CN110717472B - Fault diagnosis method and system based on improved wavelet threshold denoising - Google Patents
Fault diagnosis method and system based on improved wavelet threshold denoising Download PDFInfo
- Publication number
- CN110717472B CN110717472B CN201910987108.7A CN201910987108A CN110717472B CN 110717472 B CN110717472 B CN 110717472B CN 201910987108 A CN201910987108 A CN 201910987108A CN 110717472 B CN110717472 B CN 110717472B
- Authority
- CN
- China
- Prior art keywords
- coefficient
- layer
- data
- wavelet
- skewness
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Complex Calculations (AREA)
Abstract
The present disclosure discloses a fault diagnosis method and system for improved wavelet threshold denoising, comprising: acquiring TE process data to be diagnosed; carrying out standardization processing on the acquired data; carrying out wavelet transform decomposition on the data after the standardization treatment, and decomposing the data into a plurality of layers to obtain wavelet coefficients of each layer; calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data; and analyzing the data subjected to noise reduction by using the PCA model, and judging whether fault data exist in the TE process data to be diagnosed.
Description
Technical Field
The disclosure relates to the technical field of TE process fault diagnosis, in particular to a fault diagnosis method and system based on improved wavelet threshold denoising.
Background
The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.
Since twenty-first century, with the rapid development of science and technology, modern industrial systems become large and complex, and the safety and stability of the production process need to be improved, so that faults need to be detected accurately and timely. Multivariate statistical fault diagnosis methods are commonly used because they do not require knowledge of professional process knowledge, do not require complex mathematical models, and rely solely on collected industrial process data. The principal component analysis method is a powerful technique for extracting data structures from high-dimensional data, and is widely used for process monitoring and fault diagnosis due to its ability to process high-dimensional, noisy, and highly correlated data. However, the PCA method can analyze data only on a single scale, and cannot be used for processing multi-scale process data, which has great limitations. In real-life industrial processes, information about faults is randomly distributed on different scales due to the degree, orientation and cause of the fault. Multi-scale principal component analysis (MSPCA) extends the single-scale modeling approach to PCA to multiple scales. The multi-scale principal component analysis algorithm performs wavelet transformation on original data, then performs principal component analysis on each scale, selects low-frequency components possibly containing fault information through proper control limits, filters noise of high-frequency components, and finally performs signal reconstruction.
Due to sensor noise, interference, instrument degradation, human error and the like, actual production data of a chemical process inevitably contains random and noise. The noisy data interferes with the data analysis and can seriously affect the accuracy of fault diagnosis. Wavelet denoising is typically used to remove these noises. The wavelet threshold denoising method generally employs a hard threshold method and a soft threshold method to perform threshold processing on wavelet coefficients. The traditional hard threshold method can cause the denoised data to generate additional oscillation because of discontinuity at a demarcation point; although the traditional soft threshold method has good continuity, the extreme value is eliminated, so that the estimated wavelet coefficient and the noisy wavelet coefficient have constant deviation, and the problem of signal distortion after denoising is caused.
The selection of wavelets and the selection of thresholds in the wavelet threshold denoising method are difficult points and difficulties in application. Commonly used wavelets include the bior, db, and svm wavelets. Based on historical experience and experimental effect, we select db5 wavelet basis function to perform denoising processing on the original data.
The classical general threshold formula is as follows:
where n is the data length and σ is the noise standard deviation.
Conventional thresholding methods have both hard and soft thresholds. Their main idea is to delete wavelet coefficients of small amplitude and wavelet coefficients of larger contraction. The hard thresholding method compares the absolute value of the data to a threshold, with data points less than or equal to the threshold set to 0 and other data points held constant. The functional expression of the hard threshold is:
where λ (i) is the threshold and cd (i) is the input signal.
The soft threshold is a comparison of the absolute value of the data to a specified threshold. Data points less than or equal to the threshold are set to 0, while other data points need to be shrunk toward 0, setting the difference of this data point minus the threshold. The functional expression of the soft threshold method is:
the hard threshold value method and the soft threshold value method are easy to calculate, high in efficiency and widely applied in practice. However, there are also some disadvantages in these processes: for example, the soft threshold method has a constant deviation, which can cause signal distortion; the hard thresholding denoised signal is not smooth and produces additional oscillations.
The conventional threshold function has a fixed threshold value on each layer of wavelet coefficients after wavelet decomposition, which causes an "over-killing" phenomenon on the valid data signal. In practical application, it is generally determined to adopt a soft threshold method or a hard threshold method alone according to experimental effects. However, the selection of a proper threshold function according to the characteristics of wavelet coefficients of each layer of the signal is not considered, the flexibility is lacked, and the advantages of a soft threshold method and a hard threshold method cannot be comprehensively utilized.
In the course of implementing the present disclosure, the inventors found that the following technical problems exist in the prior art:
(1) The traditional threshold method fixedly adopts a soft threshold method or a hard threshold method on each scale, does not consider to select a proper threshold function according to the characteristics of wavelet coefficients of each layer, is lack of flexibility and cannot comprehensively utilize the advantages of the soft threshold method and the hard threshold method.
(2) Actual data obtained in an actual industrial production process are multi-scale in nature, so the PCA fault diagnosis method cannot be applied to fault diagnosis research of the multi-scale data. The multi-directional principal component analysis method overcomes the defect that the traditional PCA method can only diagnose sample data on a single scale. The traditional multi-scale principal component analysis method has the defects of poor analysis and detection capability on process change and high susceptibility to noise data and random error interference.
Disclosure of Invention
In order to solve the deficiencies of the prior art, the present disclosure provides a fault diagnosis method and system based on improved wavelet threshold denoising;
in a first aspect, the present disclosure provides a fault diagnosis method based on improved wavelet threshold denoising;
the fault diagnosis method based on the improved wavelet threshold denoising comprises the following steps:
acquiring data of a TE Process (Tennessee Eastman Process, wherein the TE Process is a Process for truly simulating a chemical Process) to be diagnosed; carrying out standardization processing on the acquired data;
performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data;
establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; and if the statistic of the data to be diagnosed exceeds the set control limit, indicating that fault data exist in the TE process data to be diagnosed.
In a second aspect, the present disclosure also provides a fault diagnosis system based on improved wavelet threshold denoising;
a fault diagnosis system based on improved wavelet threshold denoising comprises:
an acquisition module configured to: acquiring data of a TE Process (a Tennessee Eastman Process, wherein the TE Process is a Process for truly simulating a chemical Process) to be diagnosed; carrying out standardization processing on the acquired data;
a wavelet transform module configured to: performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
a noise reduction module configured to: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer accords with normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data;
a statistics calculation module configured to: establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
a fault diagnosis module configured to: comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; and if the statistic of the data to be detected exceeds the set control limit, indicating that fault data exists in the TE process data to be diagnosed.
In a third aspect, the present disclosure also provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method of the first aspect are completed.
In a fourth aspect, the present disclosure also provides a computer-readable storage medium for storing computer instructions which, when executed by a processor, perform the steps of the method of the first aspect.
Compared with the prior art, the beneficial effect of this disclosure is:
each layer of wavelet coefficients is denoised by introducing a normal distribution test to select soft threshold and hard threshold methods. And then combining the improved adaptive wavelet threshold denoising method with a multi-scale principal component analysis method to provide a fault diagnosis strategy based on the improved multi-scale principal component analysis (IMSPCA). Firstly, wavelet threshold denoising is improved by utilizing normal distribution inspection, then an improved wavelet threshold denoising method is combined with multi-scale principal component analysis, a comprehensive scale PCA model is established for fault diagnosis, and the reliability of fault diagnosis is improved. Finally, the feasibility and the effectiveness of the IMSPCA method are verified through simulation research of the TE process.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
FIG. 1 is a flow chart of a method of the first embodiment;
FIG. 2 is a flow chart of a modeling algorithm of the improved multi-scale principal component analysis of the first embodiment;
FIG. 3 is a fault diagnosis process based on the improved MSPCA according to the first embodiment;
FIG. 4 is a process flow diagram of the TE-based process of the first embodiment;
5 (a) -5 (h) are comparison of fault diagnosis simulation results of the fault 5 based on the conventional soft threshold method in combination with the principal component analysis method and the improved IMSPCA method according to the first embodiment, including a variable fault contribution diagram of a fault detection diagram;
6 (a) -6 (h) are comparison of fault diagnosis simulation results of the fault 7 based on the conventional soft threshold method in combination with the principal component analysis method and the improved IMSPCA method according to the first embodiment, including a variable fault contribution diagram of a fault detection diagram;
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Interpretation of terms:
fault Diagnosis algorithm (Fault Diagnosis) -Fault Diagnosis refers to detecting, separating and identifying faults occurring in a system, namely judging whether the faults occur, positioning the positions and the types of the faults, determining the size and the occurrence time of the faults and the like.
Wavelet threshold denoising algorithm (WT) -the basic idea of the WT method is to select the generated wavelet coefficients after the signal is wavelet transformed. Because the wavelet coefficient of the signal is larger after wavelet decomposition and the wavelet coefficient of the noise is smaller, by selecting a proper threshold value, the wavelet coefficient larger than the threshold value is considered to be generated by the signal and should be reserved, and the wavelet coefficient smaller than the threshold value is considered to be generated by the noise and is set as zero, so that the purpose of denoising is achieved.
The multi-scale principal component analysis algorithm (MSPCA) -the MSPCA method combines the capability advantages of principal component analysis for removing correlation among variables and wavelet transformation for extracting local characteristics of the variables and approximate decomposition variable autocorrelation, decomposes noise into a plurality of scales, establishes a PCA model on each scale, sets a threshold value, screens out a low-frequency part possibly containing fault information, removes high-frequency noise, and enables reconstructed data to be smooth.
In a first aspect, the present disclosure provides a fault diagnosis method based on improved wavelet threshold denoising;
as shown in fig. 1, fig. 2 and fig. 3, the fault diagnosis method based on improved wavelet threshold denoising includes:
s1: acquiring data of a TE Process (a Tennessee Eastman Process, wherein the TE Process is a Process for truly simulating a chemical Process) to be diagnosed; carrying out standardization processing on the acquired data;
s2: performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
s3: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer;
judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method;
after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data;
s4: establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
s5: comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; and if the statistic of the data to be detected exceeds the set control limit, indicating that fault data exists in the TE process data to be diagnosed.
As one or more embodiments, the method further comprises:
if the primary diagnosis result of the TE process data to be diagnosed in the S5 is that fault data exists, inputting the statistic output value of each PCA model into a total PCA model to obtain total statistic, comparing the total statistic with the set total control limit, and if the total statistic is smaller than the set total control limit, indicating that the primary diagnosis result is wrong and not giving an alarm; if the total control limit is larger than the set total control limit, the preliminary diagnosis result is correct, and an alarm is given.
As one or more embodiments, the method further comprises:
s6: identifying fault types by using the contribution graph;
s7: and determining a fault source according to the variable with the maximum contribution value corresponding to the statistic.
As one or more embodiments, in S1, the TE process data to be diagnosed includes: operating parameters of the reactor, condenser, circulator, compressor, separator and stripper.
As one or more embodiments, in S1, the normalizing the acquired data; the method comprises the following specific steps: the Z-score normalization process normalizes the data. The Z-score method is such that the standard data is equal to the data minus the mean of the data divided by the standard deviation of the data.
A normalization processing formula:
as one or more embodiments, in S2, the normalized data is decomposed into a plurality of layers by performing wavelet decomposition; the method comprises the following specific steps:
adopting a wavelet base function db5;
determining the optimal decomposition layer number m by adopting a mode maximum value mode;
and performing wavelet decomposition on the normalized data to decompose the normalized data into m layers.
As one or more embodiments, in S3, for the wavelet coefficients of each layer, a skewness coefficient and a kurtosis coefficient are calculated;
the skewness coefficient calculation method comprises the following steps:
wherein SK 1 The coefficient of the skewness is the coefficient of the skewness,
is the average value of the samples, m 2 Is the second-order center distance, m, of the sample 3 Is the third-order center distance, x, of the sample i Denotes the ith sample and n denotes the total number of samples.
Wherein, the step of calculating the kurtosis coefficient is as follows:
wherein, γ 2 Denotes the kurtosis coefficient, k 4 Representing the fourth-order center distance, k, of the sample 2 Representing the second order sample center distance, n representing the total number of samples, x i Denotes the ith sample and x denotes the sample mean.
It should be understood that skewness measures the asymmetry of the probability distribution of a data set, which is a measure of the degree of asymmetry relative to the mean. By calculating the skewness coefficient, the degree of asymmetry and the distribution direction of the data can be determined.
It should be understood that KURT (KURT) measures the flatness of the data distribution and is a statistic of the steep or smooth distribution of the study data. The kurtosis value of a normal distribution is 3, and the kurtosis is defined as the fourth-order center-to-center distance divided by the square of the square difference minus three.
As one or more embodiments, in S3, whether the wavelet coefficient of the current layer conforms to normal distribution is determined according to the skewness coefficient and the kurtosis coefficient of each layer; the specific steps of the judgment comprise:
calculating the standard deviation of the skewness coefficient of each layer according to the skewness coefficient of each layer; calculating the Z score value of the skewness coefficient based on the standard deviation of the skewness coefficient of the layer;
calculating the standard deviation of the kurtosis coefficient of each layer according to the kurtosis coefficient of each layer; calculating the Z score value of the kurtosis coefficient based on the standard deviation of the kurtosis coefficient of the layer;
if the Z score value of the skewness coefficient and the Z score value of the kurtosis coefficient are both in the set range, the wavelet coefficient of the current layer is in accordance with normal distribution; otherwise, it indicates that the wavelet coefficient of the current layer does not conform to the normal distribution.
It should be understood that the Z score value of the skewness factor (Sk-Z) is calculated by the formula:
wherein, delta Skewness Skewness is a Skewness factor, which is the standard deviation of kurtosis, and is used to measure the asymmetry of data distribution.
It should be understood that the Z score value of the kurtosis coefficient (Kurt-Z) is calculated by the formula:
wherein, delta Kurtosis Kurtosis is a Kurtosis coefficient, which is a standard deviation of skewness, and is used to measure the steepness or smoothness of the data distribution.
It is understood that at the test level of α =0.05, whether Kurt-Z and Sk-Z satisfy a range of variables bounded by the hypothetical conditions-1.96 to 1.96, both of which are considered normal distributions, and one of which is not. Therefore, whether the data conforms to the normal distribution or not can be judged according to the wavelet coefficient of the layer. If the normal distribution characteristic is met, a hard threshold value method is adopted; otherwise, judging the data to be in non-normal distribution, and adopting a soft threshold value method. And adaptively determining a soft threshold value and a hard threshold value according to the characteristics of the data.
It should be understood that, in S3, if the normal distribution is satisfied, denoising is performed by using a hard threshold method; otherwise, denoising by adopting a soft threshold method;
the improved adaptive threshold processing method selects a soft threshold and a hard threshold method by introducing normality test, carries out denoising processing on the wavelet coefficient of each layer and fully utilizes the advantages of the soft threshold and the hard threshold processing method. For each layer of wavelet coefficient, if the wavelet coefficient of the layer conforms to normal distribution, the high-frequency part of the signal is considered to be mainly noise, and the noise and the signal distribution have great difference. In order to maintain the local edge characteristics of the signal, denoising by a hard threshold method; if the wavelet coefficient of the layer does not conform to normal distribution, the fact that the high-frequency part of the data is mainly signals and secondarily noise or the noise and the signals are difficult to distinguish is proved, denoising processing is carried out on the wavelet coefficient by adopting a soft threshold method, and smoothness and continuity of denoising are guaranteed.
It should be understood that the specific steps of reconstructing the data signal using the processed high frequency coefficients and low frequency coefficients to obtain the noise-reduced data include:
and (3) performing data reconstruction on the wavelet low-frequency coefficient and each layer of high-frequency coefficient by using a wavelet inverse transformation method to obtain denoised data.
As one or more embodiments, in S4, the denoised data for each layer is input into a corresponding PCA model to obtain statistics of each layer; the method comprises the following specific steps:
and respectively establishing corresponding PCA models for the denoised data of each layer, and then solving the statistic of the PCA models.
It should be understood that the statistics are an indicator of whether an anomaly has occurred in the detection process. The control limit refers to a limit of a control range of the fault detection.
As one or more embodiments, in S5, if the statistic of the data to be detected is smaller than the set control limit, it indicates that no fault occurs, and continues fault detection; and if the statistic of the data to be detected exceeds the set control limit, indicating that fault data exists in the TE process data to be diagnosed.
As one or more embodiments, in S6, the identifying the fault type by using the contribution graph; the method comprises the following specific steps:
the contribution graph gives the respective monitored process variable versus the detection statistic (typically T) 2 Or SPE). Calculating the contribution degree of each variable to the fault by using a contribution graph method, sequencing the contribution degrees of each variable, and determining the faultAnd (4) determining the fault type according to the variable with the largest influence of the fault.
As one or more embodiments, in S7, the fault source is determined according to the variable with the largest contribution value corresponding to the statistic; the method comprises the following specific steps:
and calculating the contribution degree of each process variable to the fault by using an accumulative variance contribution rate method, sequencing the contribution degrees of each variable, determining the variable with the largest influence on the fault, and further determining a fault source.
As one or more embodiments, the set control limit is calculated by:
sa1: acquiring historical TE process data; carrying out standardization processing on the acquired data;
sa2: performing wavelet decomposition on the standardized data to decompose the data into a plurality of layers;
sa3: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; reconstructing data by using a wavelet inverse transformation method by combining a high-frequency coefficient and a low-frequency coefficient obtained after each layer of wavelet denoising to obtain denoised data;
sa4: establishing a corresponding PCA model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
sa5: based on the statistics for each layer, the set control limits are calculated.
Principal component analysis is carried out on the reconstructed data, the number of the principal components is determined, and Q and T are calculated 2 Control limit Q of statistic limit And
thereby obtaining a comprehensive scale PCA model.
The embodiment is as follows:
the present embodiment uses a TE process data set. TE processes were developed by isemann chemical company for the purpose of providing a realistic industrial process for evaluating process control and monitoring methods. TE process data has become popular for use in the field of fault diagnosis and process monitoring as a data source for comparing various methods. The TE process includes five main unit operations: the reactor, condenser, cycle, compressor, separator and stripper had 21 failures in the TE process. The process has a total of 12 operating variables, 41 measured variables, including 15 known disturbances and 6 operating modes. The entire test data set includes a training set and a test set. The data in the dataset consisted of 22 different simulated data, with 52 observed variables for each sample in the dataset. Where each training set test sample represents a fault. Fig. 4 is a TE process flow diagram.
The experimental environment is a Windows 7 flagship edition 64-bit system, MATLABR2016a, and the experimental data set is a TE process data set. We use the Q Statistic (SPE) and T 2 Statistics are monitored to detect faults, and control limits are indicated by dashed lines. When Q or T of the test data 2 Above the control limit, a fault is indicated. And after the fault is detected, further utilizing the contribution graph to identify the fault. The source and type of the fault may be defined by the pair T 2 And determining the variable with the largest Q statistic contribution value.
We have selected fault 5 and fault 7 for the fault diagnosis experiment. The improved IMSPCA method proposed herein was compared to the PCA method which combines the traditional soft threshold method. Experimental results are as follows, fig. 5 (a) -5 (h) are simulation results of failure 5, and fig. 6 (a) -6 (h) are simulation results of failure 7.
Fig. 5 (a), 5 (b), 5 (e) and 5 (f) are conventional soft thresholding-PCA fault diagnosis methods, and fig. 5 (c), 5 (d), 5 (g) and 5 (h) are adaptive thresholding-modified IMSPCA fault diagnosis methods; fig. 5 (a) -5 (d) are fault detection diagrams, and fig. 5 (e) -5 (h) are diagrams of the contribution of variables to a fault.
Fig. 6 (a), 6 (b), 6 (e) and 6 (f) are conventional soft thresholding-PCA fault diagnosis methods, and fig. 6 (c), 6 (d), 6 (g) and 6 (h) are adaptive thresholding-modified IMSPCA fault diagnosis methods; fig. 6 (a) -6 (d) are fault detection diagrams, and fig. 6 (e) -6 (h) are diagrams of the contribution of variables to a fault.
The experimental results of the faults 5 and 7 show that the fault detection effect of the proposed IMSIPCA fault diagnosis algorithm is superior to that of the traditional soft threshold method combined with the PCA fault diagnosis algorithm. The new algorithm is significantly better than the conventional algorithm in finding the variables that have the greatest impact on the fault. As can be seen from fig. 6 (a) -6 (h), the conventional soft threshold-PCA method detects no fault in fault 5 using the Q statistic, but the improved adaptive threshold method, i.e., the Q statistic of IMSPCA, detects a fault.
Meanwhile, as can be seen from fig. 5 (a) -5 (h) and fig. 6 (a) -6 (h), the conventional soft threshold-PCA algorithm cannot detect the variables that have the greatest contribution to the fault at the same time, or detect several variables that have a large influence on the fault, and cannot determine which variable causes the fault; while in the contribution graph of each variable of the new algorithm to the fault, T 2 The statistic and the Q statistic can simultaneously and accurately detect the variable which has the largest influence on the fault. The experimental results verify the effectiveness of the adaptive threshold method and the IMSCA fault detection model provided by the method. Simulation results show that the method is simple and effective, and the fault diagnosis effect is improved. The new algorithm is an obvious traditional algorithm in the aspect of finding the variable which has the largest influence on the fault.
Aiming at the defects existing in the traditional soft threshold and hard threshold methods which are independently used, an adaptive threshold method based on the normality distribution test is provided. The method utilizes a normal distribution test to adaptively select a soft threshold and a hard threshold method to denoise wavelet coefficients of each layer. Then, the improved wavelet threshold denoising method and the multi-scale principal component analysis method are combined, and a fault diagnosis strategy based on the improved multi-scale principal component analysis (IMSPCA) is provided. Firstly, wavelet threshold denoising is improved by utilizing normal distribution inspection, then an improved wavelet threshold denoising method is combined with multi-scale principal component analysis, a comprehensive PCA model is established for fault diagnosis, noise is effectively removed, and the reliability of fault diagnosis is improved. Finally, the feasibility and the effectiveness of the improved IMSPCA method are verified through simulation research of the TE process.
The second embodiment also provides a fault diagnosis system based on improved wavelet threshold denoising;
a fault diagnosis system based on improved wavelet threshold denoising comprises:
an acquisition module configured to: acquiring data of a TE Process (a Tennessee Eastman Process, wherein the TE Process is a Process for truly simulating a chemical Process) to be diagnosed; carrying out standardization processing on the acquired data;
a wavelet transform module configured to: performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
a noise reduction module configured to: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data;
a statistics calculation module configured to: establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
a fault diagnosis module configured to: comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; and if the statistic of the data to be diagnosed exceeds the set control limit, indicating that fault data exist in the TE process data to be diagnosed.
In a third embodiment, the present embodiment further provides an electronic device, which includes a memory, a processor, and computer instructions stored in the memory and executed on the processor, where the computer instructions, when executed by the processor, implement the steps of the method in the first embodiment.
In a fourth embodiment, the present embodiment further provides a computer-readable storage medium for storing computer instructions, and the computer instructions, when executed by a processor, perform the steps of the method in the first embodiment.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made to the present application by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (8)
1. The fault diagnosis method based on improved wavelet threshold denoising is characterized by comprising the following steps:
acquiring TE process data to be diagnosed; carrying out standardization processing on the acquired data;
performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer; the specific steps of the judgment comprise:
calculating the standard deviation of the skewness coefficient of each layer according to the skewness coefficient of each layer; calculating the Z score value of the skewness coefficient based on the standard deviation of the skewness coefficient of the layer;
calculating the standard deviation of the kurtosis coefficient of each layer according to the kurtosis coefficient of each layer; calculating the Z score value of the kurtosis coefficient based on the standard deviation of the kurtosis coefficient of the layer;
if the Z score value of the skewness coefficient and the Z score value of the kurtosis coefficient are both in the set range, the wavelet coefficient of the current layer is in accordance with normal distribution; otherwise, the wavelet coefficient of the current layer does not conform to normal distribution;
establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; if the statistic of the data to be diagnosed exceeds the set control limit, indicating that fault data exists in the TE process data to be diagnosed; the set control limit calculation process comprises the following steps:
sa1: acquiring historical TE process data; carrying out standardization processing on the acquired data;
sa2: performing wavelet decomposition on the standardized data to decompose the data into a plurality of layers;
sa3: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; reconstructing data by using a wavelet inverse transformation method by combining a high-frequency coefficient and a low-frequency coefficient obtained after each layer of wavelet denoising to obtain denoised data;
sa4: establishing a corresponding PCA model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
sa5: based on the statistics for each layer, the set control limits are calculated.
2. The method of claim 1, wherein the normalizing the acquired data; the method comprises the following specific steps: the Z-score normalization process normalizes the data.
3. The method according to claim 1, wherein the normalized data is decomposed into a plurality of layers by wavelet decomposition; the method comprises the following specific steps:
adopting a wavelet base function db5;
determining the optimal decomposition layer number m by adopting a mode maximum value mode;
and performing wavelet decomposition on the normalized data to decompose the normalized data into m layers.
4. The method of claim 1, wherein for the wavelet coefficients of each layer, skewness coefficients and kurtosis coefficients are calculated;
wherein, the step of calculating the skewness coefficient is as follows:
wherein SK 1 The coefficient of the skewness is the coefficient of the skewness,
is the average value of the samples, m 2 Is the second order centre distance of the sample, m 3 Is the third-order center distance, x, of the sample i Denotes the ith sample and n denotes the total number of samples.
5. The method of claim 4, wherein the step of calculating the kurtosis factor comprises:
wherein, γ 2 Denotes the kurtosis coefficient, k 4 Representing the fourth-order center distance, k, of the sample 2 Representing the second order sample center distance, n representing the total number of samples, x i Denotes the ith sample and x denotes the sample mean.
6. The fault diagnosis system based on the improved wavelet threshold denoising is characterized by comprising the following components:
an acquisition module configured to: acquiring TE process data to be diagnosed; carrying out standardization processing on the acquired data;
a wavelet transform module configured to: performing wavelet transform decomposition on the normalized data, decomposing the normalized data into a plurality of layers to obtain wavelet coefficients of each layer, wherein the wavelet coefficients comprise: a low frequency coefficient or a high frequency coefficient;
a noise reduction module configured to: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; after denoising, obtaining a high-frequency coefficient and a low-frequency coefficient of each layer of wavelet coefficient after being processed, and reconstructing a data signal by using the processed high-frequency coefficient and low-frequency coefficient to obtain denoised data; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer; the specific steps of the judgment comprise:
calculating the standard deviation of the skewness coefficient of each layer according to the skewness coefficient of each layer; calculating the Z score value of the skewness coefficient based on the standard deviation of the skewness coefficient of the layer;
calculating the standard deviation of the kurtosis coefficient of each layer according to the kurtosis coefficient of each layer; calculating the Z score value of the kurtosis coefficient based on the standard deviation of the kurtosis coefficient of the layer;
if the Z score value of the skewness coefficient and the Z score value of the kurtosis coefficient are both in the set range, the wavelet coefficient of the current layer is in accordance with normal distribution; otherwise, the wavelet coefficient of the current layer does not conform to normal distribution;
a statistics calculation module configured to: establishing a corresponding Principal Component Analysis (PCA) model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
a fault diagnosis module configured to: comparing the statistic of each layer with the set control limit, if the statistic of the data to be detected is smaller than the set control limit, indicating that no fault occurs, and continuing fault detection; if the statistic of the data to be diagnosed exceeds the set control limit, indicating that fault data exists in the TE process data to be diagnosed; the calculation process of the set control limit comprises the following steps:
sa1: acquiring historical TE process data; carrying out standardization processing on the acquired data;
sa2: performing wavelet decomposition on the standardized data, and decomposing the data into a plurality of layers;
sa3: calculating skewness coefficients and kurtosis coefficients for the wavelet coefficients of each layer; judging whether the wavelet coefficient of the current layer conforms to normal distribution or not according to the skewness coefficient and the kurtosis coefficient of each layer, and if so, denoising by adopting a hard threshold method; otherwise, denoising by adopting a soft threshold method; combining the high-frequency coefficient and the low-frequency coefficient obtained after the wavelet denoising of each layer, and reconstructing data by using a wavelet inverse transformation method to obtain denoised data;
sa4: establishing a corresponding PCA model for each layer of wavelet coefficient, and inputting the denoised data of each layer into the corresponding PCA model to obtain the statistic of each layer;
sa5: based on the statistics for each layer, the set control limits are calculated.
7. An electronic device comprising a memory and a processor and computer instructions stored on the memory and executable on the processor, the computer instructions when executed by the processor performing the steps of the method of any one of claims 1 to 5.
8. A computer-readable storage medium storing computer instructions which, when executed by a processor, perform the steps of the method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910987108.7A CN110717472B (en) | 2019-10-17 | 2019-10-17 | Fault diagnosis method and system based on improved wavelet threshold denoising |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910987108.7A CN110717472B (en) | 2019-10-17 | 2019-10-17 | Fault diagnosis method and system based on improved wavelet threshold denoising |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110717472A CN110717472A (en) | 2020-01-21 |
| CN110717472B true CN110717472B (en) | 2023-02-03 |
Family
ID=69212744
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910987108.7A Active CN110717472B (en) | 2019-10-17 | 2019-10-17 | Fault diagnosis method and system based on improved wavelet threshold denoising |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110717472B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111797943B (en) * | 2020-07-28 | 2024-02-02 | 中车青岛四方机车车辆股份有限公司 | Urban rail vehicle and passenger compartment door fault diagnosis method thereof |
| CN111707908B (en) * | 2020-07-29 | 2022-05-17 | 中国科学技术大学先进技术研究院 | Multi-load loop series fault arc detection method and device and storage medium |
| CN114136600A (en) * | 2021-11-19 | 2022-03-04 | 神华准格尔能源有限责任公司 | Equipment fault monitoring method, system and storage medium |
| CN114722885B (en) * | 2022-06-09 | 2022-08-16 | 山东山矿机械有限公司 | Intelligent detection method and system for abnormal operation of carrier roller carrying trolley |
| CN116588282B (en) * | 2023-07-17 | 2023-10-13 | 青岛哈尔滨工程大学创新发展中心 | AUV intelligent operation and maintenance system and method |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103837345B (en) * | 2014-03-25 | 2016-06-22 | 上海电机学院 | Fault Diagnosis of Gear Case method and device |
| CN105334269A (en) * | 2015-10-19 | 2016-02-17 | 江苏大学 | Pipeline defect type determination method based on neural network and guided wave characteristic database |
| CN106017879B (en) * | 2016-05-18 | 2018-07-03 | 河北工业大学 | Omnipotent breaker mechanical failure diagnostic method based on acoustic signal Fusion Features |
| CN108416106B (en) * | 2018-02-05 | 2022-02-08 | 江苏方天电力技术有限公司 | Water feeding pump fault detection method based on multi-scale principal component analysis |
-
2019
- 2019-10-17 CN CN201910987108.7A patent/CN110717472B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110717472A (en) | 2020-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110717472B (en) | Fault diagnosis method and system based on improved wavelet threshold denoising | |
| CN108804832B (en) | VMD-based interval threshold partial discharge denoising method | |
| RU2488815C2 (en) | Method and apparatus for classifying sound-generating processes | |
| US20150219530A1 (en) | Systems and methods for event detection and diagnosis | |
| US20170052060A1 (en) | Method and system for automatically detecting faults in a rotating shaft | |
| CN108304348B (en) | Bearing residual life prediction method based on binary wiener process | |
| CN108535635B (en) | EEMD and HMM based analog circuit intermittent fault diagnosis method | |
| CN113375939B (en) | Mechanical part fault diagnosis method based on SVD and VMD | |
| CN116955938B (en) | Dry-type waste gas treatment equipment monitoring method and system based on data analysis | |
| CN112955839A (en) | Abnormality detection device, abnormality detection method, and program | |
| CN114692758A (en) | Power communication fault analysis method and device, terminal equipment and medium | |
| CN112506752A (en) | Operation and maintenance method, device and equipment based on time series data trend prediction | |
| Sun et al. | Data-driven fault diagnosis method based on second-order time-reassigned multisynchrosqueezing transform and evenly mini-batch training | |
| Sha et al. | A novel noise reduction method for natural gas pipeline defect detection signals | |
| CN113051092B (en) | Fault diagnosis method based on optimized kernel density estimation and JS divergence | |
| JP2009110262A (en) | Filter device, filtering method for filter device and filtering program | |
| Xu et al. | Rolling bearing fault feature extraction via improved SSD and a singular-value energy autocorrelation coefficient spectrum | |
| CN112033656A (en) | Mechanical system fault detection method based on broadband spectrum processing | |
| CN116609106A (en) | Drilling and testing integrated equipment operation state monitoring system | |
| CN112016800A (en) | Feature selection method and system based on effectiveness index | |
| CN115310490B (en) | Rotary equipment fault analysis method based on multi-domain feature and sensitive feature selection | |
| CN114676593A (en) | Abnormity detection method of textile equipment and related device | |
| CN114090949A (en) | Robust vibration signal characteristic value calculation method | |
| CN112237433B (en) | Electroencephalogram signal abnormity monitoring system and method | |
| CN112232419A (en) | Rotating equipment fault monitoring method and system based on big data and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |