Cast-in-place pile integrity detection system based on multi-mode ultrasonic array

Abstract

The invention relates to the technical field of cast-in-place pile detection, and discloses a cast-in-place pile integrity detection system based on a multi-mode ultrasonic array, and the system comprises a deformable multi-mode ultrasonic probe which is used for adaptively adjusting the posture through an intelligent material driver; the intelligent coupling injection module is used for automatically cleaning the inner wall of an internal channel of the cast-in-place pile; the multi-source data processing module is used for receiving the original ultrasonic waveform data and the real-time position information of the ultrasonic probe; the deep learning recognition module is used for generating a recognition result including defect categories and preliminary features; the defect positioning and evaluating module is used for constructing a defect model comprising defect space coordinates, geometric dimensions and severity parameters; and the diagnosis decision module is used for providing targeted repair suggestions. The ultrasonic signals are transmitted and received through the integrated longitudinal wave transducer, the transverse wave transducer and the Rayleigh wave transducer, the internal structure information of the pile body is obtained, in this way, the ultrasonic beam coverage range is expanded, and the detection blind area of the area between the sounding pipes in a traditional method is effectively made up.

Description

Technical Field

[0001] This invention relates to the field of cast-in-place pile testing technology, specifically a cast-in-place pile integrity testing system based on a multimodal ultrasonic array. Background Technology

[0002] Cast-in-place piles are a common type of foundation in civil engineering, widely used in high-rise buildings, bridges, ports, and other structures. Their construction quality directly determines the safety and durability of the project. However, due to complex geological conditions and the difficulty in controlling construction techniques, cast-in-place piles are highly susceptible to various defects during the pile formation process, such as segregation, diameter reduction, mud inclusion, voids, or cracks. If these internal defects are not detected and assessed in a timely and accurate manner, they will pose serious threats to structural safety and may even lead to engineering accidents.

[0003] To ensure the construction quality of cast-in-place piles, the engineering community has developed various testing technologies. Among them, ultrasonic testing is widely used due to its non-destructive and efficient characteristics. Traditional ultrasonic testing is often carried out through a sonic logging tube pre-embedded in the pile body. During testing, an ultrasonic transducer is placed in the sonic logging tube, and information about the interior of the pile is obtained by emitting and receiving ultrasonic waves. Some methods also use sound wave transmission or reflection methods, relying on the sound wave propagation time or echo signal intensity to determine the integrity of the pile body.

[0004] However, existing ultrasonic testing technologies rely on acoustic logging tubes, which makes it difficult for the ultrasonic beam to cover large areas between the tubes. This often leads to the omission of internal cracks or voids. Deposits or attached air bubbles on the inner wall of the acoustic logging tube, or non-ideal contact between the probe and the medium, can all cause severe attenuation or distortion of the ultrasonic signal. Furthermore, the analysis and interpretation of the large amount of ultrasonic data collected is time-consuming and labor-intensive, and highly susceptible to the influence of the operator's subjective judgment. Therefore, this invention provides a cast-in-place pile integrity testing system based on a multimodal ultrasonic array to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a cast-in-place pile integrity detection system based on a multimodal ultrasonic array, which solves the problems of limited detection range, insufficient accuracy, low automation, and strong dependence on construction conditions in the detection of defects in cast-in-place piles.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a cast-in-place pile integrity detection system based on a multimodal ultrasonic array, comprising:

[0008] A deformable multimodal ultrasonic probe is designed for deployment within the internal channels of cast-in-place piles (e.g., pre-embedded acoustic logging tubes or boreholes). It integrates a smart material actuator that receives external control signals to drive the ultrasonic transducer units within the probe to perform millimeter-level radial displacement and tilt adjustments. This adaptive adjustment allows the probe to close minute gaps with the inner wall of the channel, achieving automatic adaptive adjustment of the probe's posture and ensuring stable contact between the ultrasonic transducer and the channel wall. The probe contains multiple ultrasonic transducer units, including longitudinal wave transmitter-receiver wafers, transverse wave transmitter-receiver wafers, and surface wave excitation-receiver wafers. These wafers can simultaneously transmit and receive ultrasonic signals in longitudinal, transverse, and Rayleigh wave modes. By independently controlling the amplitude and phase of the excitation signal for each ultrasonic transducer unit, a beamforming controller can achieve fan-shaped or holographic scanning modes of the ultrasonic beam to generate raw ultrasonic waveform data containing multidimensional acoustic features.

[0009] The intelligent coupling injection module is used to pre-treat and inject coupling medium into the inner wall of the channel inside the cast-in-place pile before the ultrasonic probe emits ultrasonic signals. Specifically, it includes a high-pressure microfluidic jetting device for automatically cleaning the inner wall of the channel during the descent of the ultrasonic probe, removing adhering mud, air bubbles, or sediment. Simultaneously, the module is equipped with a composite gel storage unit for storing a composite gel with adjustable acoustic properties. This gel precisely adjusts its acoustic impedance by incorporating micron-sized air bubbles or nano-sized particles into the matrix material, matching it to the acoustic impedance of the concrete. The pumping device precisely controls the injection volume and rate of the composite gel based on the descent speed of the ultrasonic probe and the inner diameter of the channel inside the cast-in-place pile, ensuring the formation of a bubble-free, continuous coupling layer between the probe and the inner wall of the channel, thereby constructing a stable acoustic coupling environment. The composite gel remains in a fluid state for a set time to ensure full filling and gradually solidifies or decomposes after testing, facilitating subsequent processing.

[0010] The multi-source data processing module receives and preprocesses various types of data acquired by the system. The data acquisition unit synchronously acquires the raw waveform data, frequency spectrum information, and phase information generated by the deformable multimodal ultrasonic probe. The real-time position tracking unit continuously acquires the real-time depth, rotation angle, and deformation parameters of the intelligent material actuator within the internal channel of the cast-in-place pile. The data preprocessing unit performs wavelet transform noise reduction, baseline correction, and amplitude normalization on these raw ultrasonic waveform data to improve signal quality and data consistency, forming a preprocessed dataset for subsequent analysis.

[0011] The deep learning-based identification module performs in-depth analysis on the preprocessed dataset to identify defects. It employs a hybrid architecture combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). Specifically, the multimodal feature extraction layer processes longitudinal, transverse, and Rayleigh wave signals in parallel, extracting features of each mode in the time and frequency domains. The sequence information processing layer processes sequential features along the depth direction of the pile to capture the longitudinal continuity, distribution trends, and evolution patterns of defects. The attention weighting layer dynamically weights features from different ultrasonic modes and different depth locations through an attention mechanism, enabling the deep learning network to focus on key signal regions related to defects, thereby improving the accuracy of defect identification and generating identification results including defect category and preliminary features.

[0012] The defect location and assessment module is used to accurately locate and quantitatively assess the internal defects of the cast-in-place pile in three dimensions based on the defect identification results output by the deep learning recognition module and the real-time position information of the ultrasonic probe. Its three-dimensional spatial positioning algorithm unit combines generalized cross-correlation (GCC) or minimum variance distortion-free response (MVDR) beamforming algorithms, and utilizes the ultrasonic wave arrival time difference Δt between each transducer unit of the ultrasonic probe. jk Invert the spatial coordinates of the defect (X) d ,Y d Z d ).

[0013] The relationship between the propagation time T of ultrasound in the medium and the propagation distance L and the speed of sound c can be expressed as:

[0014]

[0015] Based on this, by analyzing the time difference of arrival of multiple received signals, the localization algorithm can determine the precise location of the defect. The sound velocity analysis unit calculates the average propagation velocity c of ultrasonic waves within the defect area. defect The sound velocity c in a normal concrete area concrete deviation rate To assess the density or degree of damage of materials. The amplitude attenuation analysis unit analyzes the ultrasonic amplitude attenuation coefficient α in the defect area. defect Compared with normal region α concrete The difference is used to estimate the severity or volume of the defect. The formula for calculating the attenuation coefficient α is as follows:

[0016]

[0017] Where L is the distance the ultrasonic wave travels, and A ref It is a reference amplitude (e.g., the received amplitude in a defect-free area), A obsThese are the observed amplitudes within or after passing through the defect area. Through these analyses, a defect model is constructed, including the defect's spatial coordinates, geometric dimensions, and severity parameters.

[0018] Diagnostic Decision Module: This module, based on the defect model built by the defect location and assessment module, utilizes a reinforcement learning (RL) framework for intelligent diagnosis and decision-making. Its environment construction unit uses the acoustic model of the cast-in-place pile as the state space S, and defect classification, severity rating, and targeted repair suggestions as the action space A. The algorithm training unit employs reinforcement learning algorithms such as Deep Q-Network (DQN), autonomously learning and optimizing defect diagnosis and repair suggestion strategies for the cast-in-place pile through interaction with the environment (i.e., training based on a large amount of simulated and actual detection data). Finally, the diagnostic report generation unit automatically generates a standardized comprehensive diagnostic report, which includes a detailed list of defects, an overall health assessment of the pile, and targeted repair suggestions based on the reinforcement learning model, such as recommended grouting range, grouting material type, or whether further supplementary testing is needed.

[0019] A second aspect of the present invention provides a method for detecting the integrity of cast-in-place piles based on a multimodal ultrasonic array. This method, applied to the aforementioned system, includes the following steps:

[0020] S1. Lower the deformable multimodal ultrasonic probe into the internal channel of the cast-in-place pile, and receive control signals through the intelligent material actuator to drive the ultrasonic transducer unit in the probe to perform radial displacement and tilt adjustment, so as to adaptively adjust the probe posture and achieve stable contact with the inner wall of the channel.

[0021] S2. Before the ultrasonic probe emits an ultrasonic signal, the high-pressure microfluidic jetting device in the intelligent coupling injection module automatically cleans the inner wall of the channel inside the grouting pile to remove adhering mud, air bubbles, or sediment. Subsequently, the pumping device injects acoustically adjustable composite gel from the composite gel storage unit between the probe and the inner wall of the channel to create a stable acoustic coupling environment.

[0022] S3. Through the ultrasonic transducer unit in the deformable multimodal ultrasonic probe, ultrasonic signals of longitudinal wave, transverse wave and Rayleigh wave modes are transmitted and received synchronously or alternately, and the original ultrasonic wave data containing multidimensional acoustic features are generated by combining the fan-shaped scanning or holographic scanning mode realized by the beamforming controller.

[0023] S4. The multi-source data processing module receives the raw ultrasonic waveform data and the real-time position information of the ultrasonic probe. The data preprocessing unit performs wavelet transform noise reduction, baseline correction, and amplitude normalization on the raw ultrasonic waveform data to form a preprocessed dataset for use by the deep learning module.

[0024] S5. The deep learning recognition module uses a hybrid convolutional neural network and long short-term memory network fusion architecture to extract deep features based on the obtained preprocessed dataset, and combines an attention mechanism to identify the types of defects inside the cast-in-place pile, generating recognition results including defect categories and preliminary features.

[0025] S6. The defect location and assessment module, based on the obtained defect identification results and the real-time position information of the ultrasonic probe, and combined with array signal processing algorithms (e.g., generalized cross-correlation function or minimum variance distortionless response beamforming algorithm), utilizes the ultrasonic wave arrival time difference between each transducer unit of the ultrasonic probe to invert the three-dimensional spatial coordinates (X) of the identified internal defects of the cast-in-place pile. d ,Y d Z d Simultaneously, the deviation rate between the average propagation velocity of ultrasound in the defect area and the sound velocity in the normal concrete area is calculated using the sound velocity analysis unit to assess the material's density or degree of damage. The difference between the ultrasound amplitude attenuation coefficient in the defect area and the normal area is analyzed using the amplitude attenuation analysis unit to estimate the defect's geometric dimensions and severity. Finally, a defect model is constructed, including the defect's spatial coordinates, geometric dimensions, and severity parameters.

[0026] S7. The diagnostic decision module, based on the constructed defect model, autonomously learns and optimizes diagnostic strategies using a reinforcement learning framework. It interacts with a predefined acoustic model environment of the cast-in-place pile and adjusts its strategy according to the obtained reward signals, ultimately generating a comprehensive diagnostic report and providing targeted repair suggestions.

[0027] This invention provides a cast-in-place pile integrity detection system based on a multimodal ultrasonic array. It possesses the following features:

[0028] Beneficial effects:

[0029] 1. This invention employs a deformable multimodal ultrasonic probe, whose integrated longitudinal wave, transverse wave, and Rayleigh wave transducers can synchronously or alternately transmit and receive ultrasonic signals of different modes, acquiring information about the internal structure of the pile from multiple physical dimensions. Combined with the probe's adaptive attitude adjustment achieved by an intelligent material actuator and the fan-shaped and holographic scanning modes of the beamforming controller, the coverage range of the ultrasonic beam is significantly expanded, effectively compensating for the detection blind spots in the area between acoustic tubes using traditional methods, and improving the ability to identify various complex defects and the accuracy of three-dimensional positioning.

[0030] 2. The intelligent coupling injection module introduced in this invention automatically cleans the inner wall of the channel inside the grouting pile through a high-pressure microfluidic jet device and precisely injects a composite gel with adjustable acoustic properties, ensuring optimal acoustic coupling between the probe and the pile body. This active and intelligent coupling mechanism effectively solves the problems of signal loss and misjudgment caused by poor coupling, medium attenuation, or bubble interference in traditional detection. The high signal-to-noise ratio raw ultrasonic waveform data provides high-quality input for accurate analysis by subsequent deep learning algorithms, thereby ensuring the stability and reliability of the entire detection process.

[0031] 3. The deep learning-based recognition and diagnostic decision-making modules constructed in this invention utilize a hybrid convolutional neural network and long short-term memory network fusion architecture for deep feature extraction and defect identification, and introduce an attention mechanism to focus on key signals. In particular, the application of a reinforcement learning framework enables the system to autonomously learn and optimize diagnostic strategies, progressing from identifying defect types and locating sizes to generating comprehensive diagnostic reports and providing targeted repair suggestions. This reduces reliance on human experience and improves diagnostic efficiency and the scientific rigor of decision-making. Attached Figure Description

[0032] Figure 1 This is a diagram of the integrity detection system for cast-in-place piles according to the present invention.

[0033] Figure 2 This is a schematic diagram of the deformable multimodal ultrasonic probe structure of the present invention;

[0034] Figure 3 This is a schematic diagram of the intelligent coupling injection module structure of the present invention;

[0035] Figure 4 This is a schematic diagram of the multi-source data processing module structure of the present invention;

[0036] Figure 5 This is a schematic diagram of the deep learning recognition module structure of the present invention;

[0037] Figure 6 This is a schematic diagram of the defect location and assessment module structure of the present invention;

[0038] Figure 7 This is a schematic diagram of the diagnostic decision module structure of the present invention;

[0039] Figure 8 This is a flowchart of the method for detecting the integrity of cast-in-place piles according to the present invention.

[0040] Among them, 100 is a deformable multimodal ultrasonic probe; 200 is an intelligent coupling injection module; 300 is a multi-source data processing module; 400 is a deep learning recognition module; 500 is a defect location assessment module; and 600 is a diagnostic decision module. Detailed Implementation

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] See attached document Figure 1 , Figure 1 This is an architecture diagram of a cast-in-place pile integrity testing system according to an embodiment of the present invention. The present invention provides a cast-in-place pile integrity testing system based on a multimodal ultrasonic array, comprising: a deformable multimodal ultrasonic probe 100, an intelligent coupling injection module 200, a multi-source data processing module 300, a deep learning recognition module 400, a defect location assessment module 500, and a diagnostic decision module 600.

[0043] A deformable multimodal ultrasonic probe 100 is deployed inside the channel of a cast-in-place pile. It adjusts its orientation using an internal intelligent material actuator to achieve close contact with the channel wall. The probe 100 simultaneously transmits and receives ultrasonic signals in longitudinal, transverse, and Rayleigh wave modes, generating raw ultrasonic waveform data.

[0044] The intelligent coupling injection module 200 works in conjunction with the deformable multimodal ultrasonic probe 100. Before the probe 100 transmits ultrasonic signals, the module 200 automatically cleans the inner wall of the channel inside the grouting pile. After cleaning, the module 200 injects a composite gel with adjustable acoustic properties to create a stable acoustic coupling environment between the probe 100 and the inner wall of the channel, ensuring effective transmission of ultrasonic signals.

[0045] The multi-source data processing module 300 receives raw ultrasonic waveform data generated by the deformable multimodal ultrasonic probe 100 and the real-time position information of the probe 100 through a data interface. The module 300 performs noise reduction and normalization processing on the received raw ultrasonic waveform data to form a preprocessed dataset for subsequent analysis.

[0046] The deep learning recognition module 400 receives a preprocessed dataset from the multi-source data processing module 300. Module 400 utilizes its internal hybrid convolutional neural network and long short-term memory network fusion architecture to perform deep feature extraction on the preprocessed dataset. Simultaneously, combining an attention mechanism, module 400 identifies the types of defects present within the cast-in-place pile and outputs recognition results containing defect categories and preliminary features.

[0047] The defect location assessment module 500 receives the defect identification results output by the deep learning recognition module 400 and the real-time position information of the ultrasonic probe 100 provided by the multi-source data processing module 300. Module 500 combines array signal processing algorithms to perform three-dimensional spatial location of the identified defects inside the cast-in-place pile. Furthermore, module 500 estimates the geometric dimensions and severity of the defects using an acoustic physical model, ultimately constructing a defect model that includes the defect's spatial coordinates, geometric dimensions, and severity parameters.

[0048] The diagnostic decision module 600 receives the defect model output by the defect location and evaluation module 500. Module 600 utilizes a reinforcement learning framework to autonomously learn and optimize its diagnostic strategy based on the defect model. Finally, module 600 generates a comprehensive diagnostic report and provides targeted remediation suggestions. This diagnostic report can be output to an external display or printing device via a data interface.

[0049] The workflow of this detection system is as follows: The deformable multimodal ultrasonic probe 100 first enters the internal channel of the cast-in-place pile and performs adaptive attitude adjustment. The intelligent coupling injection module 200 cleans the inner wall of the channel and injects coupling medium to establish an acoustic coupling channel. The probe 100 synchronously transmits and receives multimodal ultrasonic signals and transmits the raw data to the multi-source data processing module 300 for preprocessing. The preprocessed data is used by the deep learning recognition module 400 to identify the defect type. The defect location and assessment module 500 performs three-dimensional location and quantitative assessment of the defect based on the recognition results and probe position information, and constructs a defect model. Finally, the diagnostic decision module 600 uses reinforcement learning to perform intelligent diagnosis of the defect model and generates a comprehensive report and repair suggestions.

[0050] See attached document Figure 2 , Figure 2 This is a schematic diagram of a deformable multimodal ultrasonic probe according to an embodiment of the present invention. The deformable multimodal ultrasonic probe 100 includes multiple ultrasonic transducer units, a micro-driver, and a beamforming controller.

[0051] The ultrasonic transducer units are arranged in a ring or spiral array on the external structure of the probe 100. Each unit integrates a longitudinal wave transmitting and receiving crystal, a transverse wave transmitting and receiving crystal, and a surface wave excitation and receiving crystal. These crystals can selectively transmit or receive ultrasonic waves of different modes as needed.

[0052] A micro-actuator is connected to each ultrasonic transducer unit and receives control signals from the beamforming controller via an internal control circuit. This micro-actuator, fabricated from smart materials such as shape memory alloys or piezoelectric ceramics, is capable of generating precise millimeter-level displacements in response to control signals. Specifically, the micro-actuator enables radial displacement and tilt adjustment of the ultrasonic transducer unit, achieving adaptive fitting of the probe 100 within the internal channel of the cast-in-place pile. This adjustment bridges the minute gaps between the outer surface of the probe and the inner wall of the cast-in-place pile channel, ensuring full contact between the ultrasonic transducer unit and the coupling medium.

[0053] A beamforming controller is connected to each ultrasonic transducer unit to control the transmission and reception of ultrasonic waves. The controller independently controls the amplitude A of the excitation signal for each ultrasonic transducer unit. i and phase φ i By precisely adjusting these parameters, the beamforming controller can achieve fan-shaped or holographic scanning modes of the ultrasonic beam. For example, to achieve this at the target point (x... f ,y f ,z f A focal point is formed at (x), for those located at (x) i ,y i ,z i The excitation signal delay time t of the i-th transducer unit is given. d,i Calculated based on the propagation speed c of ultrasound in the medium:

[0054]

[0055] In the formula, t d,i The delay time of the excitation signal for the i-th transducer unit; (x f ,y f ,z f The spatial coordinates of the target point; (x i ,y i ,z i The spatial coordinates of the i-th transducer unit; c is the propagation speed of ultrasound in the medium.

[0056] By continuously or discretely adjusting the focus target point (x) f ,y f ,z f This allows for ultrasonic beam scanning of a specified area inside the cast-in-place pile.

[0057] Furthermore, the beamforming controller also controls the synchronous or alternating transmission of different modes of ultrasound. In receiving mode, the beamforming controller synchronously acquires and preprocesses the echo signals received by each ultrasonic transducer unit. For the received ultrasonic signal S... recv (t), which can be described as the transmitted signal S emit(t) Response of the medium through the pile The convolution result is then superimposed with noise N(t):

[0058]

[0059] In the formula, and D represents the spatial position vectors of the transmitting and receiving transducers, respectively; medium This represents the internal medium characteristics and defect distribution of the pile. By analyzing the amplitude, propagation time, and frequency characteristics of the received signal, acoustic feature data of different modes can be extracted.

[0060] See attached document Figure 3 , Figure 3 This is a schematic diagram of a smart coupling injection module according to an embodiment of the present invention. The smart coupling injection module 200 includes a high-pressure microfluidic jetting device, a composite gel storage unit, and a pumping device.

[0061] A high-pressure microfluidic jetting device is positioned below the deformable multimodal ultrasonic probe 100 or along the probe's movement path. During the lowering of the probe 100, this device jets a high-pressure microfluidic fluid (e.g., deionized water or diluted cleaning agent) into the inner wall of the channel within the cast-in-place pile. The jetting fluid has sufficient pressure to effectively remove mud, air bubbles, or sediment adhering to the channel wall, providing a clean interface for the subsequent injection of the coupling medium.

[0062] The composite gel storage unit is used to store composite gels with tunable acoustic properties. The composite gel is prepared by incorporating a specific proportion of micron-sized bubbles or nano-sized particles into a matrix material. By precisely controlling the type and content of these dopants, the density ρ of the composite gel can be adjusted. g and the speed of ultrasonic wave propagation c g This results in its acoustic impedance Z g As close as possible to the acoustic impedance Z of the cast-in-place pile concrete c A perfect match. The acoustic impedance Z is defined as:

[0063] Z = ρ·c;

[0064] In the formula, ρ is the density of the medium, and c is the propagation speed of ultrasound in the medium. By adjusting the Z-axis of the gel... g To minimize the reflection coefficient of ultrasound at the coupling interface

[0065] Reducing the R-axis helps maximize the energy of the ultrasonic waves transmitted from the probe through the coupling medium into the pile, ensuring effective signal transmission. The composite gel is designed to remain fluid for a set time to ensure complete filling of minute gaps, and to gradually solidify or decompose after testing for easy subsequent processing.

[0066] The pumping device is connected to the composite gel storage unit and receives control signals from the system control unit. Based on the descent speed of the deformable multimodal ultrasonic probe 100 and the inner diameter of the channel within the grouting pile, the pumping device precisely controls the injection volume and rate of the composite gel. Through a precision pumping mechanism such as a peristaltic pump or syringe pump, the composite gel is ensured to be injected at a constant flow rate into the annular gap between the probe 100 and the inner wall of the channel, forming a continuous, bubble-free coupling layer, thereby constructing a stable acoustic coupling environment.

[0067] See attached document Figure 4 , Figure 4 This is a schematic diagram of a multi-source data processing module according to an embodiment of the present invention. The multi-source data processing module 300 includes a data acquisition unit, a real-time location tracking unit, and a data preprocessing unit.

[0068] The data acquisition unit is connected to the data output port of the deformable multimodal ultrasonic probe 100. This unit synchronously receives and acquires the raw ultrasonic waveform data, frequency spectrum information, and phase information generated by the probe 100 during the detection process. This data is stored in digital format and used as input for subsequent processing.

[0069] The real-time position tracking unit is connected to the position sensor (e.g., encoder, inclinometer, gyroscope) in the deformable multimodal ultrasonic probe 100. This unit acquires the depth D of the probe 100 within the internal channel of the cast-in-place pile in real time. p Rotation angle α p and the deformation parameters of each micro actuator (e.g., radial displacement Δr) i and tilt angle β i This position and orientation information is crucial for the subsequent three-dimensional localization of defects.

[0070] The data preprocessing unit receives the raw ultrasonic waveform data transmitted from the data acquisition unit. This unit performs a series of preprocessing operations on the raw ultrasonic waveform data. First, wavelet transform denoising or similar algorithms are used to remove environmental and system noise from the signal. Second, baseline correction is performed to eliminate DC bias or baseline drift. Finally, amplitude normalization is performed to unify the waveform amplitude to a preset range, such as [0,1]. The normalized signal W′ norm (t) can be represented as:

[0071]

[0072] In the formula, W(t) is the original waveform data, W min and W maxThese are the minimum and maximum values ​​of the waveform data within a specific time period. After these preprocessing steps, the data preprocessing unit generates a high-quality preprocessed dataset, which is then transmitted to the deep learning recognition module 400.

[0073] See attached document Figure 5 , Figure 5 This is a schematic diagram of a deep learning recognition module structure according to an embodiment of the present invention. The deep learning recognition module 400 receives a preprocessed dataset output by the multi-source data processing module 300 and includes a multimodal feature extraction layer, a sequence information processing layer, and an attention weighting layer.

[0074] The multimodal feature extraction layer processes the P-wave, S-wave, and Rayleigh wave signals contained in the preprocessed dataset in parallel. This layer employs multiple independent convolutional neural network branches, each processing ultrasound signals for a specific modality. The convolutional layer performs convolution operations on the input signal through filters to extract the time-domain and frequency-domain features of each modality. For example, for the input feature map X, the output feature map Y of the convolution operation can be represented as:

[0075] Y i,j =∑ m ∑ n X i+m,j+n ·K m,n ;

[0076] In the formula, K is the convolution kernel; Y t,j X represents the value of the output feature map at position (t,j) of the convolution operation; i+m,j+n The input feature map has the value at position (i+m, j+n); K m,n Σ is the value of the convolution kernel at position (m,n); m Σ n This represents the summation operation applied to all elements of the convolution kernel. Through multiple layers of convolution and pooling operations, local and global features in the signal are extracted, such as waveform envelope, spectral energy distribution, and specific frequency components.

[0077] The sequence information processing layer receives feature data output from the multimodal feature extraction layer. This layer employs a Long Short-Term Memory (LSTM) network or other recurrent neural network structures to process the sequential features along the depth direction of the cast-in-place pile. The LSTM network, through its unique gating mechanism (input gate, forget gate, output gate), can capture the longitudinal continuity, distribution trend, and evolution law of defects. For the input x at time step t... t The hidden state h of the previous time step t-1 The LSTM unit calculates the current hidden state h. t :

[0078] h t =LSTM(x t ,ht-1 );

[0079] This structure helps to identify and correlate multiple signal features that may indicate the same defect and appear at different depth locations.

[0080] The attention-weighted layer follows the sequence information processing layer. This layer dynamically weights features from different ultrasonic modalities and depth locations through an attention mechanism. The attention mechanism calculates a weight distribution that indicates which feature regions the model should focus on during defect identification. For example, by learning an attention weight vector 'a', different features can be weighted and summed to obtain the final fused feature vector F. fused :

[0081]

[0082] In the formula, f k These are different feature vectors (e.g., features from different modalities or different depth locations); F fused This represents the final fused feature vector; N is the total number of feature vectors; a k The corresponding attention weight represents the importance of the k-th feature vector. This dynamic weighting mechanism enables the deep learning network to focus on key signal regions related to defects, thereby improving the accuracy and robustness of defect identification. Finally, the deep learning identification module 400 outputs identification results including defect category and preliminary features, and transmits them to the defect localization and evaluation module 500.

[0083] See attached document Figure 6 , Figure 6 This is a schematic diagram of a defect location assessment module according to an embodiment of the present invention. The defect location assessment module 500 receives the defect identification result output by the deep learning recognition module 400 and the real-time position information of the ultrasonic probe provided by the multi-source data processing module 300. The module 500 includes a three-dimensional spatial positioning algorithm unit, a sound velocity analysis unit, and an amplitude attenuation analysis unit.

[0084] The three-dimensional spatial positioning algorithm unit utilizes the time difference of ultrasonic wave arrival between the 100 transducer units of the ultrasonic probe to invert the three-dimensional spatial coordinates (X, Y, F) of the identified internal defects in the cast-in-place pile. d ,Y d Z d This unit can employ the generalized cross-correlation GCC function or the minimum variance distortion-free response (MVDR) beamforming algorithm.

[0085] When using the generalized cross-correlation (GCC) function, it is used to estimate the time delay τ between two received signals x(t) and y(t). The GCC function R... xy (τ) is defined as:

[0086]

[0087] In the formula, X(f) and Y(f) are the Fourier transforms of x(t) and y(t) respectively, Y*(f) is the complex conjugate of Y(f), and W(f) is the weighting function. By finding R... xy The peak value of (τ) can be used to obtain the precise time difference Δt between the signal propagation from the defect to different transducer units. jk By combining the known position of the ultrasonic transducer unit and the propagation speed of ultrasonic waves in the medium, the spatial coordinates of the defect can be determined using triangulation or multi-point positioning methods.

[0088] When employing the Minimum Variance Distortionless Response (MVDR) beamforming algorithm, it minimizes the output noise power by optimizing the array weight vector w, while maintaining a gain of 1 in a specific direction. Its output power P MVDR Represented as:

[0089] P MVDR =w H R x w;

[0090] In the formula, R x It is the covariance matrix of the received signal, w H It is the conjugate transpose of w. By traversing spatial directions and finding the direction corresponding to the minimum output power, the defect source can be located with high resolution.

[0091] Based on the defect location results, the sound velocity analysis unit calculates the average propagation velocity c of ultrasonic waves within the defect area. defect By comparing c defect The sound velocity c in a normal concrete area concrete deviation rate Δc rate To assess the density or degree of damage of the material. The deviation rate is defined as:

[0092]

[0093] A large positive deviation rate usually indicates a decrease in sound velocity in the defect area, a decrease in material density, or the presence of damage.

[0094] The amplitude attenuation analysis unit analyzes the amplitude attenuation of ultrasonic waves after passing through a defect area. This unit calculates the ultrasonic amplitude attenuation coefficient α in the defect area. defect The formula for calculating the attenuation coefficient α is as follows:

[0095]

[0096] In the formula, L is the propagation distance of the ultrasonic wave along a specific path, and A... refIt is a reference amplitude (e.g., the received amplitude in a defect-free or normal area), A obs It refers to the observed amplitude in or after passing through the defect region. By comparing α defect Attenuation coefficient α of the normal region concrete The differences can be used to estimate the geometric dimensions and severity of the defect. Finally, the defect location assessment module 500 constructs a defect model including the defect's spatial coordinates, geometric dimensions, and severity parameters, and transmits it to the diagnostic decision module 600.

[0097] See attached document Figure 7 , Figure 7 This is a schematic diagram of a diagnostic decision module structure according to an embodiment of the present invention. The diagnostic decision module 600 receives a defect model output by the defect localization and evaluation module 500. The module 600 includes a reinforcement learning framework, a diagnostic report generation unit, and a repair suggestion generation unit.

[0098] The reinforcement learning framework receives a defect model and uses it as the state input to the environment. The framework defines the defect states of the cast-in-place pile (e.g., the size of voids, the length of cracks, the degree of segregation, etc.) as the state space S. The defined action space A contains a series of possible diagnostic and repair strategies, such as minor defects (no treatment required), moderate defects (local repair), severe defects (reinforcement required), or extremely severe defects (reconstruction recommended), etc. The framework learns the optimal diagnostic strategy by interacting with the simulated environment, performing a series of actions, and observing the results. At each time step t, the agent starts from the current state s. t Choose an action a t After this action is performed, the environment transitions to the new state s. t+1 and receive a reward r t The reward function R(s,a) is designed to guide the model in learning effective diagnostic and remedial decisions. For example, accurately identifying the type and severity of defects will result in a positive reward, while misclassification or missed detection will result in a negative reward. The goal of reinforcement learning is to learn a policy π(s) that maximizes the long-term cumulative reward. The discounted cumulative reward G... t The calculation method is as follows:

[0099]

[0100] In the formula, r t+k+1 γ is the reward at future time step t+k+1; γ is the discount factor (0≤γ≤1); T is the length of the trajectory. A reinforcement learning framework is used to train a decision model using algorithms such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) to generate corresponding diagnostic results and remediation suggestions based on the specific parameters of the defect model.

[0101] The diagnostic report generation unit automatically generates a comprehensive diagnostic report based on the diagnostic results output by the reinforcement learning framework. This report includes an overall integrity assessment of the cast-in-place pile, a detailed list of identified defects (including type, 3D location, geometry, and severity), comparative analysis of historical inspection data (if available), and a confidence assessment of the inspection results. The report is output in structured text and charts for easy user reading.

[0102] The repair suggestion generation unit receives the decision results from the reinforcement learning framework and, combined with a pre-set repair knowledge base, provides targeted repair suggestions for each identified defect. These suggestions include, but are not limited to, specific construction schemes such as local grouting, crack repair, grouting reinforcement, or replacement, and may include recommended materials and construction techniques. Suggestions are intelligently matched based on the type, size, severity of the defect, and the structural importance of the cast-in-place pile. The diagnostic report and repair suggestions are ultimately output through the system interface.

[0103] See attached document Figure 8 , Figure 8 This is a flowchart of a method for detecting the integrity of cast-in-place piles according to an embodiment of the present invention. The present invention provides a method for detecting the integrity of cast-in-place piles based on a multimodal ultrasonic array, the specific implementation steps of which are as follows:

[0104] Step S1: Place the deformable multimodal ultrasonic probe 100 at the entrance of the channel inside the grouting pile to be tested. During the lowering process, the probe 100 adaptively adjusts the radial position and tilt angle of the ultrasonic transducer unit according to the geometry and size of the channel through its internal micro-actuator, ensuring that the outer surface of the probe 100 fits tightly against the inner wall of the channel.

[0105] Step S2: During the lowering of the probe 100, the high-pressure microfluidic jetting device of the intelligent coupling injection module 200 performs high-pressure microfluidic jetting on the inner wall of the channel inside the grouting pile to remove adhering substances and air bubbles. After cleaning, the pumping device injects an acoustically adjustable composite gel into the annular space between the probe 100 and the inner wall of the channel according to the position of the probe 100 and the inner diameter of the channel, thus constructing a stable acoustic coupling medium layer.

[0106] Step S3: The beamforming controller of the deformable multimodal ultrasonic probe 100 controls the ultrasonic transducer unit to synchronously transmit and receive ultrasonic signals in longitudinal, transverse, and Rayleigh wave modes. As the probe 100 moves along the depth direction, multimodal ultrasonic shape data is continuously acquired. Simultaneously, the real-time position tracking unit of the multi-source data processing module 300 synchronously acquires the real-time depth, rotation angle, and attitude information of the probe 100 inside the cast-in-place pile.

[0107] Step S4: The data acquisition unit of the multi-source data processing module 300 receives the raw ultrasonic waveform data and real-time position information from the probe 100. The data preprocessing unit performs noise reduction processing on the raw ultrasonic waveform data to remove environmental noise and system noise; then performs baseline correction; subsequently, it performs amplitude normalization processing to form a preprocessed dataset for subsequent analysis.

[0108] Step S5: The deep learning recognition module 400 receives the preprocessed dataset output by the multi-source data processing module 300. The multimodal feature extraction layer extracts features from each modality of the ultrasonic signal in parallel; the sequence information processing layer processes the serialized features to capture the longitudinal information of the defects; and the attention weighting layer dynamically weights different features. Finally, module 400 identifies the types of defects present inside the cast-in-place pile and outputs recognition results containing the defect category and preliminary features.

[0109] Step S6: The defect location and assessment module 500 receives the defect identification results output by the deep learning recognition module 400 and the real-time position information of the ultrasonic probe 100 provided by the multi-source data processing module 300. The three-dimensional spatial positioning algorithm unit uses the ultrasonic time difference of arrival or beamforming algorithm to perform three-dimensional spatial positioning of the identified defect. The sound velocity analysis unit analyzes the ultrasonic velocity deviation in the defect area, and the amplitude attenuation analysis unit analyzes the ultrasonic amplitude attenuation. Based on the above analysis results, module 500 estimates the geometric size and severity of the defect and constructs a defect model.

[0110] Step S7: The diagnostic decision module 600 receives the defect model output by the defect location and evaluation module 500. The reinforcement learning framework autonomously learns and optimizes the diagnostic strategy based on the defect model, generating corresponding diagnostic results. The diagnostic report generation unit automatically generates a comprehensive diagnostic report based on the diagnostic results, and the repair suggestion generation unit provides targeted repair suggestions. This diagnostic report and repair suggestions can be output through a data interface.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cast-in-place pile integrity detection system based on a multimodal ultrasonic array, characterized in that, include: A deformable multimodal ultrasonic probe is used to adaptively adjust its attitude in the internal channel of a cast-in-place pile through a smart material actuator, and simultaneously transmit and receive longitudinal wave, transverse wave and Rayleigh wave mode ultrasonic signals to generate raw ultrasonic waveform data containing multidimensional acoustic features. The intelligent coupling injection module is used to automatically clean the inner wall of the channel inside the grouting pile before the ultrasonic probe emits ultrasonic signals, and inject a composite gel with adjustable acoustic properties to create an acoustic coupling environment. The multi-source data processing module is used to receive the original ultrasonic waveform data and the real-time position information of the ultrasonic probe, and to perform noise reduction and standardization processing on the original ultrasonic waveform data to form a preprocessed dataset. The deep learning recognition module is used to perform deep feature extraction based on the preprocessed dataset using a hybrid convolutional neural network and long short-term memory network fusion architecture, and to identify the defect types inside the cast-in-place pile by combining an attention mechanism, generating recognition results including defect categories and preliminary features. The defect location assessment module is used to locate the internal defects of the identified cast-in-place pile in three dimensions based on the defect identification results and the real-time position information of the ultrasonic probe, combined with the array signal processing algorithm. It also estimates the geometric size and severity through an acoustic physical model and constructs a defect model including the spatial coordinates, geometric size and severity parameters of the defect. The diagnostic decision module is used to autonomously learn and optimize diagnostic strategies based on the defect model using a reinforcement learning framework, generate a comprehensive diagnostic report, and provide targeted remediation suggestions.

2. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The deformable multimodal ultrasonic probe includes: An ultrasonic transducer unit is used to integrate a longitudinal wave transmitting and receiving chip, a transverse wave transmitting and receiving chip, and a surface wave excitation and receiving chip. A miniature actuator is used to adjust the radial displacement and tilt angle of the ultrasonic transducer unit to adaptively conform to the inner wall of the channel inside the grouting pile. A beamforming controller is used to perform fan-shaped or holographic scanning of an ultrasonic beam by independently controlling the amplitude and phase of the excitation signal of the ultrasonic transducer unit.

3. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 2, characterized in that, Upon receiving a control signal, the micro-driver causes the ultrasonic transducer unit to make millimeter-level displacement adjustments within the internal channel of the grouting pile, thereby bridging the tiny gaps between the probe and the inner wall of the grouting pile channel.

4. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The intelligent coupling injection module includes: High-pressure microfluidic jetting device is used to automatically clean the inner wall of the channel inside the grouting pile during the lowering of the ultrasonic probe, removing attached mud, air bubbles or sediment; A composite gel storage unit is used to store composite gels with tunable acoustic properties, and the acoustic impedance is adjusted by incorporating micron-sized bubbles or nano-sized particles into the matrix material. A pumping device is used to control the injection volume and rate of the composite gel based on the descent speed of the ultrasonic probe and the inner diameter of the channel inside the grouting pile, ensuring the formation of a bubble-free continuous coupling layer.

5. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 4, characterized in that, The composite gel in the intelligent coupling injection module remains in a fluid state for a set time to ensure full filling, and gradually solidifies or decomposes after the test is completed.

6. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The multi-source data processing module includes: The data acquisition unit is used to synchronously acquire the raw waveform data, frequency spectrum information, and phase information generated by the ultrasound probe. The real-time position tracking unit is used to acquire the real-time depth, rotation angle, and deformation parameters of the intelligent material actuator of the ultrasonic probe within the internal channel of the grouting pile. The data preprocessing unit is used to perform wavelet transform noise reduction, baseline correction, and amplitude normalization on the raw ultrasonic waveform data.

7. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The deep learning recognition module includes: A multimodal feature extraction layer is used to process longitudinal wave, transverse wave and Rayleigh wave signals in parallel and extract the time domain and frequency domain features of each mode; The sequence information processing layer is used to process the serialized features along the depth direction of the cast-in-place pile, and to capture the longitudinal continuity, distribution trend and evolution law of defects. The attention-weighted layer is used to dynamically weight features of different ultrasound modalities and features at different depth locations through an attention mechanism, so that the deep learning network can focus on the key signal regions of the defect.

8. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The defect location and assessment module includes: The three-dimensional spatial positioning algorithm unit is used to combine the generalized cross-correlation or minimum variance distortionless response beamforming algorithm and utilize the ultrasonic arrival time difference between each transducer unit of the ultrasonic probe to invert the spatial coordinates of the defect. The sound velocity analysis unit is used to assess the density or degree of damage of a material by calculating the deviation rate between the average propagation velocity of ultrasonic waves in the defect area and the sound velocity in the normal concrete area. The amplitude attenuation analysis unit is used to estimate the severity or volume of a defect by analyzing the difference between the ultrasonic amplitude attenuation coefficient in the defect area and that in the normal area.

9. The cast-in-place pile integrity detection system based on a multimodal ultrasonic array according to claim 1, characterized in that, The diagnostic decision module includes: The environment building unit is used to define the acoustic model of the cast-in-place pile as the state space, and the defect classification, severity rating and repair suggestions as the action space. The algorithm training unit is used to autonomously learn and optimize the defect diagnosis and repair suggestion strategy for cast-in-place piles through interaction with the environment using a deep Q-network; The diagnostic report generation unit is used to automatically generate standardized inspection reports, including a list of defects, pile health assessment, and repair recommendations based on a reinforcement learning model.

10. A method for detecting the integrity of cast-in-place piles based on a multimodal ultrasonic array, applied to the cast-in-place pile integrity detection system based on a multimodal ultrasonic array as described in any one of claims 1-9, characterized in that, Includes the following steps: S1. Lower the deformable multimodal ultrasonic probe into the internal channel of the grouting pile, and adaptively adjust the probe posture through the intelligent material actuator; S2. Before the ultrasonic probe emits an ultrasonic signal, the intelligent coupling injection module automatically cleans the inner wall of the channel inside the grouting pile and injects a composite gel with adjustable acoustic properties to construct an acoustic coupling environment. S3. Simultaneously transmit and receive longitudinal wave, transverse wave and Rayleigh wave mode ultrasonic signals through a deformable multimodal ultrasonic probe to generate raw ultrasonic waveform data containing multidimensional acoustic features. S4. Receive the raw ultrasonic waveform data and the real-time position information of the ultrasonic probe, and perform noise reduction and standardization on the raw ultrasonic waveform data to form a preprocessed dataset. S5. Based on the preprocessed dataset, deep feature extraction is performed using a hybrid convolutional neural network and long short-term memory network fusion architecture, and attention mechanism is combined to identify the defect types inside the cast-in-place pile, generating identification results including defect categories and preliminary features. S6. Based on the defect identification results and the real-time position information of the ultrasonic probe, the internal defects of the identified cast-in-place pile are located in three-dimensional space by combining the array signal processing algorithm. The geometric size and severity are estimated by the acoustic physical model, and a defect model including the spatial coordinates, geometric size and severity parameters of the defect are constructed. S7. Based on the constructed defect model, the system uses a reinforcement learning framework to learn and optimize diagnostic strategies, generate a comprehensive diagnostic report, and provide targeted remediation suggestions.

Images