Intelligent ultrasonic detection system and method for cracks of uneven coal mine drill rod

Abstract

The invention relates to an intelligent ultrasonic detection system and method for cracks of an uneven coal mine drill rod, and belongs to the technical field of coal mine drilling. The system comprises an excitation system, a multi-channel parallel acquisition module and a real-time signal processing unit, the excitation system is used for realizing high-voltage narrow pulse output; the multi-channel parallel acquisition module receives pulse signals sent by the excitation system, high-speed sampling and parallel processing are achieved, and signal processing time delay is remarkably shortened. And the real-time signal processing unit realizes diagnosis of drill rod faults through a robust releasable signal transformation network model under variable working conditions. The method shows excellent crack detection performance under the conditions of strong mechanical vibration, temperature gradient and multi-physical field interference, realizes breakthrough in the aspects of detection precision and stability compared with a traditional method, meets the real-time response requirement at the same time, and is successfully applied to health monitoring of complex structure areas such as drill rod thread joints.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mine drilling, and relates to a non-planar intelligent ultrasonic detection system and method for cracks in coal mine drill pipes. Background Art

[0002] Based on an in-depth analysis of the current situation of drill pipe structural health monitoring technology in coal mine gas drainage operations, the current industry faces severe technical challenges. In the complex underground working conditions, drill pipes are subjected to alternating loads and multi-directional geological stresses for a long time, and micro-crack defects are likely to occur on their surfaces and inside. Especially in the threaded joint parts, the incidence of fracture accidents caused by stress concentration is high. This phenomenon directly threatens the safety of coal mine production. According to statistical data, the gas drainage interruption accidents caused by drill pipe failure account for more than 35% of the total underground equipment failures. Traditional non-destructive testing technologies have shown significant limitations in dealing with such engineering problems: Eddy current testing methods are easily interfered by magnetic field distortion on non-planar structure surfaces (such as drill pipe thread areas), and the false detection rate generally remains in the range of 15 - 30%; The detection rate of ultrasonic testing technology for micro-cracks less than 0.5mm is less than 40%, and the signal attenuation rate in the mud medium exceeds 60dB / m, seriously affecting the detection reliability. Although existing intelligent diagnostic algorithms (such as convolutional neural networks, support vector machines) can achieve an AUC value of more than 0.85 in the laboratory simulation environment, in the actual underground working conditions, due to strong mechanical vibrations (peak acceleration ≥ 15g), temperature gradients (the temperature difference at a well depth of 3000m reaches 80°C), and electromagnetic-thermal-mechanical multi-physical field coupling interference, the model performance shows a cliff-like decline, and the AUC value drops suddenly to below 0.7.

[0003] The current technical bottlenecks are mainly reflected in three dimensions: First, there are inherent defects in signal feature extraction methods. The frequency band selection mechanism of traditional wavelet transform lacks self-adaptability and is difficult to effectively separate crack features from background noise, resulting in a high degree of redundancy in the feature space; Second, the cross-condition model migration ability is weak. The distribution differences between laboratory simulation data and underground measured data are significant (KL divergence > 2.3), and traditional transfer learning methods fail to establish effective domain-invariant feature representations during the feature alignment process; Third, the real-time performance of the system cannot meet the engineering requirements. The single-frame processing time of existing detection systems exceeds 500ms and cannot adapt to the continuous operation scenario of gas drainage. These problems severely restrict the industrial application process of intelligent detection technology, forcing enterprises to still rely on the regular shutdown detection mode, resulting in a high rate of unplanned shutdowns.

[0004] In recent years, research institutions at home and abroad have attempted to break through the existing dilemmas through multi-technology integration paths. Typical solutions include the development of a three-source synchronous acquisition system for vibration, acoustic emission, and temperature, but the problem of synergistic optimization of dynamic allocation of feature weights and noise suppression has not been solved, and the efficiency of multi-modal data fusion is insufficient. Although the dynamic regulation technology of gas drainage negative pressure has improved the drainage efficiency, its technical route is completely focused on the optimization of gas flow and does not involve the dimension of drill pipe structural health monitoring. The shock wave permeability enhancement technology forms a coal body fracture network through multi-stage shocks, which improves the gas desorption rate, but this technology only focuses on the effect of reservoir transformation and lacks monitoring means for the structural state of the drill pipe body. The existing borehole trajectory measurement system (such as the combination of non-magnetic drill pipe and magnetoresistive sensor) can feedback the borehole spatial position in real time, but its technical architecture does not reserve a crack diagnosis function interface, forming a detection blind area. The common problem of these technical routes is that they overly focus on the improvement of a single technical index, lack systematic consideration of the coupled action of multi-physical fields under complex working conditions, and have not established a dynamic mapping model between detection data and structural health status.

[0005] The pain points of the industry are also reflected in the technical lag in the signal processing link. The improvement amplitude of the signal-to-noise ratio of the traditional sliding threshold filtering algorithm in a non-stationary noise environment is less than 3 dB, making it difficult to cope with underground random impact noise (the peak value can reach 120 dB). In the frequency domain analysis method, the fast Fourier transform is limited by the fixed time-frequency resolution and cannot capture transient crack characteristics; although wavelet packet decomposition has the ability of multi-scale analysis, the selection of basis functions depends on empirical parameters and has poor adaptability under the variable characteristics of coal and rock media. In the time-frequency analysis field, although the Hilbert-Huang transform can process non-stationary signals, there are problems of mode mixing and end effects, and the boundary distortion rate is high in a continuous vibration environment. These technical defects directly lead to a high information loss rate in the feature extraction link, seriously restricting the performance upper limit of the subsequent diagnosis model.

[0006] In terms of the intelligent diagnosis model architecture, the existing deep learning methods face severe domain adaptation challenges. The supervised learning model relies on a large amount of labeled data, while the acquisition cost of underground actual working condition samples is extremely high, and there are technical problems that the working condition conditions cannot be reproduced. In the scenario of laboratory-field data migration of traditional transfer learning methods (such as domain adversarial training), the feature alignment fails due to excessive domain differences, and the accuracy of the model after migration decreases. More seriously, the existing models lack an interpretability mechanism, and it is difficult to trace the reasons for misjudgment through feature visualization in engineering applications, restricting the iterative optimization of technical solutions. In terms of model real-time performance, although the parallel computing architecture based on GPU can improve the processing speed, it is limited by the computing resource constraints of underground equipment (the computing power of a typical embedded platform is <4 TOPS) and it is difficult to meet the millisecond-level response requirements.

[0007] The special underground environment poses stringent requirements on the detection hardware. The excitation voltage of existing ultrasonic detection systems is generally lower than 400V, resulting in a penetration depth of sound waves in coal and rock media of less than 1.5m, unable to meet the deep-hole detection requirements. The sampling rate of the data acquisition system is mostly limited below 20MHz, making it difficult to capture the characteristics of high-frequency crack echoes (>15MHz). Due to power consumption limitations, the embedded processing platform cannot carry complex signal processing algorithms, forcing the system to adopt a split architecture of "front-end acquisition + back-end processing", introducing a communication delay of over 200ms. The power management system lacks stability in the wide temperature range from -20°C to 80°C, and the battery life is difficult to exceed 8 hours in continuous working mode, seriously affecting the continuity of detection operations.

[0008] The lag in the development of the industry standard system further exacerbates the difficulty of technology application. The current JB / T 12463-2015 "General Rules for Ultrasonic Testing in Non-Destructive Testing" does not formulate adaptable clauses for the special underground environment in coal mines, lacking specific specifications in key links such as probe coupling and signal calibration. The compensation method for non-planar detection surfaces in the ASTM E317-2016 standard has large application errors in the drill pipe thread area. In the field of intelligent diagnosis, the ISO 13374-2003 "Condition Monitoring and Diagnosis of Machines" standard has not established a performance evaluation system for transfer learning models, resulting in a lack of a unified benchmark for horizontal comparison of technical solutions. The lag in the standard system makes the promotion of new technologies face compliance risks and indirectly raises the technology adoption costs of enterprises.

[0009] The changes in the market demand side put forward higher requirements for technological innovation. With the acceleration of the intelligent mine construction process, enterprises' demand for detection technology has shifted from single defect identification to full-life health management, requiring the system to have the ability of trend prediction and remaining life assessment. Existing technical solutions are mostly limited to post-event diagnosis, with obvious shortcomings in the dimension of predictive maintenance - the traditional method based on threshold alarm has a high false alarm rate, while the prediction model based on machine learning has a large error in long-term prediction over three months. At the same time, coal mining enterprises require the detection system to achieve data fusion with the existing digital twin platform, but the communication protocol compatibility of the existing system is insufficient, and the standardization degree of data interfaces is low, resulting in an increase in system integration costs.

[0010] In view of the systematic defects of the existing technical system in dimensions such as signal processing, model generalization, and hardware design, a new intelligent ultrasonic detection method and system for cracks in non-planar coal mine drill pipes are urgently needed. Summary of the Invention

[0011] In view of this, the purpose of the present invention is to provide a non-planar coal mine drill pipe crack intelligent ultrasonic detection system and method, aiming at the technical bottleneck of drill pipe micro-crack detection in complex underground working conditions, and proposing an innovative solution that combines multi-modal signal fusion and adaptive transfer learning. By overcoming key technical problems such as adaptive noise suppression, cross-domain feature alignment, and embedded real-time processing, the present invention significantly improves the reliability and applicability of the detection system under complex working conditions, and provides a new generation of intelligent detection solutions for coal mine safety production.

[0012] To achieve the above object, the present invention provides the following technical solutions:

[0013] A non-planar coal mine drill pipe crack intelligent ultrasonic detection system includes an excitation system, a multi-channel parallel acquisition module, and a real-time signal processing unit (i.e., the host computer);

[0014] The excitation system is used to achieve high-voltage narrow pulse output; the multi-channel parallel acquisition module receives the pulse signal sent by the excitation system, realizes high-speed sampling and parallel processing, and significantly shortens the signal processing delay; the real-time signal processing unit realizes the diagnosis of drill pipe faults through the Lubang releasable signal transformation network model under variable working conditions.

[0015] Preferably, the excitation system includes a clock management module (CMU), a waveform generation module, a Marx generator circuit, and a communication module, etc.;

[0016] The excitation signal is generated by the waveform generation module in the FPGA, sent to the Marx generator circuit through the communication module, and the signal is amplified by the Marx generator circuit and then outputs a high-frequency and high-voltage excitation pulse to drive the ultrasonic body wave probe; the clock management module (CMU) is used to provide a precise high-frequency clock signal, and the current clock frequency can reach up to 800 MHz at most, providing guarantee for the stable operation of the waveform generation module and the Marx generator.

[0017] Preferably, the waveform generation module realizes precise pulse control through digital logic, including the setting of pulse width, frequency, and amplitude, ensuring the time resolution and amplitude accuracy of the excitation signal.

[0018] Preferably, the Marx generator circuit includes cascaded capacitors and avalanche triodes, which are used to stack the input signals with lower amplitudes step by step and finally output high-voltage pulses;

[0019] The working principle of the Marx generator is to store energy in capacitors step by step. When the input signal triggers, the cascaded avalanche triodes conduct in turn, quickly stacking the charges stored in each capacitor and outputting them, thereby generating high-voltage pulses.

[0020] Preferably, the multi-channel parallel acquisition module includes an ADC conversion module, a clock management unit (CMU), an acquisition control module, a buffer module, an external memory, a communication module, etc.;

[0021] The signal generated by the excitation system enters the storage module via the ADC conversion module, the acquisition control module, the buffer module, and the processing module, and then is transmitted to the real-time signal processing unit (i.e., the host computer) using the communication module according to requirements, completing the entire process of data acquisition.

[0022] Preferably, the signal transformation network model of Lubang releasability under variable working conditions includes a segmented embedding module, multiple Transformer modules, and a fully connected layer;

[0023] The system cuts the collected ultrasonic signals of coal mine drill pipe cracks into multiple signal segments and sorts them through the segmented embedding module; the sorted signals are respectively input into multiple Transformer modules to extract crack features, and finally the crack signals are adaptively weighted through the fully connected layer to obtain the state of the drill pipe.

[0024] Preferably, the Transformer module includes a normalization module, a multi-head self-attention mechanism module, and a convolutional mapping module.

[0025] Preferably, the fully connected layer includes a linear layer, a convolutional layer, a normalization layer, a non-linear mapping layer, etc.

[0026] Preferably, the clock management module (CMU), the waveform generation module, and the communication module in the excitation system are designed based on FPGA; the clock management unit (CMU), the acquisition control module, the buffer module, the external memory, and the communication module in the multi-channel parallel acquisition module are designed based on FPGA.

[0027] The beneficial effects of the present invention are as follows:

[0028] Due to problems such as poor adaptability to non-flat surfaces, severe signal attenuation in noisy environments, and insufficient generalization ability of cross-condition models in traditional detection technologies, the present invention constructs a multi-source signal synchronous acquisition system, integrates an adaptive noise suppression module, combines time-frequency domain feature fusion methods with multi-scale attention mechanisms to strengthen crack-sensitive feature extraction, and significantly improves signal recognition.

[0029] The embedded hardware platform developed by the present invention includes a high-frequency pulse excitation system, a multi-channel parallel acquisition module, and a real-time signal processing unit, combined with a cloud intelligent diagnosis system to form a complete technical chain. The excitation system uses a cascaded capacitor and an avalanche triode design to achieve high-voltage narrow pulse output, and the multi-channel parallel acquisition module uses FPGA control to achieve high-speed sampling and parallel processing, significantly shortening the signal processing delay.

[0030] The method of the present invention exhibits excellent crack detection performance under conditions of strong mechanical vibration, temperature gradient, and multi-physical field interference, achieving breakthroughs in detection accuracy and stability compared to traditional methods. At the same time, it meets the requirements of real-time response, and has been successfully applied to the health monitoring of complex structural areas such as drill pipe thread joints, filling the gaps in the prior art for non-planar surface detection and cross-condition migration applications, providing a reliable means for structural safety assessment in coal mine gas drainage operations, and having the potential to be extended to fields such as oil drilling and geological exploration.

[0031] Other advantages, objectives, and features of the present invention will to some extent be described in the subsequent specification, and to some extent will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. Brief Description of the Drawings

[0032] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0033] Figure 1 It is a framework diagram of the Lupang release signal transformation network model under variable working conditions;

[0034] Figure 2 It is a control schematic diagram of the drill pipe ultrasonic signal acquisition module. Detailed Embodiments

[0035] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention schematically. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0036] Among them, the drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and should not be construed as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the drawings will be omitted, enlarged, or reduced, and do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0037] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the accompanying drawings. This is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only used for exemplary illustration and should not be construed as a limitation of the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0038] Embodiment 1:

[0039] In response to the challenges that the probability distribution of the actual test samples of the drill pipe system is unpredictable and the assumption of the same distribution of the sample data is not valid due to factors such as changes in the system working conditions, the optimization criteria of the unsupervised domain adaptation model are studied. Based on the domain adaptation method of adversarial learning, by introducing a multi-level game competition mechanism, the feature rearrangement in the deep network training process is realized, and a transfer learning model with implicit constraints is established. Considering that the diagnostic accuracy of the trained model will be severely affected when the working conditions change, which limits the application of the deep learning model in the actual industry. Therefore, this embodiment provides an intelligent detection method for drill pipe fatigue cracks, which uses the Signal-Transformer network model for variable working conditions, an intelligent fault diagnosis model focusing on variable working conditions problems, as Figure 1 shown. First, signal embedding is used to complete the segmentation and embedded representation of one-dimensional signals, thus enriching the information in the high-dimensional space. Then, these embedded sub-signals are further processed through the multi-head self-attention mechanism to explore the deep features of the signal states in different spaces. Finally, an attention interpretable method is proposed for vibration signals to increase the interpretability of the proposed model and overcome the shortcomings of the black box. Through experimental verification, the proposed model is superior to other advanced deep learning methods in terms of diagnostic accuracy under unknown working conditions. According to the principle of this model, the key weights of the signal are further explained in an interpretable manner. In addition, it can be predicted that Signal-Transformer can achieve a robust recognition effect in signal denoising with its interpretable feature extraction ability.

[0040] Embodiment 2:

[0041] This embodiment provides a non-destructive detection system for coal mine drill pipe fatigue cracks, including a high-frequency pulse excitation system, a multi-channel parallel acquisition module, and a real-time signal processing unit.

[0042] The excitation system is realized by the cooperation of multiple functional modules, including a clock management module (CMU), a waveform generation module, a Marx generator circuit, and a communication interface, etc. The excitation signal is generated by the waveform generation module in the FPGA, transmitted to the Marx generator circuit via high-speed signals, and outputs high-frequency and high-voltage excitation pulses for driving the ultrasonic body wave probe.

[0043] The excitation signal is generated by the waveform generation module in the FPGA. After being amplified by the Marx generator circuit, it can output a signal with an amplitude of 600V and a narrow pulse of 50ns. The clock management module (CMU) of the FPGA provides accurate high-frequency clock signals. The current clock frequency can reach up to 800MHz at most, providing guarantee for the stable operation of the waveform generation module and the Marx generator. The waveform generation module realizes precise pulse control through digital logic, including the setting of pulse width, frequency, and amplitude, ensuring the time resolution and amplitude accuracy of the excitation signal.

[0044] The Marx generator circuit is the core component of this excitation system. It is designed with cascaded capacitors and the avalanche triode FMMT417TD, and is used to stack the input signals with lower amplitudes step by step and finally output high-voltage pulses. FMMT417TD is a high-gain, fast-switching avalanche triode, which can withstand high voltage instantly and release the stored energy quickly, and is suitable for the generation of high-frequency and high-voltage pulse signals. The working principle of the Marx generator is to store energy step by step through capacitors. When triggered by the input signal, the cascaded avalanche triodes are turned on in sequence, and the charges stored in each capacitor are quickly stacked and output, thus generating high-voltage pulses. In order to meet the requirement of a 50ns pulse width, the charging and discharging circuit paths of the capacitors are optimized, and the efficient switching characteristics of FMMT417TD are adopted to shorten the pulse rise and fall times. The whole circuit design is compact, can achieve a large voltage output with low power consumption, and has the ability to operate at a high repetition frequency.

[0045] The signal acquisition end designed by using the FPGA is realized by the cooperation of multiple modules, such as a clock management unit (CMU), an acquisition control module, a buffer module, an external memory, and a communication interface, etc. The external signal enters the storage module after passing through the ADC conversion module, the acquisition control module, the buffer module, and the processing module, and then is transmitted to the upper computer according to the requirement by using the communication module, completing the whole process of data acquisition.

[0046] The multi-channel parallel data acquisition module is controlled by the acquisition control module. The digital signals passing through the ADC conversion circuit are input into the FPGA chip through the pins of the circuit board. Using a common 8-bit ADC conversion circuit, up to 10-channel data acquisition can be achieved. The sampling speed is determined by the clock frequency of the ADC conversion chip. The currently used chip can reach up to 32 MHz at most. The required clock signal is generated by the clock management unit in the FPGA. This unit can generate a high-speed clock signal of up to 800 MHz at most. After the signal enters the chip, it is temporarily stored in the buffer module. The buffer module uses internal memories of the chip such as FIFO or RAM, and the on-chip data storage can reach up to 16 Mbit at most. By taking advantage of the characteristics of fast storage and retrieval of on-chip storage, the acquisition and processing delay can be reduced. There is circuit noise in the signals in the buffer module. To improve the data quality, a data processing module is designed to perform fast data filtering and denoising on the signals. Since the FPGA has the advantage of highly parallel computing, it can achieve a computing speed superior to that of the CPU and is also easier to deploy compared to the GPU. The processed data is stored in an off-chip large-capacity memory such as SDRAM, Flash, etc., and data storage of up to 8 Gbit can be achieved. The data bandwidth reaches 25 Gb / s. At the same time, when the acquisition speed is too fast, off-chip storage can also be used as a buffer. When data reading is required, various communication modules can be used for data exchange, such as SPI, I2C, UART, Ethernet, USB, etc., and the data can be transmitted to external memories such as industrial control computers, SD cards, USB flash drives, etc. Specifically, as Figure 2 shown.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An intelligent ultrasonic detection system for cracks in non-flat coal mine drill pipes, characterized in that It includes an excitation system, a multi-channel parallel acquisition module, and a real-time signal processing unit; The excitation system is used to achieve high-voltage narrow pulse output; the multi-channel parallel acquisition module receives the pulse signal sent by the excitation system to complete high-speed sampling and parallel processing; the real-time signal processing unit realizes the diagnosis of drill pipe faults through the Lubang releasability signal transformation network model under variable working conditions.

2. The intelligent ultrasonic detection system for non-planar coal mine drill pipe cracks according to claim 1, characterized in that, The excitation system includes a clock management module, a waveform generation module, a Marx generator circuit, and a communication module; The excitation signal is generated by the waveform generation module and sent to the Marx generator circuit through the communication module. After being amplified by the Marx generator circuit, a high-frequency and high-voltage excitation pulse is output to drive the ultrasonic body wave probe; the clock management module is used to provide an accurate high-frequency clock signal to ensure the stable operation of the waveform generation module and the Marx generator.

3. The intelligent ultrasonic detection system for non-flat coal mine drill pipe cracks according to claim 2, characterized in that The waveform generation module realizes precise pulse control through digital logic, including the setting of pulse width, frequency, and amplitude, to ensure the time resolution and amplitude accuracy of the excitation signal.

4. The intelligent ultrasonic detection system for non-flat coal mine drill pipe cracks according to claim 2, wherein The Marx generator circuit includes cascaded capacitors and avalanche triodes, which are used to stack the low-amplitude input signals step by step and finally output high-voltage pulses; The working principle of the Marx generator is to store energy in capacitors step by step. When triggered by the input signal, the cascaded avalanche triodes are turned on in sequence, and the charges stored in each capacitor are quickly stacked and output, thereby generating high-voltage pulses.

5. The intelligent ultrasonic detection system for non-flat coal mine drill pipe cracks according to claim 1, wherein The multi-channel parallel acquisition module includes an ADC conversion module, a clock management unit, an acquisition control module, a buffer module, an external memory, and a communication module; The signal generated by the excitation system enters the storage module after passing through the ADC conversion module, the acquisition control module, the buffer module, and the processing module, and then is transmitted to the real-time signal processing unit through the communication module according to requirements to complete the whole process of data acquisition.

6. The intelligent ultrasonic detection system for cracks in non-flat coal mine drill pipes according to claim 1, characterized in that The Lubang releasability signal transformation network model under variable working conditions includes a segmented embedding module, multiple Transformer modules, and a fully connected layer The system cuts the collected ultrasonic signals of coal mine drill pipe cracks into multiple signal segments and sorts them through the segmented embedding module; the sorted signals are respectively input into multiple Transformer modules to extract crack features, and finally the crack signals are adaptively weighted through the fully connected layer to obtain the state of the drill pipe.

7. The intelligent ultrasonic detection system for cracks of non-flat coal mine drill pipes according to claim 6, characterized in that, The Transformer module includes a normalization module, a multi-head self-attention mechanism module, and a convolutional mapping module.

8. The intelligent ultrasonic detection system for non-planar coal mine drill pipe cracks according to claim 6, wherein The fully connected layer includes a linear layer, a convolutional layer, a normalization layer, and a non-linear mapping layer.

9. The intelligent ultrasonic detection system for non-flat coal mine drill pipe cracks according to claim 1, wherein The clock management module, the waveform generation module, and the communication module in the excitation system are designed based on FPGA; the clock management unit, the acquisition control module, the buffer module, the external memory, and the communication module in the multi-channel parallel acquisition module are designed based on FPGA.

Images