Defect detection method and system for prefabricated reinforced concrete pile

Abstract

The invention relates to the technical field of industrial automatic monitoring and control, in particular to a defect detection method and system for a prefabricated reinforced concrete pile, and the method comprises the steps: obtaining a pile body vibration signal, and carrying out the wavelet packet decomposition; calculating a kurtosis value and an energy concentration factor of each node; fusing the kurtosis value and the energy concentration factor to obtain a defect likelihood value; performing an adaptive optimal basis search based on the defect likelihood value to determine an optimal basis that highlights the defect signal; and reconstructing a defect signal based on the optimal basis so as to extract and analyze the defect. According to the invention, through multi-feature fusion and adaptive search, defect features can be effectively separated from strong background interference and a response action can be triggered, so that the adaptive ability, the detection sensitivity and the automation level of quality monitoring of the monitoring device in a complex industrial environment are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation monitoring and control technology, and in particular to a defect detection method and system for precast reinforced concrete piles. Background Technology

[0002] In the industrialized automated monitoring and non-destructive testing of precast reinforced concrete piles, stress waves are typically emitted into the pile body using automated acquisition devices, and the echo signals received by sensors are then analyzed to determine whether defects such as cracks and voids exist within the pile. However, in complex industrial construction or production environments, the actual received echo signals are highly complex and non-stationary signals, containing both high-energy low-frequency global resonance signals from the pile itself and weak high-frequency transient impact signals caused by defects. Therefore, accurately extracting weak defect features from strong background interference becomes a key point in the hardware and control logic design of the monitoring system.

[0003] Existing technologies typically employ time-frequency analysis methods such as wavelet packet transform to decompose complex signals into multiple fine frequency bands for analysis. The core of this approach lies in how to select the optimal basis that can highlight defects from a large number of frequency bands.

[0004] Currently, the control logic of existing monitoring devices generally relies on a single mathematical criterion, and its logic thresholds are usually preset at the factory, making real-time adjustment impossible based on actual working conditions. For example, the commonly used minimum Shannon entropy criterion, derived from information theory, aims to find the most concentrated signal energy and the most ordered expression of information. However, in the automated monitoring and detection scenario of precast reinforced concrete piles, the most concentrated energy is usually the resonant signal of the pile itself, while the transient impact caused by defects, due to its extremely low energy proportion, is often misjudged as disordered noise under the rigid entropy criterion logic. It is evident that this unidirectional control logic based on fixed parameters results in the monitoring system lacking adaptive adjustment capabilities for complex working conditions. The ultimately selected optimal basis may overwhelm subtle defect characteristics, leading to low sensitivity and frequent missed detections in automated monitoring devices, making it difficult to meet the automated requirements of full-process quality control in industrial settings. Summary of the Invention

[0005] To address the technical problems of existing monitoring devices, such as weak adaptive control capabilities, rigid detection logic, and difficulty in real-time calibration in complex industrial environments, which lead to missed defect detection, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for detecting defects in precast reinforced concrete piles, the method comprising the steps of: The vibration response signal of the pile is acquired and decomposed into a wavelet packet tree containing multiple nodes; each node is a coefficient sequence; the kurtosis value of the coefficient sequence of each node is calculated; for each node, the energy concentration factor of its coefficient sequence is calculated based on the energy entropy and total energy; the defect likelihood value of each node is calculated based on the kurtosis value and the energy concentration factor; when the kurtosis value is greater than zero, the defect likelihood value is the product of the square of the normalized value of the energy concentration factor and the logarithm of the kurtosis value; when the kurtosis value is not greater than zero, the defect likelihood value is the product of the normalized value of the energy concentration factor and the kurtosis value; an adaptive optimal basis search is performed on the wavelet packet tree based on the defect likelihood value to determine the optimal basis; the wavelet packet tree is sparsified based on the optimal basis to obtain a sparsed wavelet packet tree; an inverse wavelet packet transform is performed on the sparsed wavelet packet tree to obtain the defect signal; and corresponding control commands are generated based on the defect signal.

[0007] This invention first uses kurtosis values ​​to initially screen signals, then uses energy concentration factors to distinguish between real defects and isolated noise in high-kurtosis signals. Simultaneously, the maximum value of the energy concentration factor is introduced for normalization, and a defect likelihood value is constructed by combining piecewise functions and nonlinear enhancement. This defect likelihood value can effectively distinguish between defect, resonance, and noise signals, enabling subsequent adaptive optimal basis search based on the defect likelihood value to accurately pinpoint the defect component. Finally, the defect signal obtained through sparse reconstruction eliminates interference from strong resonance backgrounds and noise, improving the ability to identify weak anomalies and triggering accurate quality control commands. This achieves deep coupling between detection logic and automated execution actions, solving the problem of missed detections caused by logical rigidity in monitoring systems under complex environments.

[0008] Preferably, for each node, calculating the energy concentration factor of its coefficient sequence based on the energy entropy and total energy of the coefficient sequence includes: calculating the total energy of the coefficient sequence; calculating the energy entropy of the coefficient sequence; the energy concentration factor is the product of the negative natural exponent of the energy entropy and the total energy.

[0009] This invention multiplies the true total energy of a node with the negative natural exponent of the energy entropy to achieve collaborative signal screening: the total energy term, as the basic value of physical intensity, can effectively suppress isolated noise with extremely low total energy; the negative exponent of the energy entropy, through nonlinear conversion characteristics, effectively suppresses resonant signals with dispersed energy. After multiplying the two, only true defect signals that simultaneously satisfy energy concentration and have a certain total energy can obtain high scores, thereby achieving accurate differentiation between defects and noise.

[0010] Preferably, calculating the energy entropy of the coefficient sequence includes: calculating the energy percentage of each coefficient in the coefficient sequence relative to the total energy of the sequence; and calculating the Shannon entropy of the coefficient sequence based on the energy percentage to obtain the energy entropy.

[0011] This invention assesses the degree of energy dispersion on the time axis by calculating the energy percentage and introducing Shannon entropy. This utilizes the essential difference between resonant signals and transient signals in the energy time distribution structure. By using Shannon entropy, the physical distribution characteristics of the signal are accurately mapped into an evaluable index, providing a basis for subsequent steps to distinguish between resonant signals and transient signals.

[0012] Preferably, the step of performing an adaptive optimal basis search on the wavelet packet full tree based on the defect likelihood value to determine the optimal basis includes: recursively making decisions layer by layer from the bottom layer of the wavelet packet full tree upwards; for any parent node, comparing the defect likelihood value of the parent node with the sum of the defect likelihood values ​​of its two child nodes; when the defect likelihood value of the parent node is not less than the sum of the defect likelihood values ​​of its two child nodes, retaining the parent node and excluding the two child nodes; when the defect likelihood value of the parent node is less than the sum of the defect likelihood values ​​of its two child nodes, excluding the parent node and retaining the two child nodes.

[0013] This invention achieves adaptive optimal basis search by using a bottom-up, layer-by-layer recursive decision-making method that compares the sum of the defect likelihood values ​​of parent nodes and child nodes. This decision rule ensures that the algorithm always makes the optimal choice between continuing decomposition and retaining the current value, enabling the search process to automatically find the best balance between insufficient and excessive decomposition. This allows for the accurate aggregation of all node combinations with high defect likelihood values ​​in the wavelet packet full tree, providing defect information for the final defect signal reconstruction.

[0014] Preferably, the acquisition of the defect likelihood value includes: taking any node as the current node, its defect likelihood value satisfies the following relationship: ; in, The current node The defect likelihood value; The current node Energy concentration factor; The current node kurtosis value; It is the maximum value of the energy concentration factor among all nodes in the wavelet packet full tree; It is the preset third minute value.

[0015] This invention achieves nonlinear fusion and complementary advantages of kurtosis value and energy concentration factor through a specific piecewise relation: in branches with kurtosis value greater than zero, the contribution weight of the real defect is nonlinearly enhanced, while the interference caused by excessively high kurtosis value is suppressed by the logarithmic function term; in branches with kurtosis value less than or equal to zero, the resonance signal is automatically suppressed by the negative or zero value of kurtosis value, ultimately providing a robust and clear decision basis for optimal basis search.

[0016] Preferably, the sparsification process of the wavelet packet full tree based on the optimal basis includes: setting all the node coefficients of the wavelet packet full tree that do not belong to the optimal basis to zero, and retaining the nodes in the optimal basis.

[0017] Preferably, the step of generating corresponding control commands based on the defect signal includes: determining that there is a defect in the pile body when there is a pulse in the defect signal with an amplitude exceeding a preset threshold; calculating the defect depth based on the arrival time of the pulse and the sound velocity of the pile material; and triggering control commands based on the defect depth and amplitude.

[0018] Preferably, obtaining the kurtosis value includes: taking any node as the current node, its kurtosis value satisfies the following relationship: ; in, The current node kurtosis value; The current node The first coefficient sequence in the series One element; The current node The total length of the coefficient sequence; , These are the current nodes. The mean and standard deviation of all coefficients within the coefficient sequence; This is a preset correction option.

[0019] Preferably, acquiring the vibration response signal of the pile includes: acquiring the original vibration signal of the pile; filtering the original vibration signal to obtain the vibration response signal.

[0020] In a second aspect, the present invention provides a defect detection system for precast reinforced concrete piles. The defect detection system for precast reinforced concrete piles includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement a defect detection method for precast reinforced concrete piles according to the first aspect of the present invention.

[0021] By adopting the above technical solution, a defect detection method for precast reinforced concrete piles according to the first aspect of the present invention is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0022] The beneficial effects of this invention are as follows: This invention constructs a multi-stage, nonlinear optimal basis search criterion: First, the kurtosis value of the nodes is calculated to initially screen for spike signals, and then the energy concentration factor is calculated to distinguish between real defects and isolated noise in high kurtosis signals; at the same time, a piecewise nonlinear defect likelihood value is designed, and by normalizing and nonlinearly enhancing the energy concentration factor, combined with logarithmic suppression of the kurtosis value, the maximum separation of the three types of signals—defect, resonance, and noise—is achieved; finally, the defect signal is reconstructed based on the optimal basis obtained from the defect likelihood value search, and it is used as the accurate input basis for generating control commands, effectively filtering out strong background interference, realizing an adaptive closed loop from feature perception to decision control, and improving the logic response speed and control reliability of the monitoring device under industrial field conditions. Attached Figure Description

[0023] Figure 1 A flowchart illustrating a defect detection method for precast reinforced concrete piles provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a defect detection system for precast reinforced concrete piles provided in an embodiment of the present invention. Detailed Implementation

[0024] The first aspect of this invention provides a method for detecting defects in precast reinforced concrete piles, such as... Figure 1 As shown, the method includes steps S100-S500: Step S100: Obtain the vibration response signal of the pile and perform wavelet packet decomposition on it to obtain a wavelet packet full tree containing multiple nodes; each node is a coefficient sequence.

[0025] It should be noted that when performing structural testing on soil piles, sensors are typically deployed on the piles. The vibration signals collected are highly complex time-domain signals. This complexity stems primarily from the superposition and aliasing of the pile's own global resonance signal and the transient impacts caused by defects in the time domain. Direct analysis of the vibration signals can easily mask weak defect impacts with the strong resonance background, leading to missed detections. Therefore, this step requires transforming the vibration signals from the one-dimensional time domain to the time-frequency domain, which can effectively separate different characteristics. Wavelet packet transform, as a time-frequency decomposition tool, not only decomposes the low-frequency portion of the signal but also performs the same iterative decomposition on the high-frequency portion, providing a finer and more uniform time-frequency resolution across the entire frequency band. Considering that defect impacts often exhibit high-frequency transient characteristics, while pile resonance is concentrated in the low frequency range, the fine resolution of wavelet packet transform across the entire frequency band can precisely separate these two types of signals. Therefore, this invention uses it to process the collected vibration signals.

[0026] Specifically, firstly, a low-strain dynamic testing method is employed, using a handheld impact hammer to strike the top of the pile, while simultaneously placing an accelerometer at the pile top to collect vibration signals from the pile body. The sampling frequency is [missing information]. The frequency can be set to 20kHz and adjusted according to actual testing needs. To eliminate DC components and high-frequency noise unrelated to the testing, the vibration signal of the pile body also needs to be filtered to obtain the filtered signal. .

[0027] Then, wavelet packet transform is used to perform full-tree decomposition on the filtered signal to completely cover the full frequency band information of the signal. In a feasible implementation, it is preferred to use... Wavelet basis filtered signal conduct Level-by-level tree decomposition. The decomposition process is as follows: at level 1... Decomposed into low-frequency components and high frequency components ; Enter the second layer, and iterate over the two components of the first layer respectively. Decomposed into and , Decomposed into and This process is repeated layer by layer to decompose the entire tree until the first level is completed. Layer decomposition. Ultimately, a total of [number] layers are generated. There are nodes, each corresponding to a branch of the wavelet packet full tree, and all nodes are coefficient sequences. This time series represents the components of the vibration signal in a specific time-frequency range, where, It is the first The first layer The signal components corresponding to each node. The current node The first coefficient sequence in the series One coefficient, The current node The total length of the coefficient sequence.

[0028] It should be noted that the number of decomposition layers... Based on the signal sampling frequency The number of decomposition layers is determined by the characteristic frequency range of the defect signal of interest. Should meet And to ensure that the bandwidth resolution of the lowest layer is sufficient to distinguish defect features, the present invention preferably has 3-5 layers.

[0029] To visually illustrate the effectiveness of this invention, a set of data is introduced. Assume there is a tiny crack defect at a distance of 5 meters in the pile body. Vibration signals are collected from this pile body and preprocessed to obtain... ,exist The signal exhibits significant aliasing characteristics: the overall signal is a continuous, high-energy low-frequency oscillation, appearing only in... The system constantly superimposes extremely weak and short-duration spike defect echoes, and also mixes in transient electronic noise. Because the defect echoes are masked by the resonance signal, they cannot be directly observed with the naked eye. At that time, the tiny crack was almost indistinguishable. Performed The layer decomposition yielded 8 nodes. Three representative nodes were selected, and the nodes are as follows: The sequence exhibits stable periodic oscillations with uniform energy distribution and is statistically flat. The sequence energy only shows a pulse at point 4, and also exhibits a peak characteristic in statistical morphology; The energy of this sequence is highly concentrated at point 5, and it exhibits obvious peak characteristics in statistical morphology.

[0030] At this point, the wavelet packet full tree containing multiple nodes has been obtained.

[0031] Step S200: Calculate the kurtosis value of the coefficient sequence for each node.

[0032] It should be noted that, in order to locate pile defects from the entire wavelet packet tree, this step requires screening: considering that pile defects manifest as transient impacts in vibration signals, and transient impacts have high kurtosis values, kurtosis values ​​can effectively distinguish the distribution differences of different signals. Therefore, this invention introduces kurtosis values ​​to screen each node. The global resonance signal energy of the pile itself is stable, and the coefficient distribution is close to a Gaussian normal distribution, with kurtosis values ​​approaching 0 or being negative; the transient impact response caused by small defects such as cracks and cavities manifests as a few extreme peaks in the time domain, and the coefficient distribution exhibits heavy-tailed or peaked characteristics, corresponding to extremely high positive kurtosis values. Based on this difference, kurtosis values ​​can be used to mark all nodes in the frequency band that may contain defect peak characteristics.

[0033] Specifically, based on nodes For the current node, its kurtosis value satisfies the following relation: ; in, The current node kurtosis value; The current node The first coefficient sequence in the series One element; The current node The total length of the coefficient sequence; , These are the current nodes. The mean and standard deviation of all coefficients within the coefficient sequence; This is a preset correction term used to normalize the kurtosis value of the standard Gaussian normal distribution to 0. Since the ratio of the fourth central moment to the fourth power of the standard deviation of the standard Gaussian distribution is 3, it can... Set to 3, subtract After that, the kurtosis of the Gaussian distribution is 0.

[0034] In this relation, Used for the current node Each coefficient value in the coefficient sequence is standardized to eliminate the dimensional influence of coefficient values ​​at different nodes. The fourth power of this value is used to nonlinearly amplify peak values ​​far from the mean, while suppressing stationary values ​​close to the mean. The result obtained through this formula... It can accurately assess the spike level of the current node's signal. The signal is represented by a spike. This indicates that the signal is flat. The larger the value, the greater the likelihood that the node contains transient impact characteristics.

[0035] For example, calculating the nodes mentioned above , , The kurtosis value, in terms of nodes For example, the average value of its coefficient sequence is The standard deviation is By substituting each element, mean, and standard deviation of the coefficient sequence into the kurtosis value formula, the following can be calculated: kurtosis value Similarly, the nodes are calculated. kurtosis value ,node kurtosis value Observations revealed that kurtosis values ​​can distinguish between resonant and non-resonant signals by being positive or negative, but cannot distinguish between real defect signals and isolated noise signals, which are both positive.

[0036] At this point, the kurtosis values ​​of all nodes in the wavelet packet full tree have been obtained.

[0037] Step S300: For each node, calculate the energy concentration factor of its coefficient sequence based on the energy entropy and total energy of the coefficient sequence.

[0038] It should be noted that while kurtosis values ​​can filter out high-kurtosis signals, they cannot distinguish between genuine defect signals and isolated noise. This limitation affects the accuracy of defect localization. Therefore, this step needs to construct a specific index to differentiate between the two, addressing this limitation of kurtosis values. From a physical perspective, the transient impacts generated by genuine defects not only have high kurtosis values ​​but also possess a certain amount of energy, which is highly concentrated over time; while isolated noise, although energy is concentrated, has extremely low total energy. Based on this fundamental difference, this step first assesses the temporal dispersion of signal energy, then combines it with the true total energy of the node, and, based on kurtosis value filtering, achieves accurate identification of genuine defects.

[0039] First, the energy entropy of the signal energy over time is calculated. It should be noted that the energy distribution structures of resonance, defect, and noise signals are significantly different over time. The energy of a resonance signal is steadily distributed across the entire time axis, while the energy of defects and noise is instantaneously concentrated at a few points in time. Based on this difference, this invention constructs an energy entropy index to provide an evaluation basis for distinguishing resonance signals from defects and noise by assessing the distribution characteristics of energy over time.

[0040] Based on the above logic, the energy entropy of the current node satisfies the following relationship: ; in, The current node The energy entropy; , These are the current nodes. The first coefficient sequence in the series The, the One element; The current node The total length of the coefficient sequence; It is a preset first tiny value used to prevent It is 0, and it has the same properties as For quantities with the same dimensions, it can be set to 0.001; It is a preset second tiny value used to prevent The input to the function is 0, which can be set to 0.001.

[0041] In this formula, energy distribution is evaluated by weighted logarithmic summation based on energy proportion. Calculate the total energy of the current node; Calculate the current node's... The proportion of each element's energy to the total energy is used to assess the energy contribution weight at a single time point, and is compared with... The calculations combine to convert the energy percentage into the energy distribution contribution at the corresponding time point; the energy percentage at all time points is multiplied by the energy distribution contribution and summed, then the result is negative to ensure... This forms the final energy entropy assessment result. It can accurately assess the degree of energy dispersion of the current node on the time axis. The higher the value, the more dispersed the energy distribution along the time axis, corresponding to a resonance signal; The lower the value, the more concentrated the energy, corresponding to defects or noise signals.

[0042] Then, the temporal dispersion of signal energy is combined with the true total energy of the node. It should be noted that although both real defects and isolated noise exhibit low energy entropy, they differ significantly in their physical energy: the transient impact generated by a real defect possesses a certain total energy, while the total energy of isolated noise is extremely low. Based on this, this invention combines energy entropy with the true total energy of the node to construct an energy concentration factor, thereby achieving accurate identification of real defects.

[0043] Based on the above logic, the energy concentration factor satisfies the following relationship: ; in, The current node Energy concentration factor; ; The current node The first coefficient sequence in the series One element; The current node The total length of the coefficient sequence; It is a natural exponential function.

[0044] In this relation, Used to convert the energy entropy of the current node to The concentration coefficient between them, if the energy entropy is very high, will approach 0 after transformation, and if the energy entropy is very low, will approach 1 after transformation; This represents the total energy of the current node. This value serves as a baseline for evaluating the physical strength of the signal. The total energy of isolated noise tends towards 0, regardless of its concentration; the calculated energy concentration factor also tends towards 0. Multiplying these two values ​​distinguishes between resonance, defect, and noise signals. For resonance signals, their high energy is multiplied by a concentration factor approaching 0, thus leading to... Very small; for real defects, the energy is multiplied by a concentration coefficient approaching 1, resulting in The concentration factor is very large; however, for isolated noise, although its concentration factor also approaches 1, its total energy approaches 0, leading to... It is also very small.

[0045] For example, calculating the nodes mentioned above , , Energy concentration factor, node of , Substituting into the relationship of energy concentration factor, the calculation is obtained. Energy Concentration Factor Similarly, the nodes are calculated. Energy Concentration Factor ; Energy Concentration Factor Observations revealed that by combining kurtosis value screening with energy concentration factor, three types of signals—resonance, defects, and noise—can be clearly distinguished: kurtosis value excludes resonance signal nodes. The energy concentration factor will detect defect signals with high kurtosis values. It can be distinguished from noise signals with high kurtosis values, thus overcoming the limitations of a single kurtosis value index in identification.

[0046] At this point, the energy concentration factor of each node has been obtained.

[0047] Step S400: Based on the kurtosis value and energy concentration factor, calculate the defect likelihood value of each node; perform adaptive optimal basis search on the wavelet packet full tree based on the defect likelihood value to determine the optimal basis.

[0048] It should be noted that kurtosis is good at filtering transient signals with high kurtosis, while energy concentration factor is focused on distinguishing defects and noise in high kurtosis signals. Both have limited information when used alone. Therefore, it is necessary to first complement the advantages of kurtosis and energy concentration factor through nonlinear fusion, and then use adaptive search to select the optimal basis from the node tree, ultimately achieving maximum separation of defective signals.

[0049] Specifically, firstly, the kurtosis value and energy concentration factor are nonlinearly fused to obtain the defect likelihood value of the current node, wherein the defect likelihood value of the current node satisfies the following relationship: ; in, The current node The defect likelihood value; The current node Energy concentration factor; The current node kurtosis value; It is the maximum value of the energy concentration factor among all nodes in the wavelet packet full tree; It is a preset third micro value used to prevent It is 0, and has the same dimensions as the energy concentration factor, which can be set to 0.01.

[0050] This relation is a piecewise function, and its logical judgment is based on kurtosis values. Based on, At that time, transient signals with high kurtosis values ​​are used to process real defects or isolated noise. The term is squared. Perform nonlinear enhancement to ensure The contribution of [the subject / entity] has a higher weight in the final value; The term is used to constrain kurtosis values. The gain of isolated noise often exhibits high kurtosis values. The function exhibits increasing but gradually decreasing characteristics, which nonlinearly suppresses excessive gain caused by excessively large kurtosis values. This suppression characteristic prevents excessively large kurtosis values ​​generated by noise from affecting the energy concentration factor. Interference from contribution weights. At that time, the non-transient signal corresponding to the low kurtosis value is used to process pile resonance. The energy of the pile resonance signal is stable, and the coefficient distribution is close to Gaussian, which makes it The value is negative or zero; in this case, the signal is not a real defect or isolated noise. If the value is always positive or zero, the product of the two will necessarily be negative or zero. This will automatically suppress non-defect signals such as resonance, causing them to... Negative or zero values ​​are excluded from subsequent searches.

[0051] For example, calculating the nodes mentioned above , , The defect likelihood value, node The kurtosis value is If it is less than 0, then its defect likelihood value ;node The kurtosis value is If the value is greater than 0, then its defect likelihood value is... ;node The kurtosis value is If the value is greater than 0, then its defect likelihood value is... As can be observed, the defect node The defect likelihood value is much higher than that of the noise node. and resonance nodes This indicates that the segmented defect likelihood value can effectively assess the defect feature intensity of different signals. Through this differentiated assessment, a clear basis is provided for subsequent optimal basis search and defect signal reconstruction.

[0052] Then, an optimal basis is selected from the node tree through adaptive search. It should be noted that the wavelet packet tree contains a large number of nodes, with different nodes corresponding to signal components in different frequency bands. The energy of the actual defect signal is often dispersed across several associated nodes, and the defect likelihood value of a single node cannot fully capture the defect features. Therefore, it is necessary to find the optimal combination of nodes to maximize the total defect likelihood value of the combination, thereby integrating the distribution information of the defect signal across multiple frequency bands. This invention employs an adaptive search algorithm based on a greedy strategy: starting from the root node, this algorithm compares the defect likelihood contributions of child nodes layer by layer, prioritizing the retention of nodes with high defect likelihood values ​​and recursively expanding them. It has the advantages of high computational efficiency and rapid convergence to local optima, making it suitable for handling the hierarchical structure of wavelet packet trees. Simultaneously, its greedy nature avoids redundant computation caused by traversing all node combinations, improving search efficiency while ensuring the integrity of defect features.

[0053] Specifically, the search algorithm is a bottom-up dynamic programming process, from... Start with one layer and iterate upwards. Up to the top-level root node, the decision rules include: In each Layer, for any parent node of that layer Compare its own defect likelihood value with that of its two child nodes. and The sum of the defect likelihood values: when When the probability of a node's defect is higher than the sum of its child nodes, it indicates that decomposition is not worthwhile, and the parent node should be retained. As a candidate member of the optimal basis, all its descendant nodes are excluded. When When the sum of the child nodes is higher, it indicates that the decomposition is worthwhile. In this case, the child nodes are retained and the parent nodes are excluded. .

[0054] By adaptively searching and recursively executing the above decisions, the final output is an adaptive nonlinear optimal basis, which is the set of all nodes that are retained, have no parent nodes, and have the highest defect likelihood value.

[0055] Thus, the optimal basis has been obtained.

[0056] Step S500: Based on the optimal basis, the wavelet packet full tree is sparsified to obtain a sparsed wavelet packet full tree. The sparsed wavelet packet full tree is then subjected to inverse wavelet packet transform to obtain a defect signal. A corresponding control command is generated based on the defect signal.

[0057] It should be noted that, in order to transform the optimal basis into directly analyzable defect features, signal reconstruction and analysis are required to achieve the visual identification and localization of defects. Since the optimal basis has integrated all nodes with high defect likelihood values ​​and eliminated interference nodes such as resonance and noise, the interference can be removed from the original complex signal by reconstructing the signal through inverse transformation, highlighting the pure defect features and providing an intuitive basis for subsequent analysis.

[0058] Specifically, the wavelet packet tree is first sparsified by setting all node coefficients not belonging to the optimal basis to zero, retaining only the nodes in the optimal basis. Then, an inverse wavelet packet transform is performed on the sparsified wavelet packet tree. This transform, as the inverse operation of the wavelet packet transform, relies on the reconstruction characteristics of orthogonal filter banks to convert the frequency domain coefficients back to a one-dimensional time domain signal without distortion, ensuring the fidelity of the reconstructed signal in terms of time domain shape and timestamp, laying the foundation for accurate defect location. After obtaining the defect signal, the waveform characteristics are used to determine the pile defect condition and trigger corresponding control actions. If the amplitude of the defect signal does not exceed the preset threshold at any time, it is determined to be a flat line close to zero, indicating that there is no valid defect signal and the pile body is determined to be defect-free; the system generates a quality qualified signal, allowing the precast pile to enter the next construction process or the factory delivery process.

[0059] If the defect signal is at a certain moment If the amplitude exceeds the preset threshold, it can be determined as a clear spike pulse, indicating a defect. This is combined with the sound velocity of the pile material. Through formula Calculate the depth of the defect The system executes the following control logic based on the depth and magnitude of the defect: Minor defect handling: If the defect parameters are within the repairable range, a marking repair command is output, and the marking device is controlled to mark at the corresponding depth of the pile; If the defect parameters exceed the preset threshold, a process blocking command is immediately triggered, the pile status is locked as unqualified, and a real-time alarm is sent to the central control room to forcibly stop subsequent transportation or construction actions.

[0060] Furthermore, the statistical patterns of detected defects can be fed back to the preceding production control system. For example, if multiple piles have defects at the same location, a production parameter optimization command is triggered, prompting the production line to adjust the concrete pouring speed or vibration frequency, thus achieving closed-loop optimization of the production process.

[0061] It should be noted that the preset threshold can be determined based on the statistical characteristics of the background noise. For example, three times the standard deviation of the signal amplitude obtained by a normal, defect-free pile signal under the same processing procedure can be taken as the preset threshold.

[0062] For example, the resonant signal in the original signal and noise signals Submerged defect signals After the logical filtering process of this invention, only nodes with high defect likelihood values ​​are retained. The final reconstructed defect signal exhibits The isolated spike signal at any given moment is directly used as the trigger input for the system's automated diagnostic unit, enabling the monitoring device to lock onto defect features from complex background interference and generate high-confidence positioning commands and alarm control signals, thereby improving the identification accuracy and response reliability of the industrial site automatic monitoring system.

[0063] The second aspect of this embodiment provides a defect detection system for precast reinforced concrete piles, such as... Figure 2 As shown, the defect detection system for precast reinforced concrete piles includes a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement a defect detection method for precast reinforced concrete piles according to the first aspect of the present invention.

[0064] The defect detection system for precast reinforced concrete piles also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their setup and functions are known in the art and will not be described in detail here.

[0065] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0066] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for defect detection in precast reinforced concrete piles, characterized in that, Including the following steps: The vibration response signal of the pile is acquired and decomposed into wavelet packet to obtain a full wavelet packet tree containing multiple nodes; each node is a coefficient sequence. Calculate the kurtosis value of the coefficient sequence for each node; For each node, the energy concentration factor of its coefficient sequence is calculated based on the energy entropy and total energy of the coefficient sequence. Based on the kurtosis value and energy concentration factor, calculate the defect likelihood value for each node; When the kurtosis value is greater than zero, the defect likelihood value is the product of the square of the normalized value of the energy concentration factor and the logarithm of the kurtosis value. When the kurtosis value is not greater than zero, the defect likelihood value is the product of the normalized value of the energy concentration factor and the kurtosis value; Based on the defect likelihood value, perform an adaptive optimal basis search on the wavelet packet full tree to determine the optimal basis; The wavelet packet full tree is sparsified based on the optimal basis to obtain a sparsed wavelet packet full tree. An inverse wavelet packet transform is then performed on the sparsed wavelet packet full tree to obtain a defect signal. A corresponding control command is then generated based on the defect signal.

2. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, For each node, the energy concentration factor of its coefficient sequence is calculated based on the energy entropy and total energy of the coefficient sequence, including: Calculate the total energy of the coefficient sequence; Calculate the energy entropy of the coefficient sequence; The energy concentration factor is the product of the negative natural exponent of energy entropy and the total energy.

3. The defect detection method for precast reinforced concrete piles according to claim 2, characterized in that, The calculation of the energy entropy of the coefficient sequence includes: Calculate the percentage of energy of each coefficient in the coefficient sequence relative to the total energy of the sequence. The energy entropy is obtained by calculating the Shannon entropy of the coefficient sequence based on the energy percentage.

4. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The step of performing an adaptive optimal basis search on the wavelet packet full tree based on the defect likelihood value to determine the optimal basis includes: The decision is made recursively from the bottom layer of the wavelet packet full tree upwards. For any parent node, compare the defect likelihood value of the parent node with the sum of the defect likelihood values ​​of its two child nodes; If the defect likelihood value of the parent node is not less than the sum of the defect likelihood values ​​of the two child nodes, retain the parent node and exclude the two child nodes. If the defect likelihood value of the parent node is less than the sum of the defect likelihood values ​​of the two child nodes, the parent node is excluded and the two child nodes are retained.

5. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The acquisition of the defect likelihood value includes: Taking any node as the current node, its defect likelihood value satisfies the following relation: ； in, The current node The defect likelihood value; The current node Energy concentration factor; The current node kurtosis value; It is the maximum value of the energy concentration factor among all nodes in the wavelet packet full tree; It is the preset third minute value.

6. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The sparsification process of the wavelet packet full tree based on the optimal basis includes: Set all coefficients of nodes in the wavelet packet full tree that do not belong to the optimal basis to zero, and retain the nodes in the optimal basis.

7. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The step of generating corresponding control commands based on the defect signal includes: If there is a pulse in the defect signal with an amplitude exceeding a preset threshold, it is determined that there is a defect in the pile body, and the defect depth is calculated based on the arrival time of the pulse and the sound velocity of the pile body material. Control commands are triggered based on the defect depth and amplitude.

8. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The kurtosis value is obtained by: Taking any node as the current node, its kurtosis value satisfies the following relationship: ； in, The current node kurtosis value; The current node The first coefficient sequence in the series One element; The current node The total length of the coefficient sequence; , These are the current nodes. The mean and standard deviation of all coefficients within the coefficient sequence; This is a preset correction option.

9. The defect detection method for precast reinforced concrete piles according to claim 1, characterized in that, The acquisition of the vibration response signal of the pile includes: Collect the original vibration signal of the pile; The original vibration signal is filtered to obtain the vibration response signal.

10. A defect detection system for precast reinforced concrete piles, characterized in that, The defect detection system for precast reinforced concrete piles includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a defect detection method for precast reinforced concrete piles according to any one of claims 1-9.

Images