Manufacturing industry processing equipment information digitization system

Abstract

The invention discloses a manufacturing industry processing equipment information digitization system comprising a data acquisition layer which acquires data through a sensor which is connected to a data acquisition terminal in a wired or wireless mode; the data acquisition terminal is used for receiving the data acquired by the data acquisition layer, respectively marking the acquired data as key monitoring parameters and non-key monitoring parameters, and then transmitting the key monitoring parameters and the non-key monitoring parameters to the server; the invention relates to the technical field of manufacturing industry data processing, and can realize equipment fault point judgment when equipment fails, can more comprehensively capture subtle changes of equipment operation states in fault prediction, greatly improves the fault feature identification degree, and improves the production continuity by predicting the fault occurrence time and type. And the risks of capacity loss and sharp rise of maintenance cost caused by faults are reduced. Impact of faults on equipment is reduced, and the service life of the equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of manufacturing data processing, and particularly to a digital system for manufacturing processing equipment information. Background Art

[0002] In today's highly industrialized era, various complex equipment is widely used in many key fields such as manufacturing, energy, and transportation, becoming the core elements to support production operations and ensure the stable operation of the economy and society. However, during the long-term operation of the equipment, it will inevitably be affected by multiple factors such as mechanical wear, electrical aging, environmental erosion, and operating condition fluctuations, resulting in frequent failures and bringing many severe challenges to enterprise production.

[0003] In the face of the drawbacks of the traditional maintenance mode, fault prediction technology has emerged, aiming to detect the signs of equipment failures in advance and create "opportunities" for preventive maintenance.

[0004] However, for the existing prediction technologies, on the one hand, some prediction technologies focus on a single monitoring parameter dimension and judge faults only based on the changes of a certain type of physical quantity of the equipment, ignoring that the operation of the equipment is the result of the coordinated action of multiple parameters. On the other hand, in the feature extraction link, the existing methods often stay at the simple data statistics level and do not deeply explore the high-order spectrum of the parameters, so the accuracy of the prediction needs to be improved. Summary of the Invention

[0005] To solve the technical problems in the background art, the present invention proposes a digital system for manufacturing processing equipment information.

[0006] A digital system for manufacturing processing equipment information proposed by the present invention includes:

[0007] Data acquisition layer: Collect data through sensors, and the sensors are connected to the data acquisition terminal in a wired or wireless manner;

[0008] Data acquisition terminal: Used to receive the data collected by the data acquisition layer, mark the collected data as key monitoring parameters and non-key monitoring parameters respectively, and then transmit the key monitoring parameters and non-key monitoring parameters to the server;

[0009] The server includes:

[0010] Data storage module: Used to receive and store the key monitoring parameters and non-key monitoring parameters transmitted by the data acquisition terminal;

[0011] Equipment fault point judgment module: Used to judge the equipment fault point by analyzing the key monitoring parameters in the data storage module;

[0012] Key monitoring parameter data observation range adjustment module: used to classify key monitoring parameters into slow-changing parameters and fast-changing parameters, select an observation time interval for each of the slow-changing parameters and fast-changing parameters, and then standardize and de-trend the slow-changing parameters and fast-changing parameters within their respective time intervals to obtain a new time series data of slow-changing parameter data and a new time series data of fast-changing parameter data respectively;

[0013] Feature vector generation module: by analyzing the new time series data of slow-changing parameter data and the new time series data of fast-changing parameter data generated by the key monitoring parameters in the key monitoring parameter data observation range adjustment module, generate the feature vectors of the key monitoring parameters, and generate the feature vectors of non-key monitoring parameters by analyzing the non-key monitoring parameters in the data storage module;

[0014] Matrix generation module: used to construct a three-dimensional hypermatrix from the fault type dimension, the monitoring parameter dimension, and the feature dimension of the feature vectors generated by the feature vector generation module;

[0015] Fault prediction module: through the analysis of the feature vector generation module and the matrix generation module, used to predict the future fault occurrence time point and the corresponding fault type.

[0016] Preferably, in the device fault point judgment module, the device fault point is judged as follows:

[0017] Suppose the device has n key monitoring parameters, denoted as p1, p2,..., p n ;

[0018] Suppose the device has m possible fault types, denoted as F1, F2,..., F m ;

[0019] For each key monitoring parameter P i , i = 1, 2,..., n; set the normal operating range of each key monitoring parameter P i as [L i , U i , where L i is the lower limit value and U i is the upper limit value;

[0020] By analyzing the historical fault data, construct a fault feature matrix A, and the dimension of the fault feature matrix A is m × n;

[0021] The element a ij in the fault feature matrix A represents the typical change characteristic value of the key monitoring parameter P j when the fault type F i occurs, j = 1, 2,..., m;

[0022] When the device fails, obtain the actual value of each key monitoring parameter P at the current moment i and calculate the deviation value d of each key monitoring parameter P i from the normal operating range, where i = 1, 2,..., n; the calculation formula is as follows: i , i = 1, 2,..., n; the calculation formula is as follows:

[0023]

[0024] Evaluate the similarity between the current device state and each possible fault type, and evaluate the similarity degree S(F j ), and its calculation formula is as follows:

[0025]

[0026] where w i is the weight coefficient of each key monitoring parameter;

[0027] Then the predicted fault type is:

[0028] Calculate the comprehensive evaluation function value S(F j ) for each possible fault type F j . After comparing these values, find the fault type corresponding to the maximum value.

[0029] Preferably, the determination of the weight coefficient w i :

[0030] Adopt an adaptive weight adjustment algorithm based on the information entropy method, and the weight coefficient w i is determined by the reciprocal of the information entropy.

[0031] Preferably, in the key monitoring parameter data observation range adjustment module, for n key monitoring parameters, within a set historical period, for each key monitoring parameter P i , i = 1, 2,..., n; for each key monitoring parameter P i construct a data set R, and the data points in the data set R are arranged in chronological order. Calculate the change rate between adjacent data points, and use the clustering algorithm to divide the data set R into a slow-changing parameter subset R1 and a fast-changing parameter subset R2 according to the change rate;

[0032] At time t, for the slow-changing parameter subset R1, select the data points in the past time interval [t a , t] to form a subset R1a, and perform normalization and detrending processing on the subset R1a to obtain a new time series data Da of the slow-changing parameter data;

[0033] At time t, for the fast-changing parameter subset R2, select the past time interval [tb The data points within [t, t] form a subset R2b. The subset R2b is normalized and detrended to obtain a new time series data Db for the fast-varying parameter data;

[0034] wherein, the time length of the time interval [t, t] is greater than the time length of the time interval [t, t]. a , t] is greater than the time length of the time interval [t b , t].

[0035] Preferably, in the eigenvector generation module, for the slow-varying parameter data Da, first perform the Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t), and the intrinsic mode function is IMF;

[0036] IMF a，k (t), where k = 1, 2,..., k a ; k a is the number of the decomposed intrinsic mode functions;

[0037] By performing the Hilbert transform on each IMF a，k (t), obtain the corresponding instantaneous frequency f a，k (t) and the instantaneous amplitude A a，k (t), and thus construct the spectral feature matrix F a as:

[0038]

[0039] k a Among the k IMFs, calculate the phase correlation coefficient matrix between each pair of IMFs. For any two IMFs, IMF a，j (t) and IMF a，k (t), j ≠ k, the calculation formula for their phase correlation coefficient r a (j, k) is:

[0040]

[0041] In the formula, and are the means of the corresponding IMF a，j (t) and IMF a，k (t) respectively, and T is the data length;

[0042] Form the phase correlation matrix P with the phase correlation coefficient values of all pairwise IMF combinations a :

[0043]

[0044] For the fast-varying parameter data Db, short-time Fourier transform is used to perform time-frequency analysis on Db to obtain the time-frequency distribution matrix S b :

[0045] S b =[S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component;

[0046] From the time-frequency distribution matrix S b the proportion of high-frequency band energy in different time periods is statistically calculated to reflect the dynamic changes of high-frequency energy of fast-varying parameters during operation. Abnormal fluctuations often occur before a fault. The calculation formula is:

[0047]

[0048] where o high defines the starting index of the high-frequency band, and O is the total number of frequency components;

[0049] Perform wavelet packet decomposition on Db to obtain the energy entropy feature H in different wavelet packet sub-bands b ;

[0050] At time t, construct the feature vector of the key monitoring parameter as:

[0051]

[0052] where F a，t : The spectral feature matrix obtained based on the slow-varying parameter data Da after processing, which represents the energy distribution and oscillation characteristic changes of the slow-varying parameter at different time and frequency scales at time t;

[0053] P a，t : It is the phase correlation matrix constructed based on the slow-varying parameter data Da, which is used to measure the cooperative operation law of each intrinsic mode function after the slow-varying parameter is decomposed at the phase level;

[0054] S b，t : It is the time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-varying parameter data Db using short-time Fourier transform, which reflects the energy amplitude distribution of the fast-varying parameter at different time points and different frequency components at time t;

[0055] represents at time t, from the time-frequency distribution matrix S b the proportion of high-frequency band energy in different time periods is statistically calculated;

[0056] H b，t: It refers to the wavelet packet energy entropy obtained after wavelet packet decomposition of Db at time t;

[0057] Suppose the device has h non-critical monitoring parameters. For each non-critical monitoring parameter P hj , j = 1, 2,..., h, based on the working conditions of the actual operation of the device, set the relative change rate threshold At time t, calculate the relative change rate C of the working condition of each non-critical monitoring parameter P hj ; hj，t : Mark the cases that exceed the corresponding threshold ;

[0058] At the same time, carry out the Granger causality test process to determine the non-critical monitoring parameters that have a causal relationship with the critical monitoring parameters, and form a subset S causal ;

[0059] For the element P hj ∈S causal , record the causal lag order τ hj , the causal strength I hj , the fusion skewness Sk hj，t , the kurtosis Hu hj，t , and form the comprehensive feature vector of non-critical monitoring parameters as:

[0060]

[0061] Preferably, in the matrix generation module, construct a three-dimensional hypermatrix M, and its dimension is set as:

[0062] m×(n + h)×l;

[0063] In the formula, the three dimensions include:

[0064] m: The dimension of fault types. Suppose the device has m possible fault types, which are respectively denoted as F1, F2,..., F m ;

[0065] (n + h): The dimension of monitoring parameters, including n critical monitoring parameters and h non-critical monitoring parameters, a total of (n + h) parameters;

[0066] l: The dimension of features, including F a , P a , S b , H b , C hj，t , S causal , τ hj , I hj , Sk hj，t , Ku hj，t ;

[0067] Through the fault type F i , for i = 1, 2,..., m, assign values to the matrix elements m ijk where, in m ijk , i represents the fault type, j represents the monitoring parameter, and k represents the feature.

[0068] Preferably, in the fault prediction module, at the monitoring time t, splice the feature vectors and to generate a spliced feature vector

[0069]

[0070] Next perform a three-dimensional dot product operation with the three-dimensional hypermatrix M layer by layer to derive the matching score vector for each fault type

[0071] The operation formula is as follows:

[0072]

[0073] This score comprehensively integrates multi-dimensional, multi-context features of multiple parameters and their internal complex associations with each fault type, accurately reflecting the degree to which the current operating state of the device matches each fault type, and being more sensitive to capturing subtle fault precursors compared to traditional methods

[0074] Set the score threshold τ. When the maximum value of the calculated S i，t is greater than τ, immediately trigger a potential fault warning

[0075] Fault time and type prediction:

[0076] After the potential fault warning is activated, continuously calculate the matching score vector at the future prediction time t'. Build a dynamic Bayesian network model for each fault type, and update the fault probability distribution according to Bayes' theorem with new data

[0077] As the fault probability distribution is updated according to Bayes' theorem with new data at each future prediction time t' After N consecutive updates, assume that the of N consecutive predictions are all less than the set standard deviation threshold, and for a certain fault type F i , its prediction probability exceeds the set prediction probability threshold for N consecutive times, and at the same time, the absolute value of the difference in the matching scores for N consecutive times is less than the set threshold;

[0078] Then the earliest t' moment reached is the predicted fault occurrence time, and the fault type corresponding to the maximum probability at this time is the predicted fault type.

[0079] A method for judging fault points and predicting faults of manufacturing processing equipment, comprising the following steps:

[0080] Divide the collected data into key monitoring parameters and non-key monitoring parameters;

[0081] When the equipment fails, judge the fault point of the equipment, including the following steps:

[0082] Suppose the equipment has n key monitoring parameters, denoted as p1, p2,..., p n ;

[0083] Suppose the equipment has m possible fault types, denoted as F1, F2,..., F m ;

[0084] For each key monitoring parameter P i , i = 1, 2,..., n; set the normal operating range of each key monitoring parameter P i as [L i , U i , where L i is the lower limit value and U i is the upper limit value;

[0085] Through the analysis of historical fault data, construct a fault feature matrix A, and the dimension of the fault feature matrix A is m×n;

[0086] The element a ij in the fault feature matrix A represents the typical change characteristic value of the key monitoring parameter P j when the fault type F i occurs, j = 1, 2,..., m;

[0087] When the equipment fails, obtain the actual value of each key monitoring parameter P i at the current moment, and calculate the deviation value d i between each key monitoring parameter P i and the normal operating range, i = 1, 2,..., n; the calculation formula is as follows:

[0088]

[0089] Evaluate the similarity between the current equipment state and each possible fault type, evaluate the similarity S(F j ), and its calculation formula is as follows:

[0090]

[0091] Among them, w iis the weight coefficient of each key monitoring parameter, and an adaptive weight adjustment algorithm based on information entropy is adopted. The weight coefficient w i is determined by the reciprocal of the information entropy;

[0092] Then the predicted fault type is:

[0093] Calculate the comprehensive evaluation function value S(F j ) for each possible fault type F j . After comparing these values, find the fault type corresponding to the maximum value;

[0094] The fault prediction is:

[0095] It includes the following steps:

[0096] For n key monitoring parameters, within a set historical period, for each key monitoring parameter P i , i = 1, 2,..., n; for each key monitoring parameter P i construct a data set R. The data points in the data set R are arranged in chronological order. Calculate the change rate between adjacent data points, and use the clustering algorithm to divide the data set R into a slow-varying parameter subset R1 and a fast-varying parameter subset R2 according to the change rate;

[0097] At time t, for the slow-varying parameter subset R1, select the data points in the past time interval [t a , t] to form a subset R1a. Perform standardization and detrending processing on the subset R1a to obtain a new time series data Da of the slow-varying parameter data;

[0098] At time t, for the fast-varying parameter subset R2, select the data points in the past time interval [t b , t] to form a subset R2b. Perform standardization and detrending processing on the subset R2b to obtain a new time series data Db of the fast-varying parameter data;

[0099] Among them, the time length of the time interval [t a , t] is greater than the time length of the time interval [t b , t];

[0100] For the slow-varying parameter data Da, first perform Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t). The intrinsic mode function is IMF;

[0101] IMF a，k (t), where k = 1, 2,..., k a ; k ais the number of intrinsic mode functions obtained by decomposition;

[0102] By performing the Hilbert transform on each IMF a，k (t), the corresponding instantaneous frequency f a，k (t) and the instantaneous amplitude A a，k (t) are obtained, and based on this, the spectral feature matrix F a is:

[0103]

[0104] k a Among the k IMFs, calculate the phase correlation coefficient matrix between each pair of IMFs. For any two IMFs, IMF a，j (t) and IMF a，j (t), j≠k, the formula for their phase correlation coefficient r a (j, k) is:

[0105]

[0106] In the formula, and are the means of the corresponding IMF a，j (t) and IMF a，k (t) respectively, and T is the data length;

[0107] Form the phase correlation matrix P a with the phase correlation coefficient values of all pairwise IMF combinations:

[0108]

[0109] For the fast-varying parameter data Db, perform time-frequency analysis on Db using the short-time Fourier transform to obtain the time-frequency distribution matrix S b :

[0110] S b =[S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component;

[0111] From the time-frequency distribution matrix S b , statistically calculate the proportion of high-frequency band energy in different time periods which reflects the dynamic change of high-frequency energy of the fast-varying parameter during operation. Abnormal fluctuations often occur before a fault. The calculation formula is:

[0112]

[0113] In the formula, ohigh Define the starting index of the high-frequency band, where O is the total number of frequency components;

[0114] Perform wavelet packet decomposition on Db to obtain the energy entropy feature H in different wavelet packet sub-bands b ;

[0115] At time t, construct the feature vector of the key monitoring parameters as:

[0116]

[0117] In the formula, F a，t : The spectral feature matrix obtained after processing the slow-varying parameter data Da, which represents the energy distribution and oscillation characteristic changes of the slow-varying parameters at different time and frequency scales at time t;

[0118] P a，t : The phase correlation matrix constructed based on the slow-varying parameter data Da, which is used to measure the cooperative operation law between the intrinsic mode functions of the slow-varying parameters after decomposition at the phase level;

[0119] S b，t : The time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-varying parameter data Db using the short-time Fourier transform, which reflects the energy amplitude distribution of the fast-varying parameters at different time points and different frequency components at time t;

[0120] represents at time t, from the time-frequency distribution matrix S b calculate the high-frequency band energy ratio within different time periods;

[0121] H b，t : Refers to the wavelet packet energy entropy obtained after wavelet packet decomposition of Db at time t;

[0122] Suppose the device has h non-critical monitoring parameters, and for each non-critical monitoring parameter P hj , j = 1, 2,..., h, set the relative change rate threshold based on the actual operating conditions of the device At time t, calculate the relative change rate C of the operating condition of each non-critical monitoring parameter P hj : Mark the cases exceeding the corresponding threshold hj，t ; At the same time, carry out the Granger causality test process to determine the non-critical monitoring parameters that have a causal relationship with the key monitoring parameters, and form a subset S

[0123] ; causal ;

[0124] For the element P hj ∈S causal, record the causal lag order τ hj , causal strength I hj , fusion skewness Sk hj，t , kurtosis Ku hj，t , to form a comprehensive feature vector of non-critical monitoring parameters It is:

[0125]

[0126] Construct a three-dimensional hypermatrix M, and its dimension is set to:

[0127] m×(n + h)×l;

[0128] Wherein, the three dimensions include:

[0129] m: fault type dimension. Suppose the device has m possible fault types, which are respectively denoted as F1, F2,..., F m ;

[0130] (n + h): monitoring parameter dimension, including n key monitoring parameters and h non-critical monitoring parameters, a total of (n + h) parameters;

[0131] l: feature dimension, including F a , P a , S b , H b , C hj，t , S causal , τ hj , I hj , Sk hj，t , Ku hj，t ;

[0132] Through the fault type F i , i = 1, 2,..., m, assign values to the matrix element m ijk In m ijk , i represents the fault type, j represents the monitoring parameter, and k represents the feature;

[0133] At the monitoring moment t, splice the feature vectors and to generate a spliced feature vector

[0134] Next Perform a three-dimensional dot product operation with the three-dimensional hypermatrix M layer by layer to derive a matching score vector for each fault type

[0135] The operation formula is as follows:

[0136]

[0137] This score comprehensively integrates multi-dimensional and multi-situational features of multiple parameters and their complex internal correlations with each fault type, accurately reflecting the degree to which the current operating state of the equipment matches each fault type. It can capture the subtle precursors of faults more sensitively than traditional methods.

[0138] Set the score threshold τ. When the maximum value of the calculated S i，t is greater than τ, immediately trigger a potential fault warning.

[0139] Accurate time and type prediction:

[0140] After the potential fault warning is activated, continuously calculate the matching score vector at the predicted time t' in the future time period, build a dynamic Bayesian network model for each fault type, and update the fault probability distribution according to Bayes' theorem with new data.

[0141] As Bayes' theorem updates the fault probability distribution at each predicted time t' in the future time period with new data. After N consecutive updates, assume that the of N consecutive predictions are all less than the set standard deviation threshold, and for a certain fault type F i , its predicted probability exceeds the set predicted probability threshold for N consecutive times, and at the same time, the absolute value of the difference in the matching scores for N consecutive times is less than the set threshold.

[0142] Then the earliest t' moment reached is the predicted fault occurrence time, and the fault type corresponding to the maximum probability at this time is the predicted fault type.

[0143] In the present invention, the proposed digital information system for manufacturing processing equipment has the following beneficial technical effects:

[0144] During fault prediction, by setting the key monitoring parameter data observation range adjustment module, different observation time intervals are matched for slow-varying parameters and fast-varying parameters. Through the setting of the feature vector generation module, for key monitoring parameters, considering both slow-varying and fast-varying characteristics, rich high-order spectra and collaborative features such as the spectral feature matrix and phase correlation matrix of slow-varying parameters, and the time-frequency distribution matrix, high-frequency band energy ratio, and wavelet packet energy entropy of fast-varying parameters are extracted; for non-key monitoring parameters, the relative change rate, skewness, kurtosis, causal lag order, and intensity are also calculated. The feature set constructed by integrating these features can capture the subtle changes in the operating state of the equipment more comprehensively, greatly improving the fault feature recognition rate.

[0145] In the fault prediction module, when entering the fault prediction and evaluation stage, the matching score and the fault probability distribution of the dynamic Bayesian network are continuously updated according to newly collected data, and the prediction results are flexibly adjusted according to the real-time operation situation of the equipment to ensure the timeliness and reliability of the prediction. The algorithm accurately predicts the fault occurrence time and type.

[0146] In the equipment fault point judgment module, when the equipment fails, the equipment fault point can be judged to find the equipment fault point.

[0147] In summary, when the equipment of the present application fails, the equipment fault point can be judged, and the subtle changes in the operation state of the equipment can be more comprehensively captured in the fault prediction, greatly improving the fault feature recognition rate. By predicting the fault occurrence time and type, the production continuity is improved, and the risks of production capacity loss and soaring maintenance costs caused by faults are reduced. The impact of faults on the equipment is reduced, and the service life of the equipment is extended.

[0148] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. Brief Description of the Drawings

[0149] Figure 1 It is a principle block diagram of the system of the present invention. Detailed Embodiments

[0150] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference signs denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0151] As Figure 1 shown, a digital manufacturing processing equipment information system includes:

[0152] Data acquisition layer: Data is acquired through sensors, and the sensors are connected to the data acquisition terminal in a wired or wireless manner;

[0153] Data acquisition terminal: It is used to receive the data acquired by the data acquisition layer, mark the acquired data as key monitoring parameters and non-key monitoring parameters respectively, and then transmit the key monitoring parameters and non-key monitoring parameters to the server;

[0154] The judgment of key monitoring parameters and non-key monitoring parameters can be manually judged and then preset classified into key monitoring parameters and non-key monitoring parameters;

[0155] When the data acquisition layer acquires data through sensors, the time data of the data generation time is acquired together;

[0156] The server includes:

[0157] Data storage module: used to receive and store the key monitoring parameters and non-key monitoring parameters transmitted by the data acquisition terminal;

[0158] Device fault point judgment module: used to judge the device fault point by analyzing the key monitoring parameters in the data storage module;

[0159] Key monitoring parameter data observation range adjustment module: used to divide the key monitoring parameters into slow-changing parameters and fast-changing parameters, select an observation time interval for each of the slow-changing parameters and fast-changing parameters, and then standardize and detrend the slow-changing parameters and fast-changing parameters within their respective time intervals to obtain a new time series data of the slow-changing parameter data and a new time series data of the fast-changing parameter data respectively;

[0160] Feature vector generation module: by analyzing the new time series data of the slow-changing parameter data and the new time series data of the fast-changing parameter data generated by the key monitoring parameters in the key monitoring parameter data observation range adjustment module, generate the feature vector of the key monitoring parameters, and generate the non-key monitoring parameter feature vector by analyzing the non-key monitoring parameters in the data storage module;

[0161] Matrix generation module: used to construct a three-dimensional hypermatrix from the fault type dimension, the monitoring parameter dimension, and the feature dimension of the feature vectors generated by the feature vector generation module;

[0162] Fault prediction module: by analyzing the feature vector generation module and the matrix generation module, used to predict the future fault occurrence time point and the corresponding fault type;

[0163] In the device fault point judgment module, the device fault point is judged as follows:

[0164] Suppose the device has n key monitoring parameters, which are respectively denoted as p1, p2,..., p n ;

[0165] As an example, taking the motor device as an example, here p1 can be the current, p2 can be the temperature, and p3 can be the vibration frequency;

[0166] Suppose the device has m possible fault types, which are respectively denoted as F1, F2,..., F m ;

[0167] As an example, taking the motor device as an example, F1 can be the motor winding short circuit, and F2 can be the bearing wear;

[0168] For each key monitoring parameter P i , i = 1, 2,..., n; set each key monitoring parameter Pi The normal operating range is [L i , U i , where L i is the lower limit value and U i is the upper limit value;

[0169] Set the normal operating range of each key monitoring parameter P i . The normal operating range of each key monitoring parameter P i can be determined through the technical manual of the device and the statistical analysis of a large amount of past normal operating data;

[0170] As an example, taking a motor device as an example, for the current parameter p1 of the motor device, with the unit of amperes, the normal operating range may be [3, 10], that is, when the motor is operating normally, the current value should be between 3 amperes and 10 amperes;

[0171] By analyzing historical fault data, construct a fault feature matrix A, and the dimension of the fault feature matrix A is m×n;

[0172] The element a ij in the fault feature matrix A j represents the typical change characteristic value of the key monitoring parameter P i when the fault type F

[0173] occurs, where j = 1, 2,..., m;

[0174] Taking a motor device as an example, for the fault type F1 of motor winding short - circuit, when this fault occurs, the current parameter p1 may increase significantly. Assume the typical change characteristic value is a11 = 20, that is, the current will rise to 20 amperes; the temperature parameter p2 may also increase rapidly. Assume the typical change characteristic value is a21 = 150, that is, the temperature will rise to 150 degrees Celsius; i When the device fails, obtain the actual value of each key monitoring parameter P i at the current moment, and calculate the deviation value d i of each key monitoring parameter P

[0175]

[0176] Taking a motor device as an example, if the current value p1 of the motor device is 15 amperes at present, and the normal operating range is [3, 10], then the deviation value d1 of the current parameter is d1 = 15 - 10 = 5;

[0177] Evaluate the similarity degree between the current device state and each possible fault type, and evaluate the similarity degree S(F j ), and its calculation formula is as follows:

[0178]

[0179] Among them, w i is the weight coefficient of each key monitoring parameter, reflecting the relative importance of different parameters in judging the fault type;

[0180] The weight coefficient w i can be determined by methods such as expert scoring and historical data analysis.

[0181] Determination of the weight coefficient w i :

[0182] Adopt an adaptive weight adjustment algorithm based on the information entropy method, and the weight coefficient w i is determined by the reciprocal of the information entropy.

[0183] In this way, the lower the information entropy, the smaller the feature weight, and the higher the information entropy, the larger the feature weight.

[0184] Then the predicted fault type is:

[0185] Calculate the comprehensive evaluation function value S(F j ) for each possible fault type F j . After comparing these values, find the fault type corresponding to the maximum value.

[0186] In the key monitoring parameter data observation range adjustment module;

[0187] For n key monitoring parameters, within a set historical period, for each key monitoring parameter P i , i = 1, 2,..., n; for each key monitoring parameter P i construct a data set R. The data points in the data set R are arranged in chronological order, calculate the change rate between adjacent data points, and use the clustering algorithm to divide the data set R into a slow-changing parameter subset R1 and a fast-changing parameter subset R2 according to the change rate;

[0188] At time t, for the slow-changing parameter subset R1, select the data points in the past time interval [t a , t] to form a subset R1a, and perform standardization and detrending processing on the subset R1a to obtain a new time series data Da of the slow-changing parameter data;

[0189] At time t, for the fast-changing parameter subset R2, select the data points in the past time interval [t b , t] to form a subset R2b, and perform standardization and detrending processing on the subset R2b to obtain a new time series data Db of the fast-changing parameter data;

[0190] Among them, the time interval [t a , t] has a longer time length than the time interval [t b , t];

[0191] For the slow-varying parameter data Da, first perform the Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t), and the intrinsic mode function is the IMF;

[0192] IMF a，k (t), where k = 1, 2,..., k a ; k a is the number of intrinsic mode functions obtained by decomposition;

[0193] By performing the Hilbert transform on each IMF a，k (t), the corresponding instantaneous frequency f a，k (t) and the instantaneous amplitude A a，k (t) are obtained, and a spectral feature matrix F a is constructed as follows:

[0194]

[0195] This matrix clearly shows the energy distribution state of the slow-varying parameter at different time and frequency scales and the change trend of the oscillation characteristics.

[0196] k a Among the individual IMFs, calculate the phase correlation coefficient matrix between each pair of IMFs to measure the phase co-variation between different IMFs. For any two IMFs, IMF a，j (t) and IMF a，k (t), j ≠ k, the calculation formula for their phase correlation coefficient r a (j, k) is as follows:

[0197]

[0198] In the formula, and are the means of the corresponding IMF a，j (t) and IMF a，k (t) respectively, and T is the data length;

[0199] Form a phase correlation matrix P a from the phase correlation coefficient values of all pairwise IMF combinations:

[0200]

[0201] So as to reflect the cooperative operation law of each oscillation component inside the slow-varying parameter at the phase level and provide a strong reference for subsequent fault correlation analysis

[0202] For the fast-varying parameter data Db, perform time-frequency analysis on Db using the short-time Fourier transform to obtain the time-frequency distribution matrix S b :

[0203] S b =[S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component;

[0204] From the time-frequency distribution matrix S b statistically calculate the proportion of high-frequency band energy in different time periods to reflect the dynamic change of high-frequency energy of the fast-varying parameter during operation, and abnormal fluctuations often occur before a fault. The calculation formula is:

[0205]

[0206] where o high defines the starting index of the high-frequency band, and O is the total number of frequency components;

[0207] Perform wavelet packet decomposition on Db to obtain the energy entropy feature H of different wavelet packet sub-bands b ;

[0208] H b can measure the disorder degree of the energy distribution of the fast-varying parameter in different frequency bands. The entropy value often deviates from the normal level under fault conditions, and potential fault signals can be captured by this

[0209] At time t, construct the feature vector of the key monitoring parameter as:

[0210]

[0211] where F a，t : The spectral feature matrix obtained based on the slow-varying parameter data Da after processing, which represents the energy distribution and oscillation characteristic changes of the slow-varying parameter at different time and frequency scales at time t;

[0212] P a，t : Is the phase correlation matrix constructed based on the slow-varying parameter data Da, which is used to measure the cooperative operation law between the intrinsic mode functions of the slow-varying parameter after decomposition at the phase level;

[0213] S b，t: It is the time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-changing parameter data Db using the short-time Fourier transform, which reflects the energy amplitude distribution of the fast-changing parameters at different time points and different frequency components at time t;

[0214] represents, at time t, from the time-frequency distribution matrix S b the proportion of high-frequency band energy in different time periods is statistically calculated;

[0215] H b，t : It refers to the wavelet packet energy entropy obtained after wavelet packet decomposition of Db at time t;

[0216] Suppose the device has h non-critical monitoring parameters, and for each non-critical monitoring parameter P hj , j = 1, 2,..., h, based on the working conditions of the actual operation of the device, the relative change rate threshold is set At time t, calculate the relative change rate C hj of the working condition where each non-critical monitoring parameter P hj，t : Mark the cases exceeding the corresponding threshold ;

[0217] At the same time, carry out the Granger causality test process to determine the non-critical monitoring parameters that have a causal relationship with the key monitoring parameters, and form a subset S causal ;

[0218] For the element P hj ∈ S causal , record the causal lag order τ hj , the causal strength I hj , the fusion skewness Sk hj，t , the kurtosis Ku hj，t , and form the comprehensive feature vector of non-critical monitoring parameters as:

[0219]

[0220] To highlight its unique attribute as a potential fault inducement

[0221] can be obtained through methods such as expert scoring and historical data analysis in the existing technology;

[0222] The working conditions of the actual operation of the device include the startup stage, the starting stage, the running stage, and the stage of running with load reduction;

[0223] Through the above design of high-order spectrum and feature extraction, mining the key monitoring parameter features based on Da and Db, analyzing the device operation state information contained in the parameter changes from different angles, strengthening the ability of early fault identification and prediction, and assisting in the formulation of subsequent accurate fault prediction and preventive maintenance strategies

[0224] In the matrix generation module, a three-dimensional hypermatrix M is constructed, and its dimensions are set as:

[0225] m×(n + h)×l;

[0226] In the formula, the three dimensions include:

[0227] m: the dimension of fault types. Suppose the device has m possible fault types, denoted as F1, F2,..., F m ;

[0228] (n + h): the dimension of monitoring parameters, including n key monitoring parameters and h non-key monitoring parameters, a total of (n + h) parameters;

[0229] l: the dimension of features, including F a , P a , S b , H b , C hj，t , S causal , τ hj , I hj , Sk hj，t , Ku hj，t ;

[0230] Through the fault type F i , i = 1, 2,..., m, assign values to the matrix element m ijk m ijk In m, i represents the fault type, j represents the monitoring parameter, and k represents the feature;

[0231] The constructed three-dimensional hypermatrix fully considers the complex non-linear associations among fault types, monitoring parameters, and feature dimensions. Based on the assignment of matrix elements from massive historical data and expert experience, it accurately maps the intricate connections between different faults and parameters, features, laying a solid foundation for subsequent predictions, enabling predictions to directly hit the root cause of faults, avoiding misjudgments and missed judgments, and achieving early and accurate detection and warning of faults.

[0232] In the fault prediction module,

[0233] At the monitoring time t, splice the feature vectors and to generate a spliced feature vector

[0234] Next Perform a three-dimensional dot product operation with the three-dimensional hypermatrix M layer by layer to derive a matching score vector

[0235] The operation formula is as follows:

[0236]

[0237] This score comprehensively integrates multi-dimensional and multi-situational characteristics of multiple parameters and their complex internal relationships with each fault type, accurately reflecting the degree to which the current operating state of the equipment matches each fault type. It can capture subtle fault precursors more sensitively than traditional methods.

[0238] Set the score threshold τ. When the maximum value of the calculated S i，t is greater than τ, immediately trigger a potential fault warning.

[0239] Accurate time and type prediction:

[0240] After the potential fault warning is activated, continuously calculate the matching score vector at the predicted time t' in the future period. Build a dynamic Bayesian network model for each fault type, which can use the matching score and operating condition parameters as node variables, and update the fault probability distribution according to Bayes' theorem with new data.

[0241] As Bayes' theorem updates the fault probability distribution at each predicted time t' in the future period with new data. After N consecutive updates, assume that the of N consecutive predictions are all less than the set standard deviation threshold, and for a certain fault type F i , its predicted probability exceeds the set predicted probability threshold for N consecutive times, and at the same time, the absolute value of the difference in the matching scores for N consecutive times is less than the set threshold.

[0242] Then the earliest t' moment reached is the predicted fault occurrence time, and the fault type corresponding to the maximum probability at this time is the predicted fault type.

[0243] t' represents a series of time points starting from time t, and these time points constitute the future period.

[0244] Lock in potential hazards in advance, lay a scientific path for preventive maintenance, and ensure the reliable operation of the equipment.

[0245] This application also includes:

[0246] A method for judging the fault point and predicting the fault of a manufacturing processing equipment, including the following steps:

[0247] Divide the collected data into key monitoring parameters and non-key monitoring parameters.

[0248] When the equipment fails, conduct fault point judgment on the equipment, including the following steps:

[0249] Suppose the equipment has n key monitoring parameters, which are respectively denoted as p1, p2,..., pn ;

[0250] Suppose the device has m possible failure types, denoted as F1, F2,..., F m ;

[0251] For each key monitoring parameter P i , i = 1, 2,..., n; Set the normal operating range of each key monitoring parameter P i as [L i , U i , where L i is the lower limit value and U i is the upper limit value;

[0252] By analyzing historical failure data, construct a failure feature matrix A, and the dimension of the failure feature matrix A is m × n;

[0253] The element a ij in the failure feature matrix A represents the typical change characteristic value of the key monitoring parameter P j when the failure type F i occurs, j = 1, 2,..., m;

[0254] When the device fails, obtain the actual value of each key monitoring parameter P i at the current moment, and calculate the deviation value d i between each key monitoring parameter P i and the normal operating range, i = 1, 2,..., n; The calculation formula is as follows:

[0255]

[0256] Evaluate the similarity between the current device state and each possible failure type, evaluate the similarity S(F j ), and its calculation formula is as follows:

[0257]

[0258] where w i is the weight coefficient of each key monitoring parameter, and the adaptive weight adjustment algorithm based on the information entropy method is adopted. The weight coefficient w i is determined by the reciprocal of the information entropy;

[0259] Then the predicted failure type is:

[0260] After calculating the comprehensive evaluation function value S(F j ) for each possible failure type F j , by comparing these values, find the failure type corresponding to the maximum value;

[0261] The fault prediction is as follows:

[0262] It includes the following steps:

[0263] For n key monitoring parameters, within a set historical period, for each key monitoring parameter Pi, i , where i = 1, 2,..., n; for each key monitoring parameter Pi i construct a data set R. The data points in the data set R are arranged in chronological order. Calculate the change rate between adjacent data points. Use the clustering algorithm to divide the data set R into a slow-changing parameter subset R1 and a fast-changing parameter subset R2 according to the change rate;

[0264] At time t, for the slow-changing parameter subset R1, select the data points in the past time interval [t a , t] to form a subset R1a. Perform normalization and detrending processing on the subset R1a to obtain a new time series data Da of the slow-changing parameter data;

[0265] At time t, for the fast-changing parameter subset R2, select the data points in the past time interval [t b , t] to form a subset R2b. Perform normalization and detrending processing on the subset R2b to obtain a new time series data Db of the fast-changing parameter data;

[0266] Among them, the time length of the time interval [t a , t] is greater than the time length of the time interval [t b , t];

[0267] For the slow-changing parameter data Da, first perform Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t), and the intrinsic mode function is the IMF;

[0268] IMF a，k (t), where k = 1, 2,..., k a ; k a is the number of intrinsic mode functions obtained by decomposition;

[0269] By performing the Hilbert transform on each IMF a，k (t), obtain the corresponding instantaneous frequency f a，k (t) and instantaneous amplitude A a，k (t), and thus construct the spectral feature matrix F a as:

[0270]

[0271] k aAmong the IMFs, calculate the phase correlation coefficient matrix between each pair of IMFs. For any two IMFs, IMF a，j (t) and IMF a，k (t), where j ≠ k, the phase correlation coefficient r a (j, k) is calculated as follows:

[0272]

[0273] In the formula, and are the means of the corresponding IMFs a，j (t) and IMF a，k (t) respectively, and T is the data length;

[0274] Form the phase correlation matrix P a with the phase correlation coefficient values of all pairwise IMF combinations:

[0275]

[0276] For the fast-changing parameter data Db, perform time-frequency analysis on Db using the short-time Fourier transform to obtain the time-frequency distribution matrix S b :

[0277] S b = [S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component;

[0278] From the time-frequency distribution matrix S b , statistically calculate the proportion of high-frequency band energy in different time periods to reflect the dynamic change of high-frequency energy of the fast-changing parameter during operation. Abnormal fluctuations often occur before a fault. The calculation formula is:

[0279]

[0280] In the formula, o high defines the starting index of the high-frequency band, and O is the total number of frequency components;

[0281] Perform wavelet packet decomposition on Db to obtain the energy entropy feature H b in different wavelet packet sub-bands;

[0282] At time t, construct the feature vector of the key monitoring parameter as:

[0283]

[0284] In the formula, Fa，t : The spectral feature matrix obtained after processing the slow-varying parameter data Da represents the energy distribution and oscillation characteristic changes of the slow-varying parameters at different time and frequency scales at time t;

[0285] P a，t : It is a phase correlation matrix constructed based on the slow-varying parameter data Da, which is used to measure the collaborative operation law of each intrinsic mode function after the slow-varying parameters are decomposed at the phase level;

[0286] S b，t : It is a time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-varying parameter data Db using the short-time Fourier transform, which reflects the energy amplitude distribution of the fast-varying parameters at different time points and different frequency components at time t;

[0287] represents at time t, from the time-frequency distribution matrix S b Among them, the proportion of high-frequency band energy in different time periods is statistically calculated;

[0288] H b，t : It refers to the wavelet packet energy entropy obtained after wavelet packet decomposition of Db at time t;

[0289] Suppose the device has h non-critical monitoring parameters, and for each non-critical monitoring parameter P hj , j = 1, 2,..., h, based on the actual operating conditions of the device, set the relative change rate threshold At time t, calculate the relative change rate C of the operating conditions of each non-critical monitoring parameter P hj : Mark the situations that exceed the corresponding threshold hj，t : ;

[0290] At the same time, carry out the Granger causality test process to determine the non-critical monitoring parameters that have a causal relationship with the critical monitoring parameters, and form a subset S causal ;

[0291] For the element P hj ∈ S causal , record the causal lag order τ hj 、causal strength I hj 、fusion skewness Sk hj，t 、kurtosis Ku hj，t , and form the comprehensive feature vector of non-critical monitoring parameters is:

[0292]

[0293] Construct a three-dimensional supermatrix M, and its dimension is set to:

[0294] m×(n + j)×l;

[0295] In the formula, the three dimensions include:

[0296] m: the dimension of fault types. Suppose the device has m possible fault types, denoted as F1, F2,..., F m ;

[0297] (n + h): the dimension of monitoring parameters, including n key monitoring parameters and h non - key monitoring parameters, a total of (n + h) parameters;

[0298] l: the dimension of features, including F a 、P a 、S b 、 H b 、C hj，t 、S causal 、τ hj 、I hj 、Sk hj，t 、Ku hj，t ;

[0299] Through the fault type F i , i = 1, 2,..., m, assign values to the matrix element m ijk m ijk In, i represents the fault type, j represents the monitoring parameter, and k represents the feature;

[0300] At the monitoring time t, splice the feature vectors and to generate a spliced feature vector

[0301] Next Perform a three - dimensional dot - product operation with the three - dimensional hyper - matrix M layer by layer to derive the matching score vector for each fault type

[0302] The operation formula is as follows:

[0303]

[0304] This score comprehensively integrates multi - dimensional, multi - context features of multiple parameters and the internal complex relationships with each fault type, accurately reflecting the degree to which the current operating state of the device matches each fault type, and is more sensitive in capturing subtle fault precursors compared to traditional methods

[0305] Set the score threshold τ. When the maximum value of the calculated S i，t is greater than τ, immediately trigger a potential fault warning,

[0306] Accurate time and type prediction:

[0307] After the potential fault warning is activated, continuously calculate the matching score vector at the predicted time t' in the future time period, build a dynamic Bayesian network model for each fault type, and update the fault probability distribution according to Bayes' theorem with new data

[0308] As Bayes' theorem updates the fault probability distribution at each predicted time t' in the future time period with new data After N consecutive updates, assume that the of N consecutive predictions are all less than the set standard deviation threshold, and for a certain fault type F i , its predicted probability exceeds the set predicted probability threshold for N consecutive times, and at the same time, the absolute value of the matching score difference in N consecutive is less than the set threshold;

[0309] Then the earliest reached t' moment is the predicted fault occurrence time, and the fault type corresponding to the maximum probability at this time is the predicted fault type.

[0310] During fault prediction, by setting the key monitoring parameter data observation range adjustment module, different observation time intervals are matched for slow-changing parameters and fast-changing parameters. By setting the feature vector generation module, for key monitoring parameters, considering both slow-changing and fast-changing characteristics, the spectral feature matrix and phase correlation matrix of slow-changing parameters, and the time-frequency distribution matrix, high-frequency band energy ratio, and wavelet packet energy entropy of fast-changing parameters are extracted, etc., rich high-order spectra and collaborative features; for non-key monitoring parameters, the relative change rate, skewness, kurtosis, causal lag order, and intensity are also calculated. The feature set constructed by integrating these features can capture the subtle changes in the device operation state more comprehensively, greatly improving the fault feature recognition rate;

[0311] In the fault prediction module, when entering the fault prediction and evaluation stage, continuously update the matching score and the dynamic Bayesian network fault probability distribution according to the newly collected data, and flexibly adjust the prediction result according to the real-time situation of the device operation to ensure the timeliness and reliability of the prediction. The algorithm accurately predicts the fault occurrence time and type.

[0312] In the device fault point judgment module, when the device fails, the device fault point can be judged to find the device fault point.

[0313] In summary, when the device of the present application fails, the device fault point can be judged, and the subtle changes in the device operation state can be captured more comprehensively during fault prediction, greatly improving the fault feature recognition rate. By predicting the fault occurrence time and type, the production continuity is improved, and the risk of soaring production capacity loss and maintenance cost caused by faults is reduced. The impact of faults on the device is reduced, and the service life of the device is extended.

[0314] At the same time, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0315] In the embodiments provided by the present invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the above-described invention embodiments are merely illustrative. For example, the division of modules is only a logical function division, and there can be other division methods in actual implementation.

[0316] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0317] In addition, in each embodiment of the present invention, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0318] For those skilled in the field of operation and maintenance, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the basic features of the present invention, the present invention can be implemented in other specific forms.

[0319] As mentioned above, only the preferred specific implementation manners of the present invention are described, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent replacements or changes, and should be covered by the protection scope of the present invention.

Claims

1. A digital system for manufacturing processing equipment information, characterized in that Including: Data acquisition layer: Data is acquired through sensors, and the sensors are connected to the data acquisition terminal in a wired or wireless manner. Data acquisition terminal: It is used to receive the data acquired by the data acquisition layer, mark the acquired data as key monitoring parameters and non-key monitoring parameters respectively, and then transmit the key monitoring parameters and non-key monitoring parameters to the server. The server includes: Data storage module: It is used to receive and store the key monitoring parameters and non-key monitoring parameters transmitted by the data acquisition terminal. Device fault point judgment module: It is used to judge the device fault point by analyzing the key monitoring parameters in the data storage module. Key monitoring parameter data observation range adjustment module: It is used to classify the key monitoring parameters in the data storage module into slow-changing parameters and fast-changing parameters, select an observation time interval for each of the slow-changing parameters and fast-changing parameters, and then standardize and detrend the parameters within their respective time intervals. Feature vector generation module: By analyzing the data in the key monitoring parameter data observation range adjustment module, it generates the feature vector of the key monitoring parameters, and by analyzing the non-key monitoring parameters in the data storage module, it generates the non-key monitoring parameter feature vector. Matrix generation module: It is used to construct a three-dimensional hypermatrix from the fault type dimension, the monitoring parameter dimension, and the feature dimension of the feature vectors generated by the feature vector generation module. Fault prediction module: By analyzing the feature vector generation module and the matrix generation module, it is used to predict the future fault occurrence time point and the corresponding fault type.

2. The digitalization system for manufacturing processing equipment information according to claim 1, characterized in that In the device fault point judgment module, the device fault point is judged as follows: Suppose the device has n key monitoring parameters, denoted as p1, p2, ..., p n ; Suppose the device has m possible failure types, denoted as F1, F2, ..., F m ; For each key monitoring parameter P i , i = 1, 2, ..., n; Set the normal operating range of each key monitoring parameter P i to be [L i , U i , where L i is the lower limit value and U i is the upper limit value; By analyzing the historical fault data, a fault feature matrix A is constructed, and the dimension of the fault feature matrix A is m×n. The element a in the fault feature matrix A ij indicates that when the fault type F j occurs, for j = 1, 2,..., m, the typical change characteristic values of the key monitoring parameter P i ; When the device fails, obtain the actual value of each key monitoring parameter P at the current moment i and calculate the deviation value d i between each key monitoring parameter P i and the normal operating range, where i = 1, 2,..., n; the calculation formula is as follows: Evaluate the similarity between the current device status and each possible failure type. The similarity S(F j ) is calculated as follows: where, w i is the weight coefficient of each key monitoring parameter; Then the predicted fault type is: Calculate the comprehensive evaluation function value S(F j ) for each possible fault type F j . After comparing these values, find the fault type corresponding to the maximum value.

3. The digital system for manufacturing processing equipment information according to claim 2, characterized in that Weight coefficient w i Determination of: Adopt an adaptive weight adjustment algorithm based on the method of information entropy, with the weight coefficient w i Determined by the reciprocal of the information entropy.

4. The digitalization system for manufacturing processing equipment information according to claim 1, wherein In the key monitoring parameter data observation range adjustment module, for n key monitoring parameters, within a set historical period, for each key monitoring parameter P i , i = 1, 2,..., n; for each key monitoring parameter P i Construct a data set R. The data points in the data set R are arranged in chronological order. Calculate the change rate between adjacent data points. Use a clustering algorithm to divide the data set R into a slow-changing parameter subset R1 and a fast-changing parameter subset R2 according to the change rate; At time t, for the subset R1 of slow-varying parameters, select the data points within the past time interval [t a , t] to form the subset R1a. Standardize and detrend the subset R1a to obtain a new time series data Da of the slow-varying parameter data; At time t, for the subset R2 of fast-varying parameters, select the data points within the past time interval [t b , t] to form the subset R2b. Standardize and detrend the subset R2b to obtain a new time series data Db of the fast-varying parameter data; Among them, the time length of the time interval [t a , t] is greater than the time length of the time interval [t b , t].

5. The digital system for manufacturing processing equipment information according to claim 4, wherein In the eigenvector generation module, for the slow-varying parameter data Da, first perform the Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t), and the intrinsic mode function is IMF; IMF a，k In (t), k = 1, 2,..., k a ; k a is the number of intrinsic mode functions obtained by decomposition; By performing the Hilbert transform on each IMF a，k (t), the corresponding instantaneous frequency f a，k (t) and the instantaneous amplitude A a，k (t) are obtained, and based on this, the spectral feature matrix F a is constructed as follows: k a In the kth IMF, calculate the phase correlation coefficient matrix between each IMF. For any two IMFs, IMF a，j (t) and IMF a，k (t), where j ≠ k, the formula for its phase correlation coefficient r a (j, k) is as follows: In the formula, and are the means corresponding to IMF a，j (t) and IMF a，k (t) respectively, and T is the data length; Form a phase correlation matrix P with the phase correlation coefficient values of all pairwise IMF combinations a : For the fast-varying parameter data Db, perform time-frequency analysis on Db using the short-time Fourier transform to obtain the time-frequency distribution matrix S b : S b = [S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component; From the time-frequency distribution matrix S b calculate the proportion of high-frequency band energy in different time periods to reflect the dynamic change of high-frequency energy of the fast-changing parameter during operation. Abnormal fluctuations often occur before a fault. The calculation formula is: where, o high defines the starting index of the high-frequency band, and O is the total number of frequency components; Perform wavelet packet decomposition on Db to obtain the energy entropy feature H in different wavelet packet sub-bands b ; At time t, construct the feature vector of the key monitoring parameters It is: where, F a，t : the spectral feature matrix obtained after processing the slow-varying parameter data Da, which represents the energy distribution and the change of oscillation characteristics of the slow-varying parameter at different time and frequency scales at time t; P a，t : It is a phase correlation matrix constructed based on the slow-varying parameter data Da, and is used to measure the collaborative operation law among the intrinsic mode functions after the slow-varying parameters are decomposed at the phase level; S b，t : It is a time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-varying parameter data Db using the short-time Fourier transform, reflecting the energy amplitude distribution of the fast-varying parameters at different time points and different frequency components at time t; Denote, at time t, from the time-frequency distribution matrix S b to statistically calculate the proportion of the high-frequency band energy within different time periods; H b，t : The wavelet packet energy entropy obtained by wavelet packet decomposition of Db at time t; Suppose the device has h non-critical monitoring parameters, and for each non-critical monitoring parameter P hj , j = 1, 2,..., h, based on the operating conditions of the device during actual operation, set the relative change rate threshold At time t, calculate the relative change rate G of the operating conditions of each non-critical monitoring parameter P hj ; for the cases exceeding the corresponding threshold hj，t : mark them; ​ Meanwhile, carry out the Granger causality test process to determine the non-critical monitoring parameters that have a causal relationship with the key monitoring parameters, and form a subset S causal ; For the element P hj ∈S causal , record the causal lag order τ hj , the causal intensity I hj , the fusion skewness Sk hj，t , the kurtosis Ku hj，t , and form the comprehensive feature vector of non-critical monitoring parameters as follows:

6. The digitalization system for manufacturing processing equipment information according to claim 5, characterized in that, In the matrix generation module, a three-dimensional hypermatrix M is constructed, and its dimension is set as: m×(n + h)×l; Wherein, the three dimensions include: m: Fault type dimension. Assume that the device has m possible fault types, denoted as F1, F2,..., F m ; (n + h): Monitoring parameter dimension, including n key monitoring parameters and h non-key monitoring parameters, a total of (n + h) parameters. l: Feature dimension, including F a , P a , S b , , H b , C hj，t , S causal , τ hj , I hj , Sk hj，t , Ku hj，t ; Through fault type F i , i = 1, 2, ..., m, for matrix element m ijk assignment, m ijk where i represents the fault type, j represents the monitoring parameter, and k represents the feature.

7. The digital system for manufacturing processing equipment information according to claim 6, characterized in that, In the fault prediction module, at the monitoring time t, the splicing feature vectors and are spliced to generate a spliced feature vector Next Perform a three-dimensional dot product operation with the three-dimensional supermatrix M layer by layer to derive a matching score vector for each fault type The operation formula is as follows: Set the scoring threshold τ. When the maximum value of S i，t calculated is greater than τ, immediately trigger a potential fault warning. Accurate time and type prediction: After the potential fault warning is activated, continuously calculate the matching score vector at the predicted moment t' in the future time period, build a dynamic Bayesian network model for each fault type, and update the fault probability distribution according to Bayes' theorem with new data As the Bayesian theorem updates with new data, predicting the failure probability distribution at each future time point t' After N consecutive updates, let the of N consecutive predictions be less than the set standard deviation threshold, and for a certain failure type F i , its prediction probability exceeds the set prediction probability threshold for N consecutive times, and at the same time, the absolute value of the difference in matching scores for N consecutive times is less than the set threshold; Then the earliest reached t' moment is the predicted fault occurrence time, and at this time, the fault type with the highest probability is the predicted fault type.

8. The method for judging the fault points and predicting the faults of the manufacturing processing equipment according to any one of claims 1-7, characterized in that, Including the following steps: The acquired data is divided into key monitoring parameters and non-key monitoring parameters. When the device fails, the device fault point is judged, including the following steps: Suppose the device has n key monitoring parameters, denoted as p1, p2, ..., p n ; Suppose the device has m possible failure types, denoted as F1, F2, ..., F m ; For each key monitoring parameter P i , i = 1, 2, ..., n; set the normal operating range of each key monitoring parameter P i as [L i , U i , where L i is the lower limit value and U i is the upper limit value; By analyzing the historical fault data, a fault feature matrix A is constructed, and the dimension of the fault feature matrix A is m×n. The element a in the fault feature matrix A ij indicates that when the fault type F j occurs, for j = 1, 2,..., m, the typical change characteristic values of the key monitoring parameter P i ; When the device fails, obtain the actual value of each key monitoring parameter P i at the current moment, and calculate the deviation value d i between each key monitoring parameter P i and the normal operating range, where i = 1, 2,..., n; the calculation formula is as follows: Evaluate the similarity between the current device status and each possible failure type. Evaluate the similarity degree S(F j ), and its calculation formula is as follows: where, w i is the weight coefficient of each key monitoring parameter, and an adaptive weight adjustment algorithm based on information entropy is adopted. The weight coefficient w i is determined by the reciprocal of the information entropy; Then the predicted fault type is: Calculate the comprehensive evaluation function value S(F j ) for each possible fault type F j . After comparing these values, find the fault type corresponding to the maximum value; Fault prediction is: Including the following steps: For n key monitoring parameters, within a set historical period, for each key monitoring parameter P i , i = 1, 2, ..., n; for each key monitoring parameter P i Construct a data set R. The data points in the data set R are arranged in chronological order. Calculate the change rate between adjacent data points. Use a clustering algorithm to divide the data set R into a slow-changing parameter subset R1 and a fast-changing parameter subset R2 according to the change rate; At time t, for the subset R1 of the slow-varying parameters, select the data points in the past time interval [t a , t] to form the subset R1a. Standardize and detrend the subset R1a to obtain a new time series data Da of the slow-varying parameter data; At time t, for the subset R2 of fast-varying parameters, select the data points within the past time interval [t b , t] to form the subset R2b. Standardize and detrend the subset R2b to obtain a new time series data Db of the fast-varying parameter data; wherein, the time length of the time interval [t a , t] is greater than the time length of the time interval [t b , t]; For the slow-varying parameter data Da, first perform the Hilbert-Huang transform to decompose Da into multiple intrinsic mode functions, denoted as IMF a，k (t) and a residual term R a (t), where the intrinsic mode functions are the IMFs; IMF a，k In (t), k = 1, 2, ..., k a ; k a is the number of intrinsic mode functions obtained by decomposition; By performing the Hilbert transform on each IMF a，k (t), the corresponding instantaneous frequency f a，k (t) and the instantaneous amplitude A a，k (t) are obtained, and based on this, the spectral feature matrix F a is constructed as follows: k a Among the IMFs, calculate the phase correlation coefficient matrix between each IMF. For any two IMFs, IMF a，j (t) and IMF a，k (t), j≠k, the formula for its phase correlation coefficient r a (j, k) is as follows: wherein, and are the means corresponding to IMF a，j (t) and IMF a，k (t) respectively, and T is the data length; Form a phase correlation matrix P with the phase correlation coefficient values of all pairwise IMF combinations a : For the fast-varying parameter data Db, perform time-frequency analysis on Db using the short-time Fourier transform to obtain the time-frequency distribution matrix S b : S b = [S b (o, u)], where o represents the frequency index and u represents the time index, and the matrix element S b (o, u) represents the energy amplitude at the u-th time point and the o-th frequency component; From the time-frequency distribution matrix S b statistically analyze the proportion of high-frequency band energy in different time periods to reflect the dynamic change of high-frequency energy of the fast-changing parameter during operation. Abnormal fluctuations often occur before a fault. The calculation formula is as follows: where o high defines the starting index of the high-frequency band, and O is the total number of frequency components; Perform wavelet packet decomposition on Db to obtain the energy entropy feature H in different wavelet packet sub-bands b ; At time t, construct the feature vector of the key monitoring parameters It is: Where, F a，t : The spectral feature matrix obtained after processing the slow-varying parameter data Da, which represents the energy distribution and the change of oscillation characteristics of the slow-varying parameters at different time and frequency scales at time t; P a，t : It is a phase correlation matrix constructed based on the slow-varying parameter data Da, and is used to measure the collaborative operation law among the intrinsic mode functions after the slow-varying parameters are decomposed at the phase level; S b，t : It is the time-frequency distribution matrix obtained by performing time-frequency analysis on the fast-varying parameter data Db using the short-time Fourier transform, which reflects the energy amplitude distribution of the fast-varying parameters at different time points and different frequency components at time t; Representing at time t, from the time-frequency distribution matrix S b In it, calculate the proportion of the high-frequency band energy in different time periods; H b，t : The wavelet packet energy entropy obtained by wavelet packet decomposition of Db at time t; Suppose the device has h non-critical monitoring parameters, and for each non-critical monitoring parameter P hj , j = 1, 2,..., h, based on the operating conditions of the device during actual operation, set the relative change rate threshold At time t, calculate the relative change rate C of the operating conditions of each non-critical monitoring parameter P hj : Mark the cases that exceed the corresponding threshold hj，t ; ​ Meanwhile, the Granger causality test process is carried out to identify non-critical monitoring parameters that have a causal relationship with critical monitoring parameters, forming a subset S causal ; For the element P hj ∈S causal , record the causal lag order τ hj , the causal intensity I hj , the fusion skewness Sk hj，t , the kurtosis Ku hj，t , and form the comprehensive feature vector of non-critical monitoring parameters as follows: Construct a three-dimensional hypermatrix M, and its dimension is set as: m×(n + h)×l; Wherein, the three dimensions include: m: Fault type dimension. Suppose the device has m possible fault types, denoted as F1, F2, ..., F m ; (n + h): Monitoring parameter dimension, including n key monitoring parameters and h non-key monitoring parameters, a total of (n + h) parameters. l: Feature dimension, including F a , P a , S b , , H b , C hj，t , S causal , τ hj , I hj , Sk hj，t , Ku hj，t ; Through fault type F i , i = 1, 2, ..., m, for matrix element m ijk assign a value, m ijk where i represents the fault type, j represents the monitoring parameter, and k represents the feature; At the monitoring moment t, splice the feature vectors and to generate a spliced feature vector Next Perform a three-dimensional dot product operation with the three-dimensional supermatrix M layer by layer to derive a matching score vector for each fault type The operation formula is as follows: This score comprehensively integrates multi-dimensional, multi-situation features of multiple parameters and their internal complex correlations with each fault type, accurately reflects the degree to which the current operating state of the device fits each fault type, and can capture subtle fault precursors more sensitively than traditional methods. Set a scoring threshold τ. When the maximum value of the calculated S i，t is greater than τ, immediately trigger a potential fault warning; Accurate time and type prediction: After the potential fault warning is activated, continuously calculate the matching score vector at the predicted time t' in the future period, build a dynamic Bayesian network model for each fault type, and update the fault probability distribution according to Bayes' theorem with new data. As the Bayesian theorem updates with new data, predicting the failure probability distribution at each future time period t' After N consecutive updates, assume that for N consecutive predictions are all less than the set standard deviation threshold, and for a certain failure type F i , its prediction probability exceeds the set prediction probability threshold for N consecutive times, and at the same time, for N consecutive times the absolute value of the matching score difference is less than the set threshold; Then the earliest achieved time t' is the predicted fault occurrence time, and the fault type corresponding to the maximum probability at this time is the predicted fault type.

Images