Sewage treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction

Abstract

The invention discloses a sewage treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction, and belongs to the technical field of equipment intelligent operation and maintenance. According to the method, after adaptive filtering and quality evaluation are carried out on a vibration signal, an adaptive decomposition strategy fusing VMD and improved EMD is adopted, and a modal number and parameters are optimized according to an information entropy criterion; extracting multi-scale features from three levels of micro-scale (instantaneous characteristics), mesoscale (local statistics) and macro-scale (global energy / entropy), and screening key features through mutual information and a two-layer fusion mechanism; a health index model is constructed, and normal, attention, warning and danger four-level dynamic early warning is achieved; and meanwhile, a self-adaptive fault knowledge base capable of being incrementally updated is established in combination with density clustering and a Markov chain. According to the method, the fault identification accuracy and the system self-adaptive capacity are remarkably improved, and the method is suitable for intelligent diagnosis of sewage treatment key equipment such as a centrifugal pump and a Roots blower.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment equipment monitoring technology, and specifically relates to a wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction. Background Technology

[0002] The core equipment of wastewater treatment plants includes rotating machinery such as centrifugal pumps, blowers, compressors, and agitators. These devices operate under complex conditions, and their fault signals often exhibit multi-scale, non-stationary, and nonlinear composite characteristics. Traditional fault diagnosis methods have the following technical shortcomings: The signal decomposition method lacks adaptability: fixed signal processing methods are difficult to adapt to the dynamic operating conditions of different devices.

[0003] Feature extraction is limited to a single scale: it lacks multi-scale analysis that ranges from microscopic instantaneous changes to macroscopic statistical trends.

[0004] Feature fusion lacks intelligence: simple feature overlay ignores the correlation and complementarity between features.

[0005] The static nature of diagnostic thresholds: Fixed thresholds cannot adapt to the dynamic degradation process of equipment performance.

[0006] Therefore, there is an urgent need for an innovative diagnostic method that can adaptively process complex signals, intelligently extract multi-scale features, and dynamically adjust diagnostic strategies. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a fault diagnosis method for sewage treatment plant equipment based on multi-scale adaptive feature extraction. The present invention significantly improves the fault identification accuracy and system adaptability, and is applicable to the intelligent diagnosis of key sewage treatment equipment such as centrifugal pumps and Roots blowers.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for fault diagnosis of wastewater treatment plant equipment based on multi-scale adaptive feature extraction, comprising the following steps: S1. Adaptive preprocessing and quality assessment are performed on the collected vibration signals of sewage treatment plant equipment. An adaptive signal quality assessment mechanism is established, and the preprocessing parameters are automatically adjusted according to the signal quality indicators. S2. An adaptive signal decomposition method that fuses variational mode decomposition and improved empirical mode decomposition is used to decompose the preprocessed signal; S3. Extract multi-scale features, including micro-scale instantaneous features, meso-scale local statistical features, and macro-scale global statistical features; S4. Establish an intelligent feature fusion mechanism based on information theory, and use mutual information to evaluate feature importance to solve the problems of feature redundancy and complementarity; S5. Construct a dynamic threshold diagnostic decision system to achieve adaptive early warning based on equipment health index and degradation trend.

[0009] Preferably, in step S1, the signal quality evaluation function includes signal-to-noise ratio, signal feature strength index, and signal purity index.

[0010] Preferably, in step S2, the adaptive signal decomposition method determines the optimal number of modes using the information entropy criterion and introduces an adaptive penalty parameter to optimize the variational mode decomposition process.

[0011] Preferably, in step S3, the microscale instantaneous features include instantaneous amplitude, instantaneous frequency, and instantaneous energy; the mesoscale local statistical features include local energy density, local frequency variance, and local correlation dimension; and the macroscale global statistical features include modal energy ratio and modal entropy.

[0012] Preferably, in step S4, the intelligent feature fusion adopts a two-layer fusion strategy: the first layer performs weighted fusion of features at the same scale, and the second layer uses adaptive weights to fuse features at different scales.

[0013] Preferably, in step S5, the dynamic threshold adjustment is adaptively adjusted based on the rate of change of the device health index and the variance of the health index fluctuation, and a four-level early warning mechanism is established: normal state, attention state, warning state and danger state.

[0014] Preferably, the method further includes S6: constructing an adaptive fault knowledge base to achieve incremental learning and evolutionary modeling of fault modes.

[0015] Preferably, in step S6, the adaptive fault knowledge base uses a density-based anomaly detection algorithm to identify new fault modes and models the fault evolution path based on Markov chains.

[0016] A wastewater treatment plant equipment fault diagnosis system based on multi-scale adaptive feature extraction, employing the aforementioned wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction, includes: The signal preprocessing module is used to perform adaptive preprocessing and quality assessment on the collected vibration signals from wastewater treatment plant equipment. The signal decomposition module, connected to the signal preprocessing module, is used to decompose the preprocessed signal using a fusion method of variational mode decomposition and improved empirical mode decomposition. The feature extraction module, connected to the signal decomposition module, is used to extract microscale instantaneous features, mesoscale local statistical features, and macroscale global statistical features. The feature fusion module, connected to the feature extraction module, is used to intelligently fuse multi-scale features based on information theory. The diagnostic decision module, connected to the feature fusion module, is used to construct a dynamic threshold diagnostic decision system and achieve adaptive early warning based on the device health index. The fault knowledge base module, connected to the diagnostic decision module, is used to build an adaptive fault knowledge base to achieve incremental learning and evolutionary modeling of fault modes.

[0017] Preferably, the feature fusion module adopts a two-layer fusion strategy: the first layer performs weighted fusion of features at the same scale, and the second layer uses adaptive weights to fuse features at different scales.

[0018] Preferably, the diagnostic decision module establishes a four-level early warning mechanism, including normal state, attention state, warning state and danger state, and evaluates the credibility of the diagnostic results based on Bayesian inference.

[0019] The present invention can achieve the following beneficial effects: 1. This invention integrates VMD and improved EMD, and adaptively determines the optimal number of modes and penalty parameters based on information entropy, effectively suppressing mode aliasing and over-decomposition problems, and significantly improving the ability to extract fault features under complex working conditions.

[0020] 2. This invention constructs a multi-scale feature system at the microscale (instantaneous amplitude, frequency), mesoscale (local energy, correlation dimension), and macroscale (modal energy ratio, entropy) to comprehensively characterize the operating status of equipment and improve the sensitivity to early minor faults.

[0021] 3. This invention adopts a two-layer fusion strategy based on mutual information. First, it weights and aggregates features of the same scale, and then adaptively fuses features across scales. This effectively removes redundancy, retains key information, and improves the efficiency and accuracy of subsequent diagnosis.

[0022] 4. This invention establishes a four-level dynamic threshold early warning mechanism based on equipment health index, and introduces Bayesian inference to evaluate the credibility of diagnostic results, avoiding false alarms and missed alarms, and improving decision reliability.

[0023] 5. This invention uses an adaptive fault knowledge base, combined with density clustering and Markov chain model, to achieve incremental learning and evolutionary modeling of fault modes, enabling the system to continuously optimize diagnostic performance as operational data accumulates. Attached Figure Description

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] Preferred solutions include Figure 1As shown, a fault diagnosis method for wastewater treatment plant equipment based on multi-scale adaptive feature extraction comprises the following core steps: S1 adaptive signal preprocessing and quality assessment; S2 variational mode decomposition and adaptive signal decomposition; S3 multi-scale feature extraction; S4 intelligent feature fusion based on information theory; S5 construction of a dynamic threshold diagnostic decision system; and S6 construction of an adaptive fault knowledge base. These steps are linked together to form a complete fault diagnosis closed loop, achieving fully automated processing from raw signal acquisition to fault early warning. The specific method is as follows: 1. Adaptive signal preprocessing and quality assessment In vibration monitoring of wastewater treatment plant equipment, the raw signals are often affected by environmental noise and fluctuations in operating conditions. To improve signal quality, this invention first establishes an adaptive signal quality evaluation mechanism.

[0026] For core equipment such as centrifugal pumps and fans, accelerometers are installed in three directions—horizontal, vertical, and axial—to construct a multi-dimensional monitoring network. The sampling frequency is dynamically adjusted according to the equipment's rotational speed and structural characteristics to ensure the capture of key frequency components.

[0027] Signal model establishment: ; in, In order to collect signals, These are fault characteristic signals. It is random noise. This is an interference signal due to operating conditions.

[0028] Signal quality indicators: Define the signal quality evaluation function: .

[0029] in: Weighting coefficients, satisfying .

[0030] For signal-to-noise ratio, This represents the power of the fault characteristic signal s(t). The power of random noise n(t).

[0031] Var(s) represents the variance of the fault characteristic signal s(t), reflecting the degree of dispersion of s(t); Var(n) represents the variance of the random noise n(t), reflecting the fluctuation range of the noise.

[0032] Rxs represents the correlation coefficient between the acquired signal x(t) and the fault characteristic signal s(t); Rxi represents the correlation coefficient between the acquired signal x(t) and the operating condition interference signal i(t). When the fault characteristic is dominant, the SCI approaches 1; when the interference is dominant, the SCI approaches 0.

[0033] Adaptive filtering strategy: The optimal filtering parameters are automatically selected based on the signal quality index Q value. .

[0034] in, The cutoff frequency, As the reference frequency, and These are adaptive parameters. When the signal quality index... At lower frequencies, the cutoff frequency decreases, and the filter passband narrows to more strictly suppress noise; when At higher frequencies, the cutoff frequency approaches the reference value, retaining more signal components.

[0035] 2. Variational Mode Decomposition and Adaptive Signal Decomposition: Traditional signal decomposition methods struggle to adapt to the complex signal characteristics of wastewater treatment equipment. This invention proposes an adaptive signal decomposition method that integrates Variational Mode Decomposition (VMD) and Improved Empirical Mode Decomposition (EMD). After signal preprocessing, it is necessary to further extract physically meaningful modal components from the acquired signal to provide more accurate features for subsequent fault feature identification. Traditional single signal decomposition methods suffer from problems such as mode aliasing and reliance on experience for decomposition results, making them unsuitable for the signal characteristics of complex industrial scenarios. To address this, an adaptive signal decomposition method based on the fusion of Variational Mode Decomposition and Improved Empirical Mode Decomposition is proposed. This method adaptively determines the optimal number of modes, optimizes VMD decomposition parameters, and performs multi-scale feature extraction to achieve feature extraction and construction from the preprocessed signal.

[0036] 2.1 Determination of the number of adaptive modes: The number of modes is a core parameter in signal decomposition. Too few modes lead to the merging of useful information, making it impossible to separate fault-feature modes; too many modes result in redundant modes and even over-decomposition, introducing false features. Therefore, an optimal mode number selection algorithm based on the information entropy criterion is established: .

[0037] in: K: Decomposition mode number, indicating that the original signal is decomposed into K modal components.

[0038] : Optimal mode number, which is the K value that minimizes the objective function obtained by solving the above formula.

[0039] For total information entropy, The probability percentage of the i-th mode component is usually defined as the ratio of the energy of that mode to the total energy of all modes.

[0040] To avoid excessive decomposition of penalty items, This represents the i-th, j-th modal component.

[0041] For intermodal information, Representing modes cross-correlation coefficients, Modal The autocorrelation coefficient.

[0042] 2.2 Variational Mode Decomposition Optimization: Traditional Variational Mode Decomposition (VMD) decomposes a signal into multiple modal components with specific center frequencies by constructing a variational model. However, its penalty parameter is a fixed value, which cannot adapt to the decomposition requirements of signals with different quality levels. Therefore, an adaptive penalty parameter is introduced, and the VMD objective function is optimized to achieve adaptive adjustment of decomposition parameters based on signal quality. A modal energy regularization term is introduced based on the standard VMD objective function. This is to constrain the energy distribution of each mode during the decomposition process and avoid excessive energy concentration in a single mode.

[0043] VMD objective function optimization: .

[0044] Constraints: .

[0045] in: The modal component (IMF) is obtained from the k-th decomposition.

[0046] Let be the center frequency of the k-th mode.

[0047] It is the Dirac function (unit impulse function).

[0048] The imaginary unit is used to represent complex signals.

[0049] Let be the first-order partial derivative with respect to time t.

[0050] Introduce adaptive penalty parameters: .

[0051] in: Used as the base penalty parameter.

[0052] This is for adjusting the coefficient.

[0053] This is a signal quality indicator.

[0054] This is the quality threshold.

[0055] 2.3 Multi-scale instantaneous feature extraction: After completing the adaptive VMD decomposition, each modal component These all contain signal information within a specific frequency range, which may include features describing equipment faults. To comprehensively extract these features, feature extraction is performed on each modal component from three dimensions: microscale, mesoscale, and macroscale, forming multi-dimensional features to provide input for subsequent fault diagnosis models.

[0056] Microscale features (instantaneous characteristics): Instantaneous amplitude: ,in Indicates to Perform a Hilbert transform.

[0057] Instantaneous frequency: ,in This represents the instantaneous phase of the k-th modal component.

[0058] Instantaneous energy: .

[0059] Mesoscale features (local statistics): In the sliding window Internal calculation: Local energy density: .

[0060] Local frequency variance: .

[0061] Local correlation dimension: .

[0062] in, The relevant integral function is defined as:

[0063] In the formula, The number of sample points within the window. , This represents the state vector after reconstructing the phase space. For Heaviside step function, This is the distance threshold.

[0064] Macro-scale features (global statistics): Calculate over the total duration T of the signal: Modal energy ratio: .

[0065] Modal entropy: ,in Represents the instantaneous amplitude of the k-th modal component The probability density function.

[0066] 3. Information Theory-Based Intelligent Feature Fusion: After multi-scale instantaneous feature extraction, high-dimensional features covering micro, meso, and macro scales are obtained. However, these features may contain information redundancy, and some features contribute little to fault classification. Directly using the complete feature set for fault diagnosis leads to increased model complexity, decreased computational efficiency, difficulty in model fitting, and even the curse of dimensionality. To address this, an information theory-based intelligent feature fusion mechanism is established. By quantifying the correlation between features and faults, optimizing feature combinations, and dynamically selecting key features, the mechanism solves the problems of feature redundancy and complementarity, transforming the high-dimensional original feature set into a dimensionality-reduced core feature set, providing efficient input for subsequent fault diagnosis models.

[0067] 3.1 Feature Importance Assessment: Mutual information, in information theory, is an indicator that measures the degree of information correlation between two random variables. It can effectively quantify the nonlinear correlation between features and fault categories. A feature importance evaluation algorithm based on mutual information is adopted: .

[0068] in: For the first One characteristic.

[0069] This refers to the fault category.

[0070] Features Mutual information with fault categories.

[0071] Features Values Marginal probability distribution at time.

[0072] Fault Category for Marginal probability distribution at time.

[0073] Features The value is And fault category for The joint probability distribution at time .

[0074] Assessing feature importance solely based on mutual information may overestimate features highly correlated with other features. For example, instantaneous energy at the microscale is highly correlated with local energy density at the mesoscale, and both have high mutual information; however, retaining both simultaneously would lead to redundancy. Therefore, feature importance weights are defined and adjusted as follows: .

[0075] in, The maximum correlation coefficient, This is the correlation threshold. This represents the total number of features. When the correlation between a feature and other features exceeds a threshold, its weight is set to zero to eliminate redundancy.

[0076] 3.2 Multi-level feature fusion strategy: Multi-level feature fusion adopts the steps of first integrating within the same scale and then optimizing between different scales. This not only preserves the complementary information of features within the same scale, but also achieves synergistic enhancement of features between different scales, thus avoiding the limitations of single-scale features.

[0077] First layer: Intra-scale fusion Weighted fusion of features at the same scale: .

[0078] in, For the first A feature set at each scale This is a standardized feature.

[0079] Second layer: Inter-scale fusion: Adaptive weighting is used to fuse features at different scales: .

[0080] The adaptive weights are: .

[0081] For the first Quality evaluation indicators at various scales This refers to the temperature parameter.

[0082] 3.3 Dynamic Feature Selection Mechanism After importance assessment and multi-level fusion, although the feature dimensionality has been significantly reduced, a small number of redundant features may still exist, and different fault types have different feature requirements. Therefore, a dynamic selection algorithm based on recursive feature elimination is established to dynamically filter the optimal feature subset according to fault classification requirements, further reducing dimensionality and improving model efficiency.

[0083] .

[0084] in: For feature subsets Classification loss.

[0085] The number of features.

[0086] This is a regularization parameter used to balance classification loss and the number of features. The larger the value, the heavier the penalty for the number of features, and the more likely it is to select fewer features; The smaller the value, the more emphasis is placed on reducing classification loss, which may result in the preservation of more features.

[0087] 4. Dynamic threshold diagnostic decision system: The health status of industrial equipment gradually deteriorates over time. Using fixed thresholds for diagnosis can easily lead to missed or false diagnoses. Therefore, a dynamic threshold diagnostic decision-making mechanism based on equipment degradation trends is established. This mechanism quantifies equipment health status by constructing a comprehensive health index, dynamically adjusts diagnostic thresholds based on degradation trends, and establishes a multi-level early warning system. This enables dynamic health assessment and provides decision support for the full lifecycle health management of equipment.

[0088] 4.1 Equipment Health Index Construction: The Equipment Health Index (HI) transforms high-dimensional core features into single-valued health indicators, integrating information from multiple core features to intuitively reflect the degree of equipment degradation from normal failure: .

[0089] in: This indicates the number of features involved in the health index calculation.

[0090] This represents the weight of the i-th feature in the health index.

[0091] This represents the value of the i-th feature at time t.

[0092] For the first Normalized mapping function for each feature: .

[0093] in: This represents the normal state threshold for the i-th feature, i.e., the critical value when the feature is within the normal range.

[0094] The fault state threshold of the i-th feature is the critical value at which the feature reaches the fault state.

[0095] 4.2 Dynamic threshold adaptive adjustment: Dynamic threshold update algorithm based on historical device data: .

[0096] in: The baseline threshold is determined based on historical data under normal device conditions and serves as the initial reference value for the dynamic threshold.

[0097] The rate of change of the health index reflects the speed at which the health status of the equipment deteriorates.

[0098] The variance of the health index is a measure of its volatility. Stability.

[0099] and The parameters are used to adjust the degree of influence of the rate of change of the health index and the variance of fluctuation on the dynamic threshold.

[0100] 4.3 Multi-level early warning mechanism: Equipment failures typically go through multiple stages, and a single warning level cannot meet the decision-making needs of different stages. Therefore, a four-level health status and three-level warning system is established, with warnings triggered based on the HI value and HI change rate.

[0101] Normal state: .

[0102] Note the status: .

[0103] Warning status: .

[0104] Dangerous situation: .

[0105] Warning trigger conditions: .

[0106] in: , , These correspond to the health index thresholds for three different warning levels.

[0107] , , These correspond to the thresholds for the rate of change of the health index at each of the three warning levels.

[0108] 5. Construction of an adaptive fault knowledge base: While dynamic threshold diagnostic decision-making systems can achieve real-time assessment and early warning of equipment health status, as equipment operating time increases and operating conditions change, new faults that are not covered by historical samples may occur. Traditional fixed knowledge bases cannot identify such faults, easily leading to missed diagnoses. To address this, an adaptive fault knowledge base is constructed. Through fault mode self-learning, dynamic modeling of fault evolution paths, and reliability assessment of diagnostic results, the knowledge base is updated and optimized, providing a continuously iterative fault cognition foundation for the dynamic threshold diagnostic system and ensuring continuous improvement of diagnostic capabilities throughout the equipment's lifecycle.

[0109] 5.1 Fault Mode Self-Learning Mechanism The fault mode self-learning mechanism based on incremental learning identifies new faults through anomaly detection and dynamically updates the knowledge base, enabling the system to autonomously recognize new faults. New Failure Mode Detection: The density-based anomaly detection algorithm DBSCAN is used to identify new fault modes. By calculating the local density of samples, it identifies anomalous sample clusters that differ significantly from known fault modes and determines them as new fault modes. .

[0110] in: For the sample The local density.

[0111] for The neighborhood set.

[0112] Samples and samples eigenvectors.

[0113] This represents the kernel width parameter, which controls the decay rate of distance weights in local density calculations.

[0114] To ensure that the knowledge base always reflects the latest fault characteristics of the equipment, a dynamic update strategy is adopted, which involves adding new faults incrementally and removing outdated faults.

[0115] Knowledge base update strategy: .

[0116] in: For a moment The fault knowledge base.

[0117] For a moment The fault knowledge base.

[0118] This is a newly identified fault mode.

[0119] This is an outdated failure mode.

[0120] 5.2 Fault Evolution Path Modeling: Equipment failures do not develop randomly, but rather follow a gradual evolutionary path from minor to severe. By modeling failure evolution paths based on Markov chains and analyzing historical failure state transition patterns, future failure trends can be predicted, providing a basis for developing maintenance plans in advance.

[0121] Markov chain model: .

[0122] in: For the equipment at all times The fault status.

[0123] This represents the total number of fault states.

[0124] From state Transition to state The probability of.

[0125] The number of observed transitions.

[0126] Fault development trend prediction:

[0127] in: For a moment The state probability vector, whose th... element Indicates that the device is in a certain state. The probability of; Let be the state transition probability matrix, whose elements are... Determined by the aforementioned Markov chain model; For a moment The predicted state probability vector.

[0128] 5.3 Assessment of diagnostic reliability: The fault type judgment output by the dynamic threshold diagnostic system may be uncertain due to characteristic fluctuations and operating condition interference. A diagnostic reliability assessment based on Bayesian inference quantifies the reliability of the diagnostic results by fusing prior fault probabilities with real-time evidence, avoiding over-maintenance or under-maintenance caused by misjudgments.

[0129] Establish a diagnostic reliability assessment based on Bayesian inference: .

[0130] in: For the first Types of faults.

[0131] This is the observation evidence vector.

[0132] In order to observe evidence The equipment experienced the first Type of fault The posterior probability.

[0133] Let be the likelihood probability, representing the probability at the th... Type of fault When it happened, evidence was observed. The probability of.

[0134] For the first Type of fault The prior probability is determined based on historical failure statistics.

[0135] Credibility Indicators: .

[0136] Example 1: Multi-scale adaptive diagnostics for centrifugal pumps.

[0137] To address the fault diagnosis needs of centrifugal pump equipment in wastewater treatment plants, and considering their operating environment (humid, high sediment content, large load fluctuations) and common faults (impeller cavitation, bearing wear, rotor imbalance, etc.), a multi-scale adaptive diagnostic scheme is designed. The specific implementation steps are as follows: 1. Adaptive signal acquisition and preprocessing Sensor configuration: Accelerometers are installed in three directions on the pump body: horizontal, vertical, and axial. The horizontal direction primarily collects radial vibrations caused by rotor imbalance and misalignment; the vertical direction captures the vertical impact of bearing outer ring wear; and the axial direction monitors axial pulsations caused by impeller cavitation. The sampling frequency is adaptively adjusted according to the pump speed. .

[0138] in, Rotational speed (rpm) This represents the number of leaves.

[0139] Signal quality assessment: Calculating real-time signal quality indicators ,when The filter parameters are automatically adjusted in real time.

[0140] 2. Implementation of variational mode decomposition: Separate fault-related modal components (such as bearing impact and cavitation pulse) from complex vibration signals to avoid modal aliasing.

[0141] Adaptive selection of modal number: The number of decomposition modes K is dynamically adjusted according to the operating conditions of the centrifugal pump.

[0142] Normal operating conditions: It mainly decomposes the fundamental frequency component (1st harmonic, reflecting rotor rotation), the 2nd harmonic component (misalignment characteristics), and the high-frequency noise component (>1kHz, environmental interference).

[0143] Fault conditions: This adds fault characteristic modes (such as 5-8kHz high-frequency pulses of cavitation and periodic impacts of bearing failure (approximately 300-500Hz, depending on the bearing model)).

[0144] Decomposition effect verification: through modal aliasing index To assess the quality of the decomposition, the following requirements must be met. For example: among the three modes decomposed under normal operating conditions, the fundamental frequency mode and the high-frequency noise mode... , The decomposition effect is satisfactory; if It automatically increases the number of modes K and re-decomposes them.

[0145] 3. Multi-scale feature extraction and fusion: Extracting feature vectors: .

[0146] in: Microscale (instantaneous) characteristics (8): Based on instantaneous characteristics, focusing on the instantaneous abrupt changes of faults, including instantaneous amplitude change rate, instantaneous frequency standard deviation, instantaneous phase fluctuation rate, instantaneous energy peak, etc.; Mesoscale (local) features (6): Local statistics based on sliding window capture short-term trends of faults, including local energy density, local frequency variance, local correlation dimension, etc. Macro-scale (global): 4 features including modal energy ratio and modal entropy.

[0147] For example, when the outer ring of a bearing is worn, the local energy density of the vertical mode at the impact location will be 3-5 times higher than normal; when the impeller is unbalanced, the frequency variance of the horizontal mode will increase with the rotational speed; during the cavitation development stage, the local correlation dimension will increase significantly.

[0148] 4. Dynamic diagnostic decision-making Establish a centrifugal pump health index model: .

[0149] Set four warning thresholds: normal:

[0150] Notice: (Initiating cavitation monitoring) warn: (Check impeller balance) Danger: (Shutdown for maintenance) Example 2: Roots blower adaptive diagnostics To address the fault diagnosis needs of Roots blowers, and considering their significant low-frequency vibrations (low speed, low gear meshing frequency) and the core faults being poor gear meshing and bearing wear, an adaptive diagnostic scheme is designed. The specific implementation steps are as follows: 1. Low-frequency signal acquisition and feature enhancement processing Because the low-frequency characteristics (<30Hz) of Roots blowers, resulting from their relatively low operating speed (500-2000rpm), are easily masked by environmental vibrations, a special low-frequency enhancement strategy is adopted. Sensor configuration and installation: Three low-frequency response accelerometers were selected and installed at key fault locations: Horizontal direction of gearbox housing: collecting gear meshing vibration; Vertical direction of the drive end bearing housing: Monitor the wear of the bearing outer ring; Rotor housing axial: Captures low-frequency axial vibrations (<20Hz) caused by rotor imbalance or rubbing. Adaptive frequency band allocation: .

[0151] Low-frequency feature enhancement processing: To address the issue of low signal-to-noise ratio in low-frequency signals from Roots blowers, and to enhance low-frequency characteristics, a bandpass filter (bandwidth ±50Hz) with a center frequency of 326Hz (meshing frequency) is selected for the gearbox vibration signal to extract the gear fault sideband modulated at the meshing frequency.

[0152] 2. Gear meshing fault feature extraction The key focus is on extracting the gear meshing frequency and its sideband characteristics:

[0153] Sideband energy ratio: .

[0154] The numerator is the sum of the energies of the sidebands on both sides of the meshing frequency (excluding the meshing frequency itself), and the denominator is the energy at the meshing frequency. The problem was initially diagnosed as a gear malfunction.

[0155] 3. Multiscale mode decomposition and feature fusion Adaptive selection of modal number: The vibration signal of a Roots blower contains three main components: rotor low frequency (<20Hz), bearing characteristics (80-110Hz), and gear meshing (300-350Hz). Therefore, the modal number K is set as follows: Normal operating conditions: K=4 (rotor mode, bearing mode, gear meshing mode, noise mode); Fault condition: K=6 (new gear sideband mode and bearing impact mode).

[0156] The number of modes is optimized using the information entropy criterion to ensure that the correlation coefficient between gear modes and bearing modes is <0.2 (no aliasing).

[0157] Multi-scale feature vector construction: .

[0158] There are 16 features in total, mainly including: Microscale (instantaneous): instantaneous amplitude fluctuations during gear meshing, instantaneous phase abrupt changes during bearing impact; Mesoscale (local): bearing characteristic frequency energy density and gear sideband power ratio within the sliding window; Macroscale (global): Gear modal energy ratio, bearing modal entropy.

[0159] 4. Diagnostic decision-making and early warning mechanism: Based on the health index to quantify the equipment status, graded early warnings are set for gear and bearing failures to avoid sudden shutdowns.

[0160] Health Index Model: Combining the condition of gears, bearings, and rotors:

[0161] in: Gear health index, normal ≥0.9; Bearing health index, normal ≥0.85; Rotor health index, normal ≥0.8.

[0162] Level 4 Early Warning Thresholds and Responses: Normal state: Data is recorded every 30 minutes, and a device normal operation report is pushed out. Note the status: This triggers gearbox oil temperature monitoring; if the oil temperature is >80℃, the cooling fan will start automatically. Warning status: Push bearing inspection early warning: Display the characteristic frequency energy trend graph of the bearing outer ring; Dangerous situation: Emergency warning: If SBR>0.2, shutdown protection will be triggered.

[0163] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of wastewater treatment plant equipment based on multi-scale adaptive feature extraction, characterized in that, Includes the following steps: S1. Adaptive preprocessing and quality assessment are performed on the collected vibration signals of sewage treatment plant equipment. An adaptive signal quality assessment mechanism is established, and the preprocessing parameters are automatically adjusted according to the signal quality indicators. S2. An adaptive signal decomposition method that fuses variational mode decomposition and improved empirical mode decomposition is used to decompose the preprocessed signal; S3. Extract multi-scale features, including micro-scale instantaneous features, meso-scale local statistical features, and macro-scale global statistical features; S4. Establish an intelligent feature fusion mechanism based on information theory, and use mutual information to evaluate feature importance to solve the problems of feature redundancy and complementarity; S5. Construct a dynamic threshold diagnostic decision system to achieve adaptive early warning based on equipment health index and degradation trend.

2. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, In step S1, the signal quality evaluation function is composed of a weighted combination of signal-to-noise ratio, signal feature strength index, and signal purity index.

3. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, In step S2, the adaptive signal decomposition method determines the optimal number of modes using the information entropy criterion and introduces an adaptive penalty parameter to optimize the variational mode decomposition process.

4. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, In step S3, the microscale instantaneous features include instantaneous amplitude, instantaneous frequency, and instantaneous energy; the mesoscale local statistical features include local energy density, local frequency variance, and local correlation dimension; and the macroscale global statistical features include modal energy ratio and modal entropy.

5. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, In step S4, the intelligent feature fusion adopts a two-layer fusion strategy: the first layer performs weighted fusion of features at the same scale, and the second layer uses adaptive weights to fuse features at different scales.

6. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, In step S5, the dynamic threshold adjustment is adaptively adjusted based on the rate of change of the device health index and the variance of the health index fluctuation, and a four-level early warning mechanism is established: normal state, attention state, warning state and dangerous state.

7. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 1, characterized in that, It also includes S6, building an adaptive fault knowledge base to achieve incremental learning and evolutionary modeling of fault modes.

8. The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to claim 7, characterized in that, In step S6, the adaptive fault knowledge base uses a density-based anomaly detection algorithm to identify new fault modes and models the fault evolution path based on Markov chains.

9. A wastewater treatment plant equipment fault diagnosis system based on multi-scale adaptive feature extraction, characterized in that, The wastewater treatment plant equipment fault diagnosis method based on multi-scale adaptive feature extraction according to any one of claims 1-8 is adopted, including: The signal preprocessing module is used to perform adaptive preprocessing and quality assessment on the collected vibration signals from wastewater treatment plant equipment. The signal decomposition module, connected to the signal preprocessing module, is used to decompose the preprocessed signal using a fusion method of variational mode decomposition and improved empirical mode decomposition. The feature extraction module, connected to the signal decomposition module, is used to extract microscale instantaneous features, mesoscale local statistical features, and macroscale global statistical features. The feature fusion module, connected to the feature extraction module, is used to intelligently fuse multi-scale features based on information theory. The diagnostic decision module, connected to the feature fusion module, is used to construct a dynamic threshold diagnostic decision system and achieve adaptive early warning based on the device health index. The fault knowledge base module, connected to the diagnostic decision module, is used to build an adaptive fault knowledge base to achieve incremental learning and evolutionary modeling of fault modes.

10. The wastewater treatment plant equipment fault diagnosis system based on multi-scale adaptive feature extraction according to claim 9, characterized in that, The feature fusion module adopts a two-layer fusion strategy: the first layer performs weighted fusion of features at the same scale, and the second layer uses adaptive weights to fuse features at different scales.

11. The wastewater treatment plant equipment fault diagnosis system based on multi-scale adaptive feature extraction according to claim 9, characterized in that, The diagnostic decision module establishes a four-level early warning mechanism, including normal state, attention state, warning state, and danger state, and evaluates the credibility of the diagnostic results based on Bayesian inference.

Images