Space division production line health scoring method and system based on multivariate heterogeneous model

Abstract

The invention relates to the technical field of air separation production line health scoring, and provides an air separation production line health scoring method based on a multivariate heterogeneous model, and the method comprises the steps: S1, collecting the state data, working condition data and process data of an air separation production line; s2, in order to prevent the influence of abnormal data on the model construction of the system, carrying out abnormal value preprocessing on the collected data; s3, based on the preprocessed data, performing multi-dimensional early warning through adaptive threshold early warning, trend early warning and AI early warning; s4, after parameters in the air separation production line equipment trigger early warning, fault diagnosis is carried out only for rotary equipment and only in the vibration parameter or temperature parameter early warning scene of the bearing of the equipment; and S5, respectively calculating scores of the threshold model, the trend model, the A model and the mechanism model, and calculating a total health degree score of the system through weight configuration. The health management system for the space division industry is provided, and the overall operation condition of a production line can be effectively evaluated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air separation production line health scoring, and particularly relates to an air separation production line health scoring method and system based on a multi-element heterogeneous model. BACKGROUND

[0002] Air separation equipment is mainly used for gas separation and purification processes in industrial production. They separate different gas components in air to obtain high-purity oxygen, nitrogen and other industrial gases. These high-purity gases play an important role in many industrial fields.

[0003] With the continuous development of industrial production, air separation production lines play a crucial role in industrial production. However, due to the complexity and multi-element heterogeneity of air separation production lines, the health management and maintenance of them pose challenges. In order to solve this problem, an air separation production line health management system based on multi-element heterogeneity is developed. The system aims to use advanced technologies and methods to comprehensively monitor and manage air separation production lines to ensure their efficient operation and long-term stability. Through the health management of air separation production lines, the downtime due to failures can be minimized, the production efficiency can be improved, and the maintenance cost can be reduced. The air separation production line health management system based on multi-element heterogeneity adopts advanced sensor technology, big data analysis and artificial intelligence algorithms, which can monitor and diagnose the running state of the production line equipment in real time, and predict potential failure risks. In addition, the system also provides remote monitoring and intelligent maintenance functions, so that the operating personnel can take timely measures to ensure the normal operation of the production line equipment. Therefore, the research and development of the air separation production line health management system based on multi-element heterogeneity have important practical significance and broad application prospects. Through the application of the system, the stability and reliability of industrial production can be improved, and the industrial production can be pushed towards the direction of intelligentization and automation.

[0004] The current health degree method of the equipment is based on big data and trains a prediction model, but in actual industrial field, it is difficult to obtain the whole life cycle data of the equipment, and the actual application cannot be carried out. SUMMARY

[0005] In view of the above problems, the purpose of the present application is to provide an air separation production line health scoring method and system based on a multi-element heterogeneous model, which provides a health management system for the air separation industry, can effectively evaluate the overall operation of the production line, the mechanism module can diagnose equipment failure to assist in troubleshooting, and the change of the production line parameter is quantified into a score to reflect the health change and guide the point inspection.

[0006] The above invention purpose of the present application is realized by the following technical scheme:

[0007] An air separation production line health scoring method based on a multi-element heterogeneous model, comprising the following steps:

[0008] S1: Collect status data, operating data and process data of the air separation production line;

[0009] S2: To prevent abnormal data from affecting the system model building, the collected data is preprocessed for abnormal values;

[0010] S3: Based on pre-processed data, multi-dimensional warnings are provided through adaptive threshold warnings, trend warnings, and AI warnings;

[0011] S4: When parameters in the air separation production line equipment trigger an early warning, fault diagnosis is performed only for rotating equipment and only in the scenario where the vibration parameters or temperature parameters at the equipment bearings are warned. Before diagnosis, equipment information must be entered in advance. Fault diagnosis is performed on the equipment based on the mechanism rule library pre-set by the platform.

[0012] S5: Calculate the scores of the threshold model, trend model, AI model, and mechanism model respectively, and calculate the overall health score of the system through weight configuration.

[0013] Furthermore, in step S1, the status data, operating condition data and process data of the air separation production line are collected, specifically:

[0014] The state data includes a vibration displacement signal and a vibration acceleration signal of the moving equipment;

[0015] The operating condition data includes equipment start and stop status data, full load operation status data, 80% load status data, speed data, valve opening data, etc.

[0016] The process data include data on temperature, pressure, flow rate, and pressure difference during production;

[0017] In the air separation production line, if the key units are equipped with a vibration monitoring and protection system, waveform data is collected from the front panel of the system. At the same time, the data acquisition module includes multiple industrial protocol connections including Modbus, Profibus, and OPC to collect data from the industrial air separation production line, including PLC and DCS.

[0018] Furthermore, in step S2, in order to prevent abnormal data from affecting the model construction of the system, the collected data is processed for outliers, specifically:

[0019] S21: Assume that the collected data set is X={x1,x2,…,x n}, where x i (i=1,2,…,n) represents the i-th data point in the collected data, n is the total number of collected data, and the root mean square (rms) of X is calculated using the formula:

[0020]

[0021] here, The root mean square of the data is obtained by summing the squares of each data point in the data set X, dividing it by n, and then taking the square root. It is used to measure the degree of discreteness of the data.

[0022] S22: Calculate X1=|X|, that is, for each data point x in the data set X, i Take the absolute value and get the new data set X1={|x1|,|x2|,…,|x n |}, the purpose is to unify the positive and negative characteristics of the data to facilitate subsequent outlier judgment;

[0023] S23: Setting the root mean square coefficient k, where the value of k is between 3 and 8. k is used to adjust the sensitivity of abnormal value judgment. For different air separation production line scenarios or data characteristics, an appropriate value can be selected in this range as needed.

[0024] S24: Count the number num of data in X1 that is greater than k×rms, that is, traverse each data|x in X1 i |, if |x i |>k×rms, then the count is increased by 1, and finally the number of data points that meet the conditions is obtained; and calculate P represents the proportion of abnormal data points in the data set X to the total data points;

[0025] S25: If P is greater than the preset threshold th1, it is judged as data abnormality. th1 is a proportional threshold preset according to standards including actual operation experience of air separation production line and data quality requirements, and is used to judge whether the data is abnormal. At the same time, the position of the data greater than k×rms in X1 is used to convert the data x corresponding to the position in X into i Replace with the mean of X This can correct abnormal data and ensure the data quality for subsequent model construction.

[0026] Furthermore, in step S3, based on the pre-processed data, multi-dimensional warnings are performed through adaptive threshold warnings, trend warnings, and AI warnings, specifically:

[0027] S31: Use adaptive threshold warning for a single parameter. The adaptive threshold is calculated as follows:

[0028] S311: Perform data screening to obtain data for a period of time under constant working conditions. The total number of data is j, and the data point is y1 (i = 1, 2, ... j). The middle data from the fifth to the ninety-fifth percentile are taken to form a set Y = {y1, y2, ..., y j};

[0029] S312: Perform statistical calculations and calculate the mean of Y in Sum the data points in Y and calculate the standard deviation of Y y j -y mean is the deviation of the data point from the mean;

[0030] S313: Set the threshold value, set the attention threshold value th note =y mean +3σ, warning threshold th alarm =y mean +6σ;

[0031] S314: Trigger the alarm, set the latest data as y, when th note ≤y <th alarm And for m consecutive times, the health management system issues an alarm, when y≥th alarm And it will sound an alarm m times in a row, where the number of times is set by yourself;

[0032] S32: Perform trend warning to determine whether the data continues to rise. Use a sliding window to calculate the data mean in segments. By determining the number of times the current value exceeds the preset threshold, determine whether the current data trend is rising. The calculation method is as follows:

[0033] S321: Get data and set the current value to z k+1 , from z k+1 Get the trend data of a single working condition Z=[z1,z2,…,z k ],z i (i=1,2,…,k) is the trend data point, k is the total number of data points;

[0034] S322: Set the window, the window data is Z, and the step size is 1;

[0035] S323: Perform threshold calculation and calculate the mean of window data Z And set two threshold lines

[0036] S324: Perform trend judgment and alarm, pre-define rising edge z line , when z k+1 ≥z line And z mean1 ≤z k+1 <z mean2 , record trend attention 1 time, when it is triggered a times continuously, the trend alarm is triggered, and the alarm level is attention; similarly, when z k+1 ≥z line , and z mean1 ≤z k+1 <z mean2, trend warning 1 time, when it is triggered a times continuously, it is a trend alarm, and the level is warning;

[0037] S325: Parameter optimization is performed, and the number of consecutive triggers a and the window length are optimized by the particle swarm optimization algorithm;

[0038] S33: Perform AI warning, set parameter A in the air separation unit as the important parameter, and B, C, D, E, and F are parameters related to A:

[0039] S331: Perform model input and prediction. Input the real-time data sequence A, B, C, D, E, and F into the trained LSTM model and predict the value of parameter A, which is recorded as the predicted value.

[0040] S332: Perform difference calculation and alarm, calculate the difference between the predicted value and the actual value of parameter A, dif = |predicted value - actual value|;

[0041] S333: Preset two thresholds th based on expert experience al , th ah If th al ≤dif≤th ah , the alarm level is attention, if dif≥tH ah , the alarm level is warning.

[0042] Furthermore, in step S4, when parameters in the air separation production line equipment trigger an early warning, fault diagnosis is performed only for rotating equipment and only in the early warning scenario of vibration parameters or temperature parameters at the bearings of the equipment. Before diagnosis, equipment information must be entered in advance. Fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform. Specifically:

[0043] S41: Collecting original waveform data. When an alarm occurs at a certain position of the device, obtain the original waveform data S corresponding to the sensor. S is the vibration or temperature time domain waveform data sequence collected by the sensor at the device alarm position.

[0044] S42: Extract characteristic frequencies. Using fast Fourier transform (FFT), combined with the pre-recorded rotational speed of the device, the signal's rotational frequency and its harmonics are extracted. The rotational frequency is the rotational frequency of the device's main shaft, and the harmonics are integer multiples of the rotational frequency. Similarly, for bearing faults, the characteristic frequency of the bearing fault is extracted. For impeller-type devices, the blade pass frequency is extracted. The bearing fault characteristic frequency is calculated based on the bearing model and structural parameters. The blade pass frequency is related to the number of impeller blades and the rotational speed.

[0045] S43: Perform rule matching to determine whether the extracted features meet the pre-set rules in the platform's preset mechanism rule library. The rules are formulated based on the differences in features between normal operation and failure of the device.

[0046] S44: Determine the alarm level. Two levels of values ​​are pre-set in the rules. l , v h , v l <v h , v l is the low-level alarm threshold, v h is the high-level alarm threshold, and the value currently calculated according to the rule is v alue , v alue is the parameter value obtained by calculating the characteristic frequency for determining the fault level. alue Location l With v h The alarm level is determined based on the location.

[0047] Furthermore, in step S5, the scores of the threshold model, trend model, AI model and mechanism model are calculated respectively, specifically:

[0048] S51: Calculation of health of threshold model:

[0049] S511: The attention threshold th has been calculated according to the adaptive threshold part note , warning threshold th alarm , using the same principle to calculate the optimal interval (t l ,t r ), t l is the ideal minimum value, t r For the ideal maximum health management system, the scores corresponding to each threshold are reserved at the same time, and the scores and thresholds are combined to form 4 points. The left boundary point of the optimal interval (t l ,S l ), the left boundary point of the optimal interval (t r ,S r ), attention threshold corresponding to the health point (th note ,S note ), the health point corresponding to the warning threshold (th alarm ,S alarm ), where S l For parameters at t l The corresponding basic health score, S r For parameters at t r The corresponding basic health score, S note For parameters in th note The corresponding basic health score, S alarm For parameters in th alarm The corresponding basic health score;

[0050] S512: Construct a score piecewise function and a weight piecewise function based on the four points in step S511:

[0051]

[0052] S513: Calculate S based on the real-time data x of a specific operating parameter in the air separation production line t , w t ;

[0053] S514: Calculate the threshold model scores for all segmented health intervals t represents the tth health interval;

[0054] S52: For the trend model, the score and weight calculation method is:

[0055] S521: Similar to the threshold model, three points are constructed: trend health benchmark point (tt1, ss1), trend warning point (z mean1 ,ss2), trend warning point (z mean2 ,ss3), where tt1 <z mean1 <z mean2 ;ss3 <ss2<ss1,其中tt1为设备参数无异常趋势时的基准值,对应健康度最高分ss1,z mean1 is the threshold value of mild abnormal trend in trend warning, corresponding to the medium health score ss2, z mean2 It is the threshold value of severe abnormal trend in trend warning, corresponding to the lowest health score ss3;

[0056]

[0057] S522: For all trend model scores S core2 , z is the segment index of the trend model:

[0058]

[0059] S53: For the AI ​​model, the score of the entire AI model is calculated as:

[0060] S531: For AI models, each model score has two levels

[0061]

[0062] S532: In the AI ​​module, the score of the entire AI model:

[0063]

[0064] Among them, s aq is the score of a single sub-model in the AI ​​model, s ah When the prediction deviation dif is in the attention threshold interval th al≤dif≤th ah When the health score of a single AI sub-model is s al When the prediction deviation dif is in the warning threshold interval dif≥th ah When the health score of a single AI sub-model is al is the attention threshold, th ah is the warning threshold, q is the number of AI sub-models;

[0065] S54: For the mechanism module, the score calculation method for all mechanism models is the same as that for AI models:

[0066] S541: Calculation score of a single mechanism model

[0067]

[0068] S542: Score of all mechanism models in the mechanism module

[0069]

[0070] Among them, s mr Calculate the score for a single mechanistic model, s mh The characteristic parameter v for the device operation alue In the health threshold range v l ≤v alue ≤v h When the health score of a single mechanism sub-model is s ml When the device is running, the characteristic parameter v alue In the failure risk threshold interval v alue ≥v h When the health score of a single mechanism sub-model is l is the lower limit of the health threshold used for judgment in the mechanism model, v h is the fault risk threshold used for judgment in the mechanism model, and r is the number of mechanism models.

[0071] Furthermore, in step S5, the total health score of the system is calculated by weight configuration, specifically:

[0072] There are configuration places in the health calculation module, which are used to configure the weights of the early warning model, trend model, AI model, and mechanism model. The weight configurations are α1, α2, α3, and α4 respectively; α1+α2+α3+α4=1;

[0073] The total system score in the health management system is:

[0074]

[0075] Among them, αu For each model weight, S coreu The score for each model.

[0076] A multivariate heterogeneous model-based air separation line health scoring system for executing the above-mentioned multivariate heterogeneous model-based air separation line health scoring method is characterized by comprising:

[0077] Data acquisition module, used to collect status data, operating data and process data of the air separation production line;

[0078] The data preprocessing module is used to preprocess the collected data for abnormal values ​​in order to prevent the influence of abnormal data on the system model construction;

[0079] The early warning module is used to provide multi-dimensional early warnings based on pre-processed data through adaptive threshold warnings, trend warnings, and AI warnings;

[0080] The fault diagnosis module is used to perform fault diagnosis on rotating equipment only when parameters in the air separation production line trigger an early warning, and only in the scenario where vibration parameters or temperature parameters at the bearings of the equipment are warned. Before diagnosis, equipment information must be entered in advance, and fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform;

[0081] The health calculation module is used to calculate the scores of the threshold model, trend model, AI model and mechanism model respectively, and calculate the total health score of the system through weight configuration.

[0082] A computer device includes a memory and one or more processors, wherein the memory stores computer code, and when the computer code is executed by the one or more processors, the one or more processors execute the above method.

[0083] A computer-readable storage medium stores computer code. When the computer code is executed, the above method is performed.

[0084] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0085] (1) Adapt to the health management system of the air separation industry and accurately evaluate the overall operation of the production line

[0086] Based on the fusion of multiple heterogeneous models (threshold + trend + AI + mechanism model), by collecting the full data of the air separation production line (status / operating condition / process data), pre-processing to eliminate abnormal interference, and then using multi-dimensional warnings (adaptive threshold to identify steady-state anomalies, trend warning to capture parameter degradation, AI warning to mine hidden risks), and finally through weighted fusion of the health of each model (covering the entire scenario of equipment start-up and shutdown, load fluctuations), the operating status of the entire process equipment of the air separation production line, such as "compressors, distillation towers, pumps and valves", is evaluated, and risks such as "parameter out-of-limit, trend degradation, and hidden faults" are accurately identified, so that the overall health of the production line is upgraded from "empirical judgment" to "data quantification" to support scientific decision-making.

[0087] (2) In-depth diagnosis of the mechanism module, enabling efficient inspection and troubleshooting

[0088] Relying on proprietary diagnostic logic for rotating equipment, when a bearing vibration / temperature warning is triggered, an FFT transform is used to analyze the original waveform. Combined with pre-set information such as the equipment's shaft speed and bearing model, this system accurately extracts mechanism characteristics such as rotational frequency harmonics, bearing fault characteristic frequencies (inner race / outer race / roller frequencies), and impeller blade pass frequency. Furthermore, using a mechanism rule library (matching fault frequency amplitudes and combination relationships), it directly locates fault types and locations such as misalignment, imbalance, bearing wear, and impeller corrosion. Without complex spectrum analysis, inspectors can quickly determine the "fault root cause, location, and level," improving troubleshooting efficiency by over 50% and avoiding the blindness of traditional empirical troubleshooting.

[0089] (3) Quantitative scoring of parameter changes to dynamically guide inspection strategies

[0090] Through piecewise function + weighted fusion, the fluctuations of production line parameters such as "vibration, temperature, pressure, and flow" are mapped to a health curve with a score of 0-100 (high score in the optimal range, low score in the warning range). Health not only reflects the current state (such as the threshold model identifies immediate anomalies), but also captures parameter degradation trends through trend models (such as a 1.5-fold increase in the mean triggers a trend reduction), and AI models predict risks (early warning of predicted deviations exceeding the limit), making production line health changes "visual and traceable". According to the health score and the rate of change, inspection personnel can dynamically adjust the inspection frequency (weekly inspections with high scores and low risks, spot inspections with low scores and high risks), focus on key parameters (prioritize the parameters with large health reductions), and realize the transformation of spot inspections from "regular blindness" to "accurate on demand". BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is the overall structure diagram of the air separation production line health scoring method based on the multivariate heterogeneous model of the present invention;

[0092] Figure 2 This is a structural diagram of the air separation production line health management system based on the multi-source heterogeneous model of the present invention;

[0093] Figure 3 This is a structural diagram of the data acquisition module of the present invention;

[0094] Figure 4 This is a structural diagram of the data preprocessing module of the present invention;

[0095] Figure 5 This is a flow chart of the fault diagnosis module of the present invention;

[0096] Figure 6 This is a module structure diagram of the health meter of the present invention. DETAILED DESCRIPTION

[0097] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0098] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0099] First embodiment

[0100] like Figure 1 As shown, this embodiment provides an air separation line health scoring method based on a multivariate heterogeneous model, including the following steps:

[0101] S1: Collect status data, operating data and process data of the air separation production line.

[0102] In this embodiment, step S1 is specifically as follows:

[0103] The state data includes a vibration displacement signal and a vibration acceleration signal of the moving equipment;

[0104] The operating condition data includes equipment start and stop status data, full load operation status data, 80% load status data, speed data, valve opening data, etc.

[0105] The process data include data on temperature, pressure, flow rate, and pressure difference during production;

[0106] In the air separation production line, if the key units are equipped with a vibration monitoring and protection system, waveform data is collected from the front panel of the system. At the same time, the data acquisition module includes multiple industrial protocol connections including Modbus, Profibus, and OPC to collect data from the industrial air separation production line, including PLC and DCS.

[0107] S2: To prevent abnormal data from affecting the system model building, the collected data is preprocessed for abnormal values.

[0108] In this embodiment, step S2 is specifically as follows:

[0109] S21: Assume that the collected data set is X={x1,x2,…,x n}, where x i (i=1,2,…,n) represents the i-th data point in the collected data, n is the total number of collected data, and the root mean square (rms) of X is calculated using the formula:

[0110]

[0111] here, The root mean square of the data is obtained by summing the squares of each data point in the data set X, dividing it by n, and then taking the square root. It is used to measure the degree of discreteness of the data.

[0112] S22: Calculate X1=|X|, that is, for each data point x in the data set X, i Take the absolute value and get the new data set X1={|x1|,|x2|,…,|x n |}, the purpose is to unify the positive and negative characteristics of the data to facilitate subsequent outlier judgment;

[0113] S23: Setting the root mean square coefficient k, where the value of k is between 3 and 8. k is used to adjust the sensitivity of abnormal value judgment. For different air separation production line scenarios or data characteristics, an appropriate value can be selected in this range as needed.

[0114] S24: Count the number num of data in X1 that is greater than k×rms, that is, traverse each data|x in X1 i |, if |x i |>k×rms, then the count is increased by 1, and finally the number of data points that meet the conditions is obtained; and calculate P represents the proportion of abnormal data points in the data set X to the total data points;

[0115] S25: If P is greater than the preset threshold th1, it is judged as data abnormality. th1 is a proportional threshold preset according to standards including actual operation experience of air separation production line and data quality requirements, and is used to judge whether the data is abnormal. At the same time, the position of the data greater than k×rms in X1 is used to convert the data x corresponding to the position in X into i Replace with the mean of X This can correct abnormal data and ensure the data quality for subsequent model construction.

[0116] S3: Based on the pre-processed data, multi-dimensional warnings are provided through adaptive threshold warnings, trend warnings, and AI warnings.

[0117] In this embodiment, step S3 is specifically as follows:

[0118] S31: Use adaptive threshold warning for a single parameter. The adaptive threshold is calculated as follows:

[0119] S311: Perform data screening to obtain data for a period of time under constant working conditions. The total number of data is j, and the data point is y1 (i = 1, 2, ... j). The middle data from the fifth to the ninety-fifth percentile are taken to form a set Y = {y1, y2, ..., y j};

[0120] S312: Perform statistical calculations and calculate the mean of Y in Sum the data points in Y and calculate the standard deviation of Y y j -y mean is the deviation of the data point from the mean;

[0121] S313: Set the threshold value, set the attention threshold value th note =y mean +3σ, warning threshold th alarm =y mean +6σ;

[0122] S314: Trigger the alarm, set the latest data as y, when th note ≤y <th alarm And for m consecutive times, the health management system issues an alarm, when y≥th alarm And it will sound an alarm m times in a row, where the number of times is set by yourself;

[0123] S32: Perform trend warning to determine whether the data continues to rise. Use a sliding window to calculate the data mean in segments. By determining the number of times the current value exceeds the preset threshold, determine whether the current data trend is rising. The calculation method is as follows:

[0124] S321: Get data and set the current value to zk+1 , from z k+1 Get the trend data of a single working condition Z=[z1,z2,…,z k ],z i (i=1,2,…,k) is the trend data point, k is the total number of data points;

[0125] S322: Set the window, the window data is Z, and the step size is 1;

[0126] S323: Perform threshold calculation and calculate the mean of window data Z And set two threshold lines

[0127] S324: Perform trend judgment and alarm, pre-define rising edge z line , when z k+1 ≥z line And z mean1 ≤z k+1 <z mean2 , record trend attention 1 time, when it is triggered a times continuously, the trend alarm is triggered, and the alarm level is attention; similarly, when z k+1 ≥z line , and z mean1 ≤z k+1 <z mean2 , trend warning 1 time, when it is triggered a times continuously, it is a trend alarm, and the level is warning;

[0128] S325: Parameter optimization is performed, and the number of consecutive triggers a and the window length are optimized by the particle swarm optimization algorithm;

[0129] S33: Perform AI warning, set parameter A in the air separation unit as the important parameter, and B, C, D, E, and F are parameters related to A:

[0130] S331: Perform model input and prediction. Input the real-time data sequence A, B, C, D, E, and F into the trained LSTM model and predict the value of parameter A, which is recorded as the predicted value.

[0131] S332: Perform difference calculation and alarm, calculate the difference between the predicted value and the actual value of parameter A, dif = |predicted value - actual value|;

[0132] S333: Preset two thresholds th based on expert experience al , th ah If th al ≤dif≤th ah , the alarm level is attention, if dif≥th ah , the alarm level is warning.

[0133] S4: When parameters in the air separation production line equipment trigger an early warning, fault diagnosis is performed only for rotating equipment and only in the vibration parameter or temperature parameter early warning scenario at the bearing of the equipment. Before diagnosis, the equipment information must be entered in advance, and fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform.

[0134] In this embodiment, step S4 is specifically as follows:

[0135] When a certain parameter of the equipment is warned, the diagnostic model in the diagnostic module is called and enabled. The diagnostic module is applied to rotating equipment and is called only when the vibration or temperature warning is issued at the bearing of the equipment. The diagnostic module needs to enter the equipment information in advance, such as the shaft speed and bearing model at the sensor location. At the same time, the platform has a pre-installed mechanism rule library.

[0136] S41: Collecting original waveform data. When an alarm occurs at a certain position of the device, obtain the original waveform data S corresponding to the sensor. S is the vibration or temperature time domain waveform data sequence collected by the sensor at the device alarm position.

[0137] S42: Extract characteristic frequencies. Using fast Fourier transform (FFT), combined with the pre-recorded rotational speed of the device, the signal's rotational frequency and its harmonics are extracted. The rotational frequency is the rotational frequency of the device's main shaft, and the harmonics are integer multiples of the rotational frequency. Similarly, for bearing faults, the characteristic frequency of the bearing fault is extracted. For impeller-type devices, the blade pass frequency is extracted. The bearing fault characteristic frequency is calculated based on the bearing model and structural parameters. The blade pass frequency is related to the number of impeller blades and the rotational speed.

[0138] S43: Perform rule matching to determine whether the extracted features meet the pre-set rules in the platform's preset mechanism rule library. The rules are formulated based on the differences in features between normal operation and failure of the device.

[0139] S44: Determine the alarm level. Two levels of values ​​are pre-set in the rules. l , v h , v l <v h , v l is the low-level alarm threshold, v h is the high-level alarm threshold, and the value currently calculated according to the rule is v alue , v alue is the parameter value obtained by calculating the characteristic frequency for determining the fault level. alue Location l With v h The alarm level is determined based on the location.

[0140] S5: Calculate the scores of the threshold model, trend model, AI model, and mechanism model respectively, and calculate the overall health score of the system through weight configuration.

[0141] In this embodiment, step S5 is specifically as follows:

[0142] First, the scores of the threshold model, trend model, AI model, and mechanism model are calculated respectively, as follows:

[0143] S51: Calculation of health of threshold model:

[0144] S511: The attention threshold th has been calculated according to the adaptive threshold part note , warning threshold th alarm , using the same principle to calculate the optimal interval (t l ,t r ), t l is the ideal minimum value, t r For the ideal maximum health management system, the scores corresponding to each threshold are reserved at the same time, and the scores and thresholds are combined to form 4 points. The left boundary point of the optimal interval (t l ,S l ), the left boundary point of the optimal interval (t r ,S r ), attention threshold corresponding to the health point (th note ,S note ), the health point corresponding to the warning threshold (th alarm ,S alarm ), where S l For parameters at t l The corresponding basic health score, S r For parameters at t r The corresponding basic health score, S note For parameters in th note The corresponding basic health score, S alarm For parameters in th alarm The corresponding basic health score;

[0145] S512: Construct a score piecewise function and a weight piecewise function based on the four points in step S511:

[0146]

[0147] S513: Calculate S based on the real-time data x of a specific operating parameter in the air separation production line t , w t ;

[0148] S514: Calculate the threshold model scores for all segmented health intervals t represents the tth health interval;

[0149] S52: For the trend model, the score and weight calculation method is:

[0150] S521: Similar to the threshold model, three points are constructed: trend health benchmark point (tt1, ss1), trend warning point (z mean1 ,ss2), trend warning point (z mean2 ,ss3), where tt1 <z mean1 <z mean2 ;ss3 <ss2<ss1,其中tt1为设备参数无异常趋势时的基准值,对应健康度最高分ss1,z mean1 is the threshold value of mild abnormal trend in trend warning, corresponding to the medium health score ss2, z mean2 It is the threshold value of severe abnormal trend in trend warning, corresponding to the lowest health score ss3;

[0151]

[0152] S522: For all trend model scores S core2 , z is the segment index of the trend model:

[0153]

[0154] S53: For the AI ​​model, the score of the entire AI model is calculated as:

[0155] S531: For AI models, each model score has two levels

[0156]

[0157] S532: In the AI ​​module, the score of the entire AI model:

[0158]

[0159] Among them, s aq is the score of a single sub-model in the AI ​​model, s ah When the prediction deviation dif is in the attention threshold interval th al ≤dif≤th ah When the health score of a single AI sub-model is s al When the prediction deviation dif is in the warning threshold interval dif≥th ah When the health score of a single AI sub-model is al is the attention threshold, th ah is the warning threshold, q is the number of AI sub-models;

[0160] S54: For the mechanism module, the score calculation method for all mechanism models is the same as that for AI models:

[0161] S541: Calculation score of a single mechanism model

[0162]

[0163] S542: Score of all mechanism models in the mechanism module

[0164]

[0165] Among them, s mr Calculate the score for a single mechanistic model, s mh The characteristic parameter v for the device operation alue In the health threshold range v l ≤v alue ≤v h When the health score of a single mechanism sub-model is s ml When the device is running, the characteristic parameter v alue In the failure risk threshold interval v alue ≥v h When the health score of a single mechanism sub-model is l is the lower limit of the health threshold used for judgment in the mechanism model, v h is the fault risk threshold used for judgment in the mechanism model, and r is the number of mechanism models.

[0166] Secondly, the overall health score of the system is calculated by weight configuration, specifically:

[0167] There are configuration places in the health calculation module, which are used to configure the weights of the early warning model, trend model, AI model, and mechanism model. The weight configurations are α1, α2, α3, and α4 respectively; α1+α2+α3+α4=1;

[0168] The total system score in the health management system is:

[0169]

[0170] Among them, α u For each model weight, S coreu The score for each model.

[0171] Second embodiment

[0172] like Figure 2 As shown, this embodiment provides an air separation line health scoring system based on a multivariate heterogeneous model for executing the air separation line health scoring method based on a multivariate heterogeneous model as in the first embodiment, including:

[0173] Data acquisition module 1, used to collect status data, operating data and process data of the air separation production line;

[0174] Data preprocessing module 2 is used to preprocess the collected data for abnormal values ​​in order to prevent the influence of abnormal data on the system model construction;

[0175] Early warning module 3 is used to provide multi-dimensional early warnings based on pre-processed data through adaptive threshold warnings, trend warnings, and AI warnings;

[0176] Fault Diagnosis Module 4 is used to perform fault diagnosis on rotating equipment only when parameters in the air separation production line trigger an early warning, and only in the scenario where vibration parameters or temperature parameters at the bearings of the equipment are warned. Before diagnosis, equipment information must be entered in advance, and fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform;

[0177] The health calculation module 5 is used to calculate the scores of the threshold model, trend model, AI model and mechanism model respectively, and calculate the total health score of the system through weight configuration.

[0178] like Figure 3 The data acquisition module structure diagram is shown, corresponding to the technical solution in step S1, such as Figure 4 The data preprocessing module structure diagram is shown, which corresponds to the technical solution in step S2. Figure 5 The figure shows the fault diagnosis module flow chart, which corresponds to the technical solution in step S4. Figure 6 The figure shows the structure diagram of the health meter module, corresponding to the technical solution in step S5.

[0179] Third embodiment

[0180] This embodiment provides an embodiment of a specific practical application process based on the first embodiment, which is as follows:

[0181] 01. Collect vibration waveform data of equipment in the air separation production line, including key dynamic equipment such as air compressors, boosters, expanders, and oxygen compressors. The number of equipment varies with different production volumes. Collect vibration values, speeds, temperatures, etc. of dynamic equipment from the on-site DCS database. Collect process signals of static equipment: pressure, flow, temperature, and cooling water temperature in the pipeline; including but not limited to the following systems (pre-cooling, purification, water system, and storage tank).

[0182] 02. The data obtained from the DCS is processed using an outlier method and is uniquely identified by the device code and the measuring point code and position number and stored in the database. The waveform data is also uniquely identified by the device code, measuring point code and position number and is stored in the existing database.

[0183] 03. Utilize the threshold warning and trend warning methods in the technical solution to generate threshold warnings and trend warnings for the collected data. When an alarm is generated, the specific location of the device is located based on the device code and measurement point code, and the waveform data corresponding to the device location is also found. For example, when an alarm is generated at the air compressor motor drive end, the waveform data at that location is obtained.

[0184] 04. At the same time, the corresponding speed of the device at this location is found to be 2980r / min, and the rotation frequency is calculated to be 49.67Hz.

[0185] 05. Extract corresponding features according to predetermined rules. For example, if the predefined rule is: motor frequency doubled / motor frequency doubled>1, the diagnostic conclusion is that the coupling is poorly aligned.

[0186] 06. Perform FFT transformation on the waveform and extract the amplitude corresponding to the double frequency positions of 49.67Hz and 99.33Hz, record them as f1 and f2, and calculate f2 / f1>v l , then the output coupling is misaligned.

[0187] 07. For the AI ​​early warning model, key parameters such as the "secondary intercooler third-stage inlet temperature" in the air compressor system are selected. Other related parameters include ["first-stage intercooler first-stage outlet temperature", "first-stage intercooler second-stage inlet temperature", "second-stage intercooler second-stage outlet temperature", "second-stage intercooler third-stage inlet temperature", "first-stage intercooler pressure difference", "second-stage intercooler pressure difference"].

[0188] 08. Build an LSTM prediction model based on the above parameters and calculate dif = |predicted value - actual value| to implement AI alarm.

[0189] 09. Health calculation is achieved using health calculation methods based on threshold models, trend models, AI models, and fault diagnosis models.

[0190] A computer-readable storage medium stores computer code. When the computer code is executed, the above-described method is performed. A person skilled in the art will appreciate that all or part of the steps in the various methods of the above-described embodiments can be performed by a program instructing related hardware. The program can be stored in a computer-readable storage medium. The storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0191] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0192] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0193] It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention. It should be pointed out that those skilled in the art can make several improvements and modifications without departing from the principles of the present invention, and such improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A health scoring method for air separation production line based on a multivariate heterogeneous model, characterized in that: The following steps are involved: S1: Collect status data, operating data and process data of the air separation production line; S2: To prevent abnormal data from affecting the system model building, the collected data is pre-processed for abnormal values; S3: Based on pre-processed data, multi-dimensional warnings are provided through adaptive threshold warnings, trend warnings, and AI warnings; S4: When parameters in the air separation production line equipment trigger an early warning, fault diagnosis is performed only for rotating equipment and only in the scenario where the vibration parameters or temperature parameters at the equipment bearings are warned. Before diagnosis, equipment information must be entered in advance. Fault diagnosis is performed on the equipment based on the mechanism rule library pre-set by the platform. S5: Calculate the scores of the threshold model, trend model, AI model, and mechanism model respectively, and calculate the overall health score of the system through weight configuration.

2. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 1 is characterized in that: In step S1, the status data, operating condition data and process data of the air separation production line are collected, specifically: The state data includes a vibration displacement signal and a vibration acceleration signal of the moving equipment; The operating condition data includes equipment start and stop status data, full load operation status data, 80% load status data, speed data, valve opening data, etc. The process data include data on temperature, pressure, flow rate, and pressure difference during production; In the air separation production line, if the key units are equipped with a vibration monitoring and protection system, waveform data is collected from the front panel of the system. At the same time, the data acquisition module includes multiple industrial protocol connections including Modbus, Profibus, and OPC to collect data from the industrial air separation production line, including PLC and DCS.

3. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 1 is characterized in that: In step S2, in order to prevent abnormal data from affecting the system model building, the collected data is processed for outliers, specifically: S21: Assume that the collected data set is X={x1,x2,…,x n }, where x i (i=1,2,…,n) represents the i-th data point in the collected data, n is the total number of collected data, and the root mean square (rms) of X is calculated using the formula: here, The root mean square of the data is obtained by summing the squares of each data point in the data set X, dividing it by n, and then taking the square root. It is used to measure the degree of discreteness of the data. S22: Calculate X1=|X|, that is, for each data point x in the data set X, i Take the absolute value and get the new data set X1={|x1|,|x2|,…,|x n |}, the purpose is to unify the positive and negative characteristics of the data to facilitate subsequent outlier judgment; S23: Setting the root mean square coefficient k, where the value of k is between 3 and 8. k is used to adjust the sensitivity of abnormal value judgment. For different air separation production line scenarios or data characteristics, an appropriate value can be selected in this range as needed. S24: Count the number num of data in X1 that is greater than k×rms, that is, traverse each data|x in X1 i |, if |x i |>k×rms, then the count is increased by 1, and finally the number of data points that meet the conditions is obtained; and calculate P represents the proportion of abnormal data points in the data set X to the total data points; S25: If P is greater than the preset threshold th1, it is judged as data abnormality. th1 is a proportional threshold preset according to standards including actual operation experience of air separation production line and data quality requirements, and is used to judge whether the data is abnormal. At the same time, the position of the data greater than k×rms in X1 is used to convert the data x corresponding to the position in X into i Replace with the mean of X This can correct abnormal data and ensure the data quality for subsequent model construction.

4. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 1 is characterized in that: In step S3, based on the pre-processed data, multi-dimensional warnings are performed through adaptive threshold warnings, trend warnings, and AI warnings, specifically: S31: Use adaptive threshold warning for a single parameter. The adaptive threshold is calculated as follows: S311: Perform data screening to obtain data for a period of time under constant working conditions. The total number of data is j, and the data point is y1 (i = 1, 2, ... j). The middle data from the fifth to the ninety-fifth percentile are taken to form a set Y = {y1, y2, ..., y j }; S312: Perform statistical calculations and calculate the mean of Y in Sum the data points in Y and calculate the standard deviation of Y y j -y mean is the deviation of the data point from the mean; S313: Setting the threshold value, setting the attention threshold value th note =y mean +3σ, warning threshold th alarm =y mean +6σ; S314: Trigger the alarm, set the latest data as y, when th note ≤y <th alarm And for m consecutive times, the health management system issues an alarm, when y≥th alarm And it will sound an alarm m times in a row, where the number of times is set by yourself; S32: Perform trend warning to determine whether the data continues to rise. Use a sliding window to calculate the data mean in segments. By determining the number of times the current value exceeds the preset threshold, determine whether the current data trend is rising. The calculation method is as follows: S321: Get data and set the current value to z k+1 , from z k+1 Get the trend data of a single working condition Z=[z1,z2,…,z k ],z i (i=1,2,…,k) is the trend data point, k is the total number of data points; S322: Set the window, the window data is Z, and the step size is 1; S323: Perform threshold calculation and calculate the mean of window data Z And set two threshold lines S324: Perform trend judgment and alarm, pre-define rising edge z line , when z k+1 ≥z line And z mean1 ≤z k+1 <z mean2 , record trend attention 1 time, when it is triggered a times continuously, the trend alarm is triggered, and the alarm level is attention; similarly, when z k+1 ≥z line , and z mean1 ≤z k+1 <z mean2 , trend warning 1 time, when it is triggered a times continuously, it is a trend alarm, and the level is warning; S325: Parameter optimization is performed, and the number of consecutive triggers a and the window length are optimized by the particle swarm optimization algorithm; S33: Perform AI warning, set parameter A in the air separation unit as the important parameter, and B, C, D, E, and F are parameters related to A: S331: Perform model input and prediction. Input the real-time data sequence A, B, C, D, E, and F into the trained LSTM model and predict the value of parameter A, which is recorded as the predicted value. S332: Perform difference calculation and alarm, calculate the difference between the predicted value and the actual value of parameter A, dif = |predicted value - actual value|; S333: Preset two thresholds th based on expert experience al , th ah If th al ≤dif≤th ah , the alarm level is attention, if dif≥th ah , the alarm level is warning.

5. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 1 is characterized in that: In step S4, when parameters in the air separation production line equipment trigger an early warning, fault diagnosis is performed only for rotating equipment and only in the vibration parameter or temperature parameter warning scenario at the equipment bearing. Before diagnosis, equipment information must be entered in advance. Fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform. Specifically: S41: Collecting original waveform data. When an alarm occurs at a certain position of the device, obtain the original waveform data S corresponding to the sensor. S is the vibration or temperature time domain waveform data sequence collected by the sensor at the device alarm position. S42: Extract characteristic frequencies. Using fast Fourier transform (FFT), combined with the pre-recorded rotational speed of the device, the signal's rotational frequency and its harmonics are extracted. The rotational frequency is the rotational frequency of the device's main shaft, and the harmonics are integer multiples of the rotational frequency. Similarly, for bearing faults, the characteristic frequency of the bearing fault is extracted. For impeller-type devices, the blade pass frequency is extracted. The bearing fault characteristic frequency is calculated based on the bearing model and structural parameters. The blade pass frequency is related to the number of impeller blades and the rotational speed. S43: Perform rule matching to determine whether the extracted features meet the pre-set rules in the platform's preset mechanism rule library. The rules are formulated based on the differences in features between normal operation and failure of the device. S44: Determine the alarm level. Two levels of values ​​are pre-set in the rules. l , v h , v l <v h , v l is the low-level alarm threshold, v h is the high-level alarm threshold, and the value currently calculated according to the rule is v alue , v alue is the parameter value obtained by calculating the characteristic frequency for determining the fault level. alue Location l With v h The alarm level is determined based on the location.

6. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 4 is characterized in that: In step S5, the scores of the threshold model, trend model, AI model and mechanism model are calculated respectively, specifically: S51: Calculation of health of threshold model: S511: The attention threshold th has been calculated according to the adaptive threshold part note , warning threshold th alarm , using the same principle to calculate the optimal interval (t l ,t r ), t l is the ideal minimum value, t r For the ideal maximum health management system, the scores corresponding to each threshold are reserved at the same time, and the scores and thresholds are combined to form 4 points. The left boundary point of the optimal interval (t l ,S l ), the left boundary point of the optimal interval (t r ,S r ), attention threshold corresponding to the health point (th note ,S note ), the health point corresponding to the warning threshold (th alarm ,S alarm ), where S l For parameters at t l The corresponding basic health score, S r For parameters at t r The corresponding basic health score, S note For parameters in th note The corresponding basic health score, S alarm For parameters in th alarm The corresponding basic health score; S512: Construct a score piecewise function and a weight piecewise function based on the four points in step S511: S513: Calculate S based on the real-time data x of a specific operating parameter in the air separation production line t , w t ; S514: Calculate the threshold model scores for all segmented health intervals t represents the tth health interval; S52: For the trend model, the score and weight calculation method is: S521: Similar to the threshold model, three points are constructed: the trend health benchmark point (tt1, ss1), the trend attention warning point (z mean1 , ss2), and the trend warning point (z mean2 , ss3), where tt1 < z mean1 < z mean2 ; ss3 < ss2 < ss1. Here, tt1 is the benchmark value when the device parameter has no abnormal trend, corresponding to the highest health score ss1, z mean1 is the threshold for a mild abnormal trend in trend warning, corresponding to the medium health score ss2, and z mean2 is the threshold for a severe abnormal trend in trend warning, corresponding to the lowest health score ss3; S522: For all trend model scores S core2 , z is the segment index of the trend model: S53: For the AI ​​model, the score of the entire AI model is calculated as: S531: For AI models, each model score has two levels S532: In the AI ​​module, the score of the entire AI model: Among them, s aq is the score of a single sub-model in the AI ​​model, s ah When the prediction deviation dif is in the attention threshold interval th al ≤dif≤th ah When the health score of a single AI sub-model is s al When the prediction deviation dif is in the warning threshold interval dif≥th ah When the health score of a single AI sub-model is al is the attention threshold, th ah is the warning threshold, q is the number of AI sub-models; S54: For the mechanism module, the score calculation method for all mechanism models is the same as that for AI models: S541: Calculation score of a single mechanism model S542: Score of all mechanism models in the mechanism module Among them, s mr Calculate the score for a single mechanistic model, s mh The characteristic parameter v for the device operation alue In the health threshold range v l ≤v alue ≤v h When the health score of a single mechanism sub-model is s ml When the device is running, the characteristic parameter v alue In the failure risk threshold interval v alue ≥v h When the health score of a single mechanism sub-model is l is the lower limit of the health threshold used for judgment in the mechanism model, v h is the fault risk threshold used for judgment in the mechanism model, and r is the number of mechanism models.

7. The air separation production line health scoring method based on a multivariate heterogeneous model according to claim 6 is characterized in that: In step S5, the total health score of the system is calculated by weight configuration, specifically: There are configuration places in the health calculation module, which are used to configure the weights of the early warning model, trend model, AI model, and mechanism model. The weight configurations are α1, α2, α3, and α4 respectively; α1+α2+α3+α4=1; The total system score in the health management system is: Among them, α u For each model weight, S coreu The score for each model.

8. A multivariate heterogeneous model-based air separation line health scoring system for executing the multivariate heterogeneous model-based air separation line health scoring method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, used to collect status data, operating data and process data of the air separation production line; The data preprocessing module is used to preprocess the collected data for abnormal values ​​in order to prevent the influence of abnormal data on the system model construction; The early warning module is used to provide multi-dimensional early warnings based on pre-processed data through adaptive threshold warnings, trend warnings, and AI warnings; The fault diagnosis module is used to perform fault diagnosis on rotating equipment only when parameters in the air separation production line trigger an early warning, and only in the scenario where vibration parameters or temperature parameters at the bearings of the equipment are warned. Before diagnosis, equipment information must be entered in advance, and fault diagnosis is performed on the equipment based on the mechanism rule library preset by the platform; The health calculation module is used to calculate the scores of the threshold model, trend model, AI model and mechanism model respectively, and calculate the total health score of the system through weight configuration.

9. A computer device comprising a memory and one or more processors, wherein the memory stores computer code, and when the computer code is executed by the one or more processors, the one or more processors are caused to perform the method according to any one of claims 1 to 7. 10 . A computer-readable storage medium storing computer code, wherein when the computer code is executed, the method according to claim 1 is performed.

Images