Array ultrasonic based three-dimensional state visualization monitoring system and method for anaerobic fermentation

Abstract

The application provides an anaerobic fermentation three-dimensional state visualization monitoring system and method based on an array ultrasound, which comprises a multi-plane ultrasound scanning array module, a space feature extraction and state inversion module, a visualization and human-computer interaction module and a three-dimensional acoustic parameter field reconstruction module; the multi-plane ultrasound scanning array module is arranged on the outer wall of a fermentation reactor and emits and receives ultrasound signals from multiple angles and layers; the three-dimensional acoustic parameter field reconstruction module reconstructs a three-dimensional space distribution nephogram of sound velocity and sound attenuation coefficient inside the fermentation reactor according to the received ultrasound signals; the space feature extraction and state inversion module maps the three-dimensional space distribution nephogram to a three-dimensional state diagram of microbial activity, substrate concentration or gas phase distribution by using a machine learning model; and the visualization and human-computer interaction module is used for displaying the three-dimensional state diagram in real time. The application realizes three-dimensional, non-invasive and visual monitoring of the anaerobic fermentation process and has the advantages of high spatial resolution and strong early warning capability.

Description

Technical Field

[0001] This invention relates to the technical fields of bioreactor engineering and advanced sensing, and in particular to a three-dimensional state visualization monitoring system and method for anaerobic fermentation based on array ultrasound and spatial feature recognition for large-scale anaerobic fermentation reactors. Background Technology

[0002] Anaerobic fermentation technology is a core method for realizing the resource utilization of organic waste and producing biogas (such as biogas), and is widely used in agricultural waste treatment, municipal sludge disposal, and industrial organic wastewater treatment. The core of this technology lies in the metabolic activities of the microbial community within the reactor. This process involves complex mass transfer and biochemical reactions across the gas, liquid, and solid phases. Parameters such as biomass distribution, substrate concentration gradient, core physicochemical properties of the fermentation broth (volatile fatty acid concentration, ammonia nitrogen concentration), gas phase holdup, and mixing uniformity directly determine fermentation efficiency and system stability. Therefore, optimizing the anaerobic fermentation process highly depends on precise perception of the reactor's internal state (such as biomass distribution, gas-liquid-solid three-phase structure, and substrate concentration field). Existing technologies have significant shortcomings:

[0003] 1. Limitations in monitoring dimensions: As described in CN118109284A, existing systems mostly rely on single-point or related indirect parameters such as temperature and pressure, and cannot obtain spatial distribution information.

[0004] 2. Risk of invasive damage: Electrochemical probes are prone to contamination, electrode corrosion and other problems, which can lead to measurement drift and require frequent calibration or replacement.

[0005] 3. Limited application of ultrasound: CN220413365U discloses the use of ultrasound to measure the thickness of biogas residue deposits, but this technology has a limited function and is only used for measuring static sediment layers, not for the three-dimensional characterization of dynamic biochemical processes.

[0006] 4. Lack of overall state profile: Existing methods are unable to non-invasively reconstruct the three-dimensional spatial distribution of microbial activity, material mixing state and metabolic intensity inside the reactor, leading to blind process control. Summary of the Invention

[0007] To address the shortcomings of the aforementioned background technologies, this invention proposes a monitoring system and method that can non-invasively, visually, and quantitatively present the three-dimensional state distribution inside an anaerobic fermentation reactor. Through external ultrasonic array scanning and intelligent image reconstruction, it achieves a leap from "parameter monitoring" to "spatial state visualization".

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0009] A three-dimensional state visualization monitoring system for anaerobic fermentation based on array ultrasound includes: a multi-planar ultrasonic scanning array module, a spatial feature extraction and state inversion module, a visualization and human-computer interaction module, and a three-dimensional acoustic parameter field reconstruction module.

[0010] The multi-planar ultrasonic scanning array module is arranged on the outer wall of the fermentation reactor, transmitting and receiving ultrasonic signals from multiple angles and layers; the three-dimensional acoustic parameter field reconstruction module reconstructs a three-dimensional spatial distribution cloud map of sound velocity and sound attenuation coefficient inside the fermentation reactor based on the received ultrasonic signals; the spatial feature extraction and state inversion module uses a machine learning model to map the three-dimensional spatial distribution cloud map into a three-dimensional state map of microbial activity, substrate concentration, or gas phase distribution; the visualization and human-computer interaction module is used to display the three-dimensional state map in real time.

[0011] A method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound, comprising the following steps:

[0012] S1: The fermentation reactor is scanned by an external multi-planar ultrasonic scanning array module to obtain multi-angle ultrasonic signals;

[0013] S2: Reconstruct the three-dimensional distribution field of acoustic parameters inside the reactor using the three-dimensional acoustic parameter field reconstruction module;

[0014] S3: Extract spatial features from the three-dimensional distribution field through the spatial feature extraction and state inversion module;

[0015] S4: The spatial features are inverted into a three-dimensional distribution map reflecting the fermentation state through a pre-trained model;

[0016] S5: Visualize and interactively analyze the 3D distribution map through the visualization and human-computer interaction module.

[0017] Preferably, the multi-planar ultrasonic scanning array module is a multi-planar ultrasonic scanning array, which consists of a parallel array of ring probes arranged along the axial direction of the fermentation reactor. Each ring probe array contains multiple uniformly distributed ultrasonic probes, forming a multi-layer, multi-angle scanning network.

[0018] Preferably, the ultrasonic probe includes a low-frequency probe and a medium-frequency probe, which are acoustically coupled to the outer wall of the fermentation reactor through a special coupling agent; the raw signals collected by all probes are transmitted to a signal processing server for centralized processing.

[0019] Preferably, the multi-angle ultrasound signal includes:

[0020] Time to travel of sound waves:

[0021] (1)

[0022] in, For ultrasonic waves from the transmitting probe to the receiving probe The time of spread For the propagation path of sound waves, Represents any point in the medium Local sound velocity;

[0023] Acoustic signal attenuation model:

[0024] (2)

[0025] in, The initial amplitude of the transmitted signal. To receive the signal amplitude, Let be the local acoustic attenuation coefficient of the medium; taking the logarithm of equation (2) yields the linear equation:

[0026] .

[0027] Preferably, the three-dimensional acoustic parameter field reconstruction module is based on ultrasonic tomography or synthetic aperture focusing algorithm to reconstruct the sound velocity distribution cloud map, sound attenuation coefficient distribution cloud map and backscatter intensity distribution cloud map of the cross section and longitudinal section inside the reactor.

[0028] Constructing a sound velocity field based on ultrasonic tomography algorithm To optimize the objective function of the variables, the difference between the measured propagation time and the theoretical calculation value is minimized, and a regularization term is introduced to suppress noise, thereby achieving high-precision reconstruction of the sound velocity distribution. At the same time, a synthetic aperture focusing algorithm is used to coherently superimpose the signals.

[0029] The optimization objective function for ultrasound tomography reconstruction is:

[0030] (3)

[0031] The first term on the right-hand side of the function equation is the data fitting term, which measures the difference between the propagation time calculated from the reconstructed sound velocity field and the actual measured value; the second term is the regularization term, which modifies the gradient of the sound velocity field. Apply a smoothing constraint; The regularization coefficient is used.

[0032] The formula for synthetic aperture focusing imaging is:

[0033] (4)

[0034] in, To reconstruct the image at points Strength at the location; This indicates the distance from the point to the probe. and The round-trip distance; For reference speed of sound; For the Dirac function, These are the weighting coefficients.

[0035] Preferably, the spatial feature extraction and state inversion module includes a spatial state inversion unit, which extracts spatial gradient features, regional statistical features, and heterogeneity index from the reconstructed three-dimensional acoustic parameter field; and maps the three-dimensional acoustic feature field into a biological activity intensity distribution map, a substrate concentration distribution trend map, and a gas phase holding rate distribution map through a pre-trained machine learning model.

[0036] Spatial gradient feature calculation:

[0037] (5)

[0038] in, Represents the acoustic parameter field The rate of change in three-dimensional space; discretization of equation (5) using the central difference method:

[0039]

[0040] Heterogeneity index:

[0041] (6)

[0042] The reconstruction area is divided into Calculate the value for each sub-region. mean and standard deviation Then take the relative coefficient of variation. The average value; where the mean is and standard deviation It is a regional statistical characteristic;

[0043] The machine learning model is a convolutional neural network, which is used to construct a nonlinear mapping model from acoustic features to fermentation state parameters.

[0044] CNN forward propagation formula:

[0045] (7)

[0046] in, For the input feature map, For convolution kernel weights, For bias terms, Activation function; output This is the feature map for the next layer.

[0047] Preferably, the loss function for training the machine learning model is:

[0048] (8)

[0049] The first term in this loss function is the mean squared error, which measures the model's predicted value. With real labels The difference between them; the second term is the weight decay term, This is a hyperparameter.

[0050] Preferably, the visualization and human-computer interaction module displays the inverted three-dimensional state distribution data in real time on the monitoring interface in the form of pseudo-color cloud map, dynamic isosurface, and profile animation; the inverted three-dimensional state distribution data is converted into an intuitive color cloud map through a pseudo-color mapping function, and the operator can view the state distribution of any profile in real time through the interactive interface; the deviation of the state parameters of each point from the average value of the whole field is calculated simultaneously, and when the deviation exceeds the preset threshold, an early warning is triggered, and the abnormal area is highlighted in the visualization interface;

[0051] The pseudo-color mapping function is:

[0052] (9)

[0053] in, For state parameters, and These are the minimum and maximum values ​​of the current data, respectively.

[0054] The local anomaly early warning indicators are:

[0055] (10)

[0056] in, The current state parameter is the average value across the entire reactor. Its standard deviation; if a point If the deviation from the mean exceeds two standard deviations, it is considered abnormal and an alert is triggered.

[0057] The beneficial effects of this invention are:

[0058] 1) Achieve truly non-invasive three-dimensional monitoring: Measured entirely from outside the tank, avoiding any interference with the anaerobic environment.

[0059] 2) Provides intuitive spatial status information: Operators can directly "see" where the reactor is active, where it is stagnant, and whether the mixing is uniform, providing a direct basis for precise control (such as local stirring and directional feeding).

[0060] 3) Enhanced early warning capability: Localized activity abnormalities or precursors of acidification will first appear in the three-dimensional distribution map, earlier than changes in overall average parameters (such as average pH).

[0061] 4) Suitable for large industrial reactors: The array design can be adapted to tanks of different sizes and shapes, and the system has strong scalability. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. It should be noted that the drawings in the following description are only some embodiments of the present invention.

[0063] Figure 1 This is an overall flowchart of the monitoring system described in this invention, illustrating the complete data processing flow from ultrasonic signal acquisition to three-dimensional visualization and early warning;

[0064] Figure 2 This is a schematic diagram of the overall architecture and probe array arrangement of the monitoring system of the present invention, showing the multi-layer ring array arrangement of the ultrasonic probes on the outer wall of the reactor;

[0065] Figure 3 The flowchart for the reconstruction of three-dimensional acoustic parameter fields details the reconstruction process from the original ultrasonic signal to the three-dimensional distribution of sound velocity and sound attenuation coefficient.

[0066] Figure 4 This is a schematic diagram of the inversion model from acoustic parameter field to biological activity distribution map, demonstrating the state inversion mechanism based on convolutional neural network;

[0067] Figure 5 The example image shows a visual monitoring interface, including a 3D status cloud map, anomaly warnings, and an interactive query interface. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Example 1, as Figure 1 As shown, a method for visually monitoring the three-dimensional state of anaerobic fermentation based on array ultrasound includes the following steps:

[0070] S1: The reactor is scanned by an external multi-planar ultrasonic scanning array module to obtain multi-angle ultrasonic signals;

[0071] S2: Reconstruct the three-dimensional distribution field of acoustic parameters inside the reactor using the three-dimensional acoustic parameter field reconstruction module;

[0072] S3: Extract spatial features from the three-dimensional distribution field through the spatial feature extraction and state inversion module;

[0073] S4: The spatial features are inverted into a three-dimensional distribution map reflecting the fermentation state through a pre-trained model;

[0074] S5: Visualize and interactively analyze the 3D distribution map through the visualization and human-computer interaction module.

[0075] Example 2: A three-dimensional state visualization monitoring system for anaerobic fermentation based on array ultrasound and spatial feature recognition, comprising: a multi-planar ultrasound scanning array module, a spatial feature extraction and state inversion module, a visualization and human-computer interaction module, and a three-dimensional acoustic parameter field reconstruction module.

[0076] like Figure 2 As shown, the multi-planar ultrasonic scanning array module is a multi-planar ultrasonic scanning array. Parallel ring-shaped probe arrays are arranged along the reactor axis (Z-direction) on the outer wall of the fermenter. Each ring array contains no fewer than six uniformly distributed ultrasonic probes, forming a multi-layered, multi-angle scanning network for transmitting and receiving ultrasonic signals from multiple angles and layers. The operating frequencies of the ultrasonic probes cover low frequencies (100-500kHz, used for deep penetration and overall structural imaging) and mid-frequency frequencies (1-2MHz, used for local fine feature identification).

[0077] like Figure 2 As shown, the ultrasonic probe includes a low-frequency probe 2 and a medium-frequency probe 3, which are acoustically coupled to the outer wall of the fermenter via a special coupling agent 1, ensuring that the ultrasonic signal effectively penetrates the tank wall and enters the reactor. All raw signals collected by the probes are transmitted to a signal processing server 4 for centralized processing.

[0078] The ultrasonic probe array emits ultrasonic signals on the outer wall of the fermenter. The sound waves penetrate the three-phase "gas-liquid-solid" medium inside the tank and are captured by the receiving probes. Based on the physical laws of sound wave propagation in a medium, the propagation time... With medium sound speed The amplitude of the received signal is inversely proportional to the amplitude of the received signal, as shown in equation (1). Attenuation coefficient It exhibits exponential decay, as shown in equation (2). This step converts the physical properties of the medium inside the tank into quantifiable electrical signals, providing raw data for subsequent processing. Equations (1) and (2) define the mathematical relationship between basic physical parameters such as sound wave velocity and attenuation and raw measurement data (such as flight time and signal strength), ensuring that the raw data is correctly converted into physically meaningful parameters.

[0079] The formulas for ultrasound signal acquisition and its physical basis are as follows:

[0080] The formula for sound wave flight time (used for sound speed reconstruction) is:

[0081] (1)

[0082] Equation (1) is used to calculate the ultrasonic waves emitted from the transmitting probe. to the receiving probe transmission time ,in For the propagation path of sound waves, Represents any point in the medium The local sound velocity. In numerical calculations, this integration path is typically discretized into several small segments, each with a sound velocity that can be considered constant, thus allowing for solution via iterative inversion methods. Spatial distribution.

[0083] The acoustic signal attenuation model (used for attenuation coefficient reconstruction) is as follows:

[0084] (2)

[0085] This model describes the ultrasonic amplitude along the propagation path. The decay process. The initial amplitude of the transmitted signal. To receive the signal amplitude, Let be the local acoustic attenuation coefficient of the medium. Taking the logarithm yields the linear equation:

[0086]

[0087] This linear equation is used for reconstruction. The three-dimensional distribution.

[0088] like Figure 3 As shown, the three-dimensional acoustic parameter field reconstruction module is used to receive the original radio frequency signals collected by each probe, and reconstruct the sound velocity distribution cloud map, sound attenuation coefficient distribution cloud map, and backscatter intensity distribution cloud map of the internal cross section (XY plane) and longitudinal section (XZ / YZ plane) of the reactor based on ultrasonic tomography or synthetic aperture focusing algorithm.

[0089] Based on the ultrasonic computed tomography (UT) algorithm, a sound velocity field is constructed. To optimize the objective function of the variables, as shown in formula (3), the difference between the measured propagation time and the theoretical calculated value is minimized, and a regularization term is introduced to suppress noise, thereby achieving high-precision reconstruction of the sound velocity distribution. Simultaneously, the Synthetic Aperture Focusing (SAFT) algorithm can be used to coherently superimpose the signals, improving the spatial resolution of the image. This step realizes the conversion from discrete path measurement data to a continuous three-dimensional acoustic parameter field. The three-dimensional image reconstruction algorithm describes how discrete physical parameter measurements are reconstructed into two-dimensional / three-dimensional images or parameter distribution maps reflecting the internal structure of the fermenter through specific mathematical inversion methods (such as tomographic imaging algorithms).

[0090] The optimization objective function for ultrasound tomography reconstruction is:

[0091] (3)

[0092] This function represents the sound velocity field. The reconstruction provides optimization criteria. The first term is the data fitting term, which measures the difference between the propagation time calculated from the reconstructed sound velocity field and the actual measured value; the second term is the regularization term, which adjusts the gradient of the sound velocity field. Apply smoothing constraints to prevent excessive oscillations caused by noise during the reconstruction process. This is the regularization coefficient, which is usually determined through cross-validation.

[0093] The formula for synthetic aperture focusing imaging is:

[0094] (4)

[0095] This formula enables synthetic aperture focusing imaging. To reconstruct the image at points Strength at the location; This indicates the distance from the point to the probe. and The round-trip distance; The reference speed of sound (usually the average speed of sound in the medium); The Dirac function is used to select the signal component that matches the time delay; These are weighting coefficients used to improve the imaging signal-to-noise ratio.

[0096] like Figure 4 As shown, the spatial feature extraction and state inversion module includes a spatial state inversion unit, which extracts spatial gradient features, regional statistical features (such as the mean and variance of each region), and heterogeneity index from the reconstructed three-dimensional acoustic parameter field. Through a pre-trained machine learning model (such as a convolutional neural network CNN), the three-dimensional acoustic feature field is mapped into a biological activity intensity distribution map, a substrate concentration distribution trend map, and a gas phase holding rate distribution map.

[0097] The reconstructed sound velocity and attenuation fields both contain rich spatial information. This is achieved by calculating the spatial gradient of the acoustic parameters. It can identify structural features within the medium, such as phase interfaces and concentration abrupt change regions. Furthermore, the reactor is divided into several sub-regions, and the statistical characteristics (mean values) of each region are calculated. Standard deviation And synthesize the heterogeneity index. This step quantifies the spatial non-uniformity of the medium within the tank. It uses mathematical methods (such as statistical features, texture features, and shape feature calculations) to extract quantitative indicators characterizing the internal state (e.g., bubble distribution, mycelial concentration). The high-dimensional image data is then reduced to feature vectors with clear physical meaning, which are applied to the reconstructed image or parametric field data. The spatial feature extraction formula is as follows:

[0098] Spatial gradient calculation (used to identify boundaries and regions of change):

[0099] (5)

[0100] This vector represents the acoustic parameter field. The rate of change of (e.g., sound velocity, attenuation coefficient) in three-dimensional space. In practical calculations, this can be discretized using the central difference method:

[0101]

[0102] Among them, gradient magnitude It can be used to identify discontinuous interfaces or regions of abrupt concentration changes within a medium.

[0103] Regional heterogeneity index:

[0104] (6)

[0105] This index is used to quantify the spatial non-uniformity of acoustic parameters. The reconstructed region is divided into... Calculate the value for each sub-region. mean and standard deviation Then take the relative coefficient of variation. The average value. The higher the value, the more uneven the medium is.

[0106] A nonlinear mapping model from acoustic features to fermentation state parameters is constructed using a convolutional neural network (CNN). This model extracts deep features progressively through multi-layer convolution and nonlinear activation, ultimately outputting key process parameters such as VFA concentration and gas phase holdup. Model training employs a mean squared error loss function combined with weight decay regularization to improve generalization ability. This step establishes a prediction or classification model between the extracted acoustic features and target fermentation state parameters (such as biomass, viscosity, and product concentration), achieving intelligent inference from "observable acoustic features" to "non-measurable process parameters." The state inversion (machine learning model) formula is as follows:

[0107] CNN forward propagation formula (taking a single convolutional layer as an example):

[0108] (7)

[0109] This formula describes the operation of convolutional layers. Input feature map (size) ), Convolution kernel weights (size) ), For bias terms, The activation function is (e.g., ReLU). Output For the next layer of feature maps, each element is obtained by weighting the local receptive fields and then performing nonlinear activation.

[0110] The model training loss function is:

[0111] (8)

[0112] This loss function is used to optimize the parameters of the state inversion model. The first term is the mean squared error, which measures the model's predicted value. With real labels The difference between them; the second term is the weight decay term, used to prevent overfitting. This is a hyperparameter.

[0113] like Figure 5 As shown, the visualization and human-computer interaction module displays the inverted three-dimensional state distribution data in real time on the monitoring interface in the form of pseudo-color cloud maps, dynamic isosurfaces, and profile animations, and supports querying specific values ​​in spatial coordinates.

[0114] The inverted three-dimensional state parameter field is converted into an intuitive color cloud map using a pseudo-color mapping function. Operators can view the state distribution of any profile in real time through an interactive interface. The system synchronously calculates the deviation of the state parameters at each point from the overall field mean. When the deviation exceeds a preset threshold (e.g., ...), ... An alert is triggered when an anomaly occurs, and the abnormal area is highlighted in the visualization interface. This step transforms the numerical analysis results into intuitive monitoring and decision support information. The visualization and alert formulas are as follows:

[0115] The pseudo-color mapping function is:

[0116] (9)

[0117] This function will retrieve the state parameters obtained through inversion. (e.g., VFA concentration, gas phase fraction) normalized to The range is then converted into pseudo-color using a preset color mapping table (such as jet or viridis). and These are the minimum and maximum values ​​of the current data, respectively.

[0118] The local anomaly early warning indicators are:

[0119] (10)

[0120] This indicator is used to automatically detect local anomalies. The current state parameter is the average value across the entire reactor. Its standard deviation. If a certain point If the deviation from the mean exceeds two standard deviations, it is considered abnormal and an alert is triggered.

[0121] This invention guides process adjustments (such as local stirring and feed optimization) based on early warning information, and re-collects data, reconstructs images, and evaluates the control effect in subsequent monitoring cycles, forming a closed-loop operation mode of "monitoring-early warning-control-verification" to achieve refined management of the anaerobic fermentation process.

[0122] Specific example: A cylindrical anaerobic fermenter with an effective volume of 150m³ is used for specific numerical illustration.

[0123] 1) Hardware Deployment and Data Acquisition: The tank has a diameter of 4.8m and a height of 8.5m. Four layers of ring arrays are arranged axially outside the tank, with axial positions z=1.0, 3.0, 5.0, and 7.0 m respectively. Each layer has 12 waterproof and corrosion-resistant ultrasonic probes evenly distributed, and they are kept acoustically coupled to the tank wall using a special coupling agent. The arc length between adjacent probes is 1.26m. The probes adopt a dual-frequency design: a low frequency of 200kHz for overall penetration imaging and a mid-frequency of 1.2MHz for local fine identification. All probes are connected to a central signal processing cabinet. The data acquisition cycle is a full array scan every 20 minutes. Typical measured data for probes 3 to 9 in the second layer are: straight-line distance L=3.82 m; propagation time t=2.58 ms; transmitted signal amplitude A0=5.0 V; received signal amplitude A=0.85 V.

[0124] 2) Three-dimensional acoustic parameter field reconstruction: Reconstruction was performed using the Ultrasonic Tomography (UT) algorithm. Specific parameters were: the reconstruction mesh was divided into 64×64×32 cells (a total of 131,072 mesh elements); the mesh size was... The regularization coefficient is The number of iterations was 50, and the convergence residual was... .

[0125] The acoustic parameters of some areas after reconstruction are shown in the table below:

[0126]

[0127] 3) Spatial feature extraction calculation:

[0128] First, perform gradient feature calculation (taking a point at the boundary of the anomaly region as an example):

[0129]

[0130]

[0131] Gradient magnitude:

[0132]

[0133] Then, the heterogeneity index is calculated:

[0134] The horizontal cross-section of the reactor was divided into eight 45° sector regions. The coefficient of variation of sound velocity in each region was calculated and averaged.

[0135]

[0136] in and The first Standard deviation and mean of regional sound velocity.

[0137] 4) Intelligent inversion of fermentation state

[0138] A trained 3DCNN model was used to invert the current acoustic features. The model structure is as follows: input parameters are dual-channel data of sound velocity field and attenuation field (size 64×64×32×2); there are 5 convolutional layers with kernel size 3×3×3; the number of feature map channels is 32→64→128→64→32; there are 3 fully connected layers with 512, 256, and 2 neurons respectively (outputting VFA concentration and gas phase rate). According to the inversion results output by the neural network: the VFA concentration distribution throughout the tank is the mean. ,scope The overall gas phase holding rate distribution is the mean. ,scope Monitoring the state parameters of the abnormal area revealed a VFA concentration of 3420. At this point, the concentration was 32.6% higher than the average for the entire tank, while the gas phase holding rate was 4.2%, which was 54.8% lower than the average for the entire tank.

[0139] 5) Visualized early warning and control response

[0140] First, pseudo-color mapping is performed to normalize the VFA concentration to the [0,1] interval, as shown in the mapping result of the VFA in the abnormal region above:

[0141]

[0142] 0.874 corresponds to the red and yellow hues in the pseudo-color spectrum, which is highlighted in the 3D visualization interface.

[0143] Then, an anomaly warning judgment is made, and the average VFA of the entire tank is statistically analyzed. Standard deviation Calculate the deviation of outliers:

[0144]

[0145] The deviation value of the outlier reached The warning threshold is met, and the gas phase efficiency at that point is lower than [a certain threshold]. Warning line, the system has triggered a Level 3 (yellow) alert.

[0146] Subsequently, corresponding measures were taken to regulate the process: the bottom agitator was started (power 5.5kW, duration 20 minutes); the bottom slag discharge frequency was increased (from 6 hours / time to 4 hours / time); and the feed distribution was adjusted to reduce the bottom feed ratio by 10%.

[0147] Finally, the effect of the regulation was verified, and the results of the retest after 4 hours are as follows:

[0148]

[0149] The warning status has been downgraded from "yellow alert" to "blue concern," indicating that the control measures are effective and the system is returning to a stable operating trend.

[0150] As can be seen from the above specific numerical embodiments, the system of the present invention is capable of:

[0151] 1. Achieve three-dimensional visualization of the internal state of the anaerobic digester with a spatial resolution of ≤0.1m;

[0152] 2. A full-field monitoring is completed every 20 minutes, achieving near real-time status tracking;

[0153] 3. Key parameters such as VFA concentration and gas phase fraction are predicted using an intelligent inversion model, with a prediction error of <15%;

[0154] 4. Local anomalies can be detected and early warnings triggered 4-6 hours in advance, providing sufficient response time for process control;

[0155] 5. Form a closed loop of "monitoring-early warning-control-verification" to improve system operation stability and gas production efficiency.

[0156] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Specific parameters (such as a 150m³ tank, 4 layers of 12 probes, a 20-minute cycle, and a specific CNN structure) should not be construed as limiting the scope of protection of the present invention. This system and method are also applicable to reactors of other shapes (such as cubic or egg-shaped tanks), requiring only corresponding adjustments to the geometric arrangement of the probe array. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A three-dimensional state visualization monitoring system for anaerobic fermentation based on array ultrasound, characterized in that, include: Multiplanar ultrasound scanning array module, spatial feature extraction and state inversion module, visualization and human-computer interaction module, and three-dimensional acoustic parameter field reconstruction module; The multi-planar ultrasonic scanning array module is arranged on the outer wall of the fermentation reactor, transmitting and receiving ultrasonic signals from multiple angles and layers; the three-dimensional acoustic parameter field reconstruction module reconstructs a three-dimensional spatial distribution cloud map of sound velocity and sound attenuation coefficient inside the fermentation reactor based on the received ultrasonic signals; the spatial feature extraction and state inversion module uses a machine learning model to map the three-dimensional spatial distribution cloud map into a three-dimensional state map of microbial activity, substrate concentration, or gas phase distribution; the visualization and human-computer interaction module is used to display the three-dimensional state map in real time.

2. A method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound as described in claim 1, characterized in that, The steps are as follows: S1: The fermentation reactor is scanned by an external multi-planar ultrasonic scanning array module to obtain multi-angle ultrasonic signals; S2: Reconstruct the three-dimensional distribution field of acoustic parameters inside the reactor using the three-dimensional acoustic parameter field reconstruction module; S3: Extract spatial features from the three-dimensional distribution field through the spatial feature extraction and state inversion module; S4: The spatial features are inverted into a three-dimensional distribution map reflecting the fermentation state through a pre-trained model; S5: Visualize and interactively analyze the 3D distribution map through the visualization and human-computer interaction module.

3. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 2, characterized in that, The multi-planar ultrasonic scanning array module is a multi-planar ultrasonic scanning array with parallel ring probe arrays arranged along the axial direction of the fermentation reactor. Each ring probe array contains multiple uniformly distributed ultrasonic probes, forming a multi-layer, multi-angle scanning network.

4. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 3, characterized in that, The ultrasonic probes include a low-frequency probe (2) and a medium-frequency probe (3). The low-frequency probe (2) and the medium-frequency probe (3) are acoustically coupled to the outer wall of the fermentation reactor through a special coupling agent (1). The raw signals collected by all probes are transmitted to the signal processing server (4) for centralized processing.

5. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 3, characterized in that, The multi-angle ultrasound signals include: Time to travel of sound waves: （1） in, For ultrasonic waves from the transmitting probe To the receiving probe The time of spread For the propagation path of sound waves, Represents any point in the medium Local sound velocity; Acoustic signal attenuation model: （2） in, The initial amplitude of the transmitted signal. To receive the signal amplitude, Let be the local acoustic attenuation coefficient of the medium; taking the logarithm of equation (2) yields the linear equation: 。 6. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 5, characterized in that, The three-dimensional acoustic parameter field reconstruction module is based on ultrasonic tomography or synthetic aperture focusing algorithm to reconstruct the sound velocity distribution cloud map, sound attenuation coefficient distribution cloud map and backscatter intensity distribution cloud map of the reactor's internal cross-section and longitudinal section. Constructing a sound velocity field based on ultrasonic tomography algorithm To optimize the objective function of the variables, the difference between the measured propagation time and the theoretical calculation value is minimized, and a regularization term is introduced to suppress noise, thereby achieving high-precision reconstruction of the sound velocity distribution. At the same time, a synthetic aperture focusing algorithm is used to coherently superimpose the signals. The optimization objective function for ultrasound tomography reconstruction is: （3） The first term on the right-hand side of the function equation is the data fitting term, which measures the difference between the propagation time calculated from the reconstructed sound velocity field and the actual measured value; the second term is the regularization term, which modifies the gradient of the sound velocity field. Apply a smoothing constraint; The regularization coefficient is used. The formula for synthetic aperture focusing imaging is: （4） in, To reconstruct the image at points Strength at the location; This indicates the distance from the point to the probe. and The round-trip distance; For reference speed of sound; For the Dirac function, These are the weighting coefficients.

7. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 6, characterized in that, The spatial feature extraction and state inversion module includes a spatial state inversion unit, which extracts spatial gradient features, regional statistical features, and heterogeneity index from the reconstructed three-dimensional acoustic parameter field; and maps the three-dimensional acoustic feature field into a biological activity intensity distribution map, a substrate concentration distribution trend map, and a gas phase holding rate distribution map through a pre-trained machine learning model. Spatial gradient feature calculation: （5） in, Represents the acoustic parameter field The rate of change in three-dimensional space; discretization of equation (5) using the central difference method: Heterogeneity index: （6） The reconstruction area is divided into Calculate the value for each sub-region. mean and standard deviation Then take the relative coefficient of variation. The average value; where the mean is and standard deviation It is a regional statistical characteristic; The machine learning model is a convolutional neural network, which is used to construct a nonlinear mapping model from acoustic features to fermentation state parameters. CNN forward propagation formula: （7） in, For the input feature map, For convolution kernel weights, For bias terms, Activation function; output This is the feature map for the next layer.

8. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 7, characterized in that, The loss function for training the machine learning model is: （8） The first term in this loss function is the mean squared error, which measures the model's predicted value. With real labels The difference between them; the second term is the weight decay term, This is a hyperparameter.

9. The method for three-dimensional visualization monitoring of anaerobic fermentation based on array ultrasound according to claim 7, characterized in that, The visualization and human-computer interaction module displays the inverted three-dimensional state distribution data in real time on the monitoring interface in the form of pseudo-color cloud map, dynamic isosurface, and profile animation; the inverted three-dimensional state distribution data is converted into an intuitive color cloud map through a pseudo-color mapping function, and the operator can view the state distribution of any profile in real time through the interactive interface; the deviation of the state parameters of each point from the average value of the whole field is calculated simultaneously, and when the deviation exceeds the preset threshold, an alarm is triggered and the abnormal area is highlighted in the visualization interface; The pseudo-color mapping function is: （9） in, For state parameters, and These are the minimum and maximum values ​​of the current data, respectively. The local anomaly early warning indicators are: （10） in, The current state parameter is the average value across the entire reactor. Its standard deviation; if a point If the deviation from the mean exceeds two standard deviations, it is considered abnormal and an alert is triggered.

Images