CN102736546A - State monitoring device of complex electromechanical system for flow industry and method - Google Patents

Abstract

The invention relates to a state monitoring device and method for complex electromechanical systems in the process industry. The device includes a human-computer interaction module, a data acquisition module, a data preprocessing module, a data analysis module and a failure case library. By adopting the device and method of the present invention, it is possible to monitor whether the system has a fault or an abnormal state, and to give an early warning of a vehicle jump accident or other safety accidents in the process industry system. At the same time, the KPCA method of double-parameter optimization overcomes the shortcomings of the traditional KPCA method in selecting parameters based on empirical formulas, and improves the state monitoring ability. Furthermore, making full use of the fault case database established in the historical production process of the system makes the monitoring of system faults more timely and accurate.

Description

Device and method for condition monitoring of complex electromechanical system in process industry

technical field

The invention belongs to the technical field of electromechanical system monitoring, and relates to a state monitoring device and method, in particular to a state monitoring device and method for complex electromechanical systems in the process industry.

Background technique

In the process industry, due to the increasing scale and complexity of industrial processes, the safety and reliability requirements of production systems are also increasing. The long-term safe, stable and efficient operation of the production system and the avoidance of vicious safety accidents have become an important task of modern industry. Therefore, during the operation of the system, it is necessary to detect the occurrence of faults or abnormal states in time, and to judge the type of fault and locate the source of the fault to eliminate adverse factors. Traditional condition monitoring methods can be divided into three categories: analysis-based methods, knowledge-based methods and data-driven methods. Analytical-based methods are based on strict mathematical models, such as Kalman filter, parameter estimation, equivalent space and other methods; knowledge-based methods are mainly based on process knowledge to establish models, such as fault tree (FTA), Decision tree (DT), etc.; the data-based method is mainly based on the collected process data, through various data processing and analysis methods to mine the hidden system status information in the data, and then guide the production process, such as multivariate statistical methods, Cluster analysis, spectrum analysis, etc. Due to the increasing complexity of production systems in the process industry, it is difficult to obtain strict mathematical models and detailed system knowledge. Therefore, analytical and knowledge-based methods are limited; and industrial systems usually collect and analyze equipment operating status data. Records, these condition monitoring data just contain essential information such as system operating conditions and system abnormal state evolution rules, so data-driven analysis methods have been widely used in condition monitoring and fault diagnosis in the process industry.

From the perspective of scientific research, the process industry production system represented by chemical production is a kind of system composed of many large-scale power machinery equipment, chemical reaction devices and automatic control systems coupled through energy networks, fluid networks, information networks, and control networks. Distributed complex electromechanical systems. In the actual state monitoring process, such a complex electromechanical system faces three problems: (1) The number of monitoring variables is huge, and there is correlation and strong coupling between variables. It is difficult to monitor all variables at the same time by manual methods. (2) The monitoring data presents the characteristics of slow change, massiveness, nonlinearity and atypicality, etc., and lacks effective means to mine the equipment status characteristic information contained in the data. (3) The modern process industry production system is in a multi-media coupling network environment, and currently there is still a lack of device systems and methods for effective condition monitoring at the system level.

Below the concepts such as KPCA theory involved in the present invention, wavelet denoising are done following brief introduction and definition:

Kernel principal component analysis, or KPCA for short, is a commonly used method for data-driven fault detection. The basic idea of kernel principal component analysis is to first map the data matrix X of the input space to a high-dimensional feature space F through a nonlinear mapping function Ф, and then perform principal component analysis on the mapped data in the high-dimensional space, and extract the data in the high-dimensional space. The linear characteristics of space, that is, the nonlinear characteristics of data in low-dimensional space. This non-linear mapping is achieved by introducing a kernel function to calculate the inner product of the data in the input space. KPCA determines the control limit by constructing the process statistic T ² based on the subspace information of the process principal component characteristic signal and the statistic SPE of the residual information subspace information, and then realizes the state monitoring.

The traditional KPCA method has the following shortcomings in practical application: (1) The selection of KPCA kernel parameters and the number of pivots is very subjective. At present, there is no unified criterion for the selection of kernel parameters, and most of them adopt the method of empirical formula. The selection of the number of pivots in KPCA generally adopts the simple and commonly used pivot cumulative contribution rate method (Cumulative percent variance, CPV), but there is no uniform standard for the most appropriate contribution rate, and the pivotal cumulative contribution rate is used in KPCA When calculating the number of pivots, it will first be affected by the selection of kernel parameters; (2) The entire monitoring process does not use known fault case data, but establishes a KPCA model based on given parameters to detect all types of faults. However, a fixed system model cannot have a good detection effect on all faults, and can only be very sensitive to one of them, or a certain type of fault. The present invention proposes a set of devices and methods to overcome the above problems on the basis of the improved KPCA model optimized by two parameters.

Wavelet noise reduction: The data collected in the actual industrial process is often polluted and interfered by noise, such as white noise and electromagnetic interference, among which useful signals usually appear as low-frequency signals or some relatively stable signals, while noise signals usually appear as high frequency signal. When the wavelet decomposition is performed on the actual collected data, the noise part is mainly included in the high-frequency wavelet coefficients. Therefore, the wavelet coefficients can be processed in the form of threshold and threshold, and then the signal can be reconstructed to achieve noise reduction and anti-interference. The purpose is to improve data quality and improve fault detection ability and accuracy.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art, and provide a state monitoring device and method for complex electromechanical systems in the process industry, which is aimed at the characteristics of multivariable, massive, and nonlinear monitoring data of complex electromechanical systems in the process industry. The state monitoring in the production process can be realized at the level, which can improve the state monitoring ability and detect the occurrence of faults and abnormal states in time.

The purpose of the present invention is solved by the following technical solutions:

This condition monitoring device for complex electromechanical systems in the process industry, including:

Human-computer interaction module: used to realize the interaction between the user and the status monitoring system, including the input and output of system status monitoring information, calling the data acquisition module, data preprocessing module and data analysis module;

Data acquisition module: used to extract the historical state detection data of the system and the real-time data generated during the operation of the DCS control system;

Data preprocessing module: used to remove the Gaussian white noise of the monitored variable data, and standardize the collected data to remove the influence of dimension for subsequent analysis;

Data analysis module: used to model the monitoring system, compare the real-time monitoring data with the built model, and detect the abnormal state of the system;

Fault case library: used to store and manage historical fault information of the monitored system, including fault time, fault cause, and fault mode;

The human-computer interaction module is respectively connected with the data acquisition module, the data preprocessing module and the data analysis module, as the carrier of information transmission; meanwhile, the data acquisition module, the data preprocessing module and the data analysis module are respectively connected with the failure case library, Extract information from the fault case library to complete the modeling and analysis functions.

The functions in the above data analysis module include: the analysis method of the KPCA model with double parameter optimization; combining the data information of the system case library to establish the KPCA model set of the monitoring system; comparing the real-time monitoring data with the established KPCA model set to detect the abnormal state of the system ; Make judgments on the operation of the entire process industry system and provide targeted safety warning information.

The present invention also proposes a state monitoring method for complex electromechanical systems in the process industry based on the above device, comprising the following steps:

1) Data collection: extract the historical data of the normal working conditions of the monitored object from the fault case database and collect the real-time monitoring data of the monitored object; the historical data is used to establish the system model, and the real-time data is used to monitor the system status;

2) Data preprocessing: Firstly, the wavelet denoising method is used to denoise the extracted historical data of normal working conditions and the real-time monitoring data of the collected monitoring objects; then standardize the denoised data to eliminate the monitoring variables The impact of dimension inconsistency;

3) Establishing a system model: the KPCA model is established according to the historical data of normal working conditions by using the nuclear principal component analysis method of two-parameter optimization, and the kernel parameters and the number of principal components in the KPCA model are optimized by combining the data of known fault cases to obtain KPCA model set, used to detect whether the system is in an abnormal state;

4) Abnormal state monitoring: Calculate the monitoring statistics of the collected real-time monitoring data under the established model, and compare them with the upper limit of the monitoring statistics of the model. If the upper limit of the statistics is exceeded, it can be judged that the system is under statistical There is an abnormal state in the sense;

5) Display the analysis results and effective warning information through the human-computer interaction module.

Further, the above step 3) specifically includes the following steps:

a) Establish a two-parameter objective optimization problem, and find the kernel parameter σ and the number of principal components p when the statistic T ² statistic detection rate and SPE statistic detection rate are maximized, expressed by the following formula:

$\mathrm{<span\; class="notranslate">\; max</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; st</span>}<span\; class="notranslate">\; :</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; m</span>}\end{array}\right.$

in:

σ——kernel parameter;

p - the number of pivots;

n——The number of pivots when taking 85% cumulative contribution rate;

m——the dimension of the input space, that is, the number of variables;

F _t (σ,p)——the detection rate of T ² statistics under the given kernel parameters and the number of pivots;

F _s (σ,p)——the detection rate of SPE statistics under the given kernel parameters and the number of pivots;

b) Obtain the data under normal working conditions as training samples and standardize them, and establish the KPCA model with the initial kernel parameters and the number of principal components; the initial kernel parameter σ=10m, m is the dimension of the input space, that is, the number of variables; the initial principal The number of coins is selected according to the method that the cumulative contribution rate reaches 85%;

c) Obtain the T ² statistics and the upper limit of the SPE statistics at the test level α=99% from the initial number of principal components;

d) Obtain the failure case data, and standardize each variable with the standard deviation and mean of the corresponding vector of the training data;

e) Obtain the pivot vector of the fault case data under the initial parameters, and obtain ^T2 statistics and SPE statistics;

f) Compare the statistic value and the upper limit value of the statistic, calculate the percentage of samples whose ^T2 statistic and SPE statistic exceed the upper limit respectively, and obtain the average detection rate;

g) Change the initial parameters, and calculate the average detection rate of statistics under the new parameters according to the above steps, compare with the previous average detection rate, retain the kernel parameters and the number of pivots with higher average detection rate;

h) Repeat the above steps until the average detection rate meets a certain detection rate required by fault detection, or a convergent solution is obtained; the kernel parameters and the number of principal components at this time are the optimal KPCA model parameters for the fault.

The present invention has the following beneficial effects:

The state monitoring device and method for complex electromechanical systems in the process industry of the present invention can monitor whether the system has a fault or an abnormal state, and can give early warnings of vehicle jump accidents or other safety accidents in the process industry system. At the same time, the fault detection ability of the traditional KPCA monitoring method is improved by using the KPCA method optimized by two parameters. Furthermore, due to the full use of the fault case database established in the process industry process system, the monitoring of system faults is more timely and accurate.

Description of drawings

Fig. 1 is the structural representation of device described in the present invention;

Fig. 2 is a work flow chart of the present invention;

Fig. 3 is the KPCA method solution process of two-parameter optimization;

Fig. 4 is a system structure diagram of an embodiment of the present invention;

Fig. 5 is a state monitoring diagram of the system in the present invention.

Detailed ways

Referring to Fig. 1, the condition monitoring device of process industry complex electromechanical system of the present invention comprises:

Human-computer interaction module: used to realize the interaction between the user and the status monitoring system, including the input and output of system status monitoring information, calling the data acquisition module, data preprocessing module and data analysis module. It can modify and update the system fault case library, manage historical/real-time monitoring data and call the data analysis module for status monitoring.

Data acquisition module: used to extract the historical status monitoring data of the system and the real-time monitoring data generated by the DCS control system during the system operation.

Data preprocessing module: used to remove the Gaussian white noise of the monitored variable data, and standardize the collected data to remove the influence of dimension for subsequent analysis.

Data analysis module: this module is the core part of the device of the present invention. Combined with the data information of the system case library, on the basis of the KPCA modeling of the monitoring system, compare the real-time monitoring data with the built KPCA model set, quickly and effectively detect the abnormal state of the system, and analyze the operation status of the entire process industry system Make judgments and provide effective safety warning information.

Fault case library: used to store and manage the historical fault information of the monitored system, including fault time, fault cause, and fault mode and other related information.

The human-computer interaction module is respectively connected with the data acquisition module, the data preprocessing module and the data analysis module, as the carrier of information transmission; meanwhile, the data acquisition module, the data preprocessing module and the data analysis module are respectively connected with the failure case library, Extract information from the fault case library to complete the modeling and analysis functions.

The functions in the data analysis module include: the analysis method of the KPCA model optimized by two parameters; in combination with the data information of the system case library, the KPCA model set of the monitoring system is established; the real-time monitoring data is compared with the established KPCA model set, and the detection system is abnormal Status; make judgments on the operation of the entire process industry system and provide targeted safety warning information.

The present invention can use computer memory to store system fault case data information, monitoring historical data, real-time data and data analysis process, and use input and output interface to connect keyboard, external memory and display, and analyze the KPCA model set information generated during the analysis process The results can be expressed on the display in the form of human-computer interaction.

Based on the above devices, the workflow of the process industry complex electromechanical system state monitoring and analysis method of the present invention is shown in Figure 2, and the specific steps are as follows:

Step 1: Data extraction of the monitored object. According to the specific monitored object and combined with the monitoring target, the data of normal working conditions stored in the historical database can be effectively extracted, and the real-time data of the monitored variables corresponding to the historical data can be extracted.

Step 2: Preprocessing of extracted real-time/historical data; data preprocessing includes wavelet noise reduction and standardization processing (making the mean value zero and variance 1). This step specifically includes:

(a) Perform wavelet denoising on the extracted real-time/historical data. According to the principle of wavelet transform threshold denoising, wavelet transform threshold denoising usually includes the following three steps: (1) choose a suitable wavelet base and determine the decomposition level to decompose the signal by wavelet; (2) determine the detail coefficient of each layer Threshold, use soft threshold or hard threshold to process wavelet coefficients; (3) Wavelet inverse transform to reconstruct the signal.

(b) Standardize the data after wavelet transform threshold denoising. Different variables often have different dimensions and orders of magnitude. In order to compare the degree of change of different variables on the same order of magnitude, the influence of dimension needs to be eliminated, so the data are standardized. After normalization, the mean of the data is 0 and the variance is 1.

计算核矩阵K。Step 3: Establish KPCA model based on historical normal working condition data. The original input data matrix X∈R ^n×m (m observed variables, n sampling times) is n samples under normal operating conditions, and the data matrix after preprocessing in step 2 is Gaussian Radial Basis Kernel Function

Compute the kernel matrix K.

Center the kernel matrix K, solve the eigenvalue and eigenvector a _k of K, and standardize the eigenvector so that _{<a k , a k >} =1/λ _k . where λ _k is the corresponding eigenvalue.

Compute the nonlinear pivot t _k :

是标准化之后的特征向量，

是中心化之后的核矩阵K。in,

is the normalized feature vector,

is the kernel matrix K after centralization.

Step 4: Combining with the fault case library, construct a KPCA model set for two-parameter optimization. For each fault in the fault case library, the kernel parameters and the number of pivots of KPCA are optimized to obtain the KPCA model corresponding to each fault. For the specific optimization method, please refer to the description (3).

Step 5: KPCA state monitoring based on two-parameter optimization. For a new real-time monitoring sampling data sample x _new ∈R ^1×m , construct the corresponding statistics T ² and SPE and their corresponding control limit thresholds T _α ² and SPE _α to monitor the state of the system. The statistic T ² and its corresponding control limit threshold T _α ² can be determined by the following formula:

T ² =[t ₁ ,.,t _p ]Λ ^-1 [t ₁ ,…,t _p ] ^T (2)

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Where Λ ^-1 is the inverse matrix of the diagonal matrix formed by the eigenvalues corresponding to the pivot, F _α (k,nk) is the upper limit value of the F distribution with confidence degree α and degrees of freedom respectively p and np, which can be Obtained by looking up the table. SPEs are defined as:

SPE控制限服从自由度为h的χ²分布。若a，b分别为SPE的均值和方差，则g=b/2a，h＝2a²/b。When the inspection level is α, the SPE control limit is

The SPE control limits follow a ^χ2 distribution with h degrees of freedom. If a and b are the mean and variance of SPE respectively, then g=b/2a, h=2a ² /b.

Step 6: Display the analysis results and effective warning information. Comparing the statistical value of the real-time monitoring data with the upper limit of the statistical value of the KPCA model, if T _a < T or SPE _a < SPE, it indicates that the system is in an abnormal state. The analysis results are instantly displayed through the human-computer interaction module, giving the operator a reminder of the abnormal state of the system.

The KPCA method optimization process of the above two-parameter optimization is as follows:

Referring to Fig. 3, Fig. 3 is a schematic diagram of the solution process of the KPCA method for dual-parameter optimization. Combined with the fault case base, a KPCA model with two parameters optimization is constructed for each fault in the case base. Establish a two-parameter objective optimization problem, and find the kernel parameter σ and the number of principal components p when the detection rate of T ² and SPE are maximized, which can be expressed as:

$\max \frac{1}{2}(f_t(\mathrm{\&sigma;},p)+f_{\mathrm{the\; s}}(\mathrm{\&sigma;},p))$ (5)

$\mathrm{<span\; class="notranslate">\; st</span>}<span\; class="notranslate">\; :</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; m</span>}\end{array}\right.$

in:

σ——kernel parameter;

p - the number of pivots;

n——The number of pivots when taking 85% cumulative contribution rate;

m——The dimension of the input space, that is, the number of variables;

F _t (σ,p)——the detection rate of T ² statistics under the given kernel parameters and the number of pivots;

F _s (σ,p)——the detection rate of SPE statistics under the given kernel parameters and the number of pivots;

For the two-parameter objective optimization problem, the specific solution includes the following steps:

Step 1: Obtain the data under normal working conditions as a training sample and standardize it, and establish a KPCA model with the initial kernel parameters and the number of principal components. The initial kernel parameter σ=10m, m is the dimension of the input space, that is, the number of variables. The number of initial pivots is selected according to the method that the cumulative contribution rate reaches 85%.

Step 2 Obtain the upper limit of ^T2 statistics and SPE statistics at the test level α=99% from the initial number of pivots.

Step 3: Obtain the failure case data, and standardize each variable with the standard deviation and mean of the corresponding vector of the training data.

Step 4: Obtain the pivot vector of the fault case data under the initial parameters, and obtain T ² statistics and SPE statistics.

Step 5: Compare the statistic value and the upper limit value of the statistic, and calculate the percentage of samples whose T ² statistic and SPE statistic exceed the upper limit, respectively, to obtain the average detection rate.

Step 6: Change the initial parameters, and calculate the average detection rate of statistics under the new parameters according to the above steps. Compared with the previous average detection rate, keep the kernel parameters and the number of pivots with higher average detection rate.

Step 7: Repeat the above steps until the average detection rate meets a certain detection rate required for fault detection, or a convergent solution is obtained. The kernel parameters and the number of pivots at this time are the optimal KPCA model parameters for this fault.

In the process of solving this optimization problem, it is necessary to consider that the kernel parameter σ cannot be too large, so as to prevent the kernel function from being too generalized and losing the advantage of extracting nonlinear features. When calculating the number p of pivots, if the larger the number of pivots, the higher the fault detection rate, it is necessary to consider the balance between improving the effect of dimensionality reduction and improving the fault detection rate.

The double-parameter optimization KPCA method adopted in the present invention combines fault case data, and is more pertinent to known faults. When the faults similar to those in the fault case library appear during the operation of the system, the KPCA method with two-parameter optimization can make the fault detection effect reach the best.

Referring to Fig. 4-Fig. 5, Fig. 4 is a schematic structural diagram of the compressor unit. The compressor unit system consists of 5EH-8BD steam turbine, RIK100-4 radial isothermal compact air compressor, RBZ45-7 radial barrel supercharger, TX36/1C gearbox and some auxiliary devices and equipment. 70 monitoring variables that are closely related to the operating state of the compressor unit system are selected as observation variables.

Figure 5 is a diagram of the detection of system faults by double-parameter optimized KPCA. The failure is due to the pressure drop of the steam pipe network, which leads to the low load operation of the air compressor. After optimization, the kernel parameters are selected as 495, and the number of pivots is 8. The KPCA model is established and the operating status of the compressor unit is monitored. It can be seen from the figure that near the 800th sample, both statistics show obvious overruns, and both statistics can effectively detect the fault. Among them, the detection rate of T ² statistics was 94.2%, the detection rate of SPE statistics was 99.4%, and the average detection rate was 96.8%.

The system in the actual process industry production is relatively complex and there are many system failures. Therefore, it is necessary to establish a rich database of failure cases and make targeted early warnings for abnormal states of the system. On this basis, use the experience and knowledge of technicians to further choose and analyze the condition monitoring results, and make further fault diagnosis.

Claims (4)

1. A condition monitoring device for a complex electromechanical system in the process industry, comprising:

a human-computer interaction module: the system is used for realizing interaction between a user and a state monitoring system, comprises input and output of system state monitoring information, and calls a data acquisition module, a data preprocessing module and a data analysis module;

a data acquisition module: the DCS control system is used for extracting the detection data of the historical state of the system and the real-time data generated in the operation process of the DCS control system;

a data preprocessing module: the system comprises a data acquisition unit, a data processing unit and a data processing unit, wherein the data acquisition unit is used for acquiring data of a monitored variable;

a data analysis module: the system is used for modeling the monitoring system, comparing real-time monitoring data with the established model and detecting the abnormal state of the system;

a fault case library: the system comprises a data storage module, a fault analysis module and a fault analysis module, wherein the data storage module is used for storing and managing historical fault information of a monitored system, and comprises fault time, fault reason and fault mode;

the human-computer interaction module is respectively connected with the data acquisition module, the data preprocessing module and the data analysis module and is used as a carrier for information transmission; meanwhile, the data acquisition module, the data preprocessing module and the data analysis module are respectively connected with the fault case library, and information is extracted from the fault case library to complete modeling and analysis functions.

2. The condition monitoring device of process industry complex electromechanical system according to claim 1, wherein the functions in the data analysis module include: a KPCA model analysis method of two-parameter optimization; establishing a KPCA model set of the monitoring system by combining data information of a system case base; comparing the real-time monitoring data with the established KPCA model set, and detecting the abnormal state of the system; and judging the operation condition of the whole process industrial system and providing safety early warning information.

3. A method for monitoring the state of a complex electromechanical system in the process industry based on the device of claim 1, which comprises the following steps:

1) data acquisition: extracting the normal working condition historical data of the monitored object from the fault case library and collecting the real-time monitoring data of the monitored object; the historical data is used for establishing a system model, and the real-time data is used for monitoring the system state;

2) data preprocessing: firstly, denoising extracted normal working condition historical data and collected real-time monitoring data of a monitoring object by adopting a wavelet denoising method; then, carrying out standardization processing on the data subjected to noise reduction, and eliminating the influence of inconsistent dimension of each monitoring variable;

3) establishing a system model: establishing a KPCA model according to normal working condition historical data by adopting a kernel principal component analysis method of double-parameter optimization, and optimizing the kernel parameters and the number of principal components in the KPCA model by combining with known fault case data to obtain a KPCA model set for detecting whether a system is in an abnormal state;

4) monitoring an abnormal state: calculating monitoring statistics of the acquired real-time monitoring data under the established model, comparing the monitoring statistics with the upper limit value of the monitoring statistics of the model, and judging that the system is in an abnormal state in the statistical sense if the monitoring statistics exceeds the upper limit value of the statistics;

5) and displaying the analysis result and the effective early warning information through a man-machine interaction module.

4. The method for monitoring the state of the complex electromechanical system in the process industry according to claim 3, wherein the step 3) comprises the following steps:

a) establishing a two-parameter target optimization problem, solving the equation T²The kernel parameter sigma and the number p of the principal elements when the statistic relevance ratio and the SPE statistic relevance ratio are maximum are expressed by the following formula:

<math> <mrow> <mi>max</mi> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <mrow> <mo>(</mo> <msub> <mi>F</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>&sigma;</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>F</mi> <mi>s</mi> </msub> <mrow> <mo>(</mo> <mi>&sigma;</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> </mrow> </math> (5)

<math> <mrow> <mi>st</mi> <mo>:</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mi>&sigma;</mi> <mo>></mo> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mi>n</mi> <mo>&le;</mo> <mi>p</mi> <mo>&lt;</mo> <mi>m</mi> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein:

σ -nuclear parameter;

p is the number of principal elements;

n is the number of principal elements at 85% cumulative contribution rate;

m is the dimension of the input space, namely the number of variables;

F_t(σ, p) -T at given nuclear parameters and pivot numbers²A statistic detection rate;

F_s(σ, p) -SPE statistic relevance ratio under given nuclear parameters and principal component number conditions;

b) acquiring data under a normal working condition as a training sample, standardizing, and establishing a KPCA model by using initial kernel parameters and the number of principal elements; the initial kernel parameter σ =10m, where m is an input space dimension, that is, the number of variables; selecting the initial principal component number according to a method that the cumulative contribution rate reaches 85%;

c) obtaining T under the condition that the inspection level alpha is 99% according to the number of the initial principal elements²Statistics and SPE statistics upper limit values;

d) acquiring fault case data, and standardizing each variable by using the standard deviation and the mean value of the vector corresponding to the training data;

e) solving the principal component vector of the fault case data under the initial parameters to obtain T²Statistics and SPE statistics;

f) comparing the statistic value with the statistic upper limit value, and respectively calculating T²The average detection rate is obtained according to the percentage of the samples with the statistic and the SPE statistic exceeding the upper limit value;

g) changing initial parameters, calculating the average relevance ratio of statistics under the new parameters according to the steps, comparing the average relevance ratio with the previous average relevance ratio, and keeping the kernel parameters and the number of principal elements with higher average relevance ratio;

h) repeating the steps until the average detection rate meets a certain detection rate required by fault detection or a convergence solution is obtained; the kernel parameters and the number of principal elements at this time are the optimal KPCA model parameters for the fault.

Images