Health prediction method of tobacco drying equipment in cigarette factories based on big data

Abstract

The present invention relates to the technical field of data processing, and discloses a health prediction method for tobacco drying machine equipment in a cigarette factory based on big data, comprising the steps of: collecting operation data of tobacco drying machine equipment; judging whether the operation data exceeds a set threshold, and if so, inputting an intelligent fault diagnosis model to predict the fault type; inputting the operation data into an unsupervised learning model, judging in real time whether a new data classification appears, and if a new data classification appears, an expert identifies it, and if it is identified as a new fault type, the new fault type is stored in a sample data storage device; judging whether the operation data exceeds a set threshold, and if so, triggering a mechanism to obtain a mechanism result, and an expert identifies whether the mechanism result is consistent with the fault type predicted by the intelligent fault diagnosis model, and if not, and the identification mechanism result is a fault type, the mechanism result is stored as a new fault type in the sample data storage device. The present invention can improve the fault prediction accuracy of the intelligent fault diagnosis model for tobacco drying machine equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a health prediction method for tobacco factory toner equipment based on big data. Background Art

[0002] The tobacco drying process is a key link in the production of tobacco. The process of tobacco drying is very complicated. There are many factors that affect the quality and production efficiency of the final tobacco products, such as hot air temperature, barrel wall temperature, outlet moisture and other production process and equipment parameters. Therefore, it is particularly important to monitor the production parameters and the equipment operation status. At the same time, the tobacco drying machine is a key equipment on the cigarette production line, and its operation status directly affects the quality and production efficiency of cigarettes.

[0003] From a management perspective, traditional maintenance of tobacco drying equipment mainly relies on regular inspections and post-maintenance, which requires extremely high personal experience of maintenance personnel and is easily affected by data limitations, system performance, and single technology. Therefore, traditional methods are not only inefficient and time-effective, but also difficult to detect and prevent possible equipment failures in advance, affecting cigarette quality and production efficiency. From a technical perspective, the indicators that reflect the health status of equipment are often not single, and a single indicator cannot be used to judge the health status of tobacco drying equipment under different working conditions.

[0004] Therefore, it is necessary to predict the health status of tobacco drying equipment in order to achieve accurate and timely equipment maintenance and fault prevention, and improve cigarette quality and production efficiency. Summary of the invention

[0005] The purpose of the present invention is to improve the fault prediction accuracy of the intelligent fault diagnosis model for tobacco drying machine equipment, and to provide a health prediction method for tobacco drying machine equipment in a cigarette factory based on big data.

[0006] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0007] The health prediction method of tobacco factory drying machine equipment based on big data includes the following steps:

[0008] Step 1, collecting operation data of the tofu drying machine;

[0009] Step 2, determining whether the collected operating data exceeds a set threshold, and if so, inputting the operating data into an intelligent fault diagnosis model to predict the type of fault that occurs in the tofu drying machine; the intelligent fault diagnosis model is trained by the historical operating data and fault types stored in the sample data storage;

[0010] Step 3: Input the collected operation data into the unsupervised learning model to determine in real time whether a new data classification appears. If a new data classification appears, it will be identified by experts. If the new data classification is identified as a new fault type, the new fault type will be stored in the sample data storage;

[0011] Step 4, determine whether the collected operating data exceeds the set threshold. If so, trigger the mechanism diagnosis to obtain the mechanism result. The expert will identify whether the mechanism result is consistent with the fault type predicted by the intelligent fault diagnosis model. If not, and the mechanism result is identified as a fault type, the mechanism result is stored as a new fault type in the sample data storage.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] (1) The present invention trains an intelligent fault diagnosis model based on a supervised learning model using historical operating data of the tofu drying machine and corresponding identified fault types, and can predict the fault type of the tofu drying machine based on the operating data.

[0014] (2) The present invention first determines whether the operating data exceeds a set threshold. If it does not exceed the set threshold, fault prediction is not performed, thereby saving computing costs.

[0015] (3) The present invention uses an unsupervised learning model to learn in real time whether there are new fault types in the operating data. If a new fault type appears, the intelligent fault diagnosis model is retrained using the updated training data to improve the prediction accuracy of the intelligent fault diagnosis model.

[0016] (4) The present invention compares the fault type obtained by triggering mechanism diagnosis with the fault type predicted by the intelligent fault diagnosis model, thereby further improving the prediction accuracy of the intelligent fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0020] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0021] The present invention is achieved through the following technical solutions: Figure 1 As shown, the health prediction method of tobacco factory drying machine equipment based on big data includes the following steps:

[0022] Step 1: Collect the operation data of the tofu drying machine.

[0023] The operation data of the tofu drying machine equipment includes but is not limited to bearing seat vibration, bearing seat temperature, motor operating current, motor operating power, motor operating energy consumption, motor speed, and drum speed. The bearing seat is the key part of the tofu drying machine equipment. The bearing seat includes the drum support bearing seat and the motor drive bearing seat. The vibration sensor is used to collect the bearing seat vibration. Because the speed ratio of the motor and the drum is very large, the vibration sensor needs to be customized with ultra-low frequency response and ultra-long time data collection.

[0024] Parameters such as roller support bearing seat temperature (T1), motor drive broken bearing seat temperature (T2), motor operating current (I), motor operating power (P), motor operating energy consumption (E), motor speed (rpm1), roller speed (rpm2) can be directly collected and used. For the data collection characteristics of roller support bearing seat vibration and motor drive broken bearing seat vibration, the amount of data collected will be very large. If it is directly used as the input of machine learning, it will seriously affect the performance of the model, and a large amount of noise information will also affect the effect of machine learning. Therefore, it is necessary to perform feature extraction through preprocessing to reduce the dimension, and extract features from two aspects: time domain indicators and frequency indicators.

[0025] Extract the time domain indicators of bearing seat vibration: peak value (peak), intensity (rms), skewness (skew), kurtosis (kurt), waveform index (SH), pulse index (IM), peak index (CR), margin index (CL).

[0026] Extract the frequency index of bearing seat vibration: center of gravity frequency (FC), mean square frequency (MSF), sub-frequency amplitude (X 0_1_0_4 , X 0_4_0_8 ), power frequency amplitude and its harmonics (X1, X2, X3, X4, X5, X6, X7, X8), bearing inner ring fault frequency and its harmonic amplitude (BPFI1, BPFI2, BPFI3, BPFI4, BPFI5, BPFI6, BPFI7, BPFI8), bearing outer ring fault and its harmonic amplitude (BPFO1, BPFO2, BPFO3, BPFO4, BPFO5, BPFO6, BPFO7, BPFO8), roller body fault frequency and its harmonic amplitude (BSF1, BSF2, BSF3, BSF4, BSF5, BSF6, BSF7, BSF8), cage fault frequency and its harmonic amplitude (TFT1, TFT2, TFT3, TFT4, TFT5, TFT6, TFT7, TFT8), gear meshing frequency and its harmonic amplitude (GMZ1, GMZ2, GMZ3, GMZ4, GMZ5, GMZ6, GMZ7, GMZ8), high frequency amplitude (HF). However, depending on the model of the tofu drying machine or after the equipment is upgraded, the collected operating data is not limited thereto. To facilitate the understanding of the subsequent technical solutions, it is assumed that the above 68 operating data are collected in this embodiment.

[0027] Step 2, determine whether the collected operating data exceeds the set threshold. If so, input the operating data into the intelligent fault diagnosis model to predict the type of fault that occurs in the tofu drying machine equipment; the intelligent fault diagnosis model is trained by the historical operating data and fault types stored in the sample data storage.

[0028] Since the collection and transmission of the 68 operating data are not performed at the same time, the collected operating data are input into the data synchronizer for forced time alignment to ensure the reliability and integrity of the operating data.

[0029] Then, the operation data output by the data synchronizer is assembled into tensors and stored in a sample data storage device, wherein the sample data storage device stores not only the historical operation data of the tofu drying machine equipment, but also the fault types corresponding to the historical operation data.

[0030] The intelligent fault diagnosis model is constructed based on a supervised learning model and is trained by historical operation data of the tofu drying machine equipment and corresponding identified fault types, and these historical operation data and fault types are from the sample data storage.

[0031] After the operation data is assembled into tensors, it is stored in the data storage device and judged whether each operation data exceeds the preset threshold. If it exceeds, the intelligent fault diagnosis model is triggered to obtain this batch of operation data, and then the type of fault of the tofu drying machine is predicted, so that the staff can perform maintenance in advance. If it does not exceed the set threshold, it means that the operation data is normal, and there is no need to predict the intelligent fault diagnosis model, which saves computing costs.

[0032] Step 3, input the collected operating data into the unsupervised learning model to determine in real time whether new data classifications appear. If new data classifications appear, they will be identified by experts. If the new data classifications are identified as new fault types, the new fault types will be stored in the sample data storage.

[0033] After the new operation data is assembled by the data synchronizer and tensor and stored in the sample data storage, it must be synchronously input into the unsupervised learning model regardless of whether the operation data exceeds the set threshold. The unsupervised learning model uses a mixed Gaussian model to expand the Gaussian distribution to multi-dimensional space, taking each operation data as a feature vector of one dimension, and expanding the feature vector to D dimensions, namely f1, f2, ..., f D For example, the first dimension f1 is the eigenvector of the peak value, the second dimension f2 is the eigenvector of the intensity (rms), and so on, D = 68. And each dimension of the multidimensional space is completely independent, so an independent multivariate Gaussian distribution function is established:

[0034]

[0035] in, represents the multivariate Gaussian distribution function; represents a feature vector (f1,f2,...,f D ), is the multidimensional data point to be modeled; represents the average value of the D-dimensional eigenvector, describing the center position or average position of the multivariate Gaussian distribution; ψ represents the covariance matrix of all dimensions, reflecting the correlation and discreteness between the eigenvectors; |ψ| represents the determinant value of the covariance matrix ψ; e represents the base of the natural logarithm; T represents the matrix transpose; Reflected and The degree of deviation and the influence of the correlation between eigenvectors. The covariance matrix ψ is:

[0036]

[0037] The element in row d and column d in ψ represents f d Covariance with itself, d=1,2,...,D.

[0038] If there is a correlation between different dimensions, it is called a correlated multivariate Gaussian distribution. Compared with the multivariate Gaussian distribution, ψ is no longer the covariance between the eigenvector and itself only in the diagonal, but the covariance between different eigenvectors:

[0039]

[0040] Among them, V(f d ) indicates f d The covariance with itself, Con(f d ,f d` ) indicates f d With f d` The covariance of , d=1,2,...,D, d`=1,2,...,D, d≠d`.

[0041] Since there is a correlation between the extracted feature vectors, the impact on the tofu drying machine equipment needs to be analyzed through the correlation between the feature vectors. Therefore, this solution uses the correlated multivariate Gaussian distribution. The Gaussian mixture model (GMM) is constructed using the expectation maximization algorithm (EM). For all feature vectors, a large number of samples are obtained to first solve the optimal category distribution probability:

[0042]

[0043] in, Indicates the weight of the i-th sample belonging to the c-th category, c = 1, 2, ..., C, C is the total number of fault types in the sample data memory; N is the total number of samples, one type of operation data contains multiple samples, for example, a peak may have multiple data;

[0044] Represents the feature vector of the i-th sample;

[0045] α represents the classification space, Q i (α=c) represents the probability that the i-th sample belongs to the c-th class;

[0046] represents the joint probability that the i-th sample belongs to the c-th class;

[0047] Indicates that a sample appears The probability of failure type c occurring when ;

[0048] Indicates that in the classification space α, the sample The sum of the probabilities of belonging to each fault type.

[0049] Find Q i (α=c) and then find the extreme value of the following formula to obtain the updated parameters:

[0050]

[0051] in, represents the probability of the i-th sample; P(α=c) represents the prior probability of the c-th class, represents the mean vector of the cth class; represents the variance of the cth class; Represents the deviation of the i-th sample from the mean vector of the c-th class;

[0052] Represents the probability of the entire classification space; σ 2 represents the sample variance; Represents the sample mean.

[0053] When finding the extreme value, irrelevant running data is directly discarded, and the parameters to be solved are The derivatives are 0:

[0054]

[0055] right When solving, first All irrelevant operating data is removed, and then As the ratio of each fault type, there is a natural constraint that all fault types The sum is 1, so the objective function is:

[0056]

[0057] Using the Lagrange multiplier method:

[0058]

[0059] in, represents the Lagrangian function; β represents the Lagrangian multiplier; represents the prior probability of the ith sample.

[0060] To facilitate calculation, we first Derivative, and then for the fault type Derivative:

[0061]

[0062]

[0063] Both sides sum c from 1 to C, That item is gone, the formula It becomes the number of samples N, and β can be obtained, and then β is substituted back You can get the parameters Update:

[0064]

[0065] -β=N

[0066]

[0067] So far, all parameters The update is complete, and the iteration is repeated until convergence.

[0068] If the unsupervised learning model identifies a new data classification, an expert will identify the cause of the new data classification. If it is determined to be a possible fault type of the tofu drying machine equipment, the new fault type will be stored in the sample data storage device, and the intelligent fault diagnosis model will be re-trained using the fault type and the operating data of the same period to improve the universality of the intelligent fault diagnosis model.

[0069] Step 4, determine whether the collected operating data exceeds the set threshold. If so, trigger the mechanism diagnosis to obtain the mechanism result. The expert will identify whether the mechanism result is consistent with the fault type predicted by the intelligent fault diagnosis model. If not, and the mechanism result is identified as a fault type, the mechanism result is stored as a new fault type in the sample data storage.

[0070] After the new operation data passes through the data synchronizer, it is determined whether each operation data exceeds the preset threshold. If it exceeds, the mechanism diagnosis is triggered to obtain the mechanism result, and the expert identifies whether the mechanism result is consistent with the fault type predicted by the intelligent fault diagnosis model. If it is consistent, it means that the intelligent fault diagnosis model is effective and reliable, and its prediction results are highly accurate; if it is inconsistent, it means that the universality of the intelligent fault diagnosis model needs to be improved, and the mechanism result is stored as a new fault type in the sample data storage. The intelligent fault diagnosis model is retrained using the fault type and the operation data of the same period to improve the universality of the intelligent fault diagnosis model and the accuracy of the prediction results.

[0071] It should be noted that there is no strict execution order for step 2, step 3, and step 4, and they can be executed in parallel.

[0072] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A health prediction method for tobacco factory drying equipment based on big data, characterized by: The following steps are involved: Step 1, collecting operation data of the tofu drying machine; Step 2, determining whether the collected operating data exceeds a set threshold, and if so, inputting the operating data into an intelligent fault diagnosis model to predict the type of fault that occurs in the tofu drying machine; the intelligent fault diagnosis model is trained by the historical operating data and fault types stored in the sample data storage; Step 3: Input the collected operation data into the unsupervised learning model to determine in real time whether a new data classification appears. If a new data classification appears, it will be identified by experts. If the new data classification is identified as a new fault type, the new fault type will be stored in the sample data storage; The step 3 specifically comprises the following steps: Assume that D types of operation data are collected, and each type of operation data is taken as a feature vector of one dimension, namely f1, f2, ..., f D ; Establish a multivariate Gaussian distribution function: in, represents the multivariate Gaussian distribution function; represents a feature vector of dimension D; represents the average value of the D-dimensional eigenvector; ψ represents the covariance matrix of all dimensions; |ψ| represents the determinant value of the covariance matrix ψ; e represents the base of the natural logarithm; T represents the matrix transpose; the covariance matrix ψ is: Among them, V(f d ) indicates f d The covariance with itself, Con(f d ,f d` ) indicates f d With f d` covariance, d = 1, 2, ..., D, d` = 1, 2, ..., D, d ≠ d`; Solve for the category distribution probability: in, Indicates the weight of the i-th sample belonging to the c-th category, c = 1, 2, ..., C, C is the total number of fault types in the sample data memory; N is the total number of samples; Represents the feature vector of the i-th sample; α represents the classification space, Q i (α=c) represents the probability that the i-th sample belongs to the c-th class; represents the joint probability that the i-th sample belongs to the c-th class; Indicates that a sample appears The probability of failure type c occurring when ; Indicates that in the classification space α, the sample The sum of the probabilities of belonging to each fault type; Find Q i (α=c) and then find the extreme value of the following formula to obtain the updated parameters: in, represents the probability of the i-th sample; P(α=c) represents the prior probability of the c-th class, represents the mean vector of the cth class; represents the variance of the cth class; Represents the deviation of the i-th sample from the mean vector of the c-th class; Represents the probability of the entire classification space; σ 2 represents the sample variance; represents the sample mean; When finding the extreme value, irrelevant running data is discarded, and the parameters to be solved are The derivatives are 0: The objective function is: Using the Lagrange multiplier method: in, represents the Lagrangian function; β represents the Lagrangian multiplier; Represents the prior probability of the i-th sample; For fault type Derivative: Sum c from 1 to C on both sides to get the parameter Update: -β＝N If the unsupervised learning model identifies a new data classification, it is identified by an expert. If the new data classification is identified as a new fault type, the new fault type is stored in the sample data storage, and the intelligent fault diagnosis model is newly trained; Step 4, determine whether the collected operating data exceeds the set threshold. If so, trigger the mechanism diagnosis to obtain the mechanism result. The expert will identify whether the mechanism result is consistent with the fault type predicted by the intelligent fault diagnosis model. If not, and the mechanism result is identified as a fault type, the mechanism result is stored as a new fault type in the sample data storage.

2. The health prediction method for tobacco factory toner equipment based on big data according to claim 1 is characterized by: In step 1, the collected operation data includes: Drum support bearing seat temperature, motor drive bearing seat temperature, motor operating current, motor operating power, motor operating energy consumption, motor speed, drum speed; And the time domain indicators of bearing seat vibration: peak value, intensity, skewness, kurtosis, waveform index, pulse index, peak index, margin index; And the frequency indicators of bearing seat vibration: center of gravity frequency, mean square frequency, subfrequency amplitude, power frequency amplitude and its harmonics, bearing inner ring fault frequency and its harmonic amplitude, bearing outer ring fault and its harmonic amplitude, roller body fault frequency and its harmonic amplitude, cage fault frequency and its harmonic amplitude, gear meshing frequency and its harmonic amplitude, high frequency amplitude.

3. The health prediction method for tobacco factory toner equipment based on big data according to claim 1 is characterized by: In the step 2, before the step of determining whether the collected operation data exceeds the set threshold, the step further includes: assuming that D types of operation data are collected, inputting the collected operation data into a data synchronizer for forced time alignment; Then, the running data output by the data synchronizer is assembled into tensors and stored in the sample data memory.

Images