AI-based large pump station fault on-line monitoring and early warning system

Abstract

The invention provides an AI-based large-scale pump station fault on-line monitoring and early warning system, and the system comprises a multi-mode data collection module which is used for collecting a vibration signal, a current and voltage signal, an acoustic signal and temperature data of large-scale pump station equipment in real time; the edge calculation preprocessing module is connected with the multi-modal data acquisition module and is used for cleaning, denoising and aligning the acquired original data; according to the method, feature extraction and trend prediction are carried out on time series data through a cloud AI deep analysis engine by adopting an improved hybrid model of a double-attention mechanism convolutional neural network and a long-short-term memory network, and meanwhile, a dynamic threshold algorithm based on a Gaussian mixture model is introduced, so that the problem of false alarm of a fixed threshold under working condition fluctuation is solved, and the working efficiency is improved. Therefore, tiny abnormal characteristics of the pump station equipment can be captured in real time, accurate positioning and early warning of faults are realized, and the operation safety and the maintenance efficiency of the large-scale pump station are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial monitoring and artificial intelligence, and particularly relates to an AI-based large pump station fault online monitoring and early warning system. BACKGROUND

[0002] A large pump station is a core facility of a water conservancy project, a city water supply and drainage system, and an industrial and agricultural water supply system, and its operation state is directly related to production safety and social benefits. Pump station equipment (such as motors, water pumps, bearings, etc.) is usually operated in a high-load, strong-vibration and harsh environment for a long time, and is prone to mechanical looseness, bearing wear, cavitation and other faults.

[0003] The traditional pump station monitoring method is mainly based on a "threshold alarm" mechanism of a single physical quantity (such as a vibration value), however, the working condition of a large pump station is complex, and the load changes frequently, so that a fixed threshold is often difficult to adapt to a dynamic working condition, resulting in a high false alarm rate (normal working condition fluctuation triggers an alarm) or a high missing alarm rate (early weak faults are not detected because they do not reach the threshold value), and the existing simple signal processing method (such as FFT spectrum analysis) cannot effectively mine deep nonlinear correlations between multi-source heterogeneous data (vibration, current, sound, etc.), and lacks the ability to predict the fault evolution trend, therefore, the AI-based large pump station fault online monitoring and early warning system is proposed. SUMMARY

[0004] Therefore, the AI-based large pump station fault online monitoring and early warning system is provided, so as to solve or alleviate the technical problems in the prior art, and at least provide a beneficial choice.

[0005] The technical scheme of the embodiment of the present application is implemented as follows: the AI-based large pump station fault online monitoring and early warning system comprises: A multi-modal data acquisition module is configured to acquire vibration signals, current and voltage signals, acoustic signals and temperature data of a large pump station equipment in real time. An edge computing preprocessing module is connected with the multi-modal data acquisition module, and is configured to clean, denoise and align the acquired raw data. An AI deep analysis engine in the cloud is in communication connection with the edge computing preprocessing module, and is configured to receive the preprocessed data, extract spatiotemporal features through a double-attention time series neural network model, and output an equipment health state index. A dynamic early warning module is configured to calculate an abnormal confidence according to the equipment health state index and in combination with an adaptive dynamic threshold algorithm, and trigger a hierarchical early warning when the confidence exceeds a set threshold. The dual-attention temporal neural network model includes a spatial attention unit and a temporal attention unit. The spatial attention unit is used to focus on the correlation between different sensor data, and the temporal attention unit is used to capture key time steps in the fault evolution process.

[0006] In some embodiments, the edge computing preprocessing module employs an improved wavelet threshold denoising algorithm, and its threshold selection formula is as follows: in, The standard deviation of noise. For signal length, The wavelet decomposition scale, This is the maximum decomposition scale.

[0007] In some embodiments, the construction process of the dual-attention temporal neural network model in the cloud-based AI deep analysis engine includes: One-dimensional convolutional neural networks are used to extract local spatial features of multi-source signals; The features are input into a long short-term memory network for time series modeling; A spatial attention mechanism is introduced after the convolutional neural network layer, and the weight calculation formula is as follows: in, The convolutional neural network outputs a feature matrix. and This is the weight matrix. For bias terms; A temporal attention mechanism is introduced after the Long Short-Term Memory (LSTM) network layer to weight and aggregate the hidden states at different time steps.

[0008] In some embodiments, the weight calculation of the temporal attention mechanism is based on the correlation between the current hidden state and the historical context, and its correlation score. The calculation formula is: in, For the first Hidden states of the Long Short-Term Memory network at each time step The current hidden state. , and For model parameters, This represents the normalized temporal attention weights.

[0009] In some embodiments, the dynamic early warning module uses a Gaussian mixture model based on a sliding window to construct a dynamic threshold, updates the model parameters through maximum likelihood estimation, and calculates the negative log-likelihood probability of real-time data as an anomaly score.

[0010] In some embodiments, the anomaly score The calculation logic is as follows: in, The number of Gaussian components. The mixing coefficient, Representing observation data In the The probability density function under a Gaussian distribution, when continuous The number of times exceeded the dynamic threshold When this occurs, it is determined to be a fault state.

[0011] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions: This invention employs a hybrid model combining an improved dual-attention mechanism convolutional neural network and a long short-term memory network through a cloud-based AI deep analysis engine to extract features and predict trends in time-series data. Simultaneously, it introduces a dynamic threshold algorithm based on a Gaussian mixture model to solve the false alarm problem of fixed thresholds under fluctuating operating conditions. This enables the real-time capture of minute abnormal features of pumping station equipment, achieving accurate fault location and early warning, and significantly improving the operational safety and maintenance efficiency of large pumping stations.

[0012] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a block diagram of the overall architecture of the present invention. Detailed Implementation

[0015] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0016] It is important to note that terms such as "first," "second," "symmetric," "array," "set in," and "set with" are used only to distinguish between descriptive and positional descriptions and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features specified with terms such as "first" or "symmetric" may explicitly or implicitly include one or more of that feature; similarly, when the quantity of certain features is not limited by words such as "two" or "three," it should be noted that such features also explicitly or implicitly include one or more features.

[0017] In this invention, unless otherwise explicitly specified and limited, terms such as "installation," "connection," and "fixation" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral molding; they can refer to a mechanical connection, a direct connection, a welding connection, or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the accompanying drawings and specific circumstances.

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 As shown, this embodiment of the invention provides an AI-based online monitoring and early warning system for large-scale pumping station faults, including: The multimodal data acquisition module is used to acquire vibration signals, current and voltage signals, acoustic signals and temperature data of large pumping station equipment in real time. The edge computing preprocessing module is connected to the multimodal data acquisition module and is used to clean, denoise, and align the acquired raw data. The cloud-based AI deep analysis engine communicates with the edge computing preprocessing module to receive preprocessed data and extract spatiotemporal features through a dual-attention temporal neural network model to output device health status indicators. The dynamic early warning module calculates the anomaly confidence level based on the equipment health status indicators and an adaptive dynamic threshold algorithm. When the confidence level exceeds the set threshold, a tiered early warning is triggered. The dual-attention temporal neural network model includes a spatial attention unit and a temporal attention unit. The spatial attention unit is used to focus on the correlation between different sensor data, while the temporal attention unit is used to capture key time steps in the fault evolution process.

[0020] In this embodiment, specifically, the system deploys high-precision vibration acceleration sensors, Hall current sensors, acoustic sensors and thermocouples at key parts of the pumping station. Considering the complex electromagnetic interference at the pumping station site, an improved wavelet threshold denoising algorithm is used at the edge end.

[0021] In this embodiment, specifically, the edge computing preprocessing module adopts an improved wavelet threshold denoising algorithm, and its threshold selection formula is as follows: in, The standard deviation of noise. For signal length, The wavelet decomposition scale, To achieve the maximum decomposition scale, the formula, compared to the general threshold, incorporates a scale-dependent adjustment factor. This allows more information to be preserved at the low-frequency scale (approximate signal) and stronger denoising to be applied at the high-frequency scale (detail signal), thereby effectively preserving the fault impact characteristics.

[0022] In this embodiment, the specific construction process of the dual-attention temporal neural network model in the cloud-based AI deep analysis engine includes: One-dimensional convolutional neural networks are used to extract local spatial features of multi-source signals; The features are input into a long short-term memory network for time series modeling; A spatial attention mechanism is introduced after the convolutional neural network layer, and the weight calculation formula is as follows: in, The convolutional neural network outputs a feature matrix. and This is the weight matrix. For bias terms; A temporal attention mechanism is introduced after the Long Short-Term Memory (LSTM) network layer to weight and aggregate the hidden states at different time steps.

[0023] In this embodiment, specifically, the weight calculation of the time attention mechanism is based on the correlation between the current hidden state and the historical context, and its correlation score. The calculation formula is: in, For the first Hidden states of the Long Short-Term Memory network at each time step The current hidden state. , and For model parameters, This represents the normalized temporal attention weights.

[0024] In this embodiment, the detailed steps of the dual-attention temporal neural network model are as follows: S1. Spatial Feature Extraction: First, a one-dimensional convolutional neural network is used to perform convolution operations on the multi-channel sensor data to extract local spatial features. Let the input be... ( For time steps, (Number of sensors), after passing through the convolutional layer, a feature map is obtained. ; S2, Spatial Attention Mechanism: Not all sensor data is equally important for current fault diagnosis; therefore, spatial attention weights are introduced. The calculation formula is as follows: This mechanism can automatically learn the weights of different sensors (such as vibration and current) in specific fault modes, thereby enabling the focus on key signals; S3, Temporal Modeling and Temporal Attention: The weighted features are input into the Long Short-Term Memory (LSTM) network. The LTM network controls the flow of information through gating mechanisms (forget gate, input gate, and output gate), as shown in the following formula: in, It is the Sigmoid activation function. Represents element-wise product; To capture critical fault moments in long sequences, a temporal attention mechanism is introduced to calculate the current moment. With historical moments correlation score : Final time context vector for: This mechanism enables the model to automatically assign higher weights to time steps that contain key fault features when predicting the current state, while ignoring irrelevant noise fluctuations.

[0025] In this embodiment, the dynamic early warning module uses a Gaussian mixture model based on a sliding window to construct a dynamic threshold, updates the model parameters through maximum likelihood estimation, and calculates the negative log-likelihood probability of real-time data as anomaly score.

[0026] In this embodiment, specifically, the abnormal score The calculation logic is as follows: in, The number of Gaussian components. The mixing coefficient, Representing observation data In the The system updates the Gaussian mixture model parameters within a sliding window in real time, based on the probability density function under a Gaussian distribution. Thus, the threshold It can adaptively drift according to operating conditions (such as flow rate changes), when continuous The number of times exceeded the dynamic threshold When this occurs, it is determined to be a fault state.

[0027] In this embodiment, specifically applied to a large water supply pumping station, the system deployment is as follows: A triaxial vibration sensor is installed at both the motor drive end and the non-drive end, with a sampling rate set to 20kHz. A current transformer is installed in the electrical control cabinet to collect the three-phase current, and an acoustic sensor is installed on the pump casing. The data is transmitted to the edge gateway, which uses an improved wavelet threshold denoising algorithm to reduce the noise of the vibration signal and downsamples the 20kHz vibration signal to 1kHz. The data is then timestamped and aligned with the current and temperature signals, and the data is packaged and uploaded to the cloud every 5 seconds. The cloud-based AI deep analysis engine receives data and inputs it into a pre-trained hybrid model of a dual-attention mechanism convolutional neural network and a long short-term memory network. This hybrid model first extracts features through the convolutional neural network. The spatial attention mechanism identifies that the features with the highest weights are "horizontal vibration at the motor drive end" and "C-phase current," automatically suppressing other noise channels. Subsequently, the long short-term memory network analyzes data from the past hour, and the temporal attention mechanism detects a slight increase in a harmonic component that occurred 10 minutes prior.

[0028] A hybrid model combining a dual-attention convolutional neural network and a long short-term memory network outputs the current health index (HI). The dynamic early warning module calculates Gaussian mixture model parameters based on a sliding window of data from the past 24 hours. Since the pumping station is currently operating at a high water level and high load, the threshold... Automatic up-floating; although HI increased, it still did not exceed the adaptive threshold. As time went on, HI continued to rise, eventually reaching... The time exceeded 3 times in a row The system identified the issue as "early bearing wear" and immediately sent a Level 2 warning (yellow) to the central control room, recommending a maintenance plan.

[0029] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI-based online monitoring and early warning system for large-scale pumping station faults, characterized in that: include: The multimodal data acquisition module is used to acquire vibration signals, current and voltage signals, acoustic signals and temperature data of large pumping station equipment in real time. An edge computing preprocessing module, connected to the multimodal data acquisition module, is used to clean, denoise, and align the acquired raw data. The cloud-based AI deep analysis engine communicates with the edge computing preprocessing module to receive preprocessed data, extract spatiotemporal features through a dual attention temporal neural network model, and output device health status indicators. The dynamic early warning module calculates the anomaly confidence level based on the device health status indicators and an adaptive dynamic threshold algorithm. When the confidence level exceeds a set threshold, a tiered early warning is triggered. The dual-attention temporal neural network model includes a spatial attention unit and a temporal attention unit. The spatial attention unit is used to focus on the correlation between different sensor data, and the temporal attention unit is used to capture key time steps in the fault evolution process.

2. The AI-based online monitoring and early warning system for large-scale pumping station faults according to claim 1, characterized in that, The edge computing preprocessing module employs an improved wavelet threshold denoising algorithm, and its threshold selection formula is as follows: in, The standard deviation of noise. For signal length, The wavelet decomposition scale, This is the maximum decomposition scale.

3. The AI-based online monitoring and early warning system for large-scale pumping station faults according to claim 1, characterized in that, The construction process of the dual-attention temporal neural network model in the cloud-based AI deep analysis engine includes: One-dimensional convolutional neural networks are used to extract local spatial features of multi-source signals; The features are input into a long short-term memory network for time series modeling; A spatial attention mechanism is introduced after the convolutional neural network layer, and the weight calculation formula is as follows: in, The convolutional neural network outputs a feature matrix. and This is the weight matrix. For bias terms; A temporal attention mechanism is introduced after the Long Short-Term Memory (LSTM) network layer to weight and aggregate the hidden states at different time steps.

4. The AI-based online monitoring and early warning system for large-scale pumping station faults according to claim 3, characterized in that, The weight calculation of the time attention mechanism is based on the correlation between the current hidden state and the historical context, and its correlation score. The calculation formula is: in, For the first Hidden states of the Long Short-Term Memory network at each time step The current hidden state. , and For model parameters, This represents the normalized temporal attention weights.

5. The AI-based online monitoring and early warning system for large-scale pumping station faults according to claim 1, characterized in that, The dynamic early warning module uses a Gaussian mixture model based on a sliding window to construct a dynamic threshold, updates the model parameters through maximum likelihood estimation, and calculates the negative log-likelihood probability of real-time data as anomaly score.

6. The AI-based online monitoring and early warning system for large-scale pumping station faults according to claim 5, characterized in that, The abnormal score The calculation logic is as follows: in, The number of Gaussian components. The mixing coefficient, Representing observation data In the The probability density function under a Gaussian distribution, when continuous The number of times exceeded the dynamic threshold When this occurs, it is determined to be a fault state.

Images