CN106647691A - Multi-loop oscillation extracting and detecting method of industrial process - Google Patents

Abstract

The invention discloses a multi-loop oscillation extracting and detecting method of industrial process. The method comprises the following steps that in control loops to be detected, process output signals of all the control loops to be detected are collected in advance; a spectrum related waveform matching method is used to continue end points of the output signals of the loops; improved multi-dimensional empirical mode decomposition is carried out on process output data after end-point continuation; a regularity index of a zero crossing point of each decomposed sub-signal is calculated; whether each regularity index exceeds a set threshold is determined, and whether an oscillation behavior exists in two or more decomposed sub-signals of the same layer is determined; and a detection result is obtained according to all determination results. The method of the invention can be used to detect the oscillation behavior of multiple loops in the industrial process quantitatively, regularity degrees and periods of oscillation components are obtained simultaneously, and abundant data support is provided for oscillation behavior evaluation and fault source diagnosis.

Description

A kind of industrial process multi-loop oscillation is extracted and detection method

Technical field

The present invention relates to the Performance Evaluation field in Industry Control, and in particular to one kind is divided based on multidimensional empirical mode is improved The industrial process multi-loop oscillation of solution is extracted and detection method.

Background technology

Modern industry process device has big scale, synthesis degree height, manipulation complexity, variable many, and long-play is being closed The features such as under ring control.The common chemical process of industry, usually contains thousands of control loops, and these are controlled Loop is interacted due to coupled relation.Due in process control loops controller cross adjust, external disturbance and regulating valve it is non- The generally existing of the characteristics such as linear work, especially multi-loop oscillation occurs the oscillation behavior of control loop often, and this is greatly Have impact on the economic benefit and stability of industrial flow equipment operation.

Preliminary accurately oscillation test is carried out to industrial flow equipment can reduce waste product output, reduce disqualification rate, Reliability, the security in industrial flow equipment running process is improved, while reducing manufacturing cost.Many controllers are first in operation Phase can also keep good performance, but As time goes on, due to the impact of external interference factor or equipment self problem, control The performance of device processed can be gradually lowered even failure.It is embodied in control loop process and all kinds of oscillation behaviors, some of them occurs The larger vibration of intensity is also possible to propagate to other coupling circuits, multiloop plant-level oscillation is formed, so as to threaten industrial process Safe and stable operation.

Simultaneously as apparatus of load and operating mode Jing often change in real time environment, industrial process also shows non-stationary The one side of data characteristic, is embodied in the local mean value transport phenomena of process data.For important control loop, send out in time The oscillating characteristic of existing its running contributes to engineering staff carries out fault diagnosis and investigation.Therefore, in industrial control system In energy evaluation process, effective on-line monitoring means are designed, in time, accurately detect non-stationary process data in control loop All kinds of oscillationg components, and each different frequency ranges are distinguished, and whether propagation effect is to multiple loops, for controller Performance Evaluation and control loop fault diagnosis suffer from significance.

Existing process control loops oscillation test technology, the overwhelming majority is all based on the analysis method of univariate data. The either analysis method of Kernel-based methods data Time-domain Statistics, also or Kernel-based methods data auto-correlation function (ACF) point Analysis method, all can not be to process multivariable (i.e. multiloop) process data.Although univariate analysis method can be respectively right Each variable is individually analyzed in multidimensional data, but at least two kinds defects of this mode.First, each circuit process data processing with Testing result may be inconsistent；2nd, coupled relation is thoroughly upset between each loop.Most recent two decades also occur in that some many times The detection method of road vibration, such as based on spectrum principal component analysis or the detection of the multi-loop oscillation based on spectrum envelope collimation method and cluster side Method.But these methods are to the industrial process that there is non-stationary and nonlinear characteristic, it is impossible to realize automatically without detection is intervened, need Targetedly wave filter carries out data tranquilization process and oscillationg component is separated for design in advance.

In the practical application of process oscillation detection algorithm, multiloop the output of process data, effective detection can be simultaneously processed Whether process control loops have oscillation behavior, and the rule degree index of qualitative assessment oscillation behavior, and are generally applicable to presence When variable oscillation, multiple vibration, non-stationary and non-linear component process data, for Accurate Diagnosis industrial process vibration presence Property has very important Practical significance, is also beneficial to the qualitative assessment of industrial stokehold performance.

The content of the invention

The invention provides a kind of extracted and inspection based on the industrial process multi-loop oscillation for improving multidimensional empirical mode decomposition Survey method, the process control loops process of the behavior such as variable oscillation, multicycle vibration in the presence of being applied to, detection method is universal Suitable for non-stationary or stable process data, only conventional operation data need to be obtained, without the need for process mechanism knowledge.

A kind of industrial process multi-loop oscillation is extracted and detection method, including：

Step 1, gathers the output of process signal of all control loops to be detected in industrial process；

Step 2, using spectrum waveform correlation matching process, to each the output of process signal end extending is carried out；

Step 3, the multidimensional empirical mode decomposition being improved to the output of process signal after end extending obtains decomposing son Signal；

Step 4, calculates the zero cross point regularity index that each layer decomposes subsignal；

Step 5, judges whether each zero cross point regularity index exceedes threshold value, corresponding to decompose son if exceeding threshold value Signal determining is oscillator signal；

, if same layer decomposes in subsignal, there are two or more decomposition subsignals and be judged as vibration letter in step 6 Number, then judge there is multi-loop oscillation in the industrial process.

The present invention can improve the accuracy in detection and reliability of various dimensions oscillation behavior, and can be other oscillation source positioning Method provides data and supports, has important practical value at aspect of increasing economic efficiency.

Optionally, the industrial process is chemical process, and the present invention can directly using the measurable variable conduct of chemical process The output of process signal, all the output of process signals to be detected are obtained by field real-time acquisition, and are elapsed over time, constantly Collection and renewal process output signals to monitoring system.All Process History data first to being collected carry out respectively end points Continuation is processed, and then obtains decomposing subsignal set using improved multidimensional empirical mode decomposition, is calculated each and is decomposed subsignal Corresponding zero cross point regularity index, so as to deterministic process is with the presence or absence of vibration, and then judges whether multi-loop oscillation.

Preferably, in step 2, the end extending is following operation：One is taken in the left end of active procedure output signal Section waveform, find in the output of process signal with the left side waveform of this section of waveform spectrum degree of correlation highest waveform, and using the left side Side wave shape carries out continuation to the left end of active procedure output signal；In the same manner, the right-hand member in the output of process signal takes one section of waveform, Find in the output of process signal with the right waveform of this section of Waveform Matching degree highest waveform, using the right waveform to current mistake The right-hand member of journey output signal carries out continuation.

Preferably, in step 3, during the multidimensional empirical mode decomposition being improved, using random initializtion plus inverse function Ha Erdun (Halton) sequences of displacement judge form as the projection foundation of hypersphere surface sample using the amplitude fluctuations of double threshold As the shutdown standard of multidimensional intrinsic mode function screening process.

Preferably, in step 3, random initializtion adds the Ha Erdun sequence calculations that inverse function is replaced as follows：

Initial point is substituted using random positive integer in step 3-1, initial point series expansion；

The each term coefficient of step 3-2, Ha Erdun sequence is calculated with following formula：

Wherein, S_pHa Erdun sequence current coefficient values are represented, p represents a certain selected prime number, b₀,b₁,...,b_mRepresent initial Each term coefficient in point series expansion.Function π_p(b_i) be meant that：

Optimal Multiplier G is calculated using Wo Nuoke methods_p, then parameter w value as follows：

If 1. G_pIt is a primitive root of prime number p, then w values are G_p；

If 2. G_pIt is not a primitive root of prime number p, then w values are A_p.Wherein A_pNumerically most connect in the primitive root for being prime number p Nearly G_pOne.

The shutdown standard of the multidimensional intrinsic mode function screening process refer to the envelope Jing after the standardization of two norms- Amplitude ratio σ (t) meets condition：

Condition 3-1, the overwhelming majority in sequence σ (t), i.e. accounting are (1- α) × 100%；Meet condition σ (t) ＜ θ₁；

Condition 3-2, the remainder in sequence σ (t), i.e. accounting are α × 100%；Meet condition σ (t) ＜ θ₂；

Wherein α=0.05, θ₁=0.075, θ₂=0.75.

Preferably, in step 4, calculate it is each decompose subsignal zero cross point regularity index the step of it is as follows：

Step 4-1, counts the interval between all continuous zero cross points in current decomposition subsignalAnd calculate respectively IntervalAverageAnd standard deviation

Wherein k represents the level number for decomposing subsignal, and l represents the affiliated original signal of current decomposition subsignal in input process Numbering in data；

Step 4-2, using following formula zero cross point regularity index η is calculated^k,l：

Preferably, the threshold value in step 5 is 1.

The present invention has had the advantage that compared with prior art：

1st, encourage without the need for external add-in signal, will not also introduce additional disturbance to control system, non-intrusion type can be realized Detection with diagnosis.

2nd, while processing multiloop output data, the presence of plant-level oscillation is can detect, while coupling information between retention loop.

3rd, the signal decomposition method for being adopted realizes being automatically separated for non-stationary component in process data, compared to existing Other multivariate analytical techniques, its reliability is higher.

4th, the random initializtion for being proposed plus the Ha Erdun sequences of inverse function displacement can make hypersphere surface sample more homogeneous, and The double threshold amplitude fluctuations for being adopted shut down standard to noise, localised jitter more robust.

5th, quantizating index detection can be carried out to the oscillation behavior in each loop of industrial process, is commenting for loop performance to be detected Estimate and provide abundant data support with source of trouble diagnosis.

6th, without the need for process priori, without the need for being pre-designed wave filter, also it is not required to using the method for data driven type completely Carry out manual intervention.

Description of the drawings

Fig. 1 is the schematic flow sheet of chemical process in embodiment；

Fig. 2 is the output of process signal schematic representation of the control loop each to be detected gathered in embodiment；

Fig. 3 is all the output of process signals to be detected in embodiment Jing after spectrum waveform correlation matching process end extending Schematic diagram；

Fig. 4 is the schematic diagram in embodiment after the improved multidimensional empirical mode decomposition of the output of process signal；

Fig. 5 is method of the present invention flow chart.

Specific embodiment

Below by taking the fluidisation batch process device performance assessment of domestic certain factory as an example, to there is control valve viscosity property Multi-loop oscillation (plant-level oscillation) the behavioral value method of chemical process is described in detail.

As shown in figure 1, in process of production, fluid bed T4 needs to be supplied by inert gas to ensure its security, the gas Body was full of equipment before filler, kept stable in process of production, and purge gas are served as after collection phase, and via blower fan Effect, constantly circulates in a device.Inert gas enters storage tank through heat exchanger W1 heating-up temperatures, in the presence of blower fan V2 B3, after storage tank control pressure, inert gas enters fluidized bed T4, and fluid media (medium) is sprayed into after fluid bed T4 not from top atomizer It is disconnected to carry out fluid mapper process.The source of the gas flows out T4, while maintaining the pressure of product container B5.

In the production phase of batch process, because inert gas air feed equipment valve has viscosity, cause each in equipment There is same oscillation mode, i.e. multi-loop oscillation or plant-level oscillation phenomenon in pressure circuit.In order to study the inertia of the process Gas source fault propagation path, choosing 6 variables related to inert gas air pressure carries out multi-loop oscillation detection and analysis, right It is as shown in table 1 with the description of the variable of detection and numbering that these are used for mechanical shaking extraction.

Table 1

Numbering	Description	Remarks
			1	Heat exchanger W1 outlet gas pressures
2	Blower fan V2 air outlet pressure
			3	Storage tank B3 outlet gas pressures
4	Fluid bed T4 upper gas pressure	It is defined by fluid bed T4 first half pressure
			5	Fluid bed T4 lower gas pressure	By fluid bed T4 lower half, pressure is defined
6	Product container B5 internal pressures	Product container B5 tops nitrogen pressure

As shown in figure 5, a kind of process control loops detection method of oscillations based on improvement empirical mode decomposition, including：

Step 1, the output of process signal of all control loops to be detected in gatherer process gathers numbering 1-6 to strain The output of process signal of amount.

The method of gatherer process output signal is to record control loop to be detected within default each sampling interval In process data, and the process data collected in each sampling interval add previously gathered process data end End.

Sampling interval refers to the sampling interval of performance evaluation system.Process data elapses over time continuous renewal, per Jing The time span in a sampling interval is crossed, has new process data to be added to the end of previously acquired process data.Performance The sampling interval of assessment system is typically identical with the controlling cycle in industrial control system, it is also possible to select as the whole of controlling cycle Several times, limit to determine with specific reference to the requirement of real-time and memory data output of performance monitoring and industry spot.

The present embodiment gathers 6 the output of process signals after centralization as shown in Fig. 2 abscissa is sampling in Fig. 2 Point ordinal number, unit is Sample (sampling intervals of 1 Sample, one data of correspondence), ordinate be after centralization just Gas pressure under normal operating mode, unit is MPa.

Step 2, using spectrum waveform correlation matching process, to the output of process signal end extending is carried out.

Respectively left and right end extending process is carried out to each the output of process signal, it is therefore an objective to eliminate input signal to decomposition side The end effect of method affects, and its specific embodiment is：

Step 2-1, with the left end point x (t of the output of process signal₀) it is starting point, a part of waveform w of original signal is taken to the right (t₀:t₀+ l)=[x (t₀),x(t₀+1),L x(t₀+ l)], wherein l is to meet w (t₀:t₀+ l) only comprising zero cross point Maximum；

Step 2-2, finds w (t₀:t₀+ l) midpointWherein[] represent take toward just without The nearest integer in poor direction；

Step 2-3, right direction search in next sequence x (t) withValue identical point, is designated as x (t₁), with x (t₁) for intermediate point one section is extracted with w (t₀:t₀+ l) length identical waveform, it is designated as w (t₁-[l/2]:t₁+[l/2]-1)；

Step 2-4, calculates w (t₀:t₀+ l) and w (t₁-[l/2]:t₁+ [l/2] -1) spectrum waveform correlation matching degree m₁.Wherein The calculating of spectrum index of correlation can be according to prior art " Guo W, Huang L, Chen C, et al.Elimination of end effects in local mean decomposition using spectral coherence and applications for rotating machinery[J].Digital Signal Processing,2016,55:52-63. " calculate.

Step 2-5, repeat step 2-3 and step 2-4, until the output of process signal search terminates, can so obtain one Serial matching degree m=[m₁,m₂,L m_n]；

Step 2-6, finds maximum m in m_b, then corresponding wave band w (t_b-[l/2]:t_b+ [l/2] -1) it is and w (t₀:t₀ + l) most match one section of waveform, now the left end of signal x (t) can use w (t_b-l-[l/2]:t_b- [l/2]) carry out continuation；

Step 2-7, continuation is carried out using same method to the right endpoint of signal x (t).

The present embodiment is as shown in Figure 3 to the result that 6 selected the output of process signals carry out continuation.

Step 3, the multidimensional empirical mode decomposition being improved to the output of process signal after continuation obtains decomposing sub- letter Number.

The present invention is improved to multidimensional empirical mode decomposition of the prior art, remains former methodical all mathematics With calculate feature, simply hypersphere surface sample projection according to and multidimensional intrinsic mode function screening process shutdown standard (eventually Only condition) on modify, for same group of process data, compared to former method, capacity of decomposition is higher for improved decomposition method, The subsignal quantity of acquisition is less, iteration time faster, more suitable for the oscillation behavior of the former multidimensional data of analysis.Original multidimensional Empirical mode decomposition can be according to prior art " Rehman N, Mandic D P.Multivariate empirical mode decomposition[C]//Proceedings of The Royal Society of London A:Mathematical, Physical and Engineering Sciences.The Royal Society,2010,466(2117):1291- 1302. " carry out.

In improved multidimensional empirical mode decomposition, the projection foundation of hypersphere surface sample is referred to, Ha Erdun sequence structure processes In, initial point is substituted using random positive integer in initial point series expansion.Secondly, each term coefficient of Ha Erdun sequences to Lower formula is calculated：

Wherein, S_pHa Erdun sequence current coefficient values are represented, p represents a certain selected prime number, it is generally the case that prime number is from 2 Start up value successively.b₀,b₁,...,b_mRepresent each term coefficient in initial point series expansion.Permutation function π_p(b_i) implication It is：

The value of parameter w with utilize the calculated Optimal Multiplier G of Wo Nuoke methods_pRelevant, Wo Nuoke methods can foundation Prior art " Warnock T T.Computational investigations of low-discrepancy point sets II[M]//Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing.Springer New York,1995:354-361. " is carried out.Then the obtaining value method of parameter w is：

If 1. G_pIt is a primitive root of prime number p, then w values are G_p；

If 2. G_pIt is not a primitive root of prime number p, then w values are A_p.Wherein A_pNumerically most connect in the primitive root for being prime number p Nearly G_pOne.

In improved multidimensional empirical mode decomposition, the shutdown standard of multidimensional intrinsic mode function screening process is referred to, Jing bis- Envelope after norm standardization-amplitude ratio σ (t) meets condition：

Condition 3-1, the overwhelming majority (accounting is (1- α) × 100%) in sequence σ (t) meets condition σ (t) ＜ θ₁；

Condition 3-2, the remainder (accounting is α × 100%) in sequence σ (t) meets condition σ (t) ＜ θ₂。

Wherein parameter θ₁Ensure that the global low jitter of envelope signal, and parameter θ₂Signal is then allowed to there is local larger Deviant behaviour, parameter of the present invention is set to α=0.05, θ₁=0.075, θ₂=0.75.

The multidimensional empirical mode decomposition that the process data collection that the present embodiment is obtained to continuation is improved, decomposition for obtaining Signal is as shown in Figure 4.Each the output of process data are decomposed obtain 13 straton signals respectively, be show it is convenient, herein by the 1st to the 4th Layer is added together (C₁-C₄), it is added together (C by the 5th to the 9th layer₅-C₉), it is added together (C by the 11st to the 13rd layer₁₁-C₁₃), and Represent oscillating layer the 10th layer then individually shows (C₁₀).Assessed respectively using zero cross point regularity index proposed by the invention 6 decomposition subsignal { x corresponding to 10th layer^10,1,x^10,2,x^10,3,x^10,4,x^10,5,x^10,6}。

Step 4, calculates each zero cross point regularity index for decomposing subsignal, comprises the following steps that：

Step 4-1, counts the interval between all continuous zero cross points in current decomposition subsignalAnd calculate respectively IntervalAverageAnd standard deviationWherein k represents the level number for decomposing subsignal, and l represents current decomposition letter Numbering of the original signal belonging to number in input process data；

Step 4-2, using following formula zero cross point regularity index is calculated：

Zero cross point regularity index η^k,lIt is according to prior art " Thornhill N F, Huang B, Zhang H.Detection of multiple oscillations in control loops[J].Journal of Process Control,2003,13(1):Auto-correlation function zero cross point regularity index proposed in 91-100. " simplifies.From The effect of correlation function is the impact for removing noise to oscillation test, because multidimensional empirical mode decomposition filters noise level Remove, therefore can directly using the zero cross point regularity index of time domain.

The Analysis of Deep Implications of zero cross point regularity index is, for normal vibration signal, its all waveform for including Should all have identical span lengths, therefore η^k,l→ ∝, and in actual motion, due to environment and measure error etc. it is unfavorable because Element affects, party's law regulation η^k,l＞ 1 can determine that and greatly may also have in original signal oscillation behavior signal.In addition, zero cross point Regular index is also affected by zero cross point number in signal, if zero cross point number is very few in original signal, average, standard Poor estimation may all have a relatively large deviation, therefore present invention provide that, all-wave number have to be larger than and be equal in former the output of process signal 5。

In the present embodiment, all subsignals include more than 5 all-waves in the 10th layer, therefore calculate each decomposition letter respectively Number { x^10,1,x^10,2,x^10,3,x^10,4,x^10,5,x^10,6Corresponding to zero cross point regularity index { η^10,1,η^10,2,η^10,3,η^10,4, η^10,5,η^10,6}。

Step 5, if one of zero cross point regularity index for decomposing subsignal exceedes threshold value Ω, judges the control There is oscillation behavior in the corresponding subsignal that decomposes in loop processed.The zero cross point of the 10th layer of corresponding 6 decomposition subsignal is regular Index is respectively η^10,1=3.83, η^10,2=2.56, η^10,3=3.04, η^10,4=3.83, η^10,5=1.56, η^10,6=1.76.It is aobvious So the corresponding zero cross point regularity index of all decomposition subsignals exceedes given threshold value Ω=1, can affirm these decomposition Subsignal is oscillating component, as shown in Figure 4.

, when same layer decomposes in subsignal, there are two or more and decompose subsignal performance oscillation behavior in step 6, Then can determine that the process has multi-loop oscillation (plant-level oscillation).Present case detects 6 loops while there is vibration, therefore can To be judged to plant-level oscillation.

Using the inventive method, on the basis of multi-loop oscillation detection is carried out, additionally it is possible to each process control loops Oscillation behavior carries out quantitative determination, obtains regular degree and the cycle of these oscillating components, is evaluation and the failure of oscillation behavior Source diagnosis provides abundant data and supports.

Claims (7)

1. a kind of industrial process multi-loop oscillation is extracted and detection method, it is characterised in that included：

Step 1, gathers the output of process signal of all control loops to be detected in industrial process；

Step 2, using spectrum waveform correlation matching process, to each the output of process signal end extending is carried out；

Step 3, the multidimensional empirical mode decomposition being improved to the output of process signal after end extending obtains decomposing sub- letter Number；

Step 4, calculates the zero cross point regularity index that each layer decomposes subsignal；

Step 5, judges whether each zero cross point regularity index exceedes threshold value, if exceeding threshold value, corresponding decomposition subsignal It is judged to oscillator signal.

2. industrial process multi-loop oscillation as claimed in claim 1 is extracted and detection method, it is characterised in that in step 2, institute End extending is stated for following operation：One section of waveform is taken in the left end of active procedure output signal, in finding the output of process signal With the left side waveform of this section of waveform spectrum degree of correlation highest waveform, and using the left side of the left side waveform to active procedure output signal End carries out continuation；In the same manner, the right-hand member in the output of process signal takes one section of waveform, find in the output of process signal with the Duan Bo The right waveform of shape matching degree highest waveform, continuation is carried out using the right waveform to the right-hand member of active procedure output signal.

3. industrial process multi-loop oscillation as claimed in claim 1 is extracted and detection method, it is characterised in that in step 3, entered During the improved multidimensional empirical mode decomposition of row, the Ha Erdun sequences replaced using random initializtion plus inverse function are adopted as hypersphere The projection foundation of sample, judges form as the shutdown mark of multidimensional intrinsic mode function screening process using the amplitude fluctuations of double threshold It is accurate.

4. industrial process multi-loop oscillation as claimed in claim 3 is extracted and detection method, it is characterised in that in step 3, with The Ha Erdun sequence calculations of machine initialization plus inverse function displacement are as follows：

Initial point is substituted using random positive integer in step 3-1, initial point series expansion；

The each term coefficient of step 3-2, Ha Erdun sequence is calculated with following formula：

$S_p=\frac{{\mathrm{\π}}_p\left(b_0\right)}{p}+\frac{{\mathrm{\π}}_p\left(b_1\right)}{p^2}+...+\frac{{\mathrm{\π}}_p\left(b_m\right)}{p^{m+1}};$

Wherein, S_pHa Erdun sequence current coefficient values are represented, p represents a certain selected prime number, b₀,b₁,…,b_mRepresent initial point series Each term coefficient in expansion.Function π_p(b_i) be meant that：

${\mathrm{\π}}_p\left(b_i\right)={\mathrm{wb}}_i^{-}$

Optimal Multiplier G is calculated using Wo Nuoke methods_p, then parameter w value as follows：

If 1. G_pIt is a primitive root of prime number p, then w values are G_p；

If 2. G_pIt is not a primitive root of prime number p, then w values are A_p.Wherein A_pBe prime number p primitive root in numerically closest to G_p's One.

5. industrial process multi-loop oscillation as claimed in claim 3 is extracted and detection method, it is characterised in that in step 3, institute State multidimensional intrinsic mode function screening process shutdown standard refer to the envelope Jing after the standardization of two norms-amplitude ratio σ (t) is full Sufficient condition once：

Condition 3-1, the overwhelming majority in sequence σ (t), i.e. accounting are (1- α) × 100%；Meet condition σ (t) ＜ θ₁；

Condition 3-2, the remainder in sequence σ (t), i.e. accounting are α × 100%；Meet condition σ (t) ＜ θ₂；

Wherein α=0.05, θ₁=0.075, θ₂=0.75.

6. industrial process multi-loop oscillation as claimed in claim 1 is extracted and detection method, it is characterised in that in step 4, meter The step of calculating the zero cross point regularity index of each decomposition subsignal is as follows：

Step 4-1, counts the interval between all continuous zero cross points in current decomposition subsignalAnd the difference counting periodAverageAnd standard deviation

Wherein k represents the level number for decomposing subsignal, and l represents the affiliated original signal of current decomposition subsignal in input process data In numbering；

Step 4-2, using following formula zero cross point regularity index η is calculated^k,l：

${\mathrm{\η}}^{k,l}=\frac{1}{3}\×\frac{{\mathrm{\μ}}_{Z_i}^{k,l}}{{\mathrm{\σ}}_{Z_i}^{k,l}}.$

7. industrial process multi-loop oscillation as claimed in claim 1 is extracted and detection method, it is characterised in that in step 5 Threshold value is 1.