Pneumatic conveying system abnormality processing method and system based on multi-source sensing fusion

Abstract

This invention relates to the field of pneumatic conveying technology and discloses a method and system for handling anomalies in pneumatic conveying systems based on multi-source sensor fusion. It aims to solve the problems of delayed anomaly identification, inability to distinguish anomaly types, and lack of real-time online intervention in existing technologies. The solution mainly includes: constructing a three-dimensional input tensor by synchronously acquiring multi-source information from pressure, current, and acoustic sensors; using a pre-trained anomaly detection model to identify anomaly types and output confidence levels; classifying anomalies based on confidence levels within a sliding window and executing corresponding graded fault-tolerant control operations according to the anomaly level. This invention achieves accurate early diagnosis and type identification of anomalies and can automatically trigger differentiated responses based on the severity of the anomaly, improving the operational continuity of the pneumatic conveying system.

Description

Technical Field

[0001] This invention relates to the field of pneumatic conveying technology, and specifically to a method and system for handling anomalies in a pneumatic conveying system based on multi-source sensor fusion. Background Technology

[0002] Pneumatic conveying systems, as a highly efficient and closed material transport method, have been widely used in industries such as grain processing, chemicals, pharmaceuticals, and power generation. This system carries materials through pipes using airflow, offering advantages such as flexible layout, small footprint, and high automation. However, in actual industrial operation, pneumatic conveying systems often experience abnormal conditions such as pipe blockage, dry blowing (no material conveying), and pipe leaks due to factors such as changes in material characteristics (e.g., humidity, particle size), long-term pipe wear, and fluctuations in the air supply system. If these abnormalities are not diagnosed and addressed promptly and accurately, they can lead to anything from decreased material quality and energy waste to serious consequences such as equipment damage, unplanned production line shutdowns, and even safety hazards.

[0003] Currently, the industry mainly relies on the following traditional technical solutions for monitoring and handling anomalies in pneumatic conveying systems:

[0004] Single-parameter threshold monitoring: The most common monitoring scheme is based on a single pressure sensor. By setting pressure sensors at key points in the pipeline, a blockage is determined when the pressure difference across the pipeline exceeds a preset threshold. However, this method is extremely insensitive to early, minor local blockages or micro-leaks, has a severely delayed response, and cannot distinguish the specific type of anomaly.

[0005] Indirect current monitoring: This method indirectly reflects the system load by monitoring changes in the current driving the fan or feed motor. When pipe blockage occurs, the current typically increases; during dry blowing, the current may decrease. However, this method's signal is susceptible to interference from grid voltage fluctuations and the equipment's own characteristics, resulting in a low signal-to-noise ratio. It also cannot accurately distinguish between pipe blockage, dry blowing, and other mechanical faults, leading to a high false alarm rate.

[0006] Regular manual inspection and maintenance: This method relies on operators periodically tapping on pipelines to listen for sounds, observing through sight glasses, or judging the system status based on experience. This approach is not only labor-intensive but also highly subjective, slow to respond, and unable to achieve real-time, online intervention, making it difficult to meet the demands of continuous and stable production in modern industry. Summary of the Invention

[0007] This invention aims to address the problems of existing methods for monitoring and handling anomalies in pneumatic conveying systems, which rely on single sensors and coarse control, resulting in delayed anomaly identification, inability to distinguish anomaly types, and lack of real-time online intervention. The invention proposes a method and system for handling anomalies in pneumatic conveying systems based on multi-source sensor fusion.

[0008] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0009] In a first aspect, the present invention provides an anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion, the method comprising:

[0010] Multi-channel timing signals in the pneumatic conveying system are synchronously acquired by multi-source sensors, including pressure sensors, current sensors, and acoustic sensors.

[0011] The acquired multi-channel time-series signals are preprocessed, including sliding window segmentation and normalization, to construct the input tensor;

[0012] The input tensor is fed into a pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal operation.

[0013] Anomalies are classified and determined based on the confidence level, and corresponding anomaly levels are generated.

[0014] According to the anomaly level, the corresponding graded fault-tolerant control operation is executed, and after the anomaly is handled, the pneumatic conveying system automatically returns to normal operation.

[0015] Furthermore, the synchronous acquisition frequency is not less than 100Hz, the pressure sensor is arranged at the inlet, outlet and preset key nodes of the conveying pipeline, the current sensor is used to acquire the current signal of the fan or feeding motor, and the acoustic sensor is attached to the outer wall of the conveying pipeline to acquire the flow noise spectrum with a frequency range of 200Hz to 10kHz.

[0016] Furthermore, the input tensor is constructed as follows:

[0017] The acoustic signal corresponding to the flow noise spectrum is subjected to spectrum analysis to obtain the acoustic spectrum, and the frequency band energy is extracted from the acoustic spectrum to form an acoustic spectrum energy vector;

[0018] The acoustic spectrum energy vector is aligned and extended in the time dimension using an interpolation algorithm so that it has the same number of sampling points as the pressure signal corresponding to the pressure sensor and the current signal corresponding to the current sensor within the same time window.

[0019] The aligned pressure signal sequence, current signal sequence, and interpolated acoustic spectral energy vector are combined into an input tensor.

[0020] Furthermore, the window length corresponding to each sliding window segment is 1 second, and the step size is 0.2 seconds;

[0021] Let the sampling window length be T=1s, and the sampling frequency be... Then each sampling window contains N=100 sampling points;

[0022] Let the pressure signal sequence be The current signal sequence is The interpolated acoustic spectrum energy vector is ,in, Indicates the first Pressure values ​​at each sampling point Indicates the first Current values ​​at each sampling point Indicates the first Acoustic spectral energy interpolation results corresponding to each sampling point;

[0023] The input tensor is Its dimension is N×3.

[0024] Furthermore, the training method for the anomaly detection model includes:

[0025] Acquire historical datasets, which include multi-source sensor data of the pneumatic conveying system under normal and abnormal conditions, including pipe blockage, air blowing, and leakage, and label the state of each time period in the historical data.

[0026] The multi-source sensor data in the historical dataset are preprocessed, including sliding window segmentation and normalization, to generate a training sample set;

[0027] Construct an initial neural network model based on a combination of a one-dimensional convolutional neural network and a self-attention mechanism;

[0028] The initial neural network model is trained using the training sample set, with the cross-entropy loss function as the optimization objective. The model parameters are updated using the backpropagation algorithm until the model converges, thus obtaining the anomaly detection model.

[0029] Further, based on the aforementioned confidence level, anomaly classification is determined, including:

[0030] Let the current time be The confidence sequence is The sliding window length for the confidence level is ,in, Indicates a time index;

[0031] Calculate the average confidence level of the sliding window :

[0032] ;

[0033] Calculate the standard deviation of the confidence level of the sliding window. :

[0034] ;

[0035] The anomaly level is determined according to the following rules:

[0036] If the conditions are met: and If it is, then it is judged as a level three anomaly, the highest level. This indicates the first preset confidence threshold. This indicates a preset stability threshold;

[0037] If the conditions are met: If it is, it is judged as a level 2 abnormality, among which, This indicates the second preset confidence threshold;

[0038] If the conditions are met: If the number of windows that continuously meet this condition is greater than or equal to a preset continuous window count threshold, then it is judged as a Level 1 anomaly. This indicates the third pre-set confidence threshold;

[0039] If none of the above conditions are met, the condition is considered normal.

[0040] Furthermore, the sliding window length of the confidence level ;

[0041] The first preset confidence threshold The second preset confidence threshold The third preset confidence threshold The preset stability threshold The preset threshold for the number of continuous windows is 3.

[0042] Furthermore, based on the anomaly level, corresponding hierarchical fault-tolerant control operations are performed, including:

[0043] When an abnormality is detected as Level 1, the fan speed will be automatically fine-tuned, with an adjustment range of ±5% of the rated value.

[0044] When a level 2 abnormality is detected, temporarily close the feeding valve, start the backflushing procedure, and continue for 3 seconds;

[0045] When an abnormality is determined to be Level 3, the machine will be immediately shut down and an audible and visual alarm will be triggered, and an alarm message will be pushed out.

[0046] Furthermore, the method also includes:

[0047] The pressure data collected by multiple pressure sensors arranged along the pipeline is obtained. The pressure sensors include at least a first pressure sensor installed at the pipeline inlet, a second pressure sensor installed at the pipeline outlet, and a third pressure sensor installed at at least one key node in the pipeline.

[0048] When a blockage or leakage is detected, the pressure gradient between adjacent pressure sensors is calculated.

[0049] Based on the abrupt change point of the pressure differential gradient and combined with the pipeline topology information, the pipe segment range in which the anomaly occurred is determined.

[0050] Based on the length of the pipe section and the preset pressure-distance mapping relationship, the specific location of the anomaly is estimated by interpolation calculation.

[0051] Secondly, the present invention provides an anomaly handling system for a pneumatic conveying system based on multi-source sensor fusion, for implementing the anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion as described in the first aspect, the system comprising:

[0052] The signal acquisition module is used to synchronously acquire multi-channel timing signals in the pneumatic conveying system through multi-source sensors, including pressure sensors, current sensors, and acoustic sensors.

[0053] The signal preprocessing module is used to preprocess the acquired multi-channel time-series signals, including sliding window segmentation and normalization, and to construct the input tensor.

[0054] An anomaly diagnosis module is used to input the input tensor into a pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal.

[0055] The hierarchical decision-making module is used to determine the anomaly classification based on the confidence level and generate the corresponding anomaly level.

[0056] The fault-tolerant control module is used to perform corresponding graded fault-tolerant control operations according to the anomaly level, and to control the pneumatic conveying system to automatically return to normal operation after the anomaly is handled.

[0057] The beneficial effects of this invention are as follows: The pneumatic conveying system anomaly handling method and system based on multi-source sensor fusion provided by this invention firstly overcomes the limitations of single-sensor monitoring by fusing pressure, current, and acoustic multi-source sensor information. It can capture subtle changes in system state from different physical dimensions, achieving early and comprehensive perception of various abnormal operating conditions such as pipe blockage, air blowing, and leakage. Secondly, by constructing a complete automated closed loop from signal preprocessing and intelligent diagnosis to hierarchical control, it utilizes a pre-trained anomaly detection model to achieve accurate identification of anomaly types, overcoming the shortcomings of traditional methods that can only alarm but not classify. Furthermore, the construction of a hierarchical fault-tolerant control mechanism changes the traditional extensive management mode of shutting down immediately upon an anomaly. It can automatically trigger differentiated responses based on the severity of the anomaly, enabling the system to autonomously handle most minor anomalies without interrupting operation, thereby significantly improving the operational continuity, intelligence level, and overall economic benefits of the pneumatic conveying system. Attached Figure Description

[0058] Figure 1 A flowchart illustrating an anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion, provided as an example.

[0059] Figure 2 This is a schematic diagram of the anomaly handling system of the pneumatic conveying system based on multi-source sensor fusion, provided as an example. Detailed Implementation

[0060] Existing monitoring and handling solutions for pneumatic conveying systems rely on a single sensing dimension, using only physical quantities such as pressure or current. This makes it difficult to comprehensively and early detect subtle changes in the system's state, and also fails to differentiate between different types of faults, such as pipe blockage, air blowing, or leakage. Furthermore, once an alarm is triggered, the control logic often directly shuts down the system, leading to numerous unnecessary production interruptions. Therefore, this invention addresses these issues.

[0061] In this invention, firstly, pressure sensors, current sensors, and acoustic sensors deployed in the delivery pipeline synchronously acquire multi-channel time-series signals, forming a multi-dimensional perception of the system's hydrodynamic characteristics, motor load status, and acoustic propagation characteristics. Next, the original signals are standardized using sliding window segmentation and normalization. By converting the acoustic signals into spectral energy vectors and aligning them through interpolation, a three-dimensional input tensor with a uniform time-series length is constructed. This tensor is then input into a pre-trained anomaly detection model, which outputs the anomaly type and its confidence level through deep feature extraction and pattern recognition. Based on the confidence level sequence, the average confidence level and standard deviation are statistically calculated using a sliding window, and a graded judgment from minor anomalies to severe faults is achieved according to preset threshold rules. Finally, a corresponding graded fault-tolerant control strategy is triggered based on the anomaly level, completing a closed-loop control from state perception and intelligent diagnosis to precise execution, and automatically restoring system operation after the anomaly is eliminated.

[0062] The technical solutions in this embodiment 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.

[0063] Figure 1 A flowchart illustrating an anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion is shown. Please refer to [link / reference]. Figure 1 The method includes the following steps:

[0064] Step 1: Synchronously acquire multi-channel timing signals from the pneumatic conveying system using multi-source sensors, including pressure sensors, current sensors, and acoustic sensors.

[0065] In this embodiment, the synchronous acquisition frequency is not less than 100Hz, the pressure sensor is arranged at the inlet, outlet and preset key nodes of the conveying pipeline, the current sensor is used to acquire the current signal of the fan or feeding motor, and the acoustic sensor is attached to the outer wall of the conveying pipeline to acquire the flow noise spectrum with a frequency range of 200Hz to 10kHz.

[0066] Specifically, when an anomaly occurs in a pneumatic conveying system, its mechanical, electrical, and acoustic characteristics undergo coupled changes. For example, pipe blockage can lead to pressure field reconstruction, changes in load torque, and shifts in the acoustic spectrum of solid-phase collisions. This invention constructs a multi-physics coupled sensing network by simultaneously acquiring pressure, current, and acoustic signals. Through the differentiated response characteristics of different physical quantities to the same abnormal operating condition, a complementary enhancement effect is formed in the characteristic space.

[0067] In practical applications, pressure sensors are deployed at the inlet, outlet, and key nodes of the conveying pipeline; current sensors are installed in the fan or feed motor circuit; and acoustic sensors are attached to the outer wall of the pipeline. These three types of signals are synchronously acquired in hardware at a sampling frequency of no less than 100Hz. This step constructs a multi-physics coupled sensing system. Pressure signals characterize fluid dynamics, current signals reflect load torque changes, and the flow noise spectrum (200Hz-10kHz) captures the microscopic collisions and energy distribution characteristics of material flow. This forms a three-dimensional monitoring system for the operating status, achieving a leap from single-parameter monitoring to multi-source information complementarity. It provides a comprehensive and synchronous raw data foundation for subsequent anomaly diagnosis, significantly improving the ability to capture early, subtle anomalies.

[0068] Step 2: Preprocess the acquired multi-channel time-series signals, including sliding window segmentation and normalization, and construct the input tensor.

[0069] Specifically, pressure and current signals primarily reflect the macroscopic state of the system (time-domain characteristics), while the acoustic spectrum contains the microscopic dynamics of material flow (frequency-domain characteristics). This invention constructs a three-dimensional tensor through interpolation alignment, achieving the splicing of time-frequency domain features within a unified spatiotemporal framework.

[0070] In this embodiment, the input tensor is constructed as follows:

[0071] Step 21: Perform spectral analysis on the acoustic signal corresponding to the flow noise spectrum to obtain the acoustic spectrum, and extract the frequency band energy from the acoustic spectrum to form an acoustic spectrum energy vector;

[0072] Step 22: Align and extend the acoustic spectrum energy vector in the time dimension using an interpolation algorithm so that it has the same number of sampling points as the pressure signal corresponding to the pressure sensor and the current signal corresponding to the current sensor within the same time window.

[0073] Step 23: Combine the aligned pressure signal sequence, current signal sequence, and interpolated acoustic spectrum energy vector into an input tensor.

[0074] In practical applications, the time-domain acoustic signal corresponding to the flow noise spectrum can be converted into a frequency-domain acoustic spectrum using Fast Fourier Transform (FFT). This allows for the segmentation of key frequency bands and the calculation of energy values ​​for each band to form initial feature vectors. This process transforms the non-stationary time-domain acoustic signal into a frequency-domain energy distribution that better reflects the characteristics of material flow. Simultaneously, interpolation algorithms address the heterogeneity of multi-source sensor data in terms of temporal resolution, achieving data isomorphism for different physical quantities under a unified time reference. Finally, the structured reorganization of multi-source heterogeneous sensor data is completed by constructing a three-dimensional tensor, providing spatiotemporally aligned multimodal inputs for subsequent deep learning models, significantly improving the expressive power of abnormal features and the accuracy of model diagnosis.

[0075] In this embodiment, the window length corresponding to the sliding window segment is 1 second, and the step size is 0.2 seconds;

[0076] Let the sampling window length be T=1s, and the sampling frequency be... Then each sampling window contains N=100 sampling points;

[0077] Let the pressure signal sequence be The current signal sequence is The interpolated acoustic spectrum energy vector is ,in, Indicates the first Pressure values ​​at each sampling point Indicates the first Current values ​​at each sampling point Indicates the first Acoustic spectral energy interpolation results corresponding to each sampling point;

[0078] The input tensor is Its dimension is N×3.

[0079] Step 3: Input the input tensor into the pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal.

[0080] In this embodiment, the anomaly detection model is a pre-trained model based on 1D-CNN and a self-attention mechanism. In practical applications, a pre-processed N×3 input tensor is input into the anomaly detection model. Local spatiotemporal features are extracted through convolutional layers, and then significant features are weighted using an attention mechanism. Finally, the probability distributions and corresponding confidence scores of four states are output through a fully connected layer. By using convolutional kernels to slide along the temporal dimension to capture the correlation patterns of pressure, current, and acoustic energy, and by using an attention mechanism to achieve adaptive fusion of features from different sensor channels and time points, a discrimination boundary for anomaly states is constructed in a high-dimensional feature space. The above scheme achieves an accurate mapping from multi-source sensor signals to anomaly types. Not only is the anomaly type identification accuracy high, but more importantly, the output quantified confidence scores provide a reliable decision-making basis for subsequent hierarchical control, overcoming the limitation of traditional methods that can only alarm but cannot quantitatively assess the degree of anomaly.

[0081] In this embodiment, the training method for the anomaly detection model includes:

[0082] Step 31: Obtain historical datasets, which include multi-source sensor data of the pneumatic conveying system under normal and abnormal conditions. Abnormal conditions include pipe blockage, air blowing, and leakage. The status of each time period in the historical data is marked.

[0083] Step 32: Preprocess the multi-source sensor data in the historical dataset, including sliding window segmentation and normalization, to generate a training sample set;

[0084] Step 33: Construct an initial neural network model based on the combination of a one-dimensional convolutional neural network and a self-attention mechanism;

[0085] Step 34: Train the initial neural network model using the training sample set, with the cross-entropy loss function as the optimization objective, and update the model parameters through the backpropagation algorithm until the model converges to obtain the anomaly detection model.

[0086] In the actual training process, multi-source sensor data covering four states—normal, blocked, air-blowing, and leaking—was collected and accurately labeled. Training samples were constructed using a sliding window and normalization method consistent with online diagnostics. A neural network architecture containing one-dimensional convolutional layers and a self-attention mechanism was built, and backpropagation was used for iterative training with the cross-entropy loss function as the optimization objective until the model converged. Labeled data was used to drive the model to learn the complex nonlinear mapping relationship between multi-source sensor signals and abnormal states. The one-dimensional convolutional layer is responsible for extracting local feature patterns of various anomalies in time series, while the self-attention mechanism endows the model with the ability to pay differentiated attention to different sensor channels and key time segments, thereby achieving adaptive weighted fusion of multi-dimensional features. The anomaly detection model ultimately produced by this training method has the ability to identify anomaly types and confidence levels from raw multi-source data.

[0087] Step 4: Determine the anomaly level based on the confidence level and generate the corresponding anomaly level.

[0088] This embodiment sets the length to... A sliding window is used to calculate the mean and standard deviation of the continuously output confidence sequence, and a grading rule is constructed based on a preset threshold and the number of duration windows, specifically including:

[0089] Let the current time be The confidence sequence is The sliding window length for the confidence level is ,in, Indicates a time index;

[0090] Calculate the average confidence level of the sliding window :

[0091] ;

[0092] Calculate the standard deviation of the confidence level of the sliding window. :

[0093] ;

[0094] The anomaly level is determined according to the following rules:

[0095] If the conditions are met: and If it is, then it is judged as a level three anomaly, the highest level. This indicates the first preset confidence threshold. This indicates a preset stability threshold;

[0096] If the conditions are met: If it is, it is judged as a level 2 abnormality, among which, This indicates the second preset confidence threshold;

[0097] If the conditions are met: If the number of windows that continuously meet this condition is greater than or equal to a preset continuous window count threshold, then it is judged as a Level 1 anomaly. This indicates the third pre-set confidence threshold;

[0098] If none of the above conditions are met, the condition is considered normal.

[0099] Specifically, this embodiment introduces a decision-making mechanism based on statistical process control. The mean reflects the intensity characteristics of abnormal signals, the standard deviation characterizes the stability of diagnostic results, and the number of persistence windows verifies the persistence of abnormal states, thereby establishing reliable hierarchical boundaries in the three-dimensional feature space of intensity, stability, and persistence. This design enables the system to effectively distinguish between transient interference, evolving anomalies, and deterministic faults, achieving a leap from binary alarms to multi-level early warning. It not only ensures rapid response to major faults but also avoids false shutdowns caused by transient fluctuations, significantly improving the intelligence level and decision reliability of system operation.

[0100] In the above grading rules, the sliding window length of the confidence level The first preset confidence threshold The second preset confidence threshold The third preset confidence threshold The preset stability threshold The preset threshold for the number of continuous windows is 3.

[0101] This parameter configuration achieves the best balance between real-time performance and data stability by setting the sliding window length to 5; Cooperate A strict dual verification mechanism is established to ensure that the highest level alarm is triggered only when the system continuously outputs high confidence and the results are stable, effectively preventing accidental shutdowns. and The tiered thresholds form a progressive early warning range, which, combined with the judgment condition of a continuous window number threshold of 3, can both capture early abnormal trends and avoid interference from instantaneous fluctuations. This optimized parameter system enables the system to improve the automatic recovery success rate of first-level and second-level anomalies while ensuring a high recognition accuracy, realizing a leap from coarse alarm to precise handling, and significantly reducing unplanned downtime and operation and maintenance costs.

[0102] Step 5: Perform the corresponding graded fault-tolerant control operation according to the anomaly level. After the anomaly is handled, the pneumatic conveying system will automatically return to normal operation.

[0103] In this embodiment, precise fault tolerance is achieved by establishing a three-level progressive control strategy:

[0104] When an abnormality is detected as Level 1, the fan speed will be automatically fine-tuned, with an adjustment range of ±5% of the rated value.

[0105] When a level 2 abnormality is detected, temporarily close the feeding valve, start the backflushing procedure, and continue for 3 seconds;

[0106] When an abnormality is determined to be Level 3, the machine will be immediately shut down and an audible and visual alarm will be triggered, and an alarm message will be pushed out.

[0107] Specifically, when a Level 1 anomaly occurs, the system automatically adjusts the fan speed (±5%), achieving self-correction of operating parameters by altering airflow dynamics. When a Level 2 anomaly occurs, a coordinated control of feed valve interruption and airflow backlash is employed, temporarily cutting off the material source and combining it with a 3-second reverse airflow impact to clear blocked pipes online. When a Level 3 anomaly is triggered, immediate shutdown protection and multi-level alarms are implemented, achieving ultimate protection through fault isolation and personnel intervention. This hierarchical control mechanism establishes a progressive response strategy based on the anomaly evolution pattern, ensuring both continuous operation under minor anomalies and rapid safety protection during major faults. In practical applications, most minor anomalies can be autonomously eliminated without shutting down the system, significantly improving system availability and safety thresholds.

[0108] In this embodiment, after determining that an anomaly has occurred, the method further includes:

[0109] The pressure data collected by multiple pressure sensors arranged along the pipeline is obtained. The pressure sensors include at least a first pressure sensor installed at the pipeline inlet, a second pressure sensor installed at the pipeline outlet, and a third pressure sensor installed at at least one key node in the pipeline.

[0110] When a blockage or leakage is detected, the pressure gradient between adjacent pressure sensors is calculated.

[0111] Based on the abrupt change point of the pressure differential gradient and combined with the pipeline topology information, the pipe segment range in which the anomaly occurred is determined.

[0112] Based on the length of the pipe section and the preset pressure-distance mapping relationship, the specific location of the anomaly is estimated by interpolation calculation.

[0113] Specifically, this embodiment constructs a distributed pressure sensing network (containing at least three types of measuring points: inlet, outlet, and key nodes). After identifying pipe blockages or leaks, it automatically calculates the pressure gradient distribution between adjacent sensors. Utilizing the mapping relationship between pressure gradient and flow resistance in fluid mechanics, a significant pressure drop occurs before and after the blockage point when a pipe is blocked, and a characteristic pressure drop occurs at the leak point when a leak occurs. By detecting the abrupt changes in the pressure gradient and integrating the pipe topology, the anomaly location problem can be transformed into an inverse solution of the pipe's spatial coordinates. This solution achieves meter-level accuracy in locating anomalies, enabling maintenance personnel to quickly and accurately reach the fault point, shortening anomaly handling time, and significantly improving system maintainability and operational efficiency.

[0114] In summary, the anomaly handling method for pneumatic conveying systems based on multi-source sensor fusion provided in this embodiment achieves a technological breakthrough in the anomaly management mechanism of pneumatic conveying systems by constructing a collaborative architecture of multi-source sensor fusion and intelligent hierarchical control: The multi-physics sensing system significantly improves the system's sensitivity to early, weak anomaly characteristics and its ability to identify their types; the establishment of a confidence-driven hierarchical decision-making model effectively distinguishes different stages of anomaly development, providing a theoretical basis for implementing differentiated control strategies; the adoption of hierarchical progressive fault-tolerant control ensures both the system's continuous operation under minor anomaly conditions and a rapid and safe response to major faults; and the combination of differential pressure gradient positioning technology enables precise spatial positioning of anomaly locations.

[0115] Based on the above technical solutions, this embodiment also proposes an anomaly handling system for a pneumatic conveying system based on multi-source sensor fusion, used to implement the anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion as described in the embodiment. Please refer to [link to relevant documentation]. Figure 2 The system includes:

[0116] The signal acquisition module is used to synchronously acquire multi-channel timing signals in the pneumatic conveying system through multi-source sensors, including pressure sensors, current sensors, and acoustic sensors.

[0117] The signal preprocessing module is used to preprocess the acquired multi-channel time-series signals, including sliding window segmentation and normalization, and to construct the input tensor.

[0118] An anomaly diagnosis module is used to input the input tensor into a pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal.

[0119] The hierarchical decision-making module is used to determine the anomaly classification based on the confidence level and generate the corresponding anomaly level.

[0120] The fault-tolerant control module is used to perform corresponding graded fault-tolerant control operations according to the anomaly level, and to control the pneumatic conveying system to automatically return to normal operation after the anomaly is handled.

[0121] It is understood that the pneumatic conveying system anomaly handling system based on multi-source sensor fusion described in this embodiment is a system used to implement the pneumatic conveying system anomaly handling method based on multi-source sensor fusion described in the embodiment. As the system disclosed in the embodiment corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant parts, please refer to the description of the method. It will not be repeated here.

Claims

1. An anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion, characterized in that, The method includes: Multi-channel timing signals from the pneumatic conveying system are synchronously acquired using multi-source sensors, including pressure sensors, current sensors, and acoustic sensors. The synchronous acquisition frequency is no less than 100Hz. The pressure sensors are arranged at the inlet, outlet, and preset key nodes of the conveying pipeline. The current sensors are used to acquire the current signals of the fan or feed motor. The acoustic sensors are attached to the outer wall of the conveying pipeline and acquire the flow noise spectrum in the frequency range of 200Hz to 10kHz. The acquired multi-channel time-series signals are preprocessed, including sliding window segmentation and normalization, to construct an input tensor; the input tensor is constructed as follows: The acoustic signal corresponding to the flow noise spectrum is subjected to spectrum analysis to obtain the acoustic spectrum, and the frequency band energy is extracted from the acoustic spectrum to form an acoustic spectrum energy vector; The acoustic spectrum energy vector is aligned and extended in the time dimension using an interpolation algorithm so that it has the same number of sampling points as the pressure signal corresponding to the pressure sensor and the current signal corresponding to the current sensor within the same time window. The aligned pressure signal sequence, current signal sequence, and interpolated acoustic spectral energy vector are combined into an input tensor. The input tensor is fed into a pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal operation. Anomaly classification is performed based on the confidence level to generate corresponding anomaly levels, including: Let the current time be The confidence sequence is The sliding window length for the confidence level is ,in, Indicates a time index; Calculate the average confidence level of the sliding window : ； Calculate the standard deviation of the confidence level of the sliding window. : ； The anomaly level is determined according to the following rules: If the conditions are met: and If it is, then it is judged as a level three anomaly, the highest level. This indicates the first preset confidence threshold. This indicates a preset stability threshold; If the conditions are met: If it is, it is judged as a level 2 abnormality, among which, This indicates the second preset confidence threshold; If the conditions are met: If the number of windows that continuously meet this condition is greater than or equal to a preset continuous window count threshold, then it is judged as a Level 1 anomaly. This indicates the third pre-set confidence threshold; If none of the above conditions are met, it is considered to be in a normal state. According to the anomaly level, the corresponding graded fault-tolerant control operation is executed, and after the anomaly is handled, the pneumatic conveying system automatically returns to normal operation.

2. The anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion according to claim 1, characterized in that, The window length corresponding to each sliding window segment is 1 second, and the step size is 0.2 seconds. Let the sampling window length be T=1s, and the sampling frequency be... Then each sampling window contains N=100 sampling points; Let the pressure signal sequence be The current signal sequence is The interpolated acoustic spectrum energy vector is ,in, Indicates the first Pressure values ​​at each sampling point Indicates the first Current values ​​at each sampling point Indicates the first Acoustic spectral energy interpolation results corresponding to each sampling point; The input tensor is Its dimension is N×3.

3. The anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion according to claim 1, characterized in that, The training method for the anomaly detection model includes: Acquire historical datasets, which include multi-source sensor data of the pneumatic conveying system under normal and abnormal conditions, including pipe blockage, air blowing, and leakage, and label the state of each time period in the historical data. The multi-source sensor data in the historical dataset are preprocessed, including sliding window segmentation and normalization, to generate a training sample set; Construct an initial neural network model based on a combination of a one-dimensional convolutional neural network and a self-attention mechanism; The initial neural network model is trained using the training sample set, with the cross-entropy loss function as the optimization objective. The model parameters are updated using the backpropagation algorithm until the model converges, thus obtaining the anomaly detection model.

4. The anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion according to claim 1, characterized in that, The sliding window length of the confidence level ; The first preset confidence threshold The second preset confidence threshold The third preset confidence threshold The preset stability threshold The preset threshold for the number of continuous windows is 3.

5. The anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion according to claim 1, characterized in that, Perform corresponding hierarchical fault-tolerant control operations based on the anomaly level, including: When an abnormality is detected as Level 1, the fan speed will be automatically fine-tuned, with an adjustment range of ±5% of the rated value. When a level 2 abnormality is detected, temporarily close the feeding valve, start the backflushing procedure, and continue for 3 seconds; When an abnormality is determined to be Level 3, the machine will be immediately shut down and an audible and visual alarm will be triggered, and an alarm message will be pushed out.

6. The anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion according to any one of claims 1 to 5, characterized in that, The method further includes: The pressure data collected by multiple pressure sensors arranged along the pipeline is obtained. The pressure sensors include at least a first pressure sensor installed at the pipeline inlet, a second pressure sensor installed at the pipeline outlet, and a third pressure sensor installed at at least one key node in the pipeline. When a blockage or leakage is detected, the pressure gradient between adjacent pressure sensors is calculated. Based on the abrupt change point of the pressure differential gradient and combined with the pipeline topology information, the pipe segment range in which the anomaly occurred is determined. Based on the length of the pipe section and the preset pressure-distance mapping relationship, the specific location of the anomaly is estimated by interpolation calculation.

7. An anomaly handling system for a pneumatic conveying system based on multi-source sensor fusion, characterized in that, For implementing the anomaly handling method for a pneumatic conveying system based on multi-source sensor fusion as described in any one of claims 1 to 6, the system comprises: The signal acquisition module is used to synchronously acquire multi-channel timing signals in the pneumatic conveying system through multi-source sensors, including pressure sensors, current sensors, and acoustic sensors. The signal preprocessing module is used to preprocess the acquired multi-channel time-series signals, including sliding window segmentation and normalization, and to construct the input tensor. An anomaly diagnosis module is used to input the input tensor into a pre-trained anomaly detection model to obtain the anomaly type and its confidence level. The anomaly type includes pipe blockage, air blowing, leakage, or normal. The hierarchical decision-making module is used to determine the anomaly classification based on the confidence level and generate the corresponding anomaly level. The fault-tolerant control module is used to perform corresponding graded fault-tolerant control operations according to the anomaly level, and to control the pneumatic conveying system to automatically return to normal operation after the anomaly is handled.

Images