A screw vacuum pump operation data analysis system based on big data

Abstract

The application discloses a screw vacuum pump operation data analysis system based on big data, belongs to the technical field of data processing, and comprises a data acquisition unit which is used for integrating real-time high-frequency data, process parameter data and equipment static information; a hidden state estimation unit which is used for estimating a hidden state variable representing material performance degradation of pump internal components according to the real-time high-frequency data collected by the data acquisition unit and generating equivalent deposition layer thickness; and a dynamic margin attenuation prediction unit which is used for predicting an attenuation track of a core performance margin according to the equivalent deposition layer thickness generated by the hidden state estimation unit and combining process parameter data collected by the data acquisition unit and generating a dynamic anti-impact margin prediction curve. The application greatly improves the accuracy and robustness of state perception and lays a solid foundation for subsequent accurate prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and particularly to a screw vacuum pump operation data analysis system based on big data. BACKGROUND

[0002] The monitoring system in the prior art mainly relies on threshold monitoring of real-time data of apparent physical quantities such as vibration, temperature and current. The limitation of this method is that the monitored indicators are a lagging representation of the system operating state. In the early stage of slow and continuous cumulative damage to the material performance of the pump internals, the macroscopic apparent indicators of the system will not appear significantly abnormal. This performance margin brought by design and material plays a buffering role, masking the deep physical degradation process. When the performance margin is exhausted, a normal process fluctuation may be coupled with the cumulative damage, amplified nonlinearly, and then lead to a sudden catastrophic shutdown such as stuck or broken vacuum. The traditional monitoring method cannot capture the key turning point from quantitative change to qualitative change of the performance margin, and therefore cannot provide forward-looking risk warning.

[0003] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0004] The present application aims to provide a screw vacuum pump operation data analysis system based on big data to solve the problems raised in the background.

[0005] The technical solution of the present application is as follows:

[0006] The data acquisition unit is used to integrate real-time high-frequency data, process parameter data and equipment static information;

[0007] The hidden state estimation unit is used to estimate the hidden state variable representing the material performance degradation of the pump internals according to the real-time high-frequency data collected by the data acquisition unit, and generate an equivalent deposited layer thickness;

[0008] The dynamic margin decay prediction unit is used to predict the decay trajectory of the core performance margin according to the equivalent deposited layer thickness generated by the hidden state estimation unit, and combine the process parameter data collected by the data acquisition unit, and generate a dynamic impact resistance margin prediction curve;

[0009] The functional failure risk quantification unit is used to quantify the failure risk probability according to the dynamic impact resistance margin prediction curve generated by the dynamic margin decay prediction unit, and generate a functional failure risk curve;

[0010] The decision support output unit is used to trigger an early warning and generate an operation instruction when the failure risk value at any time point in the functional failure risk curve generated by the functional failure risk quantification unit exceeds a preset warning threshold, based on the functional failure risk curve generated by the functional failure risk quantification unit.

[0011] Preferably, the hidden state estimation unit is specifically used for:

[0012] Based on the rotational speed and inlet / outlet pressure data contained in the real-time high-frequency data, and combined with the preset pumping mechanism model, the reference power consumption is calculated.

[0013] Obtain the actual measured motor power consumption contained in real-time high-frequency data;

[0014] The actual measured motor power consumption is compared with the reference power consumption to generate the power consumption residual.

[0015] Preferably, the hidden state estimation unit is further used for:

[0016] The power consumption residual is used as input, and a first-order linear filter in a preset discrete-time form is used to correct the equivalent deposition layer thickness of the previous time step in order to generate the equivalent deposition layer thickness of the current time step.

[0017] Preferably, the dynamic margin decay prediction unit is specifically used for:

[0018] Obtain future process parameters contained in the process parameter data;

[0019] The future process parameters are weighted and normalized to calculate the normalized process load intensity; the normalization process uses the maximum value of each process parameter in the historical process as the normalization benchmark.

[0020] Preferably, the dynamic margin decay prediction unit is further used for:

[0021] Based on the equivalent sedimentary layer thickness generated by the hidden state estimation unit and combined with the normalized process load intensity, the evolution trajectory of the equivalent sedimentary layer thickness in the future is predicted and generated by solving the preset first-order dynamic model.

[0022] Preferably, the dynamic margin decay prediction unit is further used for:

[0023] The equivalent deposition layer thickness and the normalized process load intensity are substituted into a preset margin decay model to calculate and generate a dynamic shock resistance margin prediction curve.

[0024] Preferably, the functional failure risk quantification unit is specifically used for:

[0025] The predicted margin value in the dynamic shock margin prediction curve is modeled as a normal distribution with the predicted margin value as the mean and the preset prediction uncertainty as the standard deviation.

[0026] Compare the normal distribution with the preset functional failure threshold;

[0027] Calculate the probability that the prediction margin value is less than the functional failure threshold, and determine this probability as the functional failure risk to generate a functional failure risk curve.

[0028] Preferably, the decision support output unit is specifically used for:

[0029] Compare the risk of functional failure with a preset warning threshold;

[0030] When the risk of functional failure exceeds the preset warning threshold, the system matches and generates intelligent maintenance prescriptions or process optimization suggestions as operation instructions from the preset decision rule base.

[0031] This invention provides an improved screw vacuum pump operation data analysis system based on big data, which has the following improvements and advantages compared with the prior art:

[0032] 1. This system forms a complete technical closed loop from data acquisition to decision support through five collaborative core units. It not only integrates real-time high-frequency data reflecting instantaneous states, process parameter data revealing future loads, and equipment static information defining model boundaries, but also delves into the equipment's interior to estimate the state of performance degradation processes that cannot be directly measured. By estimating hidden state variables characterizing the material performance degradation of pump internal components and generating a quantified equivalent deposition layer thickness, the system establishes a precise quantitative correlation between macroscopic, easily measurable physical quantities and microscopic, unmeasurable internal component degradation states. This estimation of hidden states is dynamic, adaptive, and closed-loop corrected, greatly improving the accuracy and robustness of state perception and laying a solid foundation for subsequent accurate predictions.

[0033] 2. Building upon accurate state estimation, the system further demonstrates its forward-looking predictive capabilities. By weighting and normalizing future process parameters, it calculates the normalized process load intensity, achieving effective dimensionality reduction and quantification of complex process loads. Subsequently, based on the current equivalent deposition layer thickness and combined with the future normalized process load intensity, the system deduces the future evolution trajectory of the equivalent deposition layer thickness by solving a pre-defined first-order dynamic model. Finally, this evolution trajectory and load intensity are substituted into a pre-defined margin decay model to scientifically describe the coupling and cumulative effects of base time aging, hidden state deterioration, and external process loads on performance margin consumption, calculating and generating a dynamic shock margin prediction curve. This complete prediction chain ensures that the prediction is not only based on the current state but also reflects the dynamic impact of future loads, and the generated margin prediction curve highly conforms to physical reality.

[0034] 3. To make the prediction results more valuable for decision-making, this solution also introduces probabilistic risk assessment. The system models the predicted margin value in the dynamic shock margin prediction curve as a normal distribution with its own mean and a preset prediction uncertainty as the standard deviation. By comparing this normal distribution with a preset functional failure threshold, the probability that the predicted margin value is less than the functional failure threshold is calculated, thereby transforming the single margin decay curve into a more statistically significant functional failure risk curve. This cognitive upgrade from deterministic prediction to probabilistic risk management enables decision-makers to formulate maintenance plans based on acceptable risk levels.

[0035] 4. This system transforms quantified risk information into specific, executable operational instructions, completing a closed loop from analysis to decision-making. When the risk of functional failure exceeds a preset warning threshold, the system can intelligently match from a preset decision rule base and automatically generate intelligent maintenance prescriptions or process optimization suggestions as operational instructions. This design solidifies the knowledge of domain experts into the system's capabilities, greatly reducing the threshold for users to interpret analysis results, achieving a leap from risk warning to intelligent decision support, and significantly improving the efficiency, initiative, and intelligence level of equipment management. Attached Figure Description

[0036] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0037] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0039] Example 1

[0040] Please see Figure 1 This invention provides a screw vacuum pump operation data analysis system based on big data, comprising:

[0041] The data acquisition unit is used to integrate real-time high-frequency data, process parameter data, and equipment static information;

[0042] The hidden state estimation unit is used to estimate the hidden state variables characterizing the degradation of the material properties of the pump internal components based on the real-time high-frequency data collected by the data acquisition unit, and to generate the equivalent deposition layer thickness.

[0043] The dynamic margin decay prediction unit is used to predict the decay trajectory of the core performance margin based on the equivalent deposition layer thickness generated by the hidden state estimation unit and combined with the process parameter data collected by the data acquisition unit, and to generate a dynamic shock resistance margin prediction curve.

[0044] The functional failure risk quantification unit is used to quantify the failure risk probability and generate a functional failure risk curve based on the dynamic shock resistance margin prediction curve generated by the dynamic margin decay prediction unit.

[0045] The decision support output unit is used to trigger an early warning and generate an operation command when the failure risk value at any time point in the functional failure risk curve generated by the functional failure risk quantification unit exceeds a preset warning threshold, based on the functional failure risk curve generated by the functional failure risk quantification unit.

[0046] This embodiment provides a screw vacuum pump operation data analysis system based on big data, which aims to establish a digital twin model that integrates physical mechanisms and real-time operating conditions. By estimating the state of the performance degradation process of internal components that cannot be directly measured, it can predict the decay trajectory of core performance margin and achieve quantitative early warning of future functional failure risks.

[0047] The system consists of five core units forming a complete technological closed loop from data acquisition to decision support;

[0048] The data acquisition unit aims to integrate all the information required for system analysis, providing comprehensive and accurate input for subsequent analysis, estimation, and prediction modules. In this embodiment, the unit is configured to acquire information from multiple heterogeneous data sources. Real-time high-frequency data refers to physical quantities that reflect the instantaneous operating state of the pump and are continuously acquired through a sensor network. Its function is to capture the dynamic details of equipment operation. Sources include vibration sensors directly mounted on the pump body, temperature sensors at key bearing locations, current sensors on the drive motor, spindle speed sensors, and pressure sensors on inlet and outlet pipelines. Process parameter data refers to information related to production tasks acquired from the Manufacturing Execution System (MES). Its function is to provide the system with a basis for predicting future loads. Sources include production plans and process recipes in the MES system, covering gas types, flow rates, pressure setpoints, etc. Equipment static information refers to a structured database that provides basic parameters for establishing physical models. Its source is equipment archives, which record pump model specifications, screw and bearing material properties, surface coating parameters, and historical maintenance and fault records.

[0049] The hidden state estimation unit aims to infer key hidden state variables that cannot be directly measured by sensors and characterize the degradation of material properties of pump internal components. In this embodiment, the unit estimates the hidden state variables by comparing the theoretical output of the physical model with the actual measurement values ​​of the sensors based on real-time high-frequency data provided by the data acquisition unit and analyzing the resulting residuals. The core output of this unit is the quantified equivalent deposit layer thickness. The equivalent deposit layer thickness refers to a comprehensive hidden state variable, which quantifies the macroscopic performance degradation caused by various microscopic physical degradation processes such as screw surface deposit adhesion or bearing lubrication deterioration. The source is the calculation result of this unit.

[0050] The purpose of the dynamic margin decay prediction unit is to extrapolate the decay trajectory of the pump's core performance margin over time based on the current equipment health status and future production tasks. In this embodiment, the unit uses the equivalent deposition layer thickness at the current moment output by the hidden state estimation unit as the initial condition, and combines it with future process parameter data provided by the data acquisition unit to predict the evolution of the core performance margin in the future by solving a dynamic model describing the degradation process. The output of this unit is a dynamic shock resistance margin prediction curve. The dynamic shock resistance margin refers to the core performance index, which characterizes the pump's ability to withstand additional process loads in the current state without functional failures such as jamming or vacuum breakage. Its source is the calculation result of this unit.

[0051] The purpose of the functional failure risk quantification unit is to transform the technical margin prediction curve into an intuitive failure risk probability that can be used for decision-making. In this embodiment, the unit takes into account the uncertainty of the prediction model itself, and compares the dynamic shock resistance margin prediction curve generated by the dynamic margin decay prediction unit with the preset functional failure threshold to quantify the risk of functional failure at any future time point. The output of the unit is a functional failure risk curve. Functional failure risk refers to the probability that the predicted margin value is less than the functional failure threshold. Its function is to provide direct quantitative basis for risk warning and maintenance decision-making, and its source is the calculation result of this unit.

[0052] The decision support output unit aims to transform quantified risk information into specific, executable operation instructions, completing the closed loop from prediction to decision. In this embodiment, the unit continuously monitors the risk curve generated by the functional failure risk quantification unit. When the failure risk value at any point in time on the curve exceeds the preset warning threshold, the system will automatically trigger an early warning and match and generate corresponding operation instructions from the built-in decision rule base, such as intelligent maintenance prescriptions or process optimization suggestions.

[0053] This embodiment constructs an analysis system capable of penetrating apparent data and gaining insight into internal physical degradation processes through the collaborative work of the aforementioned units. It overcomes the limitations of existing technologies that rely solely on apparent physical quantities for hysteresis threshold monitoring. By estimating hidden states and predicting future margins, it captures the critical turning point of the screw vacuum pump from quantitative to qualitative change, thereby providing forward-looking functional failure risk warnings and significantly improving the reliability of equipment operation and the proactivity of maintenance decisions.

[0054] Example 2

[0055] The hidden state estimation unit is specifically used for:

[0056] Based on the rotational speed and inlet / outlet pressure data contained in the real-time high-frequency data, and combined with the preset pumping mechanism model, the reference power consumption is calculated.

[0057] Obtain the actual measured motor power consumption contained in real-time high-frequency data;

[0058] The actual measured motor power consumption is compared with the reference power consumption to generate a power consumption residual.

[0059] The hidden state estimation unit is also used for:

[0060] The power consumption residual is used as input, and a first-order linear filter in a preset discrete-time form is used to correct the equivalent deposition layer thickness at the previous time step in order to generate the equivalent deposition layer thickness at the current time step.

[0061] This embodiment is a specific implementation of the hidden state estimation unit in Embodiment 1, which aims to accurately invert the hidden state variables that characterize the health status of internal components by analyzing the power consumption deviation.

[0062] The computational logic of this hidden state estimation unit is as follows:

[0063] Calculate the reference power consumption and power consumption residual; the reference power consumption refers to the theoretical power consumption value of the pump under ideal conditions, and its function is to establish a performance benchmark under non-degradation conditions. It is derived from the calculation results based on the preset pumping mechanism model; the preset pumping mechanism model refers to a mathematical model based on the principles of fluid mechanics and thermodynamics, which describes the relationship between the power consumption of the pump and various operating parameters. It can be constructed by those skilled in the art based on the pump design drawings and well-known physical laws.

[0064] This model is a power consumption model for a screw vacuum pump. It is established by solving a set of fluid dynamics and thermodynamic equations describing the compression process of gas in the screw profile, based on operating parameters such as the type of pumped gas, rotational speed, inlet and outlet pressures, and temperature. The equations include the gas state equation, energy conservation equation, and mass conservation equation, used to calculate theoretical compression power consumption and leakage power consumption, thus obtaining the total power consumption. Using real-time high-frequency data from the data acquisition unit, including rotational speed and inlet / outlet pressure data, as input, and combining this with gas properties determined by the process formulation, a reference power consumption is calculated. The actual measured motor power consumption from the data acquisition unit is also obtained. Based on these two factors, a power consumption residual is generated. The calculation method is as follows: The actual measured motor power consumption... Compared with reference power consumption Compare them and calculate the difference between them, that is... This difference serves as a key indicator, quantifying the additional energy consumption caused by hidden physical degradation processes not considered in the ideal model.

[0065] The equivalent deposition layer thickness is corrected using power consumption residuals. This correction process employs a pre-defined discrete-time first-order linear filter. This filter borrows the update concept from Kalman filtering, where the state update algorithm corrects the state estimate based on the residual between the measured value and the model prediction. The technical motivation lies in the fact that by monitoring the deviation between easily measurable macroscopic quantities, namely power consumption, and the ideal model, the microscopic physical changes that cause this deviation, which cannot be directly measured, can be indirectly inferred. The filter uses the currently calculated power consumption residual as input to correct the previous time-series estimate of the equivalent deposition layer thickness, thereby generating the current time-series equivalent deposition layer thickness. The calculation logic is as follows:

[0066]

[0067] in, The estimated equivalent deposition layer thickness at the current moment, whose data type is length, is calculated in this step; The estimated equivalent deposition layer thickness at the previous moment, with data type length, is derived from the output of the previous calculation cycle in this unit; State update gain coefficient, which physically represents the weight of adjusting the power consumption residual on the state estimation correction magnitude, has the dimension of length / power, and is determined by adaptive optimization through backtesting of historical data with the goal of minimizing long-term prediction error;

[0068] Actual historical trajectory It can be obtained by periodically disassembling and inspecting the screw vacuum pump in a controlled experimental environment to measure physical parameters such as the thickness of deposits on the screw surface or the amount of bearing wear; or indirectly through high-precision non-destructive testing techniques, such as ultrasonic testing or X-ray tomography.

[0069] The power consumption residual at the current moment; Physical coefficients; It is not only possible to obtain this information through CFD simulation, but also through regression analysis of historical data;

[0070] State update gain coefficient The optimal value can be determined using the gradient descent method. The specific steps are as follows:

[0071] initialization value;

[0072] Using historical data sequences, through the formula in Example 2 The historical estimated trajectory of the equivalent sedimentary layer thickness was calculated through simulation. ;

[0073] Calculate this trajectory and compare it with actual historical trajectories obtained through offline experiments or advanced analysis. Long-term prediction errors, such as root mean square error:

[0074]

[0075] in, : Root mean square error; N: Total number of historical data points; Historical estimated trajectory of equivalent sedimentary layer thickness calculated using the model; Actual historical trajectories obtained through offline experiments or advanced analysis;

[0076] Through iterative adjustments The value of makes Minimize until convergence;

[0077] The power consumption residual at the current moment, whose data type is power, is calculated from the previous steps; The physical coefficient describes the additional power consumption per unit thickness of the deposited layer, with the dimension of power / length. It is obtained through offline calibration. The calibration process involves establishing a series of independent calibration conditions in a computational fluid dynamics (CFD) simulation environment, and manually setting different constant deposited layer thicknesses in each condition j. And calculate the corresponding additional power consumption. Thus, a set of data points is obtained. Finally, by performing linear regression fitting on this set of calibration data points, the determination was made. The value;

[0078] In CFD simulation, the deposition layer thickness can be set. The range is from 0.1mm to 1.0mm, with 10 operating conditions set in 0.1mm increments. In each operating condition, the gap between the screw and the pump body is reduced by the corresponding thickness, while keeping parameters such as inlet pressure and rotational speed constant. The flow field is solved, and the additional power consumption is calculated. These 10 sets of data points were fitted using linear regression. Determine The value of , i.e., the slope of the regression line;

[0079] Compared to Example 1, this example provides a specific and operable hidden state estimation method. By introducing a physical mechanism-based reference power consumption model and power consumption residual analysis, the system can establish an accurate quantitative correlation between macroscopic and easily measurable motor power consumption data and microscopic and unmeasurable internal component degradation states. Furthermore, a first-order linear filter is used for state updates, utilizing the idea of ​​Kalman filtering, making the hidden state estimation process dynamic, adaptive, and closed-loop corrected, thereby greatly improving the accuracy and robustness of hidden state estimation and laying a solid foundation for subsequent accurate prediction.

[0080] Example 3

[0081] The dynamic margin decay prediction unit is specifically used for:

[0082] Obtain future process parameters contained in the process parameter data;

[0083] The future process parameters are weighted and normalized to calculate the normalized process load intensity; the normalization process uses the maximum value of each process parameter in the historical process as the normalization benchmark.

[0084] The dynamic margin decay prediction unit is also used for:

[0085] Based on the equivalent sedimentary layer thickness generated by the hidden state estimation unit and combined with the normalized process load intensity, the evolution trajectory of the equivalent sedimentary layer thickness in the future is predicted and generated by solving the preset first-order dynamic model.

[0086] The dynamic margin decay prediction unit is also used for:

[0087] The evolution trajectory of the equivalent deposition layer thickness in the future and the normalized process load intensity are substituted into the preset margin decay model to calculate and generate a dynamic shock resistance margin prediction curve.

[0088] This embodiment is a specific implementation of the dynamic margin decay prediction unit of Embodiment 1. It aims to transform the current health status and future process plan into an accurate dynamic shock margin prediction curve through a multi-step prediction engine.

[0089] The operational logic of this dynamic margin decay prediction unit is as follows:

[0090] Quantifying future process loads: To transform the discrete, multi-dimensional future process parameters obtained from the data acquisition unit into a continuous, single load function, the future process parameters need to be weighted and normalized to calculate the normalized process load intensity. The normalized process load intensity is a dimensionless comprehensive index used to quantify the combined impact of different future processes on the pump degradation rate; it originates from the calculation results of this step. Its calculation method is as follows:

[0091]

[0092] in, Future Moments The normalized process load strength, with data type dimensionless floating-point, is calculated by this formula; : The weight coefficient of the i-th process parameter, which is a dimensionless floating-point number with a total of 1. It is derived from the corrosion and wear model in the domain knowledge base and is preset after determining the degree of influence of each parameter on equipment degradation through historical data analysis; N: The number of key process parameters involved in the calculation; The i-th process parameter at a future time The planned value; The maximum value of the i-th process parameter across all historical processes; Future moments represent a point in time or a period of time.

[0093] Weighting coefficient The determination can be accomplished through multiple linear regression analysis. Historical data on pump degradation rates, such as the increment of equivalent deposit thickness per unit time, is collected as the dependent variable, while various process parameters, such as gas type, flow rate, and pressure setpoint, are used as independent variables. Then, a regression model is established using the historical data, with the following form:

[0094]

[0095] in, Represents the rate of degradation. Represents the i-th process parameter. These are the regression coefficients; Error term; by performing regression analysis on historical data, the regression coefficient of each parameter can be obtained. Normalized weighting coefficients Can be defined as In all The proportion in the total:

[0096]

[0097] in, Normalized weighting coefficients; : Regression coefficient of the i-th process parameter; N: Total number of parameters; : Represents absolute value;

[0098] The i-th process parameter at a future time The planned value, whose data type is related to specific parameters, comes from the future production plan provided by the manufacturing execution system; The maximum value of the i-th process parameter across all historical processes serves as a normalization benchmark; its data type is similar to... The similarity stems from statistical analysis of historical process databases; N: the number of key process parameters involved in the calculation, with data type integer;

[0099] Predicting the future evolution of the hidden state: Based on the equivalent sedimentary layer thickness generated by the current hidden state estimation unit as the initial condition, and combined with the normalized process load intensity calculated in the previous step, the evolution trajectory of the equivalent sedimentary layer thickness in the future is predicted and generated by solving a preset first-order dynamic model. This preset first-order dynamic model is a differential equation borrowed from the theory of chemical reaction kinetics. This model is similar to a first-order decay model or a first-order kinetic model. The technical motivation is that it assumes that the growth rate of the sedimentary layer is proportional to the process load intensity, and that there exists a natural clearing effect proportional to the current sedimentary layer thickness. The model is as follows:

[0100]

[0101] in, Future Moments The equivalent deposition layer thickness, with data type being length, is the dependent variable to be solved in this equation; : Deposition generation rate coefficient, which physically represents the rate of deposition layer growth caused by a unit process load intensity, and its dimension is length / time; : Natural removal rate coefficient, which physically describes the natural stripping or erosion effect of the sediment layer, and has the dimension of 1 / time; and The data is obtained through fitting and calibration of a historical dataset, which contains a series of hidden state estimates at different historical moments k. Corresponding normalized process load strength By substituting these historical data points into the differential equation model for parameter identification, the parameters can be determined. and The value; Future moments; : Future process load strength; This indicates the rate of change of the equivalent deposition layer thickness;

[0102] and The calibration can also be accomplished through similar CFD simulations or controlled laboratory experiments to maintain consistency with... Consistency of calibration methods;

[0103] parameter and Fit calibration can be achieved by solving a nonlinear least squares problem. Given a set of historical data points... Numerical methods, such as the Levenberg-Marquardt algorithm, can be used to minimize the following objective function:

[0104]

[0105] in, It utilizes a differential equation model with parameters of , The prediction of the hidden state estimate at historical time k, where M is the number of historical data points; The hidden state estimate at historical time k;

[0106] The future process load strength is calculated from the previous step;

[0107] Calculate the dynamic shock resistance margin; and determine the future evolution trajectory of the equivalent sedimentary layer thickness predicted in the previous step. And the normalized process load strength calculated in the first step. The data is then substituted into a pre-defined margin decay model to calculate and generate the final dynamic impact resistance margin prediction curve. This pre-defined margin decay model draws on the proportional risk model in reliability engineering and the theory of cumulative damage from material fatigue. This model can be called the exponential decay margin model, and its technical motivation is to use an exponential decay form to describe the process of margin decreasing with time and damage accumulation. The model is as follows:

[0108]

[0109] in, The dynamic shock margin at future time t, whose physical dimensions are related to the specific application scenario, is calculated by this formula; The initial margin of the pump in a brand new condition is provided by the equipment manufacturer according to the design specifications; Damage coupling coefficient, physically representing the accelerated effect of the combined action of hidden state and external load on margin decay, has the dimension of 1 / (length·time) and is calibrated by statistical regression analysis on a large amount of historical running datasets; : Calculate the start time; t: future time; : Future hidden state trajectory; Future load intensity trajectory; : The basic time decay coefficient, whose dimension is 1 / time; : The start time for calculation, usually the current time; t: a future time, where ;

[0110] It is a constant greater than 1 to better simulate the nonlinear accelerated decay effect;

[0111] It should be noted that the margin prediction curve itself also contains a certain degree of prediction uncertainty, which will be modeled and processed in the subsequent functional failure risk quantification unit.

[0112] Damage coupling coefficient Its calibration can be accomplished using the following multivariate nonlinear regression model:

[0113]

[0114] in, : The dynamic shock margin at future time t; : Initial margin of the pump in brand new condition; : Calculate the start time; t: future time; Damage coupling coefficient; : Future hidden state trajectory; Future load intensity trajectory; : Base time decay coefficient;

[0115] By performing regression analysis on historical margin evolution records, hidden state trajectories, and load intensity sequences, the solution is obtained. value;

[0116] This dataset contains historical margin evolution records, corresponding hidden state trajectories, and load intensity sequences; The basic time decay coefficient describes the basic decay rate caused by factors such as natural aging of materials, which is independent of the load. Its dimension is 1 / time, and its source is aging test data provided by the material supplier.

[0117] Basic time decay coefficient The aging test data provided by the supplier typically includes a performance degradation curve of the material under constant environmental conditions, which can be fitted in an exponential form.

[0118]

[0119] in This is the attenuation coefficient obtained through fitting; Performance at time t; Initial performance; t: time;

[0120] and These are the future hidden state trajectory and load intensity trajectory predicted and calculated in the preceding steps, respectively. : Calculate the start time;

[0121] This embodiment details the complete prediction chain from future process to future margin. First, by weighted normalization of multidimensional process parameters, effective dimensionality reduction and quantification of complex process loads are achieved, enhancing the model's universality. A first-order dynamic model is introduced to describe the evolution of hidden states, enabling predictions not only based on the current state but also reflecting the dynamic impact of future loads, greatly improving prediction accuracy. A margin decay model based on proportional risk theory is employed to scientifically describe the coupling and cumulative effects of base time aging, hidden state deterioration, and external process loads on performance margin consumption, making the generated margin prediction curve more consistent with physical reality and providing a high-confidence input for accurate risk quantification.

[0122] Example 4

[0123] The functional failure risk quantification unit is specifically used for:

[0124] The predicted margin value in the dynamic shock margin prediction curve is modeled as a normal distribution with the predicted margin value as the mean and the preset prediction uncertainty as the standard deviation.

[0125] Compare the normal distribution with the preset functional failure threshold;

[0126] Calculate the probability that the prediction margin value is less than the functional failure threshold, and determine this probability as the functional failure risk to generate a functional failure risk curve;

[0127] This embodiment is a specific implementation of the functional failure risk quantification unit of Embodiment 1, which aims to transform the predicted margin value and its uncertainty into an intuitive and statistically significant failure risk probability.

[0128] The computational logic of this functional failure risk quantification unit is as follows:

[0129] The predicted margin value is probabilistically modeled. Considering the inherent uncertainty of any prediction model, this unit models the predicted margin value at each future time point in the dynamic shock margin prediction curve as a normal distribution. The mean of this normal distribution is the predicted margin value directly output by the dynamic margin decay prediction unit, and the standard deviation is the preset prediction uncertainty. The preset prediction uncertainty refers to the statistical index that quantifies the error of the prediction model. Its function is to incorporate the uncertainty of the prediction into the risk assessment to make the results more robust. It is modeled and calibrated by analyzing the deviation sequence between historical prediction data and actual occurrences. Moreover, this uncertainty is set as a function that monotonically increases with the increase of the prediction time span.

[0130] Standard deviation of forecast uncertainty It can be modeled as a function that varies with the prediction time span t, for example, using a linear model;

[0131]

[0132] Or power-law model:

[0133]

[0134] in, : Standard deviation of prediction uncertainty, t: prediction time span, parameters or The model parameters can be determined by performing regression analysis on the historical prediction error sequence. The method is as follows: for each prediction task in history, record its prediction error at different time spans, and use the absolute value of these errors as the dependent variable and the prediction time span as the independent variable to perform regression fitting, thereby determining the model parameters.

[0135] Calculate the functional failure risk; compare the normal distribution with a preset functional failure threshold to generate a functional failure risk curve; the preset functional failure threshold refers to the minimum margin required to maintain the stability of a specific process, defining the boundary of functional failure. It originates from key parameters explicitly defined in production process specifications or equipment operation safety procedures and has the same physical dimensions as the margin; this calculation process utilizes the cumulative distribution function (CDF) of the standard normal distribution, as follows:

[0136]

[0137] in, The risk of functional failure at future time t, a probability value between 0 and 1, is calculated using this formula. The cumulative distribution function of the standard normal distribution; Functional failure threshold, whose data type and unit are the same as the margin, is preset according to the process specification; The mean prediction margin at time t is output by the dynamic margin decay prediction unit. The standard deviation of the prediction uncertainty at time t is preset according to the aforementioned historical error analysis model;

[0138] This embodiment addresses the limitation of traditional forecasting systems that only provide a single deterministic forecast value by introducing probabilistic methods. It models the forecast margin as a normal distribution and uses forecast uncertainty as its standard deviation, so that risk assessment is no longer a simple threshold comparison, but a probabilistic calculation that considers the range of fluctuations in the forecast value. The functional failure risk curve output by this approach is more valuable for decision-making than a single margin decay curve because it directly quantifies the probability of failure, enabling decision-makers to formulate maintenance plans based on acceptable risk levels. This achieves a cognitive upgrade from deterministic forecasting to probabilistic risk management, making decision-making more scientific and robust.

[0139] The decision support output unit is specifically used for:

[0140] Compare the risk of functional failure with a preset warning threshold;

[0141] When the risk of functional failure exceeds the preset warning threshold, the system matches and generates intelligent maintenance prescriptions or process optimization suggestions as operation instructions from the preset decision rule base.

[0142] This embodiment is a specific implementation of the decision support output unit of Embodiment 1. Its purpose is to transform the abstract risk curve into executable and intelligent operation instructions, thereby opening up the application loop of the analysis system.

[0143] The decision supports the following operational logic for the output unit:

[0144] The calculated functional failure risk is continuously compared with a preset warning threshold. The preset warning threshold is a risk probability value used to trigger an early warning. Its function is to define the risk level that requires manual intervention. It is derived from the company's risk management strategy and historical data, and is optimized by balancing the statistical expectation of false alarm costs and missed alarm losses.

[0145] The optimal setting for the warning threshold can be calculated using the following formula:

[0146] Minimize expected total loss = False positive probability * False positive loss + False negative probability * False negative loss

[0147]

[0148] in, It is the warning threshold. The costs associated with false alarms include unnecessary downtime and inspection expenses. The loss is due to underreporting, such as production losses, equipment damage, or safety accidents caused by unexpected downtime. By statistically analyzing historical data, the probabilities of false alarms and underreporting at different thresholds can be obtained, thereby finding the optimal threshold that minimizes the expected total loss. False alarm probability; : Probability of missed reports;

[0149] The actual failure and non-failure events under different risk thresholds are recorded to construct a confusion matrix, and then the false alarm rate, false positive rate and false negative rate are calculated. The false negative rate is used as the probability input in the formula.

[0150] The optimization algorithm is either grid search or gradient descent;

[0151] When the value of the functional failure risk curve exceeds the preset warning threshold at some point in the future, the system will automatically trigger an early warning.

[0152] After an alert is triggered, the system will intelligently match the alert level, the time window before the risk occurs, and the current production plan from a preset decision rule base and generate corresponding operation instructions. The preset decision rule base is a knowledge base containing IF-THEN logical rules. Its technical motivation is to solidify the knowledge of domain experts and historical successful maintenance cases into logic that can be automatically executed by computers. Its source is jointly built and maintained by equipment maintenance experts, process engineers, and data analysts.

[0153] The construction of a decision rule base can combine heuristic methods based on expert knowledge with machine learning methods based on historical data;

[0154] Expert knowledge: Conduct expert interviews and translate their experience in handling different risk situations into IF-THEN rules. If the risk is high and time is tight, an emergency shutdown should be recommended.

[0155] Historical data: Collect historical early warning records and corresponding actual maintenance operations, use risk level, time window, process type, etc. as features, and maintenance instructions as labels, train classification models such as decision trees or random forests, and automatically learn from the data to generate new rules or optimize existing rules;

[0156] For example:

[0157] Rule 1: If the risk is greater than 5% and the expected occurrence time is greater than 1 month, then generate a smart maintenance prescription that recommends an internal inspection of vacuum pump XX at the next planned shutdown.

[0158] Rule 2: If the risk is greater than 20% and the expected occurrence time is less than 1 week and involves a critical process, then it is recommended to immediately adjust the production plan, reduce the process load on the pump, and arrange process optimization suggestions and high-priority maintenance work orders for emergency repairs.

[0159] The output operation instructions are presented to the user in the form of intelligent maintenance prescriptions or process optimization suggestions;

[0160] This embodiment endows the system with the ability to make automatic decisions, transforming complex analysis results into clear, concise, and highly targeted operational instructions. By introducing flexibly configurable warning thresholds and a structured decision rule base, the early warning and response mechanisms are no longer rigid, but can provide differentiated and intelligent solutions based on the urgency and importance of risks. This greatly lowers the threshold for users to interpret analysis results, transforms the tacit knowledge of experts into the explicit capabilities of the system, and achieves a leap from risk warning to intelligent decision support, significantly improving the efficiency and intelligence level of equipment management.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A screw vacuum pump operation data analysis system based on big data, characterized in that, include: The data acquisition unit is used to integrate real-time high-frequency data, process parameter data, and equipment static information; The hidden state estimation unit is used to estimate the hidden state variables characterizing the degradation of the material properties of the pump internal components based on the real-time high-frequency data collected by the data acquisition unit, and to generate the equivalent deposition layer thickness. The dynamic margin decay prediction unit is used to predict the decay trajectory of the core performance margin based on the equivalent deposition layer thickness generated by the hidden state estimation unit and combined with the process parameter data collected by the data acquisition unit, and to generate a dynamic shock resistance margin prediction curve. The functional failure risk quantification unit is used to quantify the failure risk probability and generate a functional failure risk curve based on the dynamic shock resistance margin prediction curve generated by the dynamic margin decay prediction unit. The decision support output unit is used to trigger an early warning and generate an operation command when the failure risk value at any time point in the functional failure risk curve generated by the functional failure risk quantification unit exceeds a preset warning threshold, based on the functional failure risk curve generated by the functional failure risk quantification unit. The hidden state estimation unit is specifically used for: Based on the rotational speed and inlet / outlet pressure data contained in the real-time high-frequency data, and combined with the preset pumping mechanism model, the reference power consumption is calculated. Obtain the actual measured motor power consumption contained in real-time high-frequency data; The actual measured motor power consumption is compared with the reference power consumption to generate a power consumption residual. The hidden state estimation unit is also used for: The power consumption residual is used as input, and a first-order linear filter in a preset discrete-time form is used to correct the equivalent deposition layer thickness at the previous time step in order to generate the equivalent deposition layer thickness at the current time step. The calculation logic for the equivalent deposition layer thickness is as follows: ； in, The estimated equivalent deposition layer thickness at the current moment, whose data type is length, is calculated in this step; The estimated equivalent deposition layer thickness at the previous moment, with data type length, is derived from the output of the previous calculation cycle in this unit; : State update gain coefficient; The power consumption residual at the current moment; Physical coefficients are obtained through CFD simulation or regression analysis of historical data. Historical estimated trajectories of equivalent sedimentary layer thickness compared with actual historical trajectories obtained through offline experiments or advanced analysis. The long-term prediction error between them is expressed as the root mean square error: ； in, Root mean square error; : The total number of historical data points; Historical estimated trajectory of equivalent sedimentary layer thickness calculated using the model; Actual historical trajectories obtained through offline experiments or advanced analysis; Through iterative adjustments The value of makes Minimize until convergence; The dynamic margin decay prediction unit is also used for: Based on the equivalent sedimentary layer thickness generated by the hidden state estimation unit and combined with the normalized process load intensity, the evolution trajectory of the equivalent sedimentary layer thickness in the future is predicted and generated by solving the preset first-order dynamic model. The calculation method is as follows: ； in, Future Moments The normalized process load strength, with data type dimensionless floating-point, is calculated by this formula; : The weighting coefficient of the i-th process parameter, which is a dimensionless floating-point number and has a total of 1; The number of key process parameters involved in the calculation; The i-th process parameter at a future time The planned value; The maximum value of the i-th process parameter across all historical processes; Future moments; The dynamic margin decay prediction unit is also used for: The equivalent deposition layer thickness and the normalized process load intensity are substituted into a preset margin decay model to calculate and generate a dynamic shock resistance margin prediction curve.

2. The screw vacuum pump operation data analysis system based on big data according to claim 1, characterized in that, The dynamic margin decay prediction unit is specifically used for: Obtain future process parameters contained in the process parameter data; The future process parameters are weighted and normalized to calculate the normalized process load intensity. The normalization process uses the maximum value of each process parameter in the historical process as the normalization benchmark.

3. The screw vacuum pump operation data analysis system based on big data according to claim 1, characterized in that, The functional failure risk quantification unit is specifically used for: The predicted margin value in the dynamic shock margin prediction curve is modeled as a normal distribution with the predicted margin value as the mean and the preset prediction uncertainty as the standard deviation. Compare the normal distribution with the preset functional failure threshold; Calculate the probability that the prediction margin value is less than the functional failure threshold, and determine this probability as the functional failure risk to generate a functional failure risk curve.

4. The screw vacuum pump operation data analysis system based on big data according to claim 1, characterized in that, The decision support output unit is specifically used for: Compare the risk of functional failure with a preset warning threshold; When the risk of functional failure exceeds the preset warning threshold, the system matches and generates intelligent maintenance prescriptions or process optimization suggestions as operation instructions from the preset decision rule base.

Images