Multi-source data fusion acquisition system and method for deep water explosion container

Abstract

This invention discloses a multi-source data fusion acquisition system and method for deep-water explosion containers, belonging to the field of explosion container experimental testing technology. It utilizes a genetic algorithm-based dual-objective optimization configuration of multi-type sensor arrays, deployed in key areas of the explosion container to cover the acquisition of multiple physical field parameters including pressure, strain, and temperature. This invention achieves comprehensive acquisition of multi-source data through genetic algorithm-based sensor optimization configuration and microsecond-level synchronous acquisition technology. Adaptive modulation and demodulation, joint error correction, and deep learning compression algorithms ensure stable underwater data transmission. Intelligent data processing is achieved through real-time edge computing and historical analysis using LSTM models. The system can accurately acquire multi-physical field data such as pressure, strain, and temperature under deep-water explosion conditions, providing efficient data support for explosion container experimental analysis, performance evaluation, and operational condition prediction.

Description

Technical Field

[0001] This invention relates to the field of experimental testing technology for explosive containers, and in particular to a multi-source data fusion acquisition system and method for deep-water explosive containers. Background Technology

[0002] In deep-water explosion container experiments, accurate data acquisition, stable transmission, and efficient processing are crucial for analyzing explosion effects and evaluating container performance. Existing data acquisition technologies for deep-water explosion containers typically suffer from the following problems: First, sensor deployment is often limited, making it difficult to comprehensively capture data under the coupling effects of multiple physical fields such as explosion impact, hydrostatic pressure, and temperature changes, leading to limitations in understanding the experimental process. Second, deep-water environments are characterized by high water pressure, complex water quality, and turbulent flow, making traditional data transmission methods susceptible to interference, resulting in signal attenuation, data loss, or transmission delays, compromising data integrity. Third, existing acquisition systems are mostly composed of independently functioning modules, lacking an integrated collaborative management mechanism. This leads to inconsistent data formats, poor synchronization, low data processing efficiency, and reliance on subsequent manual analysis, making it difficult to avoid subjective errors and failing to meet the demands of high-precision experiments and complex operating condition simulations. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a multi-source data fusion acquisition system and method for deep-water explosion containers. The technical solution adopted is as follows: A multi-source data fusion acquisition method for deep-water explosion containers is used to collect explosion parameters of the deep-water explosion containers using a multi-source data fusion acquisition system. The multi-source data fusion acquisition method includes the following steps: Step 1: Based on the genetic algorithm, a dual-objective optimization configuration of multiple types of sensor arrays is performed. The sensor arrays are deployed in the key areas of the explosion container to cover the acquisition of multiple physical field parameters such as pressure, strain, and temperature. Step 2: Use high-precision clock synchronization technology to achieve microsecond-level synchronous acquisition of multiple types of sensors, add microsecond-level timestamps to the acquired data and complete time dimension alignment; Step 3: The raw data collected is preprocessed in real time through edge computing nodes, and then the preprocessed data is compressed using a deep learning compression algorithm. At the same time, the data is cached and backed up locally. Step 4: Using an underwater acoustic communication module, adaptive modulation and demodulation, joint error correction coding and retransmission mechanism are employed to stably transmit the compressed data to the water surface through the underwater acoustic link. Step 5: The water surface receiving and processing unit demodulates and restores the received data, extracts key features and identifies the experimental status through real-time processing algorithms, and mines data value by combining offline analysis and deep learning prediction models to generate an analysis report.

[0004] Optionally, the specific method for the dual-objective optimization configuration in step 1 is as follows: Construct a dual objective function with the goals of maximizing information coverage and minimizing data redundancy. Where R represents data redundancy and C represents information coverage of key areas. For data redundancy weight, Coverage weight; Using the sensor placement location and number as genes, 50-100 initial populations are randomly generated. Through genetic operations such as roulette wheel selection, single-point crossover, and 0.05 probability mutation, the process is iterated until the rate of change of the objective function value is less than or equal to 0.01 for 10 consecutive generations, at which point the optimal placement scheme is output. The sensor selection meets the following requirements: water pressure resistance greater than or equal to 1.2 times the experimental water pressure, impact resistance greater than or equal to 1000g, and 72-hour data drift less than or equal to 1%.

[0005] Optionally, the specific method for microsecond-level synchronous acquisition in step 2 is as follows: the underwater data acquisition terminal has a built-in GPS and Beidou dual-mode timing module and a DS3231 real-time clock chip. The clock is calibrated every minute through satellite signals, and each sensor receives the synchronous clock signal through the RS485 bus. The collected data is formatted with a timestamp format including year, month, day, hour, minute, second, and microsecond. The time interval is adjusted to 10μs using linear interpolation for sensor data with different sampling rates.

[0006] Optionally, the specific method for real-time preprocessing in step 3 is as follows: Edge computing nodes employ adaptive wavelet threshold filtering to denoise pressure and strain data, and use local weighted regression algorithm to repair outliers in temperature data. The deep learning compression uses a lightweight CNN model, which takes 1024 points of time series data as input, extracts features through 3 convolutional layers, and maps them into low-dimensional vectors. The compression ratio is 5:1-20:1, and the quantization precision is 8 bits.

[0007] Optionally, the data restoration method in step 3 is as follows: the water surface receiver reconstructs the data through a decoder symmetrical to the compressed CNN model, and a residual compensation module is added to the decoder output. The parameters are optimized through gradient descent so that the mean square error between the decompressed data and the original data is less than or equal to 0.01%, and the error of the key parameters is less than or equal to 1%.

[0008] Optionally, the specific method of the adaptive anti-interference transmission is as follows: the underwater acoustic communication module monitors the channel signal-to-noise ratio (SNR) in real time, and when the SNR is ≥20dB, 64QAM modulation and coding rate 3 / 4 are used; When 10dB≤SNR<20dB, 16QAM modulation and coding rate of 1 / 2 are used; When SNR < 10dB, BPSK modulation is used with a coding rate of 1 / 3. Joint error correction coding combines interleaving and convolutional coding, and adds a CRC-32 checksum, reducing the transmission error rate to a minimum. The following measures are taken to ensure data integrity using a sliding window protocol and a breakpoint resume mechanism.

[0009] Optionally, the specific method for real-time processing in step 5 is as follows: Peak detection and time-domain integration algorithms are used to extract key parameters, including peak impact pressure, maximum strain, and temperature change rate. The feature parameters are input through a support vector machine model to identify experimental stages and abnormal states in real time, and trigger audible and visual alarms when abnormalities occur.

[0010] Optionally, the specific method for offline analysis and prediction in step 5 is as follows: The raw data and processing results are categorized and stored in a MySQL database according to experiment number, operating condition, and sensor data. The signal frequency characteristics are analyzed using FFT, and the correlation between multi-physics data is explored through correlation analysis. Based on the LSTM model, the pressure, strain, and temperature curves are predicted within the next 10 seconds by inputting experimental conditions and historical data.

[0011] A multi-source data fusion acquisition system is used to implement a multi-source data fusion acquisition method. The system includes a multi-type sensor array, an underwater data acquisition terminal, an edge computing node, an underwater acoustic communication module, and a surface receiving and processing unit. The multi-type sensor array is deployed in the key areas of the explosion container to collect multiple physical field parameters such as pressure, strain, and temperature. The underwater data acquisition terminal is communicatively connected to the data output terminals of each sensor in the multi-type sensor array, and has a built-in high-precision clock synchronization module. The edge computing nodes are used for multi-channel data synchronous acquisition, preprocessing, compression, and caching; The underwater acoustic communication module is connected to the underwater data acquisition terminal and uses adaptive modulation and demodulation and joint error correction coding technology to stably transmit data through the underwater acoustic link. The surface receiving and processing unit is communicatively connected to the underwater acoustic communication module and includes an underwater acoustic receiving antenna, a signal conditioning module, an industrial control computer, and a display terminal, used for data demodulation, restoration, real-time processing, offline analysis, and visualization.

[0012] Optionally, the multi-type sensor array includes 8-12 piezoelectric pressure sensors arranged circumferentially along the inner wall of the container, 6-8 water-resistant fiber optic strain sensors installed at the weld, and 4-6 thermocouple temperature sensors installed in the vicinity of the explosion source area. The underwater data acquisition terminal uses a shell with a pressure resistance of 20MPa or higher and is equipped with a processor, a 16-channel 16-bit ADC acquisition chip and a memory card. The underwater acoustic communication module adopts a modular architecture based on OFDM technology, including a modulation and demodulation unit, an encoding and decoding unit, a channel monitoring unit, a power control unit, a TDMA protocol processing unit, and a built-in pressure sensor. The modulation and demodulation unit supports adaptive switching of 64QAM / 16QAM / BPSK. The encoding and decoding unit integrates interleaving coding, convolutional coding, and CRC-32 check functions. The channel monitoring unit collects the signal-to-noise ratio, Doppler frequency shift, and channel attenuation coefficient in real time. The power control unit dynamically adjusts the transmission power according to the channel attenuation. The TDMA protocol processing unit realizes anti-interference for multi-device communication. The built-in pressure sensor monitors the water depth in real time to assist in adjusting communication parameters. The industrial control computer of the water surface receiving and processing unit is equipped with a processor, memory and solid-state drive, and the signal conditioning module supports filtering and amplification processing.

[0013] In summary, the present invention includes at least one of the following beneficial technical effects: it achieves comprehensive acquisition of multi-source data through sensor optimization configuration based on genetic algorithm and microsecond-level synchronous acquisition technology; it ensures stable underwater data transmission by adopting adaptive modulation and demodulation, joint error correction and deep learning compression algorithms; and it achieves intelligent data processing by relying on edge computing real-time processing and LSTM model historical analysis.

[0014] The system can accurately collect multi-physical field data such as pressure, strain, and temperature in deep-water explosion environments, providing efficient data support for experimental analysis, performance evaluation, and operational condition prediction of explosion containers. It is suitable for various deep-water explosion container experimental scenarios. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall architecture of the multi-source data fusion acquisition system of the present invention; Figure 2 This is a schematic diagram of the microsecond-level synchronous acquisition process of the multi-source data fusion acquisition method of the present invention; Figure 3 This is a schematic diagram of the adaptive anti-interference underwater acoustic transmission logic of the multi-source data fusion acquisition method of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings.

[0017] This invention discloses a multi-source data fusion acquisition system and method for deep-water explosion containers.

[0018] Reference Figures 1-3 Example 1: A multi-source data fusion acquisition method for deep-water explosion containers. This method uses a multi-source data fusion acquisition system to collect explosion parameters of the deep-water explosion container. The multi-source data fusion acquisition method includes the following steps: Step 1: Based on the genetic algorithm, a dual-objective optimization configuration of multiple types of sensor arrays is performed. The sensor arrays are deployed in the key areas of the explosion container to cover the acquisition of multiple physical field parameters such as pressure, strain, and temperature. Step 2: Use high-precision clock synchronization technology to achieve microsecond-level synchronous acquisition of multiple types of sensors, add microsecond-level timestamps to the acquired data and complete time dimension alignment; Step 3: The raw data collected is preprocessed in real time through edge computing nodes, and then the preprocessed data is compressed using a deep learning compression algorithm. At the same time, the data is cached and backed up locally. Step 4: Using an underwater acoustic communication module, adaptive modulation and demodulation, joint error correction coding and retransmission mechanism are employed to stably transmit the compressed data to the water surface through the underwater acoustic link. Step 5: The water surface receiving and processing unit demodulates and restores the received data, extracts key features and identifies the experimental status through real-time processing algorithms, and mines data value by combining offline analysis and deep learning prediction models to generate an analysis report.

[0019] By adopting the above technical solutions, comprehensive capture of multiple physical field parameters and a balance between data validity are prerequisites for accurate data acquisition in deep-water explosion container experiments. Explosion impact, hydrostatic pressure, and temperature changes, among other physical fields, couple onto the container. The parameter responses in key areas directly reflect the container's performance and the explosion effect. However, a single arrangement can easily lead to missing information or data redundancy. Genetic algorithms, simulating the population iteration mechanism of biological evolution, construct optimization logic with the dual objectives of maximizing information coverage and minimizing data redundancy. Through initial population generation, selection, crossover, and mutation, the optimal sensor placement scheme is gradually selected, ensuring efficient sensor distribution in key areas of the container. This comprehensively covers the three core physical field parameters of pressure, strain, and temperature while avoiding repeated acquisition of the same physical field data, laying the foundation for accurate fusion of subsequent multi-source data.

[0020] Different types of sensors have different sampling frequencies. If the data lacks temporal consistency, subsequent fusion analysis will introduce systematic errors. High-precision clock synchronization technology establishes a unified and stable time reference through the collaborative work of a dual-mode timing module and a real-time clock chip. Regular calibration ensures that the clock deviation of all sensor channels is controlled within the microsecond range. During data acquisition, a precise timestamp is added to each set of acquired data to clearly define the corresponding time node. For sensor data with significant differences in sampling frequency, the time interval is adjusted using a linear interpolation method to achieve strict alignment of different physical field data in the time dimension. This ensures the correlation and comparability of pressure, strain, and temperature data, providing time consistency support for multi-source data fusion.

[0021] The raw data collected contains environmental noise, inherent sensor noise, and occasional outliers. Direct transmission of these data increases bandwidth consumption and affects the accuracy of subsequent analysis. Edge computing nodes, located close to the data acquisition end, can process the raw data in real time. They can filter out various types of noise using filtering algorithms, correct outliers using trend fitting, and improve data quality. Simultaneously, deep learning algorithms can uncover the inherent features of the data, mapping high-dimensional raw data to low-dimensional feature vectors. This significantly reduces the data volume while retaining key information, adapting to the limited transmission bandwidth of underwater acoustic channels. A local caching backup mechanism addresses potential interruptions in underwater transmission by providing temporary data storage, preventing data loss, and further ensuring the integrity of the collected data.

[0022] Underwater channels are characterized by high attenuation, strong interference, and dynamic changes, making it difficult for traditional fixed-parameter transmission methods to balance transmission rate and reliability. The underwater acoustic communication module dynamically adjusts the modulation scheme and coding rate by monitoring channel state parameters in real time. When channel conditions are favorable, it employs high-order modulation and high code rate to improve transmission efficiency; when channel interference is severe, it switches to low-order modulation and low code rate to ensure transmission reliability. Joint error correction coding technology, through the synergistic effect of data interleaving and convolutional coding, disperses sudden errors and enhances the data's anti-interference capability, reducing the transmission error rate. The retransmission mechanism and breakpoint resumption function address data loss or transmission interruption scenarios, ensuring complete data transmission through retransmission requests and breakpoint recovery, thus solving the core problem of unstable data transmission in complex underwater environments.

[0023] The surface receiving and processing unit first demodulates and restores the encoded and compressed data transmitted to the surface. Decoding algorithms recover the original data characteristics, and residual compensation optimizes the restoration accuracy, ensuring the data accurately reflects the underwater explosion and container response. Real-time processing algorithms rapidly extract key characteristic parameters such as peak impact pressure and strain rate of change. Based on a pre-defined feature-state mapping relationship, the system identifies the experimental stage and any anomalies in real time, providing timely warnings to ensure experimental safety. Offline analysis mines the frequency characteristics and correlations of historical data, combining them with deep learning prediction models. Using experimental parameters and similar historical data to train the model, it achieves accurate predictions of the container's physical field parameters for a future period. This provides data support for experimental scheme optimization, container performance evaluation, and operational risk prediction, realizing a closed-loop process from data acquisition to value extraction.

[0024] Example 2, the specific method for dual-objective optimization configuration in step 1 is as follows: Construct a dual objective function with the goals of maximizing information coverage and minimizing data redundancy. Where R represents data redundancy and C represents information coverage of key areas. For data redundancy weight, Coverage weight; Using the sensor placement location and number as genes, 50-100 initial populations are randomly generated. Through genetic operations such as roulette wheel selection, single-point crossover, and 0.05 probability mutation, the process is iterated until the rate of change of the objective function value is less than or equal to 0.01 for 10 consecutive generations, at which point the optimal placement scheme is output. The sensor selection meets the following requirements: water pressure resistance greater than or equal to 1.2 times the experimental water pressure, impact resistance greater than or equal to 1000g, and 72-hour data drift less than or equal to 1%.

[0025] By adopting the above technical solutions, the effectiveness of sensor placement in deep-water explosion container experiments directly determines the comprehensiveness of multiphysics parameter acquisition and the efficiency of data processing. On the one hand, the distribution of physical fields such as pressure, strain, and temperature generated by the explosion varies in different regions of the container. The parameter response in key regions is the core basis for analyzing the explosion effect and container performance, and it is necessary to ensure that no parameters are missed. On the other hand, excessive and redundant sensor placement will generate a large amount of redundant data, increasing the burden of subsequent transmission and processing, and reducing the overall efficiency of the system. Therefore, a balance needs to be established between information coverage and data redundancy.

[0026] The construction of a dual-objective function is the core logic for balancing the above requirements. The 1-R term in the function directly maps to the objective of minimizing data redundancy; the smaller R is, the larger the value of this term, representing a lower degree of redundancy. The C term directly corresponds to the objective of maximizing information coverage in key areas; the larger C is, the better the comprehensiveness of parameter collection. The introduction of weights α and β is to adjust the priority of the two objectives according to the actual needs of the experiment, ensuring that the optimization result matches the specific experiment's requirements for information integrity while avoiding resource waste caused by redundant data, thus achieving synergistic optimization of the two core objectives.

[0027] The application of genetic algorithms is a key means to achieve the aforementioned dual-objective optimization. Treating the sensor placement and quantity as genes essentially transforms the optimization problem into a process of selecting gene combinations. Randomly generating 50-100 initial populations ensures diversity in initial solutions, preventing the algorithm from converging in a single direction and leading to local optima, thus laying the foundation for selecting the globally optimal solution. The roulette wheel selection mechanism gives higher probability of selection to solutions with better objective function values, ensuring that excellent genes are preserved; single-point crossover generates new solutions that combine the core advantages of different excellent solutions; and a mutation probability of 0.05 appropriately introduces new gene changes, breaking the potential for gene solidification in the population, preventing the algorithm from getting trapped in local optima, and continuously expanding the optimization space.

[0028] The iteration termination condition is set as the rate of change of the objective function value being less than or equal to 0.01 for 10 consecutive generations. This is because when the change of the objective function value tends to level off, it indicates that the solution in the population is close to the global optimum, and it is difficult to obtain better results by continuing the iteration. Terminating at this time can reduce unnecessary computational consumption while ensuring the optimization effect, thus ensuring optimization efficiency.

[0029] All requirements for sensor selection were designed around the characteristics of the experimental environment and the need for data accuracy. Deep-water environments have high water pressure, and the explosion process is accompanied by strong impacts. Sensors must possess sufficient water pressure resistance and impact resistance to operate stably under harsh experimental conditions and avoid data acquisition interruptions due to equipment damage. The 72-hour data drift control requirement ensures the measurement accuracy of the sensors throughout the experimental period, preventing data distortion due to equipment performance degradation. This ensures that the optimized sensor layout can truly output reliable and effective multiphysics parameters, providing a solid foundation for subsequent data fusion and analysis.

[0030] Example 3, the specific method of microsecond-level synchronous acquisition in step 2 is as follows: the underwater data acquisition terminal is equipped with a GPS and Beidou dual-mode timing module and a DS3231 real-time clock chip. The clock is calibrated every minute through satellite signals, and each sensor receives the synchronous clock signal through the RS485 bus. The collected data is formatted with a timestamp format including year, month, day, hour, minute, second, and microsecond. The time interval is adjusted to 10μs using linear interpolation for sensor data with different sampling rates.

[0031] Example 4, the specific method of real-time preprocessing in step 3 is as follows: Edge computing nodes employ adaptive wavelet threshold filtering to denoise pressure and strain data, and use local weighted regression algorithm to repair outliers in temperature data. The deep learning compression uses a lightweight CNN model, which takes 1024 points of time series data as input, extracts features through 3 convolutional layers, and maps them into low-dimensional vectors. The compression ratio is 5:1-20:1, and the quantization precision is 8 bits.

[0032] Example 5, the data restoration method in step 3 is as follows: the water surface receiver reconstructs the data through a decoder symmetrical to the compressed CNN model, a residual compensation module is added to the decoder output, and the parameters are optimized through gradient descent so that the mean square error between the decompressed data and the original data is less than or equal to 0.01%, and the error of the key parameters is less than or equal to 1%.

[0033] Example 6, the specific method of the adaptive anti-interference transmission is as follows: the underwater acoustic communication module monitors the channel signal-to-noise ratio (SNR) in real time, and when the SNR ≥ 20dB, 64QAM modulation and coding rate 3 / 4 are used; When 10dB≤SNR<20dB, 16QAM modulation and coding rate of 1 / 2 are used; When SNR < 10dB, BPSK modulation is used with a coding rate of 1 / 3. Joint error correction coding combines interleaving and convolutional coding, and adds a CRC-32 checksum, reducing the transmission error rate to a minimum. The following measures are taken to ensure data integrity using a sliding window protocol and a breakpoint resume mechanism.

[0034] Example 7, the specific method for real-time processing in step 5 is as follows: Peak detection and time-domain integration algorithms are used to extract key parameters, including peak impact pressure, maximum strain, and temperature change rate. The feature parameters are input through a support vector machine model to identify experimental stages and abnormal states in real time, and trigger audible and visual alarms when abnormalities occur.

[0035] Example 8, the specific method for offline analysis and prediction in step 5 is as follows: The raw data and processing results are categorized and stored in a MySQL database according to experiment number, operating condition, and sensor data. The signal frequency characteristics are analyzed using FFT, and the correlation between multi-physics data is explored through correlation analysis. Based on the LSTM model, the pressure, strain, and temperature curves are predicted within the next 10 seconds by inputting experimental conditions and historical data.

[0036] By adopting the above technical solution, the time consistency of multi-source data is the foundation for subsequent fusion analysis. Different sensors have different sampling frequencies, necessitating a unified time reference. The GPS / BeiDou dual-mode timing module works in conjunction with the real-time clock chip, periodically calibrating to ensure the accuracy and stability of the underwater data acquisition terminal's clock. The synchronization clock signal is then transmitted to all sensors via a bus, achieving clock uniformity. Precise timestamps are added to the acquired data to clearly define the corresponding time nodes. For data with different sampling rates, linear interpolation is used to adjust the time interval, ensuring all data remain consistent in the time dimension, achieving microsecond-level synchronization and guaranteeing the correlation of multi-physics data.

[0037] Raw data acquisition contains noise and outliers, which can affect analysis accuracy, and the large data volume consumes significant transmission bandwidth. In real-time preprocessing, adaptive wavelet threshold filtering can specifically remove various types of noise from pressure-strain data, while local weighted regression algorithms can correct temperature anomalies based on data trends before and after outliers, ensuring data accuracy. Deep learning compression extracts core data features through lightweight models, transforming high-dimensional raw data into low-dimensional vectors. This reduces data volume while retaining key information, adapting to limited underwater transmission bandwidth. The quantization accuracy and compression ratio settings balance compression efficiency and data integrity.

[0038] The compressed data needs to be accurately restored before it can be used for analysis. A decoder symmetrical to the compression model is used to reverse-engineer data features, ensuring logical consistency in the restoration. The residual compensation module captures errors during the decoder's restoration process and continuously optimizes parameters through a gradient descent algorithm, constantly reducing the difference between the restored and original data, ensuring restoration accuracy, and keeping the errors of key parameters and overall data within a reasonable range to meet experimental analysis requirements.

[0039] Underwater channel conditions are complex and variable, requiring adjustments to transmission strategies based on channel conditions. Real-time monitoring of the channel signal-to-noise ratio (SNR) allows for timely assessment of channel quality. When the channel is favorable, higher-order modulation and higher code rates are used to improve transmission efficiency; when the channel is poor, lower-order modulation and lower code rates are switched to ensure reliable transmission. Joint error-correcting coding, through the combination of interleaving and convolutional coding, disperses errors during transmission, while check codes further verify data accuracy, reducing the bit error rate. A sliding window protocol and a breakpoint resume mechanism can address data loss or transmission interruption issues, ensuring complete data transmission to the surface.

[0040] The core of real-time processing is to rapidly extract key information and identify experimental states. Peak detection and time-domain integration algorithms can efficiently filter out key parameters reflecting the core state of the experiment, which are crucial for judging the experimental progress. After training, the support vector machine model can quickly match experimental stages through input feature parameters, while simultaneously identifying abnormal situations. Once an anomaly is detected, an alarm is triggered immediately, enabling timely warning of experimental risks and ensuring experimental safety.

[0041] Offline analysis aims to extract value from data and support subsequent experiments. Databases categorize and store data for easy and rapid retrieval and management. FFT analysis reveals the frequency characteristics of signals, while correlation analysis uncovers the intrinsic relationships between different physical field data. LSTM models excel at processing time-series data; after inputting experimental conditions and historical data, they can learn data variation patterns and predict physical field parameter curves in the near future, providing data references for experimental scheme optimization and container performance evaluation.

[0042] Example 9: Multi-source data fusion acquisition system, used to implement multi-source data fusion acquisition method. The system includes multi-type sensor arrays, underwater data acquisition terminal, edge computing node, underwater acoustic communication module and surface receiving and processing unit. The multi-type sensor array is deployed in the key areas of the explosion container to collect multiple physical field parameters such as pressure, strain, and temperature. The underwater data acquisition terminal is communicatively connected to the data output terminals of each sensor in the multi-type sensor array, and has a built-in high-precision clock synchronization module. The edge computing nodes are used for multi-channel data synchronous acquisition, preprocessing, compression, and caching; The underwater acoustic communication module is connected to the underwater data acquisition terminal and uses adaptive modulation and demodulation and joint error correction coding technology to stably transmit data through the underwater acoustic link. The surface receiving and processing unit is communicatively connected to the underwater acoustic communication module and includes an underwater acoustic receiving antenna, a signal conditioning module, an industrial control computer, and a display terminal, used for data demodulation, restoration, real-time processing, offline analysis, and visualization.

[0043] Example 10: The multi-type sensor array includes 8-12 piezoelectric pressure sensors arranged circumferentially along the inner wall of the container, 6-8 water-pressure resistant fiber optic strain sensors installed at the weld, and 4-6 thermocouple temperature sensors installed near the explosion source area. The underwater data acquisition terminal uses a shell with a pressure resistance of 20MPa or higher and is equipped with a processor, a 16-channel 16-bit ADC acquisition chip and a memory card. The underwater acoustic communication module adopts a modular architecture based on OFDM technology, including a modulation and demodulation unit, an encoding and decoding unit, a channel monitoring unit, a power control unit, a TDMA protocol processing unit, and a built-in pressure sensor. The modulation and demodulation unit supports adaptive switching of 64QAM / 16QAM / BPSK. The encoding and decoding unit integrates interleaving coding, convolutional coding, and CRC-32 check functions. The channel monitoring unit collects the signal-to-noise ratio, Doppler frequency shift, and channel attenuation coefficient in real time. The power control unit dynamically adjusts the transmission power according to the channel attenuation. The TDMA protocol processing unit realizes anti-interference for multi-device communication. The built-in pressure sensor monitors the water depth in real time to assist in adjusting communication parameters. The industrial control computer of the water surface receiving and processing unit is equipped with a processor, memory and solid-state drive, and the signal conditioning module supports filtering and amplification processing.

[0044] By adopting the above technical solutions, the system achieves comprehensive acquisition, efficient processing, and stable transmission of multi-source data through the collaborative work of multiple components. Multi-type sensor arrays are deployed in key areas of the explosion container to specifically capture three core physical field parameters: pressure, strain, and temperature, providing a fundamental data source for subsequent analysis. The underwater data acquisition terminal, as the core of data acquisition, aggregates data through communication connections with various sensors. Its built-in high-precision clock synchronization module ensures consistent acquisition timing across all sensors, preventing time deviations from affecting data validity. Edge computing nodes perform synchronous data acquisition, preprocessing, compression, and caching locally, improving data quality, reducing the amount of data transmitted, and preserving data even during transmission interruptions, ensuring data integrity. The underwater acoustic communication module, relying on adaptive modulation and demodulation and joint error correction coding technology, copes with the complex underwater channel environment, achieving stable data transmission. The surface receiving and processing unit, through the collaborative work of its components, completes data demodulation, reconstruction, real-time processing, and offline analysis. Simultaneously, a display terminal enables data visualization, allowing researchers to intuitively grasp the experimental status and data results, forming a complete closed loop from data acquisition to value presentation.

[0045] The configuration of multiple sensor arrays is designed around the accurate acquisition of multi-physics parameters. Piezoelectric pressure sensors are arranged circumferentially along the inner wall of the container, with the number adapted to the container's monitoring range to ensure comprehensive capture of pressure distribution; fiber optic strain sensors focus on strain-sensitive areas such as welds, and their water pressure resistance is suitable for deep-water environments, accurately monitoring structural strain changes; thermocouple temperature sensors are deployed near the explosion source area to specifically capture temperature fluctuations. The number and placement of different types of sensors are matched to their respective monitoring targets and container structural characteristics.

[0046] The pressure-resistant shell of the underwater data acquisition terminal is adapted to the high-pressure environment of deep water to avoid damage to the equipment; the processor provides data processing computing power, the 16-channel 16-bit ADC acquisition chip ensures the accuracy and speed of multi-channel data acquisition, and the memory card meets the local data caching requirements, providing support for data storage and subsequent retransmission.

[0047] The underwater acoustic communication module adopts a modular architecture, with each unit performing its specific function. The modulation and demodulation unit supports adaptive switching of multiple modulation methods to match different channel conditions; the encoding and decoding unit integrates multiple encoding and verification functions to enhance data anti-interference capabilities; the channel monitoring unit captures channel parameters in real time, providing a basis for adjusting transmission strategies; the power control unit dynamically adjusts the transmission power to adapt to channel attenuation; the TDMA protocol processing unit avoids communication interference between multiple devices; and the built-in pressure sensor monitors water depth to assist in optimizing communication parameters. The overall architecture ensures the stability and efficiency of underwater data transmission.

[0048] The industrial control computer of the surface receiving and processing unit is equipped with a high-performance processor, ample memory, and solid-state drive to meet the high-intensity computing and storage requirements for data demodulation, restoration, and analysis. The signal conditioning module improves the quality of the received signal through filtering and amplification, providing reliable input for subsequent data processing and ensuring the efficient operation of the entire system.

[0049] The following specific embodiments illustrate the implementation principle of the present invention: For water depth of 30 meters, explosive yield of 5 kg TNT, and medium density of 1100 kg / m³ 3 The deep-water explosion container experiment uses a multi-source data fusion acquisition system to achieve accurate acquisition and analysis of multiple physical field parameters such as pressure, strain, and temperature. The system components are configured and deployed according to the following specifications.

[0050] System deployment and implementation: Deployment of multiple sensor array types: Implemented using a dual-objective optimization allocation scheme based on genetic algorithms: The dual objective function is set as F=0.4×(1-R)+0.6×C, with the objectives being information coverage of key areas ≥95% and data redundancy ≤5%.

[0051] Using the sensor placement location and number as genes, 80 initial populations are randomly generated. 25 excellent schemes are selected by roulette wheel selection. The coordinates of the pressure sensor placement are exchanged by single-point cross-exchange. The number of temperature sensors is adjusted by mutation with a probability of 0.05. After 12 iterations, the rate of change of the objective function value stabilizes at 0.008, and the optimal placement scheme is output.

[0052] Sensor selection and installation: Ten piezoelectric pressure sensors are arranged at 30° intervals along the circumference of the inner wall of the container and are bonded with high-temperature resistant epoxy resin. Before bonding, the surface of the container is sanded to a roughness Ra≤1.6μm and cleaned with anhydrous ethanol. The lead wires are passed through a waterproof interface with a pressure resistance of 20MPa.

[0053] Seven water-pressure resistant fiber optic strain sensors are spot-welded to the weld seams of the container body and the connection points between the upper and lower end caps and the body. The weld diameter is 2.5mm. The optical fibers are connected by armored optical cables with a bending radius of ≥20mm.

[0054] Five thermocouple temperature sensors are inserted into the container's pre-set 3mm diameter, 10mm depth mounting holes, with the gaps filled with high-temperature sealant. Four are placed near the explosion source area, and one is placed on the external heat dissipation part.

[0055] All sensors meet the following requirements: water pressure resistance ≥36MPa (experimental water pressure 30m corresponds to 3MPa, 1.2 times safety factor), impact resistance ≥1000g, and 72-hour data drift ≤1%.

[0056] Deployment of underwater data acquisition terminals: The device uses a waterproof and pressure-resistant shell with a pressure resistance of 20MPa. Internally, it is equipped with an ARM Cortex-A9 processor (1.2GHz), a 16-channel 16-bit ADC acquisition chip (sampling rate up to 1MSPS), and a 128GB industrial-grade SD card. It also features a built-in GPS / BeiDou dual-mode timing module and a DS3231 real-time clock chip. The terminal is fixed to a 304 stainless steel bracket on top of the explosive container and securely connected to the container with bolts. The sensor cable is connected to the terminal via an RS485 bus and secured with nylon cable ties. A 5mm thick polyurethane shock-absorbing pad is attached to the terminal shell.

[0057] Edge computing node configuration: Edge computing nodes are integrated into underwater data acquisition terminals and work in conjunction with processors. They are pre-loaded with adaptive wavelet threshold filtering algorithms, local weighted regression algorithms, and lightweight CNN compression models. The models contain three convolutional layers (with kernel sizes of 3×3 and numbers of 16, 32, and 64 respectively) and one fully connected layer, supporting compression ratio adjustments from 5:1 to 20:1.

[0058] Deployment of underwater acoustic communication module: Employing a modular architecture based on OFDM technology, the module is fixed to an underwater robot, 8 meters away from the underwater data acquisition terminal, and connected to the surface control console via a waterproof umbilical cable with a 5-meter redundancy. The module operates in the frequency range of 20kHz-50kHz, with a transmission rate of 100kbps-500kbps. It features a built-in pressure sensor for real-time water depth monitoring and a power control unit that dynamically adjusts the transmission power from 0.5W to 5W.

[0059] Deployment of surface receiving and processing units: The omnidirectional transducer underwater acoustic receiving antenna is fixed to the deck support of the experimental vessel (1.8m high, away from metal obstacles). It is connected to the signal conditioning module via a cable. The signal conditioning module is connected to an industrial control computer equipped with a Core i7 processor, 16GB of memory, and a 1TB solid-state drive. The computer is connected to an external display terminal to complete the data visualization display.

[0060] Multi-source data fusion acquisition implementation process: Step 1: Sensor Optimization Configuration By using a genetic algorithm for iterative selection, the number, type, and placement of sensors are determined, enabling comprehensive coverage of multi-physics parameters and control of data redundancy.

[0061] Step 2: Microsecond-level synchronous acquisition: The underwater data acquisition terminal's clock synchronization function is activated. The GPS and Beidou dual-mode timing module receives satellite signals every minute to calibrate the DS3231 real-time clock chip, ensuring that the terminal clock accuracy is ≤10ns. Each sensor receives the synchronization clock signal through the RS485 bus, and the clock deviation is controlled within 0.8μs.

[0062] The acquisition parameters are set through the graphical interface of the water surface receiving and processing unit: pressure sensor sampling frequency 50kHz, strain sensor 5kHz, temperature sensor 5Hz, acquisition duration 60s, and triggering method is external signal triggering (associated with explosion ignition device).

[0063] After the explosion ignition device is triggered, all sensor channels start collecting data synchronously. The terminal adds a timestamp in the format of year, month, day, hour, minute, second, and microsecond to each channel's data. The time interval of the 5Hz sampling data from the temperature sensor is adjusted to 10μs using linear interpolation, achieving time dimension alignment with the data from the pressure and strain sensors.

[0064] Step 3: Edge computing preprocessing and deep learning compression: Edge computing nodes process raw data in real time: Adaptive wavelet threshold filtering (wavelet basis db4) is used to filter out environmental noise and sensor inherent noise in pressure and strain data. Local weighted regression algorithm is used to utilize the trend of 5 data points before and after the outlier to repair abrupt outliers in temperature data.

[0065] The lightweight CNN model divides the preprocessed data into blocks according to the time series, with each block containing 1024 data points. After feature extraction through 3 convolutional layers, the data is mapped to a low-dimensional vector. The compression ratio is set to 10:1, and 8-bit quantization is used. The compressed data and the original preprocessed data are stored together on a 128GB SD card.

[0066] Step 4: Adaptive Anti-interference Underwater Acoustic Transmission The underwater acoustic communication module's encoding and decoding unit encapsulates the compressed data in the format of sensor number-timestamp-data value-check code, and uses interleaved encoding (interleaving depth 16) and convolutional encoding (generator polynomials G1=133oct, G2=171oct) for joint processing, and adds a CRC-32 check code.

[0067] The channel monitoring unit collects the channel signal-to-noise ratio in real time. In the early stage of the experiment, the channel is stable with an SNR of 23dB. 64QAM modulation and a coding rate of 3 / 4 are used. During the explosion impact stage, bubble interference occurs, and the SNR drops to 14dB. It automatically switches to 16QAM modulation and a coding rate of 1 / 2. Later, water flow disturbance causes the SNR to be 8dB. It then switches to BPSK modulation and a coding rate of 1 / 3.

[0068] During transmission, the surface receiver monitors the data transmission status through a sliding window protocol. After detecting the loss of 3 data packets, it sends a retransmission request. The underwater terminal responds and completes the retransmission. The transmission is interrupted for 5 seconds due to water flow. After recovery, the transmission resumes from the interrupted position through the breakpoint resume mechanism to ensure data integrity.

[0069] Step 5: Water Surface Demodulation and Intelligent Analysis After the underwater acoustic receiving antenna receives the signal, it is amplified by the signal conditioning module, and 50Hz power frequency interference and high frequency noise are filtered out. The industrial control computer demodulates the signal and reconstructs the data through a decoder symmetrical to the compressed CNN model. The residual compensation module at the output of the decoder calculates the error between the decompressed data and the original sample. The parameters are optimized through gradient descent. Finally, the mean square error between the decompressed data and the original data is 0.008%, and the error of the key parameters is 0.8%.

[0070] Real-time processing: The peak impact pressure is extracted using a peak detection algorithm, and the maximum strain and temperature change rate are calculated using a time-domain integration algorithm. A total of 12 feature parameters are extracted and input into a pre-trained support vector machine model (trained with 1000 sets of historical data with an accuracy of 96%). The system can identify in real time that the experiment is in the explosion impact stage and no abnormal state is detected. The display terminal shows the time-domain curves and key parameters of pressure, strain and temperature in real time.

[0071] Offline analysis: Raw data, preprocessed data, and real-time processing results are categorized and stored in a MySQL database according to experiment number, operating condition, and sensor data, using a local backup + cloud backup mechanism; the frequency characteristics of the pressure signal are analyzed by FFT to generate a spectrum; correlation analysis is used to discover the positive correlation between strain change and pressure peak value; based on the LSTM model, inputting the operating condition parameters of this experiment and 50 sets of historical similar operating condition data, the pressure, strain, and temperature change curves within the next 10 seconds are predicted, with a pressure peak prediction error of 2.5% and a maximum strain prediction error of 4.2%.

[0072] The system automatically exports data files in Excel format and analysis reports in PDF format. The reports include experimental conditions, data acquisition overview, key parameter statistics, spectrum analysis results, and prediction curves, providing data support for subsequent experimental scheme adjustments and container performance evaluation.

[0073] System operation guarantee: During the experiment, the system monitoring and management module monitored the status of each device in real time: the underwater terminal's battery level was ≥85%, the underwater acoustic communication module's signal strength was ≥-65dBm, the sensor connections were normal, and there were no abnormal alarms. After the experiment, the sensors were disassembled and their surfaces were cleaned. The sealing rings of the underwater data acquisition terminal and the underwater acoustic communication module were checked for signs of aging. The sensors were calibrated, the calibration curves were updated, and data backup and system maintenance were completed.

[0074] 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 multi-source data fusion acquisition method for deep-water explosion containers, characterized in that, The explosion parameters of the deep-water explosion container were collected using a multi-source data fusion acquisition system. The multi-source data fusion acquisition method includes the following steps: Step 1: Based on the genetic algorithm, a dual-objective optimization configuration of multiple types of sensor arrays is performed. The sensor arrays are deployed in the key areas of the explosion container to cover the acquisition of multiple physical field parameters such as pressure, strain, and temperature. Step 2: Use high-precision clock synchronization technology to achieve microsecond-level synchronous acquisition of multiple types of sensors, add microsecond-level timestamps to the acquired data and complete time dimension alignment; Step 3: The raw data collected is preprocessed in real time through edge computing nodes, and then the preprocessed data is compressed using a deep learning compression algorithm. At the same time, the data is cached and backed up locally. Step 4: Using an underwater acoustic communication module, adaptive modulation and demodulation, joint error correction coding and retransmission mechanism are employed to stably transmit the compressed data to the water surface through the underwater acoustic link. Step 5: The water surface receiving and processing unit demodulates and restores the received data, extracts key features and identifies the experimental status through real-time processing algorithms, and mines data value by combining offline analysis and deep learning prediction models to generate an analysis report.

2. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 1, characterized in that, The specific method for dual-objective optimization configuration in step 1 is as follows: Construct a dual objective function with the goals of maximizing information coverage and minimizing data redundancy. Where R represents data redundancy and C represents information coverage of key areas. For data redundancy weight, Coverage weight; Using the sensor placement location and number as genes, 50-100 initial populations are randomly generated. Through genetic operations such as roulette wheel selection, single-point crossover, and 0.05 probability mutation, the process is iterated until the rate of change of the objective function value is less than or equal to 0.01 for 10 consecutive generations, at which point the optimal placement scheme is output. The sensor selection meets the following requirements: water pressure resistance greater than or equal to 1.2 times the experimental water pressure, impact resistance greater than or equal to 1000g, and 72-hour data drift less than or equal to 1%.

3. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 2, characterized in that, The specific method for microsecond-level synchronous acquisition in step 2 is as follows: the underwater data acquisition terminal has a built-in GPS and Beidou dual-mode timing module and a DS3231 real-time clock chip. The clock is calibrated every minute through satellite signals, and each sensor receives the synchronous clock signal through the RS485 bus. The collected data is formatted with a timestamp format including year, month, day, hour, minute, second, and microsecond. The time interval is adjusted to 10μs using linear interpolation for sensor data with different sampling rates.

4. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 3, characterized in that, The specific method for real-time preprocessing in step 3 is as follows: Edge computing nodes employ adaptive wavelet threshold filtering to denoise pressure and strain data, and use local weighted regression algorithm to repair outliers in temperature data. The deep learning compression uses a lightweight CNN model, which takes 1024 points of time series data as input, extracts features through 3 convolutional layers, and maps them into low-dimensional vectors. The compression ratio is 5:1-20:1, and the quantization precision is 8 bits.

5. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 4, characterized in that, The data restoration method in step 3 is as follows: the water surface receiver reconstructs the data through a decoder symmetrical to the compressed CNN model, and a residual compensation module is added to the decoder output. The parameters are optimized through gradient descent so that the mean square error between the decompressed data and the original data is less than or equal to 0.01%, and the error of the key parameters is less than or equal to 1%.

6. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 5, characterized in that, The specific method of the adaptive anti-interference transmission is as follows: the underwater acoustic communication module monitors the channel signal-to-noise ratio (SNR) in real time, and when the SNR is ≥20dB, 64QAM modulation and coding rate of 3 / 4 are used. When 10dB≤SNR<20dB, 16QAM modulation and coding rate of 1 / 2 are used; When SNR < 10dB, BPSK modulation is used with a coding rate of 1 / 3. Joint error correction coding combines interleaving and convolutional coding, and adds a CRC-32 checksum, reducing the transmission error rate to a minimum. The following measures are taken to ensure data integrity using a sliding window protocol and a breakpoint resume mechanism.

7. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 6, characterized in that, The specific method for real-time processing in step 5 is as follows: Peak detection and time-domain integration algorithms are used to extract key parameters, including peak impact pressure, maximum strain, and temperature change rate. The feature parameters are input through a support vector machine model to identify experimental stages and abnormal states in real time, and trigger audible and visual alarms when abnormalities occur.

8. The multi-source data fusion acquisition method for deep-water explosion containers according to claim 7, characterized in that, The specific methods for offline analysis and prediction in step 5 are as follows: The raw data and processing results are categorized and stored in a MySQL database according to experiment number, operating condition, and sensor data. The signal frequency characteristics are analyzed using FFT, and the correlation between multi-physics data is explored through correlation analysis. Based on the LSTM model, the pressure, strain, and temperature curves are predicted within the next 10 seconds by inputting experimental conditions and historical data.

9. A multi-source data fusion acquisition system, characterized in that, To implement the multi-source data fusion acquisition method for deep-water explosion containers as described in claim 8, the system includes a multi-type sensor array, an underwater data acquisition terminal, an edge computing node, an underwater acoustic communication module, and a surface receiving and processing unit; The multi-type sensor array is deployed in the key areas of the explosion container to collect multiple physical field parameters such as pressure, strain, and temperature. The underwater data acquisition terminal is communicatively connected to the data output terminals of each sensor in the multi-type sensor array, and has a built-in high-precision clock synchronization module. The edge computing nodes are used for multi-channel data synchronous acquisition, preprocessing, compression, and caching; The underwater acoustic communication module is connected to the underwater data acquisition terminal and uses adaptive modulation and demodulation and joint error correction coding technology to stably transmit data through the underwater acoustic link. The surface receiving and processing unit is communicatively connected to the underwater acoustic communication module and includes an underwater acoustic receiving antenna, a signal conditioning module, an industrial control computer, and a display terminal, used for data demodulation, restoration, real-time processing, offline analysis, and visualization.

10. The multi-source data fusion acquisition system according to claim 9, characterized in that, The multi-type sensor array includes 8-12 piezoelectric pressure sensors arranged circumferentially along the inner wall of the container, 6-8 water-resistant fiber optic strain sensors installed at the weld, and 4-6 thermocouple temperature sensors installed near the explosion source area. The underwater data acquisition terminal uses a shell with a pressure resistance of 20MPa or higher and is equipped with a processor, a 16-channel 16-bit ADC acquisition chip and a memory card. The underwater acoustic communication module adopts a modular architecture based on OFDM technology, including a modulation and demodulation unit, an encoding and decoding unit, a channel monitoring unit, a power control unit, a TDMA protocol processing unit, and a built-in pressure sensor. The modulation and demodulation unit supports adaptive switching of 64QAM / 16QAM / BPSK. The encoding and decoding unit integrates interleaving coding, convolutional coding, and CRC-32 check functions. The channel monitoring unit collects the signal-to-noise ratio, Doppler frequency shift, and channel attenuation coefficient in real time. The power control unit dynamically adjusts the transmission power according to the channel attenuation. The TDMA protocol processing unit realizes anti-interference for multi-device communication. The built-in pressure sensor monitors the water depth in real time to assist in adjusting communication parameters. The industrial control computer of the water surface receiving and processing unit is equipped with a processor, memory and solid-state drive, and the signal conditioning module supports filtering and amplification processing.

Images