Storage tank safety monitoring method based on Internet of Things

Abstract

The invention relates to the technical field of safety monitoring, and particularly discloses a storage tank safety monitoring method based on the Internet of Things, and the method comprises the steps: synchronously collecting a high-frequency leakage acoustic emission signal and a reference environment noise signal through a piezoelectric acoustic emission sensor disposed on the outer wall of a storage tank and an industrial microphone disposed in a storage tank region; after time delay alignment is carried out on a reference environment noise signal and a high-frequency leakage acoustic emission signal, self-adaptive filtering processing is carried out, so that environment noise interference in the high-frequency leakage acoustic emission signal is filtered out. And then, further introducing a deep learning algorithm to carry out time domain waveform analysis and adaptive content perception enhancement processing on the denoised high-frequency leakage acoustic emission signal so as to enhance key acoustic emission characteristics in the signal, thereby realizing intelligent identification of the storage tank leakage on the basis. According to the method, the accuracy and sensitivity of storage tank leakage monitoring can be effectively improved, the influence of noise interference on an industrial site on storage tank leakage identification is overcome, and early warning of tiny leakage of the storage tank is achieved.

Description

Technical Field

[0001] The present application relates to the field of safety monitoring technology, and more specifically, to a storage tank safety monitoring method based on the Internet of Things. Background Art

[0002] Storage tanks are key equipment used to store various liquid and gaseous materials in industries such as petroleum, chemical, and gas. Their safe and stable operation is crucial to the continuity and safety of the entire production system. Since many materials stored in tanks are flammable, explosive, toxic, and hazardous, a leak would not only cause significant economic losses and waste of materials, but could also trigger serious secondary accidents such as fires and explosions, severely polluting the surrounding environment and threatening the lives and property of operators and the public. Therefore, real-time and reliable monitoring of the health of storage tank structures and timely early warning of leaks are urgently needed to ensure industrial production safety and prevent major accidents.

[0003] Currently, leak detection methods for storage tanks primarily include manual inspections, pressure or liquid level monitoring, and sensor-based online monitoring. However, manual inspections lack real-time performance, are labor-intensive, and are susceptible to subjective factors, making it difficult to detect small or early-stage leaks. Monitoring methods based on process parameters such as pressure and liquid level, while capable of automation, typically trigger alarms only for larger leaks that cause significant parameter changes. They are severely insensitive for slowly developing or small leaks, and their early warning capabilities are limited.

[0004] Therefore, an optimized IoT-based tank safety monitoring method is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a tank safety monitoring method based on the Internet of Things, which synchronously collects high-frequency leakage acoustic emission signals and reference environmental noise signals through a piezoelectric acoustic emission sensor deployed on the outer wall of the tank and an industrial-grade microphone deployed in the tank area. The reference environmental noise signal is time-delay aligned with the high-frequency leakage acoustic emission signal and then adaptively filtered to filter out the environmental noise interference in the high-frequency leakage acoustic emission signal and improve the signal-to-noise ratio of the signal. Then, a deep learning algorithm is further introduced to perform time domain waveform analysis and adaptive content-aware enhancement processing on the denoised high-frequency leakage acoustic emission signal to strengthen the key acoustic emission features in the signal, thereby realizing intelligent identification of tank leaks on this basis. This method can effectively improve the accuracy and sensitivity of tank leakage monitoring, overcome the influence of noise interference in the industrial site on tank leakage identification, and realize early warning of small tank leaks.

[0006] According to one aspect of the present application, a method for monitoring storage tank safety based on the Internet of Things is provided, which includes:

[0007] High-frequency acoustic emission signals are collected by piezoelectric acoustic emission sensors deployed on the outer wall of the storage tank, and reference environmental noise signals are simultaneously collected by industrial-grade microphones deployed in the storage tank area.

[0008] Performing time delay alignment on the high-frequency acoustic emission signal and the reference ambient noise signal to obtain an aligned high-frequency acoustic emission signal and an aligned reference ambient noise signal;

[0009] Inputting the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal into a normalized LMS filter to filter out the environmental noise to obtain a denoised high-frequency acoustic emission signal;

[0010] The denoised high-frequency acoustic emission signal is input into a storage tank leakage identification model based on signal waveform feature analysis to determine whether the storage tank has a leak.

[0011] Compared with the prior art, the IoT-based tank safety monitoring method provided by this application uses a piezoelectric acoustic emission sensor deployed on the outer wall of the tank and an industrial-grade microphone deployed in the tank area to synchronously collect high-frequency leakage acoustic emission signals and reference environmental noise signals. By aligning the reference environmental noise signal with the high-frequency leakage acoustic emission signal time delay, adaptive filtering is performed to filter out the environmental noise interference in the high-frequency leakage acoustic emission signal and improve the signal-to-noise ratio of the signal. Subsequently, a deep learning algorithm is further introduced to perform time domain waveform analysis and adaptive content-aware enhancement processing on the denoised high-frequency leakage acoustic emission signal to strengthen the key acoustic emission features in the signal, thereby realizing intelligent identification of tank leaks on this basis. This method can effectively improve the accuracy and sensitivity of tank leakage monitoring, overcome the influence of noise interference in industrial sites on tank leakage identification, and achieve early warning of small tank leaks. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 Flowchart of a tank safety monitoring method based on the Internet of Things according to an embodiment of the present application.

[0014] Figure 2 Schematic diagram of data flow of a storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application.

[0015] Figure 3This is a flowchart of sub-step S2 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application.

[0016] Figure 4 This is a flowchart of sub-step S4 of the tank safety monitoring method based on the Internet of Things according to an embodiment of the present application.

[0017] Figure 5 This is a flowchart of sub-step S42 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application.

[0018] Figure 6 This is a flowchart of sub-step S422 of the tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0020] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0021] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0023] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0024] In response to the technical problems described in the above background technology, this application proposes a tank safety monitoring method based on the Internet of Things. It uses a piezoelectric acoustic emission sensor deployed on the outer wall of the tank and an industrial-grade microphone deployed in the tank area to synchronously collect high-frequency leakage acoustic emission signals and reference environmental noise signals. By aligning the time delay of the reference environmental noise signal with the high-frequency leakage acoustic emission signal and then performing adaptive filtering processing, the environmental noise interference in the high-frequency leakage acoustic emission signal is filtered out, thereby improving the signal-to-noise ratio of the signal. Then, a deep learning algorithm is further introduced to perform time domain waveform analysis and adaptive content-aware enhancement processing on the denoised high-frequency leakage acoustic emission signal to strengthen the key acoustic emission features in the signal, thereby realizing intelligent identification of tank leaks on this basis. This method can effectively improve the accuracy and sensitivity of tank leakage monitoring, overcome the impact of noise interference in industrial sites on tank leakage identification, and achieve early warning of small tank leaks.

[0025] Figure 1 Flowchart of a tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the tank safety monitoring method based on the Internet of Things includes the following steps: S1, collecting a high-frequency acoustic emission signal by a piezoelectric acoustic emission sensor deployed on the outer wall of the tank, and synchronously collecting a reference environmental noise signal by an industrial-grade microphone deployed in the tank area; S2, performing time delay alignment on the high-frequency acoustic emission signal and the reference environmental noise signal to obtain an aligned high-frequency acoustic emission signal and an aligned reference environmental noise signal; S3, inputting the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal into a normalized LMS filter to filter out environmental noise to obtain a denoised high-frequency acoustic emission signal; S4, inputting the denoised high-frequency acoustic emission signal into a tank leakage identification model based on signal waveform feature analysis to determine whether the tank has a leak.

[0026] In the above-mentioned storage tank safety monitoring method based on the Internet of Things, the step S1 collects high-frequency acoustic emission signals by a piezoelectric acoustic emission sensor deployed on the outer wall of the storage tank, and synchronously collects reference environmental noise signals by an industrial-grade microphone deployed in the storage tank area. Specifically, in recent years, acoustic emission (AE) technology has been gradually applied to the field of storage tank leakage monitoring because of its high sensitivity to high-frequency elastic waves generated by microscopic damage inside the material (such as crack extension). However, traditional acoustic emission monitoring systems face severe challenges when applied in industrial sites. The on-site environment of the storage tank area is complex, and the sound of wind and rain, pump operation, vehicle driving and other industrial activities will generate strong environmental noise and mechanical vibrations. These interference signals may overlap with the frequency band of the real leakage acoustic emission signal, which seriously affects the signal-to-noise ratio of the signal, causing the traditional monitoring method to easily generate false alarms or missed alarms, and reliability is difficult to guarantee. Therefore, in order to effectively distinguish the target signal from the interference noise, this application is based on the basic principle of signal source separation, and constructs a dual-channel data acquisition architecture to achieve synchronous and independent capture of leakage signals and environmental noise. Specifically, the present application closely deploys highly sensitive piezoelectric acoustic emission (AE) sensors on the outer wall of the storage tank, and utilizes its sensitive characteristics to internal stress waves of the material (such as the weak high-frequency vibrations generated on the tank wall when liquid leaks) to specifically capture the main signal containing real leakage information. At the same time, an industrial-grade omnidirectional microphone is deployed in an open space within the storage tank area but at a certain distance from the tank body. The microphone mainly collects reference environmental noise signals that are transmitted through the air and can represent the background sound field characteristics of the entire monitoring area. The two sensors are connected through a unified data acquisition terminal with high-precision clock synchronization function to ensure that the two signals are completely consistent in time reference. In this way, it is possible to effectively obtain acoustic emission signals that are a mixture of potential leakage signals and environmental noise, as well as relatively pure reference environmental noise signals, which provide an indispensable data basis for subsequent adaptive noise cancellation processing.

[0027] Specifically, in order to achieve effective collection of high-frequency acoustic emission signals, targeted collection is required through piezoelectric acoustic emission sensors deployed on the outer wall of the tank. Piezoelectric acoustic emission sensors work based on the piezoelectric effect and can convert tiny mechanical vibrations caused by stress changes inside the material into electrical signal outputs. They usually have high frequency components, ranging from tens of kilohertz to hundreds of kilohertz or even higher, and are particularly suitable for capturing elastic wave signals excited by transient events such as crack expansion and liquid leakage. Therefore, in the specific implementation process, the piezoelectric acoustic emission sensor is tightly installed on the surface of the outer wall of the tank so that it forms good physical contact with the metal structure of the tank, thereby ensuring that the local stress fluctuations caused by the leakage process can be efficiently captured.

[0028] The placement of the sensors has been carefully designed, usually in the lower or middle areas of the tank, which are more susceptible to liquid shock and pressure changes when liquid leaks occur, resulting in a stronger acoustic emission response. At the same time, in order to improve the integrity and representativeness of signal acquisition, multiple piezoelectric sensors are often deployed in different locations on the same tank to cover a wider range of potential leakage areas. Each sensor is connected to the data acquisition system via a low-noise shielded cable to ensure that the impact of external electromagnetic interference is minimized during signal transmission. In addition, considering the complexity of the industrial field environment, all sensor housings are made of waterproof, dustproof and corrosion-resistant materials to meet the needs of long-term operation outdoors or in semi-enclosed spaces.

[0029] At the same time, in order to effectively obtain background noise information, an industrial-grade omnidirectional microphone must be deployed in the open space within the tank area as a reference signal acquisition device. This type of microphone has a wide-band response characteristic and can cover sound wave signals from audible to ultrasonic ranges. It is suitable for collecting complex sound fields generated by the combined action of multiple noise sources in the environment. Its installation position should be as far away from the tank body as possible, maintaining a certain distance, but it must be within the acoustic influence range of the entire tank area so that it can fully reflect the overall noise characteristics of the area. For example, it can be fixed to a bracket or fence around the tank, and its direction can be adjusted to ensure maximum sensitivity to surrounding sound sources (such as pump operation, vehicle driving, windy and rainy weather, etc.).

[0030] In addition, the selection requirements for omnidirectional microphones are relatively high. They must have a high signal-to-noise ratio, a wide dynamic range, and good linear response characteristics to ensure that the collected noise signals are not distorted and have sufficient resolution. In addition, to avoid performance degradation due to environmental factors such as temperature and humidity, microphone components are usually equipped with protective covers and built-in drying devices to extend their service life and ensure the quality of collection. The connection between the microphone and the data acquisition terminal also uses an audio cable with excellent shielding performance to prevent external electromagnetic interference from introducing additional noise. In some application scenarios, wireless transmission can also be considered to increase deployment flexibility, but time synchronization and data stability issues need to be solved simultaneously.

[0031] Then, the two acquisition channels, the piezoelectric acoustic emission sensor and the omnidirectional microphone, need to be centrally managed through a unified data acquisition terminal to ensure that the two signals are strictly synchronized in the time dimension. The data acquisition terminal usually has a multi-channel input interface that supports the simultaneous access of analog or digital signals, and has a built-in high-precision clock module to provide a nanosecond-level synchronization reference. To meet the needs of high-frequency signal acquisition, the sampling rate is generally set above 1MHz, and some high-performance systems can even reach 10MHz, thereby ensuring that the detailed characteristics of the acoustic emission signal are fully captured. In addition, the terminal device is also equipped with a signal conditioning module, including a preamplifier, filter and analog-to-digital converter, to optimize signal quality and adapt to subsequent processing processes.

[0032] During the specific implementation process, the data acquisition terminal will simultaneously start the signal acquisition tasks of the two channels, so that the piezoelectric sensor and the microphone will start recording their corresponding acoustic signals at the same starting time. This synchronization mechanism not only helps the implementation of subsequent signal processing algorithms, but also effectively avoids phase misalignment problems caused by time deviation. For example, in adaptive noise cancellation technology, the degree of time alignment between the reference noise signal and the target signal directly affects the noise suppression effect, so it is necessary to ensure a high degree of consistency between the two on the time axis. To this end, some systems have also introduced external time synchronization methods such as GPS or PTP (Precision Time Protocol) to further improve the timing control accuracy of the overall system.

[0033] It's worth noting that while piezoelectric acoustic emission sensors primarily respond to stress waves within materials, while omnidirectional microphones focus on sound waves propagating through the air, and the two differ in their physical propagation paths and response mechanisms, they still form an effective complementary relationship in practical applications. On the one hand, the signal collected by the piezoelectric sensor is closer to the leak source and contains more characteristic information about the leak event itself; on the other hand, the ambient noise signal collected by the microphone reflects the background interference under the current working conditions, providing a key basis for subsequent noise separation and signal enhancement.

[0034] Furthermore, during the system debugging phase, the gain, bandwidth, and sensitivity of the two acquisition channels must be calibrated to ensure the relative proportional relationship between the signals is accurate. For example, if the piezoelectric sensor's output voltage amplitude is too low, the signal may be submerged in the acquisition system's background noise, affecting the ability to identify true leakage signals. Conversely, if the noise signal collected by the microphone is too strong, it may mask the detailed characteristics of the target signal, increasing the difficulty of post-processing. Therefore, it is usually necessary to calibrate the entire system with the help of a standard signal source or a known excitation source to ensure that each channel maintains consistency in terms of dynamic range, frequency response, and other aspects.

[0035] In the above-mentioned storage tank safety monitoring method based on the Internet of Things, the step S2 performs time delay alignment on the high-frequency acoustic emission signal and the reference environmental noise signal to obtain the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal. Specifically, due to the significant difference in the propagation speed of sound waves in different media (tank wall solid, air), and the different positions of the piezoelectric sensor and the microphone in physical space, there is a fixed delay in the time point when the interference signal generated by the same noise source reaches the two sensors. If the time delay is not corrected, direct noise filtering will lead to phase mismatch, which will reduce the noise filtering effect or even completely fail. Therefore, in order to ensure the effectiveness of noise cancellation, this application is based on the cross-correlation theory in signal processing, and calculates and compensates for the transmission delay between the two signals to achieve precise time alignment of the noise components. Among them, Figure 3 FIG is a flowchart of sub-step S2 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 3 As shown, the step S2 includes the steps of: S21, calculating the cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal, and determining the time delay based on the peak position of the cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal; S22, performing time delay alignment processing on the high-frequency acoustic emission signal and the reference ambient noise signal according to the time delay to obtain the aligned high-frequency acoustic emission signal and the aligned reference ambient noise signal.

[0036] Specifically, step S21 calculates the cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal, and determines the time delay based on the peak position of the cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal. That is, the collected high-frequency acoustic emission signal and the reference ambient noise signal are finely segmented, and a cross-correlation operation is performed on any two data segments. The cross-correlation function can measure the similarity between two time series at different time offsets, and the time offset between the two data segments corresponding to its peak is the time delay between the most correlated parts of the two signals (i.e., the same-source noise).

[0037] Specifically, in step S22, the high-frequency acoustic emission signal and the reference environmental noise signal are subjected to time delay alignment processing according to the time delay to obtain the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal. Specifically, after determining the time delay value, the present application shifts one of the signals (such as the reference environmental noise signal) on the time axis (advances or lags the time delay value) so that it is completely aligned in time with the noise component in the other signal, thereby obtaining the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal, eliminating the time asynchrony problem caused by the difference in propagation paths, and creating a prerequisite for the subsequent optimal noise suppression.

[0038] In the above-mentioned storage tank safety monitoring method based on the Internet of Things, in step S3, the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal are input into a normalized LMS filter to filter out environmental noise to obtain a denoised high-frequency acoustic emission signal. It should be understood that despite the time-delay alignment process, due to the attenuation and reflection of the sound wave during propagation and the difference in the frequency response characteristics of the two sensors themselves, the noise components in the two signals still have differences in amplitude and phase and cannot be eliminated by a simple subtraction operation. Therefore, in order to dynamically strip off the complex noise components from the main signal, the application constructs an adaptive noise canceller by introducing a normalized least mean square (NLMS) filter, using the aligned reference environmental noise signal as a reference input and the aligned high-frequency acoustic emission signal as the desired input. By iteratively updating the weight vector of the filter, it is made to gradually converge to the optimal solution, thereby filtering out components that are highly correlated with the reference noise signal from the main signal in real time. Specifically, the normalized LMS filter iteratively adjusts the weight coefficients within the filter in real time, based on the principle of minimizing output error, so that the filter's output signal waveform approximates the noise component contained in the high-frequency acoustic emission signal. Subsequently, by subtracting the optimal noise estimate generated by the filter from the original aligned high-frequency acoustic emission signal, the resulting error signal is the denoised high-frequency acoustic emission signal. This makes the weak vibration signal generated by the tank leak stand out, greatly improving the signal-to-noise ratio of the leak signal and laying the foundation for subsequent leak event identification.

[0039] In the above-mentioned method for tank safety monitoring based on the Internet of Things, in step S4, the denoised high-frequency acoustic emission signal is input into a tank leakage identification model based on signal waveform feature analysis to determine whether there is a leak in the tank. It should be understood that the present application takes into account the traditional discrimination method based on a single threshold value such as amplitude, energy, and ring count. Due to its limited feature expression ability and sensitivity to changes in working conditions, it has poor robustness and is difficult to accurately distinguish between real tiny leakage signals and system residual noise or other non-leakage sound sources. Therefore, in order to achieve intelligent and high-precision identification of leakage events, the present application is based on the powerful feature automatic learning and pattern recognition capabilities of deep learning, and constructs a tank leakage identification model based on signal waveform feature analysis to achieve deep mining and accurate classification of the inherent complex patterns of the signal. Among them, Figure 4 FIG. 4 is a flow chart of sub-step S4 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 4As shown, step S4 includes the following steps: S41, extracting the signal features of the denoised high-frequency acoustic emission signal to obtain a high-frequency acoustic emission signal feature coding vector; S42, performing adaptive content-aware enhancement processing on the high-frequency acoustic emission signal feature coding vector to obtain an enhanced high-frequency acoustic emission signal feature coding vector; S43, inputting the enhanced high-frequency acoustic emission signal feature coding vector into a classifier-based leakage detection model to obtain a leakage identification result, wherein the leakage identification result is used to indicate whether there is a leak in the storage tank.

[0040] Specifically, in a specific example of the present application, the step S41 includes: performing signal feature extraction based on the 1D-CNN model on the denoised high-frequency acoustic emission signal to obtain the high-frequency acoustic emission signal feature coding vector. It should be understood that the present application takes into account that the high-frequency acoustic emission signal after noise filtering is essentially still a complex one-dimensional time domain sequence, and the tank leakage state characterization information contained therein often exists in the form of nonlinear and non-stationary waveforms such as transient pulses, continuous emissions or specific attenuation modes. Therefore, in order to achieve adaptive and deep mining of the intrinsic pattern of the signal, the present application is based on the powerful feature learning ability of convolutional neural networks in deep learning, and constructs a one-dimensional convolutional neural network (1D-CNN) model to automatically extract the most recognizable signal features in the signal. Specifically, the 1D-CNN model contains multiple alternating stacked convolutional layers and pooling layers. In the convolutional layer, multiple convolution kernels (learnable filters) of different scales perform sliding convolution operations on the input denoised high-frequency acoustic emission signal along the time axis. Through the weight sharing mechanism, the same local waveform pattern (such as the rising edge of the pulse, the frequency of the oscillation, etc.) is captured at different time positions of the signal to generate a series of feature maps. Subsequently, the pooling layer (such as maximum pooling) downsamples the feature maps, while retaining the most significant features, while enhancing the model's invariance to small time shifts of the signal. Through multi-layer stacking, the model can gradually combine and learn higher-order and more abstract complex waveform structure features from the underlying simple waveform primitives, and finally form a compact vector representation that can characterize the core information of the original signal, that is, the high-frequency acoustic emission signal feature encoding vector.

[0041] Specifically, the step S42 performs adaptive content-aware enhancement processing on the high-frequency acoustic emission signal feature coding vector to obtain an enhanced high-frequency acoustic emission signal feature coding vector. Specifically, although the high-frequency acoustic emission signal feature coding vector contains key information of the tank leakage event, the importance and discrimination of different dimensional features within it may not be uniform. That is, some feature dimensions may correspond to the core mode of the leakage, while the other part may represent residual background disturbances or secondary vibration information unrelated to the leakage. Therefore, in order to adaptively focus on key information, the present application further performs adaptive content-aware enhancement processing on the high-frequency acoustic emission signal feature coding vector, and dynamically improves the robustness and discrimination ability of feature expression through feature content perception and attention reorganization to obtain an enhanced high-frequency acoustic emission signal feature coding vector. Among them, Figure 5 FIG4 is a flowchart of sub-step S42 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 5 As shown, the step S42 includes the steps of: S421, performing content perception on the high-frequency acoustic emission signal feature coding vector to obtain a local time-domain content-related perception coding feature map of the high-frequency acoustic emission signal; S422, inputting the local time-domain content-related perception coding feature map of the high-frequency acoustic emission signal into the spatial attention layer to obtain a local time-domain correlation content spatial significant perception coding feature map of the high-frequency acoustic emission signal; S423, performing feature shape reshaping on the local time-domain correlation content spatial significant perception coding feature map of the high-frequency acoustic emission signal to obtain the enhanced high-frequency acoustic emission signal feature coding vector.

[0042] More specifically, in a specific example of the present application, step S421 includes: performing feature segment segmentation on the high-frequency acoustic emission signal feature coding vector based on a preset window to obtain a set of local time domain feature vectors of the high-frequency acoustic emission signal; reshaping the set feature shape of the local time domain feature vector of the high-frequency acoustic emission signal into a local time domain coding feature map of the high-frequency acoustic emission signal, and performing content perception based on convolution coding to obtain the local time domain content association perception coding feature map of the high-frequency acoustic emission signal, which is expressed as follows:

[0043] X={x1,x2,...,x i ,...,x n}

[0044] F enc =C enc (Reshape(X))

[0045] Among them, X represents the set of local time domain feature vectors of high-frequency acoustic emission signals, x1, x2, x i and x nrepresents the first, second, i-th and n-th local time domain feature vectors of high-frequency acoustic emission signals in the set X, n is the number of vectors in the set of local time domain feature vectors of high-frequency acoustic emission signals, Reshape(·) represents the feature shape reshaping, C enc represents the convolutional coding operation, F enc Represents the local temporal content-related perceptual coding feature map of high-frequency acoustic emission signals.

[0046] That is, by segmenting the preset window, the one-dimensional high-frequency acoustic emission signal feature coding vector is converted into multiple high-frequency acoustic emission signal local time domain feature vectors, and then reshaped into a two-dimensional high-frequency acoustic emission signal local time domain coding feature map, so that the convolutional coding can model the waveform details within the local time domain segment and the temporal correlation between the segments, thereby generating a high-frequency acoustic emission signal local time domain content association perception coding feature map containing local time domain content association perception, strengthening the expression of key local features in the leakage signal, suppressing the interference of irrelevant noise, and providing more discriminative input for the subsequent spatial attention mechanism to focus on significant features.

[0047] Figure 6 FIG4 is a flowchart of sub-step S422 of the storage tank safety monitoring method based on the Internet of Things according to an embodiment of the present application. Figure 6 As shown, the step S422 includes the steps of: S4221, performing global mean pooling along the channel dimension on the local time domain content-related perceptual coding feature map of the high-frequency acoustic emission signal to obtain a local time domain content-related perceptual pooling coding vector of the high-frequency acoustic emission signal, and inputting the local time domain content-related perceptual pooling coding vector of the high-frequency acoustic emission signal into a multi-layer perceptron model to obtain dynamic convolution kernel parameters; S4222, based on the dynamic convolution kernel parameters, performing spatial saliency feature extraction on the local time domain content-related perceptual coding feature map of the high-frequency acoustic emission signal to obtain the spatial saliency perceptual coding feature map of the local time domain related content of the high-frequency acoustic emission signal.

[0048] In a specific example of the present application, step S4221 can be expressed as follows:

[0049] v g =GAP(F enc )

[0050]

[0051] Among them, GAP(·) represents the global mean pooling operation, v grepresents the local temporal content-aware pooling coding vector of the high-frequency acoustic emission signal, MLP(·) is a multi-layer perceptron model, W1 and b1 represent the weight matrix and bias vector of the first layer of the multi-layer perceptron model, respectively, RELU(·) represents the ReLU activation function, h represents the hidden state feature vector of the first layer of the multi-layer perceptron model, W2 and b2 represent the weight matrix and bias vector of the second layer of the multi-layer perceptron model, respectively, w flat Represents the hidden state feature vector of the second layer of the multilayer perceptron model, W dyn Represents the dynamic convolution kernel parameters.

[0052] Specifically, a multi-layer perceptron model is used to perform feature distribution perceptual coding on the local temporal content-related perceptual coding feature map of the high-frequency acoustic emission signal. This allows the model to dynamically generate dynamic convolution kernel parameters adapted to the local temporal characteristics of the current input signal based on the global statistical characteristics of the signal it represents. This dynamic generation mechanism enables subsequent convolution operations to adaptively adjust to the characteristic distribution characteristics of different signals, ensuring that key time-domain patterns related to leaks are effectively identified and enhanced during the feature extraction process, while suppressing interference from background noise and irrelevant vibrations, thereby improving the sensitivity and accuracy of detecting small leaks in storage tanks.

[0053] Preferably, when generating the dynamic convolution kernel parameter W dyn In the process, the local temporal content-related perceptual coding feature map of the high-frequency acoustic emission signal is globally averaged pooled in the channel dimension to obtain the local temporal content-perceptual pooled coding vector v of the high-frequency acoustic emission signal represented by the local system distribution between segments. g Then, the local time domain content-aware pooling encoding vector v of the high-frequency acoustic emission signal is performed using weight matrices W1 and W2 and bias vectors b1 and b2. g During the modulation process, when the characteristic distribution of the local time domain of the high-frequency acoustic emission signal depends on the parameter configuration control of the weight matrix and the bias vector, it is also expected to achieve closed-loop optimization of the parameter space, so that the dynamic convolution kernel parameter W dyn The fidelity of feature extraction can be improved. Based on this, in a preferred example of the present application, step S4221 includes: first, performing constrained optimization based on the network parameter space phase stability on the bias vector of the multilayer perceptron model to obtain an optimized bias vector; then, based on the optimized bias vector, using the multilayer perceptron model to process the local time domain content-aware pooling coding vector of the high-frequency acoustic emission signal to obtain the dynamic convolution kernel parameters.

[0054] Specifically, if W1v gand W2h as the characteristic state regulation in the weight parameter space, then the bias vectors b1 and b2 will introduce additional phase components. In order to maintain the characteristic state from local phase distortion, the characteristic latent state canonical representation is first established:

[0055]

[0056] Among them, e (·) Represents an exponential function with a natural constant as the base, ⊙ represents point multiplication by position, v1 and v2 represent the first-level high-frequency acoustic emission signal local time domain feature hidden state normalized representation vector and the second-level high-frequency acoustic emission signal local time domain feature hidden state normalized representation vector, respectively.

[0057] Then, when the transition primitives of the characteristic state transition are constructed by connecting the weight matrix, the characteristic transition evolution representation vector is defined as:

[0058] v3=(W1-W2)(W1v g -W2h)

[0059] Among them, v3 represents the characterization vector of the local time domain characteristic transition evolution of the high-frequency acoustic emission signal.

[0060] In this way, the symmetry of the matrix element effect can be determined based on the polarization selectivity of the additional phase introduced by the bias vector:

[0061]

[0062] in,(·) T represents the transpose of a vector, represents matrix multiplication, ‖·‖2 represents the calculation of the Euclidean norm, b1 ' and b ' 2 represent the optimized bias vectors of the first layer and the second layer of the multilayer perceptron model respectively.

[0063] That is, by constraining the optimization of the bias vectors b1 and b2, it is possible to achieve local connection phase rotation perception while preventing the instability of the feature hidden state caused by the additional phase defect, so that the parameter configuration control of the weight matrices W1 and W2 and the bias vectors b1 and b2 is effective for the dynamic convolution kernel parameters W dyn The generation of forms a closed-loop optimization link, thereby significantly improving the dynamic convolution kernel parameter W dyn The fidelity of feature extraction.

[0064] In a specific example of the present application, step S4222 is expressed as follows:

[0065] A=Sigmoid(W dyn *F enc )

[0066] F attn =F enc ⊙A

[0067] Where * is the convolution operation, Sigmoid(·) represents the Sigmoid activation function, A represents the spatial attention weight distribution map of the local temporal correlation content of the high-frequency acoustic emission signal, and F attn Represents the spatially significant perceptual coding feature map of the local temporal correlation content of high-frequency acoustic emission signals.

[0068] That is, the dynamic convolution kernel parameters are used to perform convolution operation on the local time domain content-related perceptual coding feature map of the high-frequency acoustic emission signal. By adaptively adjusting the weight distribution of the convolution kernel at different spatial positions, the obtained local time domain related content spatially significant perceptual coding feature map of the high-frequency acoustic emission signal can highlight the local significant feature area of ​​the leakage signal in the time domain space, make the energy distribution of the leakage-related features in the feature map more concentrated, and effectively suppress the interference of background noise, providing more accurate input for the subsequent classifier-based leakage detection model.

[0069] More specifically, in step S423, the local time-domain correlation content spatially significant perceptual coding feature map of the high-frequency acoustic emission signal is reshaped to obtain the enhanced high-frequency acoustic emission signal feature coding vector, which is expressed as:

[0070]

[0071] in, Represents the feature coding vector of the enhanced high-frequency acoustic emission signal.

[0072] That is, the spatially significant perceptual coding feature map of the local time-domain correlation content of the high-frequency acoustic emission signal is converted into a one-dimensional enhanced high-frequency acoustic emission signal feature coding vector, so that the data dimension matches the input requirements of the subsequent classifier, so that the generated enhanced high-frequency acoustic emission signal feature coding vector can more efficiently condense the key feature information of the leakage signal while maintaining a low dimension, providing an adaptive input form for subsequent classification decisions.

[0073] Specifically, in a specific example of the present application, step S43 includes: using the fully connected layer of the leak detection model to fully connect the enhanced high-frequency acoustic emission signal feature encoding vector to obtain an enhanced high-frequency acoustic emission signal feature fully connected encoding vector; inputting the enhanced high-frequency acoustic emission signal feature fully connected encoding vector into the softmax classification function of the leak detection model to obtain the probability value of the enhanced high-frequency acoustic emission signal feature fully connected encoding vector belonging to each type label, wherein the type label is used to indicate whether the tank has a leak or the tank does not have a leak; and determining the type label corresponding to the largest of the probability values ​​as the leak identification result. That is, in order to complete the conversion from feature space to decision space and achieve the final leakage state determination, the present application is based on the classification principle in supervised learning, by constructing a leak detection model based on a classifier to learn and establish a mapping relationship between the enhanced high-frequency acoustic emission signal feature encoding vector and the leakage state, and mapping it to a specific category label. Specifically, the classifier consists of a fully connected layer (Dense Layer) and an output layer. In the fully connected layer, the enhanced high-frequency acoustic emission signal feature coding vector is subjected to nonlinear transformation and dimensional compression, and is finally projected to an output layer equal to the number of preset categories (in this application, 2 neurons, representing "leakage exists" and "no leakage" respectively). The activation function of the output layer uses the Softmax function to convert the output features of the fully connected layer into a probability distribution, that is, output two numerical values, which represent the probability of the input signal being judged as a leak and no leak, respectively, and the sum of the two is 1. The training process of the model will use a large amount of labeled leakage and signal data under normal working conditions, and continuously optimize the network parameters through the back propagation algorithm until the model can accurately output high-confidence probabilities for input signals of different categories. In this way, intelligent and high-precision identification of tank leakage events can be achieved, providing operators with reliable, intuitive and easy-to-understand decision-making basis, so that corresponding maintenance measures can be taken in a timely manner.

[0074] In summary, the IoT-based tank safety monitoring method based on the embodiment of the present application is explained, which synchronously collects high-frequency leakage acoustic emission signals and reference environmental noise signals through piezoelectric acoustic emission sensors deployed on the outer wall of the tank and industrial-grade microphones deployed in the tank area. By aligning the reference environmental noise signal with the high-frequency leakage acoustic emission signal after time delay, adaptive filtering is performed to filter out the environmental noise interference in the high-frequency leakage acoustic emission signal and improve the signal-to-noise ratio of the signal. Then, a deep learning algorithm is further introduced to perform time domain waveform analysis and adaptive content-aware enhancement processing on the denoised high-frequency leakage acoustic emission signal to strengthen the key acoustic emission features in the signal, thereby realizing intelligent identification of tank leaks on this basis. This method can effectively improve the accuracy and sensitivity of tank leakage monitoring, overcome the influence of noise interference in the industrial site on tank leakage identification, and realize early warning of small tank leaks.

[0075] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0076] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0077] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0078] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0079] Finally, it should be noted that the above description has been provided for purposes of illustration and description. Furthermore, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A storage tank safety monitoring method based on the Internet of Things, characterized in that: include: High-frequency acoustic emission signals are collected by piezoelectric acoustic emission sensors deployed on the outer wall of the storage tank, and reference environmental noise signals are simultaneously collected by industrial-grade microphones deployed in the storage tank area. Performing time delay alignment on the high-frequency acoustic emission signal and the reference ambient noise signal to obtain an aligned high-frequency acoustic emission signal and an aligned reference ambient noise signal; Inputting the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal into a normalized LMS filter to filter out the environmental noise to obtain a denoised high-frequency acoustic emission signal; The denoised high-frequency acoustic emission signal is input into a storage tank leakage identification model based on signal waveform feature analysis to determine whether the storage tank has a leak.

2. The method for storage tank safety monitoring based on the Internet of Things according to claim 1 is characterized in that: Performing time delay alignment on the high-frequency acoustic emission signal and the reference ambient noise signal to obtain an aligned high-frequency acoustic emission signal and an aligned reference ambient noise signal, comprising: Calculating a cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal, and determining a time delay based on a peak position of the cross-correlation function between the high-frequency acoustic emission signal and the reference ambient noise signal; Time delay alignment processing is performed on the high-frequency acoustic emission signal and the reference environmental noise signal according to the time delay to obtain the aligned high-frequency acoustic emission signal and the aligned reference environmental noise signal.

3. The method for storage tank safety monitoring based on the Internet of Things according to claim 1, characterized in that: Inputting the de-noised high-frequency acoustic emission signal into a tank leakage identification model based on signal waveform feature analysis to determine whether the tank is leaking, including: Extracting the signal features of the denoised high-frequency acoustic emission signal to obtain a high-frequency acoustic emission signal feature coding vector; performing adaptive content-aware enhancement processing on the high-frequency acoustic emission signal feature coding vector to obtain an enhanced high-frequency acoustic emission signal feature coding vector; The enhanced high-frequency acoustic emission signal feature coding vector is input into a classifier-based leakage detection model to obtain a leakage identification result, and the leakage identification result is used to indicate whether there is a leakage in the storage tank.

4. The method for storage tank safety monitoring based on the Internet of Things according to claim 3 is characterized in that: Extracting the signal features of the denoised high-frequency acoustic emission signal to obtain a high-frequency acoustic emission signal feature coding vector includes: The denoised high-frequency acoustic emission signal is subjected to signal feature extraction based on a 1D-CNN model to obtain a feature coding vector of the high-frequency acoustic emission signal.

5. The method for storage tank safety monitoring based on the Internet of Things according to claim 4 is characterized in that: Performing adaptive content-aware enhancement processing on the high-frequency acoustic emission signal feature coding vector to obtain an enhanced high-frequency acoustic emission signal feature coding vector, comprising: Performing content perception on the high-frequency acoustic emission signal feature coding vector to obtain a local time-domain content-related perception coding feature map of the high-frequency acoustic emission signal; Inputting the local temporal content-related perceptual coding feature map of the high-frequency acoustic emission signal into the spatial attention layer to obtain the local temporal content-related spatially significant perceptual coding feature map of the high-frequency acoustic emission signal; The local time-domain correlation content space-significant perceptual coding feature map of the high-frequency acoustic emission signal is reshaped to obtain the enhanced high-frequency acoustic emission signal feature coding vector.

6. The method for storage tank safety monitoring based on the Internet of Things according to claim 5, characterized in that: Performing content perception on the high-frequency acoustic emission signal feature coding vector to obtain a local time-domain content-related perceptual coding feature map of the high-frequency acoustic emission signal includes: Performing feature segmentation on the high-frequency acoustic emission signal feature coding vector based on a preset window to obtain a set of local time domain feature vectors of the high-frequency acoustic emission signal; After reshaping the collective characteristic shape of the local time domain feature vector of the high-frequency acoustic emission signal into a local time domain coding feature map of the high-frequency acoustic emission signal, content perception based on convolutional coding is performed on it to obtain the local time domain content association perception coding feature map of the high-frequency acoustic emission signal.

7. The method for storage tank safety monitoring based on the Internet of Things according to claim 6, characterized in that: Inputting the local temporal content-related perceptual coding feature map of the high-frequency acoustic emission signal into the spatial attention layer to obtain the local temporal content-related spatially significant perceptual coding feature map of the high-frequency acoustic emission signal, comprising: Performing global mean pooling along the channel dimension on the local time domain content-associated perceptual coding feature map of the high-frequency acoustic emission signal to obtain a local time domain content-aware pooling coding vector of the high-frequency acoustic emission signal, and inputting the local time domain content-aware pooling coding vector of the high-frequency acoustic emission signal into a multi-layer perceptron model to obtain dynamic convolution kernel parameters; Based on the dynamic convolution kernel parameters, spatial saliency features are extracted from the local time-domain content-related perceptual coding feature map of the high-frequency acoustic emission signal to obtain the local time-domain content-related spatial saliency perceptual coding feature map of the high-frequency acoustic emission signal.

8. The method for storage tank safety monitoring based on the Internet of Things according to claim 7 is characterized in that: Inputting the local time-domain content-aware pooling coding vector of the high-frequency acoustic emission signal into a multi-layer perceptron model to obtain the dynamic convolution kernel parameters includes: Performing constrained optimization on the bias vector of the multilayer perceptron model based on phase stability of network parameter space to obtain an optimized bias vector; Based on the optimized bias vector, the multi-layer perceptron model is used to process the local time domain content-aware pooling coding vector of the high-frequency acoustic emission signal to obtain the dynamic convolution kernel parameters.

9. The method for storage tank safety monitoring based on the Internet of Things according to claim 3, characterized in that: Inputting the enhanced high-frequency acoustic emission signal feature encoding vector into a classifier-based leak detection model to obtain a leak identification result, which is used to indicate whether there is a leak in the storage tank, includes: Using the fully connected layer of the leakage detection model to perform fully connected encoding on the enhanced high-frequency acoustic emission signal feature coding vector to obtain an enhanced high-frequency acoustic emission signal feature fully connected coding vector; Inputting the fully connected encoding vector of the enhanced high-frequency acoustic emission signal feature into the Softmax classification function of the leakage detection model to obtain a probability value of the fully connected encoding vector of the enhanced high-frequency acoustic emission signal feature belonging to each type label, wherein the type label is used to indicate whether there is a leak in the tank or not; The type label corresponding to the largest probability value among the probability values ​​is determined as the leakage identification result.

Images