect depth position information; The first detection data and the concrete pile body quality defect detection result data are associated to form a training sample pair; The preset neural network is iteratively trained by using the training sample pair to obtain a pile construction quality real-time detection AI model capable of generating a concrete pile body quality defect prediction result according to the first detection data; The first detection data and the concrete pile body quality defect detection result data obtained by analyzing the second detection data are associated to form a training sample pair, which comprises the following steps: The concrete pile body quality defect detection result data is segmented into multiple concrete pile body quality defect detection result data segments according to the pile depth, and each concrete pile body quality defect detection result data segment comprises at least one quality state record, wherein: when a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defect-free state, the quality state record comprises a defect-free assignment; when a certain quality state record indicates that the corresponding concrete pile body quality defect detection result data segment is in a defective state, the quality state record comprises a defect type assignment, a defect bottom depth assignment and a defect top depth assignment, the defect type assignment indicates the defect type, the defect bottom depth assignment indicates the defect bottom depth position, and the defect top depth assignment indicates the defect top depth position; The first detection data is segmented into multiple first detection data segments according to the pile depth, and each first detection data segment is stored as a depth-indexed sequence or array data structure; The concrete pile body quality defect detection result data segments and the first detection data segments are aligned one by one to ensure the spatial consistency of each concrete pile body quality defect detection result data segment and each first detection data segment in the pile depth direction; From each first detection data segment, a uniform-dimension numerical feature vector is extracted as the input feature of the training sample, and the quality state record contained in the concrete pile body quality defect detection result data segment aligned with the first detection data segment to which the numerical feature vector belongs is taken as the true label of the numerical feature vector. The numerical feature vector and the true label are bound to form a training sample pair.

2. The pile construction quality real-time detection AI model training method of claim 1, wherein: During the first acoustic transmission method pile detection operation, the sound wave parameters of the cast-in-place concrete layer under the rising concrete pouring surface are collected in real time to form the first detection data.

3. The pile construction quality real-time detection AI model training method of claim 2, wherein: The first acoustic wave transmission method pile detection device used in the first acoustic wave transmission method pile detection operation comprises an acoustic wave detector, an acoustic wave transmitting probe, an acoustic wave receiving probe, a probe hanging mechanism, and a control system. The acoustic wave transmitting probe and the acoustic wave receiving probe are respectively connected with the acoustic wave detector in signal and are respectively arranged in the first acoustic measuring pipe and the second acoustic measuring pipe in a liftable manner through the probe hanging mechanism. The first acoustic measuring pipe and the second acoustic measuring pipe are arranged on the two sides of the pile hole in a vertical direction.

4. The pile construction quality real-time detection AI model training method of claim 3, wherein: The first acoustic wave transmission method pile detection operation and the second acoustic wave transmission method pile detection operation use the same acoustic wave transmission method pile detection device to ensure that the first detection data and the second detection data are consistent in depth reference. 5.The method of claim 3, wherein the method further comprises: determining the AI model based on the first data and the second data. During the first acoustic wave transmission method pile detection operation, the control system controls the acoustic wave transmitting probe and the acoustic wave receiving probe through the probe hanging mechanism to move synchronously from bottom to top at a lifting speed matched with the rising speed of the concrete pouring surface.

6. The pile construction quality real-time detection AI model training method of claim 1, wherein: The types of the acoustic wave parameters of the first detection data include any one or several of the following: acoustic time, acoustic velocity, peak amplitude, signal energy, main frequency, spectral bandwidth, attenuation coefficient, signal signal-to-noise ratio, waveform features extracted from original waveform data, and power spectral density features extracted from power spectral density (PSD) data. Furthermore, the defect types include any one or several of the following: hole, segregation, sediment, mud, reduced diameter, and expanded diameter.

7. The pile construction quality real-time detection AI model training method of claim 1, wherein: The preset neural network comprises: a learnable preprocessing layer for preprocessing the first detection data, the parameters of the learnable preprocessing layer being learned and optimized together with other parameters of the preset neural network during the training process of the preset neural network; a feature extraction layer for extracting features from the preprocessed first detection data, the feature extraction layer at least comprising a one-dimensional convolution layer; a sequence modeling layer for capturing long-distance dependencies in the feature sequence, the sequence modeling layer at least comprising a long short-term memory (LSTM) layer or a gated recurrent unit (GRU) layer; and an output layer for generating a concrete pile body quality defect prediction result based on the features and the long-distance dependencies, the output layer comprising a classification output branch for predicting the defect type and a regression output branch for predicting the defect depth position.

8. A method for real-time detection of pile construction quality, characterized in that: The method comprises: receiving first detection data, the first detection data being obtained through the first acoustic wave transmission method pile detection operation when the concrete is poured, and comprising acoustic wave parameters of the cast-in-place concrete pile body at different depths; inputting the first detection data into the pile construction quality real-time detection AI model for processing to obtain a concrete pile body quality defect prediction result; wherein the pile construction quality real-time detection AI model is obtained by training the pile construction quality real-time detection AI model using the training method of the pile construction quality real-time detection AI model according to any one of claims 1-7.

9. A real-time pile construction quality detection device, characterized in that: The base pile construction quality real-time detection device comprises a processor coupled with a memory for storing a computer program or instructions, and the processor is used for executing the computer program or instructions in the memory, so that the base pile construction quality real-time detection device executes the base pile construction quality real-time detection method as claimed in claim 8.

Images