An industrial equipment data acquisition and abnormal behavior monitoring system

Abstract

The application discloses an industrial equipment data acquisition and abnormal behavior monitoring system, and belongs to the field of data acquisition and processing. The industrial equipment data acquisition and abnormal behavior monitoring system comprises a working condition interval division and modeling module, which is used for dividing equipment running states into discrete steady working condition intervals according to collected equipment data, and independently constructing a static statistical benchmark model for each working condition interval; a nonlinear mutation signal acquisition module, which is used for deploying a high dynamic response sensor and an anti-electromagnetic interference hardware for nonlinear mutation signals, and synchronously adopting a real-time signal processing technology to eliminate environmental noise; and a data synchronous fusion module, which is used for establishing a unified time scale and synchronously aligning multi-physical quantity data through a hardware clock. The application has the beneficial effects that the application can effectively identify and monitor signals under different working conditions, and avoid misjudgment of equipment abnormal behaviors by dividing different working condition intervals and independently constructing a static statistical benchmark model and a dynamic health model.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of data acquisition and processing, and particularly relates to an industrial equipment data acquisition and abnormal behavior monitoring system. BACKGROUND

[0002] There are strong nonlinear devices in the industrial field, typical representatives including electric arc furnaces, chemical reaction kettles of certain types, nonlinear transmission systems with significant mechanical clearance or friction, and combustion processes, etc. The output parameters (such as temperature, composition, vibration) of such devices and the input control quantities (such as current, fuel, speed) present complex nonlinear relationships, often showing nonlinear dynamic characteristics such as hysteresis effect, saturation characteristics or multi-stable state phenomenon. The running state and normal parameter range of the device will change significantly with the working point (such as load, speed, formula, etc.), resulting in the same monitoring signal possibly corresponding to normal or abnormal state under different working conditions. Therefore, the traditional monitoring methods (such as statistical process control based on linear assumption, simple threshold alarm, linear regression model, etc.) often perform poorly in the abnormal detection of strong nonlinear devices, and need to be improved. SUMMARY

[0003] Therefore, it is necessary to provide an industrial equipment data acquisition and abnormal behavior monitoring system in view of the above problems.

[0004] The embodiment of the present application is implemented as follows: an industrial equipment data acquisition and abnormal behavior monitoring system, comprising:

[0005] A working condition interval division and modeling module is used to divide the equipment running state into discrete steady state working condition intervals (clustering algorithm or physical rules can be used for division, such as "70%-80% load interval" and "high temperature reaction stage") according to the collected equipment data, and to independently construct a static statistical benchmark model (such as 3σ threshold of vibration amplitude and temperature fluctuation range) for each working condition interval;

[0006] A nonlinear mutation signal acquisition module is used to deploy high dynamic response sensors (>10kHz sampling rate) and anti-electromagnetic interference hardware (such as optical fiber sensors and differential input modules) for nonlinear mutation signals (such as electric arc furnace current sudden change and chemical reaction kettle pressure peak), and to synchronously use real-time signal processing technology (such as wavelet denoising and adaptive filtering) to eliminate environmental noise (in the key action stage, such as valve opening instant, automatically switch to high resolution mode to capture transient characteristics, and ensure complete recording of nonlinear mutation signals);

[0007] A data synchronization fusion module is configured to establish a unified time scale, align multiple physical quantity data through hardware clock synchronization (such as PTP protocol), and integrate different scales and frequencies of sensor information into a unified state vector through a data fusion technology (such as Kalman filtering and deep learning feature splicing) (to solve the problem that a single signal cannot comprehensively represent a fault in a strong nonlinear device, for example, to fuse motor current harmonics and bearing vibration spectrum to identify early wear).

[0008] A feature extraction and modeling module is configured to extract a feature set sensitive to nonlinearity from the unified state vector in the same working condition interval, and to deepen and construct a dynamic health model by using a nonlinear algorithm (such as kernel principal component analysis / KPCA and long short-term memory network / LSTM): KPCA extracts a nonlinear feature combination (such as an implicit correlation between current harmonics and vibration spectrum) in the working condition interval through a kernel function mapping; LSTM learns a time sequence dependency relationship between parameters in the working condition interval (such as a thermal inertia hysteresis effect in the heating stage of a reaction kettle); and a newly built dynamic health model replaces a static statistical baseline model in the same working condition interval.

[0009] An abnormal behavior recognition module is configured to input real-time sensor data into a corresponding working condition model to generate a prediction residual, to preferentially select a dynamic health model and secondarily select a static statistical baseline model (to select the static statistical baseline model when the dynamic health model data cannot support the generation of the prediction residual, for example, when the dynamic health model is not trained, or when real-time data is out of the input range of the dynamic health model), to recognize a deviation (which can be recognized by using a nonlinear statistical control chart such as a kernel density estimation control limit, or by using a time sequence anomaly detector such as a Transformer based on an attention mechanism), and to determine a device abnormal behavior.

[0010] In one embodiment, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, further comprising:

[0011] An abnormal root cause positioning module is configured to position an abnormal root cause in combination with mechanism knowledge (such as a reaction kinetics equation) and an interpretable AI tool (such as SHAP value analysis) for a device abnormal behavior detection result, to distinguish between a control strategy adjustment (such as PID parameter tuning) and a real physical fault (such as catalyst deactivation), and to reduce false positives.

[0012] In one embodiment, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, further comprising:

[0013] A verification and optimization module is configured to build a digital twin that integrates physical equations, static statistical baseline models, and dynamic health models, inject typical fault modes (such as heat transfer coefficient attenuation and bearing gap increase) into a virtual environment to verify the effectiveness of the monitoring logic, and iteratively optimize the working condition boundaries and feature weights of the static statistical baseline models according to false positives and false negatives, and simultaneously adjust the online learning rate and decay coefficient of the dynamic health model to realize the co-evolution of the two types of models.

[0014] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, which further comprises:

[0015] A maintenance silence module is configured to automatically enter a preset silence period after the equipment performs a planned preventive maintenance operation (such as replacing a filter or lubricating a bearing), and suspend the abnormal behavior identification of related parameters (such as oil pressure and temperature transients) that are affected by the maintenance during the silence period, so as to avoid mistaking the normal adjustment process of the parameters after the maintenance as an abnormal behavior of the equipment.

[0016] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, and the working condition interval division and modeling module comprises:

[0017] A working condition authenticity verification unit is configured to verify the authenticity of each working condition interval divided, and the verification conditions include: A, there is a clear corresponding relationship on the physical rules (such as a specific process step or a load level); B, it is observed many times and stably in the historical data (set minimum occurrence frequency and duration threshold); C, the parameter combination in the working condition interval does not violate the known physical constraints of the equipment (such as mutual exclusion of the maximum pressure and the minimum temperature); the working condition interval that fails to pass the verification is marked, and after being confirmed by a human, it is decided to be retained, modified or discarded, so as to ensure that the static statistical baseline model is only established on the basis of the physically real and stable running state.

[0018] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, and the working condition interval division and modeling module comprises:

[0019] A sensor diagnosis unit is configured to continuously monitor the self-health state of a sensor (such as vibration or current), identify the aging or failure of the sensor (such as the fatigue of a piezoelectric crystal leading to a decrease in sensitivity) by analyzing technical indicators of the sensor, the technical indicators including impedance drift trend and frequency response characteristic change (such as resonance peak attenuation), and trigger an alarm and prompt to replace the sensor when the output signal of the sensor is insufficient to support the judgment of the abnormal behavior of the equipment, so as to avoid mistaking the smooth running of the equipment due to the signal distortion caused by the failure of the sensor as the abnormal behavior of the equipment.

[0020] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, the working condition interval division and modeling module comprises:

[0021] The medium physical property identification unit is used for identifying the masking effect of the change of the medium physical property (such as the viscosity of lubricating oil and the concentration of reactants) on the equipment monitoring signal (such as vibration and noise) in combination with the process medium characteristic sensor (such as dielectric constant and viscosity meter), establishing the correlation model of the medium physical property and the signal response (such as the high-viscosity lubricating oil inhibiting the abnormal sound of the bearing to cause the false normal of the vibration value), filtering the medium interference in the acquired equipment data, and improving the accuracy of the identification of the real running state of the equipment.

[0022] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, the acquired equipment data comprises load, rotating speed and formula.

[0023] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, the multi-physical quantity data comprises mechanical vibration, temperature, current and chemical concentration.

[0024] In one of the embodiments, the present application provides an industrial equipment data acquisition and abnormal behavior monitoring system, the feature set sensitive to nonlinearity comprises time domain statistics (kurtosis and margin factor), frequency domain envelope spectrum and nonlinearity index (recurrence plot entropy and Lyapunov exponent).

[0025] Compared with the prior art, the present application has the beneficial effects that: the present application can effectively identify the problem of the reaction of the monitoring signal under different working conditions by dividing different working condition intervals and independently constructing static statistical benchmark models and dynamic health models, and the false judgment of the abnormal behavior of the equipment can be avoided. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 FIG. 1 is a first part schematic diagram of an industrial equipment data acquisition and abnormal behavior monitoring system provided by an embodiment of the present application.

[0027] Figure 2 FIG. 2 is a second part schematic diagram of an industrial equipment data acquisition and abnormal behavior monitoring system provided by an embodiment of the present application.

[0028] Figure 3 FIG. 3 is a third part schematic diagram of an industrial equipment data acquisition and abnormal behavior monitoring system provided by an embodiment of the present application.

[0029] Figure 4 FIG. 4 is a fourth part schematic diagram of an industrial equipment data acquisition and abnormal behavior monitoring system provided by an embodiment of the present application.

[0030] Figure 5A first part of a substructure of a working condition interval division and modeling module provided by an embodiment of the present application is shown in the figure.

[0031] Figure 6 A second part of the substructure of the working condition interval division and modeling module provided by the embodiment of the present application is shown in the figure.

[0032] Figure 7 A third part of the substructure of the working condition interval division and modeling module provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0034] It can be understood that the terms "first", "second", etc. used in the present application can be used herein to describe various elements, but unless specifically stated, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of the present application, the first xx script can be referred to as the second xx script, and similarly, the second xx script can be referred to as the first xx script.

[0035] In one embodiment, as shown in Figure 1 An industrial equipment data acquisition and abnormal behavior monitoring system includes:

[0036] A working condition interval division and modeling module 1 is used to divide the equipment running state into discrete steady state working condition intervals (which can be divided by clustering algorithm or physical rules, such as "70%-80% load interval" and "high temperature reaction stage") according to the collected equipment data, and to independently construct a static statistical benchmark model (such as 3σ threshold of vibration amplitude and temperature fluctuation range) for each working condition interval;

[0037] A nonlinear mutation signal acquisition module 2 is used to deploy a high dynamic response sensor (>10kHz sampling rate) and an anti-electromagnetic interference hardware (such as an optical fiber sensor and a differential input module) for nonlinear mutation signals (such as arc furnace current sudden change and chemical reaction kettle pressure peak), and to synchronously use real-time signal processing technology (such as wavelet denoising and adaptive filtering) to eliminate environmental noise (in the key action stage, such as valve opening instant, automatically switch to high resolution mode to capture transient characteristics, and ensure complete recording of nonlinear mutation signals);

[0038] Data synchronization fusion module 3 is used to establish a unified time scale, align multiple physical quantity data through hardware clock synchronization (such as PTP protocol), and integrate different scales and frequencies of sensor information into a unified state vector by using data fusion technology (such as Kalman filtering and deep learning feature splicing) to solve the problem that a single signal cannot fully represent the fault in a strong nonlinear device, for example, fusing motor current harmonics and bearing vibration spectrum to identify early wear.

[0039] Feature extraction and modeling module 4 is used to extract a feature set sensitive to nonlinearity from the unified state vector within the same working condition interval, and to deepen the construction of a dynamic health model by using a nonlinear algorithm (such as kernel principal component analysis / KPCA and long short-term memory network / LSTM): KPCA extracts nonlinear feature combinations within the working condition interval (such as the implicit correlation between current harmonics and vibration spectrum) through kernel function mapping; LSTM learns the time sequence dependency between working condition interval parameters (such as the thermal inertia hysteresis effect in the heating stage of a reaction kettle); and the newly built dynamic health model replaces the static statistical baseline model in the same working condition interval.

[0040] Abnormal behavior recognition module 5 is used to input real-time sensor data into the corresponding working condition model to generate a prediction residual, to preferentially select a dynamic health model and to select a static statistical baseline model as a second choice (selecting a static statistical baseline model when the dynamic health model data cannot support the generation of a prediction residual, for example, when the dynamic health model is not trained, or when real-time data is out of the input range of the dynamic health model), to identify deviations (which can be identified by using a nonlinear statistical control chart such as kernel density estimation control limit, or by using a time sequence anomaly detector such as attention mechanism-based Transformer), and to determine the abnormal behavior of the device.

[0041] The collected device data includes load, speed, and formula; the multiple physical quantity data includes mechanical vibration, temperature, current, and chemical concentration; and the feature set sensitive to nonlinearity includes time domain statistics (kurtosis, margin factor), frequency domain envelope spectrum, and nonlinear indicators (recurrence plot entropy, Lyapunov exponent).

[0042] In the working condition interval division and modeling module 1, a clustering algorithm (such as K-means and DBSCAN) is used to automatically cluster data points (such as load and speed) based on their similarity, and each cluster represents a stable working condition interval (such as "75% load ± 3%"); or the interval boundaries are manually defined according to physical rules (such as process step settings "heating stage > 100°C" and "constant pressure reaction stage"). The independent construction of a static statistical baseline model refers to the calculation of the statistical characteristics (such as mean and standard deviation) of key parameters (such as vibration amplitude) in each divided working condition interval using only the historical normal data corresponding to the interval, and the fixed threshold (such as ±3σ range and maximum and minimum values) is set as the health baseline of the working condition.

[0043] In the nonlinear mutation signal acquisition module 2, the nonlinear mutation signal (such as current sudden change, pressure peak) has the characteristics of sharp amplitude change, extremely short duration (millisecond level) and easy to be overwhelmed by noise. The general data acquisition (the source of data obtained in the working condition interval division and modeling module 1) usually has a low sampling rate (such as 1 kHz), which is not enough to completely capture such transient details; at the same time, its hardware may not have strong anti-interference ability (such as ordinary current transformer is easy to be disturbed by electromagnetic interference). Therefore, it is necessary to specially deploy high sampling rate (>10 kHz) sensors (such as optical fiber current sensor) and anti-interference hardware (such as differential input), and apply wavelet denoising technology in real time, to ensure that the high resolution mode can be switched to in key transient events (such as valve opening), and these mutation signals can be recorded completely and accurately.

[0044] In the data synchronization fusion module 3, the hardware clock synchronization (such as PTP protocol) is used to ensure that the data from different sensors (vibration, temperature, current, etc.) have accurate and unified time scales. In the data fusion technology, Kalman filter is suitable for linear or weak nonlinear systems, and through the prediction and update cycle, it can fuse multi-source data and estimate the optimal state (such as fusing the rotating speed and vibration to estimate the shaft center position); deep learning feature splicing (such as using convolutional neural network to extract features of different sensor data, and then splicing feature vectors) can handle strong nonlinearity, and integrate heterogeneous information of different scales / frequencies (such as slow-changing temperature and high-frequency vibration) into a unified state vector containing more comprehensive information, overcoming the problem of insufficient representation ability of a single signal (for example, the fusion of current harmonics and vibration spectrum can discover wear and tear earlier).

[0045] In the feature extraction and modeling module 4, KPCA is selected because it can map the original features to a high-dimensional space through a kernel function (such as Gaussian kernel), and effectively extract nonlinear combined features (such as revealing the complex relationship between current harmonic components and specific vibration frequency amplitudes) in this space, which can better capture the nonlinear behavior of the device than linear PCA (Principal Component Analysis); LSTM is selected because its gating mechanism can effectively learn and remember the time sequence dependence in long sequences (such as the temperature rise stage of the reactor, the current temperature is not only affected by the current heating power, but also affected by the inertial hysteresis effect of the previous heat accumulation). The dynamic health model constructed by deepening these nonlinear algorithms can more finely represent the dynamic behavior and degradation trend of the device under a specific working condition than the static baseline model.

[0046] In the abnormal behavior recognition module 5, the abnormality is recognized by calculating the difference (residual error) between the predicted value generated by the real-time data input model (preferably a more accurate dynamic health model, and falling back to a static statistical benchmark model when it is not ready or the input is out of range) and the actual observation value. Under normal conditions, the residual error should be small and random, while failure will cause the residual error to increase significantly or show a certain pattern. The static statistical benchmark model is simple, stable, easy to build, but not accurate enough; the dynamic health model is more accurate, but needs enough data for training and is sensitive to input range, so two models are designed to ensure that the system provides high-precision monitoring when the dynamic health model is available, and there is still a reliable static statistical benchmark model to back up when the dynamic health model is not available, ensuring continuous monitoring.

[0047] In one embodiment, as shown in Figure 2 An industrial equipment data acquisition and abnormal behavior monitoring system further comprises:

[0048] The abnormal root cause positioning module 6 is used for positioning the abnormal root cause in combination with mechanism knowledge (such as reaction kinetics equation) and interpretable AI tools (such as SHAP value analysis, which measures the contribution of each feature) for equipment abnormal behavior detection results, distinguishing between control strategy adjustment (such as PID parameter tuning) and real physical failure (such as catalyst deactivation), and reducing false positives.

[0049] The abnormal root cause positioning module 6 provides a physical level explanation framework in combination with mechanism knowledge (such as applying reaction kinetics equation to analyze abnormal temperature / concentration data to judge whether it conforms to the reaction rate decline rule caused by deactivated catalyst); at the same time, it uses interpretable AI tools such as SHAP value analysis to calculate the contribution of each input feature (such as a specific vibration frequency, a current phase) to the abnormal detection result (such as high residual error), and finds out the abnormal root cause feature combination. Finally, if the SHAP high-contribution feature conforms to the expected mode of a certain physical failure (such as a specific wear frequency vibration contribution), it points to a real physical failure (such as bearing wear); if the high-contribution feature points to an adjustable parameter and conforms to the control logic (such as PID output saturation), it may be a control strategy problem (PID tuning is needed).

[0050] In one embodiment, as shown in Figure 3 An industrial equipment data acquisition and abnormal behavior monitoring system further comprises:

[0051] The verification and optimization module 7 is used for building a digital twin that integrates physical equations, static statistical benchmark models, and dynamic health models, injecting typical fault modes (such as heat transfer coefficient attenuation and bearing gap increase) into a virtual environment to verify the effectiveness of the monitoring logic; according to the false positives and false negatives, iteratively optimize the working condition boundary and feature weight of the static statistical benchmark model, and simultaneously adjust the online learning rate and decay coefficient of the dynamic health model, to realize the co-evolution of the two types of models.

[0052] The construction of the digital twin is achieved by integrating three core levels: the first level embeds the physical principle equations of the equipment (such as heat conduction partial differential equations, rotor dynamics equations), which accurately simulate the physical behavior of the equipment under ideal conditions; the second level integrates static statistical benchmark models, which reflect the historical normal statistical characteristics of the equipment under specific steady-state working conditions; the third level accesses dynamic health models, which capture the nonlinear evolution and degradation trend of the equipment state over time; in this virtual environment formed by the fusion of the models, typical faults are simulated by injecting fault modes, the digital twin is run, and the abnormal behavior identification results of the abnormal behavior identification module 5 (such as whether the residual exceeds the standard or whether the alarm is triggered) are observed, thereby verifying the effectiveness of the entire monitoring logic.

[0053] In one embodiment, as shown in FIG. 1, an industrial equipment data acquisition and abnormal behavior monitoring system further comprises: Figure 4

[0054] A maintenance quiet module 8 is configured to automatically enter a preset quiet period after the equipment performs a planned preventive maintenance operation (such as replacing a filter element or lubricating a bearing), and to suspend abnormal behavior identification of parameters related to the maintenance (such as oil pressure and temperature transients) during the quiet period, thereby avoiding misjudgment of the normal adjustment process of the parameters after maintenance as abnormal behavior of the equipment.

[0055] After the equipment performs a planned maintenance operation (such as replacing a filter element or lubricating a bearing), related parameters (such as oil pressure and temperature) usually need to undergo an adjustment period (such as a change in flow resistance due to a new filter element requiring the system to rebalance the oil pressure, or a temporary temperature fluctuation due to the need to circulate new lubricating oil evenly), which is a normal process within expectations, but may temporarily exceed the normal value and cause an alarm. The quiet period mechanism suspends abnormal monitoring of these directly affected parameters during the preset time period, thereby avoiding misjudgment of these normal adjustment transients as equipment failure and reducing unnecessary false alarms.

[0056] In one embodiment, as shown in FIG. 1, an industrial equipment data acquisition and abnormal behavior monitoring system further comprises: Figure 5

[0057] A working condition authenticity verification unit 11 is configured to verify the authenticity of each working condition interval divided, and the verification conditions include: A, there is a clear corresponding relationship on the physical rules (such as a specific process step, a load level); B, it is observed multiple times and stably in historical data (set minimum occurrence frequency and duration threshold); C, the parameter combination in the working condition interval does not violate known physical constraints of the equipment (such as mutual exclusion of maximum pressure and minimum temperature); working condition intervals that fail the verification are marked, and after manual confirmation, a decision is made to retain, modify, or discard, thereby ensuring that the static statistical benchmark model is only established on the basis of physically real and stable operating states.

[0058] ​​The defined operating condition intervals (especially those based on clustering algorithms) may contain pseudo-operating conditions that are purely data-driven, have unclear physical meaning, or are unstable (such as transient transition states or clusters of data anomalies). Therefore, manual verification is required: Does the operating condition correspond to a real process stage (A)? Is there sufficient historical data to support its stable existence (B)? Is its parameter combination physically feasible (C)? For example, if a clustered high-temperature, low-pressure interval violates equipment safety constraints (pressure cannot be too low at the highest temperature), manual review is needed to decide whether to correct the boundary or discard the interval. Manual verification avoids building static statistical benchmark models on unreliable or unrealistic operating conditions, ensuring model reliability.

[0059] In one embodiment, such as Figure 6 As shown, an industrial equipment data acquisition and abnormal behavior monitoring system includes a working condition interval division and modeling module 1, which comprises:

[0060] The sensor diagnostic unit 12 is used to continuously monitor the health status of sensors (such as vibration and current). By analyzing the sensor's technical indicators, including impedance drift trends and frequency response characteristics (such as resonance peak attenuation), it identifies sensor aging or failure (such as decreased sensitivity due to piezoelectric crystal fatigue). When the sensor output signal is insufficient to support the judgment of abnormal equipment behavior, it triggers an alarm and prompts the sensor to be replaced, so as to avoid the abnormal equipment behavior being mistakenly judged as stable equipment operation due to signal distortion caused by sensor failure.

[0061] The health of sensors is monitored by continuously analyzing their output technical specifications. For example, the impedance of piezoelectric vibration sensors is measured (significant impedance drift may indicate internal connection problems); their frequency response characteristics are analyzed (e.g., under standard excitation, attenuation of the resonant peak amplitude or frequency shift indicates aging / fatigue of the piezoelectric crystal leading to decreased sensitivity); baseline drift or abnormally increased noise levels are monitored in current sensors. When such indicators continue to deteriorate, indicating sensor failure or insufficient performance, an alarm is triggered to prompt replacement, preventing distorted signals from sensor malfunctions from being mistaken for normal equipment operation or masking actual equipment failures.

[0062] In one embodiment, such as Figure 7 As shown, an industrial equipment data acquisition and abnormal behavior monitoring system includes a working condition interval division and modeling module 1, which comprises:

[0063] The medium property identification unit 13 is used to combine process medium characteristic sensors (such as dielectric constant and viscometer) to identify the masking effect of changes in medium physical properties (such as lubricating oil viscosity and reactant concentration) on equipment monitoring signals (such as vibration and noise); establish a correlation model between medium properties and signal response (such as high viscosity lubricating oil suppressing bearing abnormal noise, resulting in falsely normal vibration values), filter medium interference in the collected equipment data, and improve the accuracy of identifying the true operating status of the equipment.

[0064] Real-time monitoring of process medium properties (e.g. lubricant viscosity, reactant concentration) by deploying specialized sensors (e.g. online viscometer, dielectric constant sensor). Establishing the correlation model usually involves: collecting equipment monitoring signals (vibration, noise, etc.) under different medium property conditions; applying regression analysis (e.g. multiple linear regression, support vector regression) or neural networks to learn the quantitative mapping relationship between medium properties (input) and signal features (output, such as total vibration value, envelope spectrum energy) (e.g. "for every X% increase in viscosity, the bearing vibration total value decreases by Y%"). Using the correlation model, the normal signal response can be predicted according to the current measured medium properties, and the medium interference component can be subtracted or corrected from the actual signal, so that the changes in the equipment itself can be more accurately identified.

[0065] The complete working process of the system is illustrated by monitoring a large industrial air compressor. In the working condition interval division and modeling module 1, according to its operating data (load, outlet pressure), the stable running state is divided into "70%-75% load interval", "80%-85% load interval", etc. by using clustering algorithm, and a static statistical baseline model is independently constructed for each interval. For example, in the "80%-85% load interval", the historical normal data shows that the 3σ range of the RMS value of the main bearing vibration speed is 2.0-2.8 mm / s, and the temperature fluctuates at 85±3°C, which forms the health baseline under this working condition.

[0066] In the nonlinear mutation signal acquisition module 2, high-frequency vibration acceleration sensors (>20kHz sampling rate) are deployed to capture the bearing transient impact signals that may occur when the air compressor starts or the load suddenly changes, and are matched with electromagnetic interference-resistant charge amplifiers. When the control system issues a load increase instruction instantaneously, the nonlinear mutation signal acquisition module 2 automatically switches to a high-resolution mode, and a 5ms long impact pulse with an amplitude much higher than the steady state is captured by applying wavelet denoising, ensuring that this nonlinear transient feature is not missed by the general acquisition channel.

[0067] In the data synchronization fusion module 3, the time scales of vibration, motor current, outlet pressure, cooling water temperature, etc. are accurately synchronized through PTP protocol; using deep learning feature splicing technology, the envelope spectrum features of high-frequency vibration signals (indicating impact), specific harmonic components of current signals (indicating torque fluctuation) and pressure fluctuation features are fused into a unified state vector. This vector can more comprehensively represent the overall state of the compressor under the "80%-85% load interval", for example, a slight bearing impact is often accompanied by a slight increase in the 5th harmonic of the current.

[0068] In the feature extraction and modeling module 4, for the unified state vector in the "80%-85% load interval", the time-domain kurtosis (reflecting the impact), the frequency-domain envelope spectrum entropy (reflecting the signal complexity) and other feature sets sensitive to nonlinearity are extracted; the KPCA discovers the nonlinear association combination between the energy of a certain frequency band in the vibration envelope spectrum and the current harmonics (which may correspond to the coupling effect of mechanical and electrical); at the same time, the LSTM learns the dynamic relationship that the temperature change lags behind the load change in this interval (thermal inertia effect), and based on this, a more detailed dynamic health model than the static statistical baseline model is established, which can predict the parameter values under normal conditions.

[0069] In the abnormal behavior recognition module 5, when the air compressor runs at 82% load, the real-time sensor data is input into the dynamic health model of this working condition, and the vibration envelope spectrum entropy predicted by LSTM and the actual calculated value have a significant residual (exceeding the control limit based on kernel density estimation), and the KPCA reconstruction error also abnormally rises, which determines that there is an abnormal behavior; at this time, the dynamic health model is available and the input range is compliant, so it is preferred to use it rather than the static statistical baseline model.

[0070] In the abnormal root cause positioning module 6, for the identified abnormality, combined with the mechanism knowledge of air compressor bearing fault (such as the formula for calculating the fault characteristic frequency of rolling bearing), the feature combination with the largest residual is analyzed: it is found that the frequency corresponding to the impact pulse is consistent with the fault characteristic frequency of the bearing outer ring, and the SHAP value analysis shows that the energy of this frequency band in the vibration envelope spectrum contributes the most to the abnormality determination, and the current harmonic change conforms to the expected mode of bearing jamming leading to load fluctuation, and the comprehensive judgment is that the root cause is the early wear of the bearing outer ring (real physical fault), rather than the control parameter problem.

[0071] In the verification and optimization module 7, a digital twin is constructed, in the virtual environment, a "bearing gap increase" fault mode is deliberately injected (the dynamic equation parameters are modified), and after running, the digital twin successfully simulates similar vibration impact and current harmonic change, and the abnormal behavior recognition module 5 also correctly alarms in the virtual environment, verifying the effectiveness of the monitoring logic; at the same time, according to the false alarm that has occurred in actual operation (such as sudden disturbance of cooling water temperature), the weight of the temperature feature in the static statistical baseline model is adjusted, and the online learning rate of the dynamic health model is reduced to prevent overfitting.

[0072] In the maintenance silent module 8, when the main bearing lubricating oil of the air compressor is replaced as planned, a 24-hour silent period is automatically started, during which the abnormal monitoring of lubricating oil pressure fluctuation and bearing temperature transient rise is suspended, avoiding the normal parameter adjustment in the new oil circulation and temperature balance process being misjudged as a fault.

[0073] In the working condition authenticity verification unit 11, an early clustering has divided a "75%-78% load but vibration abnormally high" interval, but the physical authenticity verification (condition C) finds that the state violates the bearing design load constraint (the vibration should not continue to exceed the standard under this load), and the historical data occurrence frequency is insufficient (condition B), which is marked and confirmed by manual confirmation as an abnormal working condition cluster, and finally discarded to ensure that the static statistical benchmark model is based on healthy data.

[0074] In the sensor diagnosis unit 12, the impedance and resonance frequency of the high-frequency vibration sensor are continuously monitored, and it is found that the resonance peak amplitude has been continuously attenuated by 15% in a few months, and the frequency response characteristic has changed. It is determined that the sensitivity of the sensor itself piezoelectric element is decreased due to aging, and timely replacement alarm is triggered to prevent the signal attenuation from covering the real bearing impact signal.

[0075] In the medium property identification unit 13, the viscosity of the newly replaced lubricating oil is monitored by an online viscometer to be 10% higher than the standard value, and based on the pre-established correlation model (high viscosity will suppress the vibration amplitude by about 8%), the vibration amplitude monitored in real time is automatically upwardly compensated and corrected, ensuring that the vibration value reduced due to the increase in viscosity is restored to a comparable level under standard viscosity, thereby more accurately exposing the real vibration rising trend caused by bearing wear, and finally completing the whole process from abnormal detection to root cause positioning.

[0076] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or sub-steps or stages of other steps.

[0077] Each technical feature of the above-described embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of each technical feature in the above-described embodiments are not described, but as long as the combination of these technical features does not exist contradictory, it should be considered as the scope of the present disclosure.

[0078] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0079] The above only describes the preferred embodiments of the present application, and is not used to limit the present application, and any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

[0080] In addition, it should be understood that although the present application is described in the embodiments, not every embodiment only contains one independent technical solution, and the description manner of the specification is only for the sake of clarity, and the skilled person in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be combined to form other embodiments which can be understood by the skilled person.

Claims

1. An industrial equipment data acquisition and abnormal behavior monitoring system, characterized by, The industrial equipment data acquisition and abnormal behavior monitoring system comprises: The working condition interval division and modeling module is used for dividing the equipment running state into discrete steady state working condition intervals according to the collected equipment data, and independently constructing a static statistical benchmark model for each working condition interval; The nonlinear mutation signal acquisition module is used for deploying a high dynamic response sensor and an anti-electromagnetic interference hardware for the nonlinear mutation signal, and synchronously using a real-time signal processing technology to eliminate environmental noise; The data synchronization fusion module is used for establishing a unified time scale, aligning multiple physical quantity data through hardware clock synchronization, and integrating different scale and frequency sensor information into a unified state vector through data fusion technology; The feature extraction and modeling module is used for extracting a feature set sensitive to nonlinearity from the unified state vector in the same working condition interval, and deepening the construction of a dynamic health model through a nonlinear algorithm: KPCA extracts a nonlinear feature combination in the working condition interval through a kernel function mapping; LSTM learns the time sequence dependence relationship between the parameters in the working condition interval; and a new dynamic health model replaces the static statistical benchmark model in the same working condition interval; The abnormal behavior identification module is used for inputting real-time sensor data into a corresponding working condition model to generate a prediction residual, preferentially selecting a dynamic health model, secondarily selecting a static statistical benchmark model, identifying a deviation, and determining an equipment abnormal behavior; The working condition interval division and modeling module comprises: The working condition authenticity verification unit is used for verifying the physical authenticity of each working condition interval divided, and the verification conditions include: A, there is a clear corresponding relationship on the physical rule; B, it is observed many times and stably in historical data; C, the parameter combination in the working condition interval does not violate the known equipment physical constraint; the working condition interval that does not pass the verification is marked, and after manual confirmation, it is decided to keep, correct or discard, so that the static statistical benchmark model is only established on the basis of the physically real and stable running state.

2. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, Further comprising: The abnormal root cause positioning module is used for positioning the abnormal root cause in combination with mechanism knowledge and an interpretable AI tool according to the equipment abnormal behavior detection result, distinguishing between control strategy adjustment and real physical failure, and reducing false positives.

3. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, Further comprising: The verification and optimization module is used for constructing a digital twin that integrates physical equations, static statistical benchmark models and dynamic health models, and injecting a typical fault mode into a virtual environment to verify the effectiveness of the monitoring logic; According to the false positive and false negative analysis, the working condition boundary and feature weight of the static statistical benchmark model are iteratively optimized, the online learning rate and decay coefficient of the dynamic health model are simultaneously adjusted, and the collaborative evolution of the two types of models is realized.

4. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, Further comprising: The maintenance silent module is used for automatically entering a preset silent period after the equipment performs a planned preventive maintenance operation, suspending the abnormal behavior identification of related parameters affected by the maintenance during the silent period, and avoiding the normal adjustment process of the parameters after the maintenance from being misjudged as an equipment abnormal behavior.

5. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, The working condition interval division and modeling module comprises: A sensor diagnosis unit is configured to continuously monitor the self-health status of the sensor, identify the aging or failure of the sensor by analyzing sensor technical indexes including impedance drift trend and frequency response characteristic change, and trigger an alarm and prompt replacement of the sensor when it is detected that the sensor output signal is insufficient to support judgment of abnormal behavior of the device, so as to avoid misjudgment of the device as running smoothly due to signal distortion caused by failure of the sensor.

6. The industrial equipment data collection and abnormal behavior monitoring system of claim 1 or 5, wherein, The working condition interval division and modeling module comprises: A medium physical property identification unit is configured to identify the masking effect of changes in physical properties of the medium on the monitoring signal of the device in combination with a process medium characteristic sensor, establish a correlation model of the medium physical property and the signal response, filter the medium interference in the collected device data, and improve the accuracy of identification of the real running state of the device.

7. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, The collected device data includes load, rotating speed and formula.

8. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, The multi-physical quantity data includes mechanical vibration, temperature, current and chemical concentration.

9. The industrial equipment data collection and abnormal behavior monitoring system of claim 1, wherein, The feature set sensitive to nonlinearity includes time domain statistics, frequency domain envelope spectrum and nonlinearity index.

Images