CN101799674B - A Method for Analyzing the Service State of NC Equipment - Google Patents

Abstract

The invention relates to a method for analyzing the service state of numerical control equipment, belonging to the technology for monitoring the service state of major equipment and forecasting the service life of the major equipment. The method comprises the following steps: firstly, recognizing the service states of a plurality of the major parts of the numerical control equipment through the characteristic physical quantity acquired by a multisensor; then, forecasting the service state of the complete machine through a classification model of a support vector machine established by the statistical learning theory; and calculating the residual service lives of the major parts and the complete machine through a 'hidden semi-Markov' random process model. The method of the invention not only can be used for recognizing the current running states of the parts but also can be used for forecasting the residual service life of the parts. The current running state and the residual service life of the complete machine are obtained by the classification forecasting method of the support vector machine according to the operation result of each part. The invention provides a new method for decision support of preventive maintenance.

Description

A kind of method for analyzing service state of numerical control equipment

Technical field

The invention belongs to great equipment service state monitoring and forecasting technique in life span, be specifically related to a kind of method that is used for the comprehensive service state of identification numerical control equipment and its remaining life of prediction, it can provide important reference for the fail-safe analysis and the maintenance decision problem of numerical control equipment.

Technical background

Numerical control equipment is as machine-tool, and Application in Manufacturing Industry is more and more universal in China.Numerical control equipment, particularly great, critical equipment are in case catastrophic failure in the course of the work with having a strong impact on the production efficiency of enterprise, brings massive losses to enterprise.In order to make the running of numerical control equipment near-zero fault, need the service reliability state of timely analyzing numerically controlled equipment, accurately predict its residue time between failures, to take rational preventative maintenance strategy in advance, prevent the generation of fault.Numerical control equipment is typically mechanical, electrical, liquid complex apparatus, along with electronic technology and development of computer, for the status monitoring means of complex apparatus great lifting has been arranged, and therefore, the running status of numerical control equipment is compared the careful possibility that is divided into.

In great equipment service state monitoring and life prediction research field, there are a large amount of papers and patent documentation to deliver.Aspect linear discriminant analysis, document [1] provides detailed introduction, and main thought is utilization folk prescription difference analysis, calculates the F test value; Document [1] [2] has provided the application of HSMM model aspect the identification of equipment degenerate state, this model is the expansion of Markov chain, it is a doubly stochastic process, promptly not only state is at random to the transfer of state, and the observation of each state also is at random, the main algorithm of HSMM model comprises forward direction-back to algorithm, and this algorithm mainly is to solve the probability that produces a certain observation sequence, and the Baum-Welch algorithm mainly is the parameter estimation problem that solves model. Document [3] describes the application of supporting vector machine model aspect numerical control equipment parts and whole aircraft reliability assessment in detail, the basic thought of this model is life-span of will obtain and the service state data input vector as model after the zero dimension processing, utilize kernel function that input vector is mapped to high-dimensional feature space, carry out The Fitting Calculation at high-dimensional feature space, obtain optimum non-linear regression function, then, combining with digital control is equipped in the probability distribution function of various influence factors under stipulated time and the rated condition, the computing unit reliability.At last, according to the reliability structure model of numerical control equipment, find out the minimal cut set of system, the computing system reliability.

Wherein the state analysis method document of concerned power generation equipment, nuclear equipment, aerospace equipment etc. is more, the analyzing service state of numerical control equipment document is less relatively, document about the prediction of numerical control equipment remaining life is rare especially, and generally there is following problem in these researchs: (1) biases toward fault diagnosis, only the state of equipment is done the simple division of " normally " and " fault ", do not consider the failure procedure of gradual change, in fact there is considerable deterioration failure state in equipment.(2) state of equipment complete machine only is concerned about in most of research, and each parts to component devices do not carry out detailed fail-safe analysis, do not have the service state and the residual life of analysis component yet.

List of references

[1]Ming?Dong，David?He.Hidden?semi-Markov?model-based?methodology?formulti-sensor?equipment?health?diagnosis?and?prognosis[J].European?Journalof?Operational?Research，2007，178：858-878.

[2] Ceng Qinghu, Qiu Jing, champion Liu. based on hidden semi-Markov model equipment degenerate state recognition methods research [J]. machine science and technology, 2008,27 (4): 429-432.

[3] Wu Jun. based on the service reliability of numerical control equipment appraisal procedure and the application [M] of performance parameter. Wuhan: the Central China University of Science and Technology, 2008.

Summary of the invention

The object of the present invention is to provide a kind of method for analyzing service state of numerical control equipment, this method can provide the new method of decision support for preventative maintenance.

A kind of method for analyzing service state of numerical control equipment is characterized in that, this method comprises the steps:

The 1st step was determined the vitals and the service state thereof of numerical control equipment to be analyzed, and establishing vitals quantity is m, and i represents the sequence number of vitals, i ∈ 1,2 ..., and m}, the number of states of i vitals is L _i, then the state set of i vitals is expressed as

Represent i vitals complete failure;

The 2nd step was utilized the characteristic signal of sensor acquisition vitals;

The characteristic signal of the 3rd step to above-mentioned collection carries out the information fusion processing, obtains the service state composite character vector Y of each vitals _i

The 4th step was discerned and predicting residual useful life the service state of each vitals of numerical control equipment:

The process of service state identification is:

Step (a1): set up model

The HSMM that is expressed as with latent-semi-Markov model of i vitals _i(λ _i)=(L _i, M _i, A _i, D _i, B _i, π _i), wherein, latent state is that the quantity of the service state of parts i is L _i, the possible observed reading number of each latent state correspondence is M _i, initial state distribution

State transition probability matrix

Represent that i vitals is from j service state s _IjJump to k service state s _IkProbability, j, k represent the sequence number of service state; Maximum rating residence time D _i, the observed value matrix

The service state of expression vitals i is s _Ij, but the service state of the vitals i that observes is s _IkProbability;

Step (a2): model training

At first to the vectorial Y of the service state composite character of vitals i _iDo vector quantization and handle the service state s that obtains dispersing _Ij, j=1,2 ..., L _i, adopt Bao Mu-Wei Erqi (Baum-Welch) algorithm to the model training then, promptly solve the parameter estimation problem of model, obtain model parameter A _i, D _i, B _i, π _iEstimated value

Successively all service states of vitals i are trained, obtain the latent-semi-Markov model of every kind of service state;

The identification of step (a3) service state:

After model training is finished, to the service state composite character vector of new collection, calculate the log-likelihood function value of each service state, the state of getting the greatest measure correspondence is the service state of current vitals;

The predicting residual useful life process is:

Step (b1): establish vitals service state s _IjResidence time D (s _Ij) the single Gaussian distribution N of obedience (μ (s _Ij), σ ²(s _Ij)), the life cycle of parts is T, satisfies:

Then to D (s _Ij) expectation value μ (s _Ij) and variances sigma ²(s _Ij) carry out parameter estimation, obtain the estimated value of average and variance

With And definition D (s _Ij) estimated value

For:

$\hat{D}\left(s_{\mathrm{ij}}\right)=\widehat{\mathrm{\μ}}\left(s_{\mathrm{ij}}\right)+\mathrm{\ρ}{\widehat{\mathrm{\σ}}}^2\left(s_{\mathrm{ij}}\right)$

Wherein $\mathrm{\ρ}=(T-\underset{j=1}{\overset{L_i}{\mathrm{\Σ}}}\widehat{\mathrm{\μ}}\left(s_{\mathrm{ij}}\right))/\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\widehat{\mathrm{\σ}}}^2\left(s_{\mathrm{ij}}\right);$

Step (b2): j service state establishing i parts is { s _Ij, i=1,2 ... m, j=1,2 ..., L _i, its remaining life is RUL _Ij, set up following recurrence equation formula, estimate the remaining life of parts;

$\left\{\begin{array}{c}{\mathrm{RUL}}_{\mathrm{ij}}={{\hat{a}}^i}_{j,j}\×[\hat{D}\left(s_{\mathrm{ij}}\right)+{\mathrm{RUL}}_{\mathrm{ij}+1}]+\underset{k=j+1}{\overset{L_i-1}{\mathrm{\Σ}}}{{\hat{a}}^i}_{j,k}\×{\mathrm{RULi}}_k\\ {\mathrm{RUL}}_{i{L}_i-1}={{\hat{a}}^i}_{L_i-1,L_i-1}\×\hat{D}\left(s_{{\mathrm{iL}}_i-1}\right)\end{array}\right.$

Wherein

With

Be state transition probability:

Expression vitals i keeps j service state s _IjProbability,

Expression vitals i is from j service state s _IjTransfer to k service state s _IkProbability.

The service state that the 5th step adopted support vector machine classification method to set up between numerical control equipment complete machine and the building block is got in touch, and finishes identification of numerical control equipment complete machine service state and predicting residual useful life.

Characteristics of the present invention are, provide a cover practical calculation method for having identification of military service running status and the remaining life prediction that a plurality of vitals and complete machine and parts all comprise the numerical control equipment of multiple failure state.The present invention can characterize physical quantity from multimode, picks out the service state of vitals, and analyzes the service state of complete machine.On this basis, can also predict the remaining life of parts and complete machine.This fail-safe analysis and maintenance decision analysis for complete machine and parts provides important reference.If words with good conditionsi, achievement of the present invention can also and the upper layer information management system of factory, as MES, system dockings such as ERP, thereby provide solid reference information for the formulation of plant produced plan and maintenance schedule.

Description of drawings

Fig. 1 is the basic process of numerical control equipment complete machine analyzing service state and remaining life prediction;

The numerical control equipment single part service state identifying of Fig. 2 is based on latent-semi-Markov model.

Embodiment

The inventive method at first by multi-sensor collection to the sign physical quantity identify the service state of a plurality of vitals of numerical control device, the support vector machine disaggregated model of setting up by Statistical Learning Theory dopes the service state of complete machine then, and go out the remaining life of vitals and complete machine, thereby the new method of decision support is provided for preventative maintenance by " latent-semi-Markov " Model Calculation.Below in conjunction with accompanying drawing and example the present invention is done explanation in further detail.

As shown in Figure 1, analytical approach of the present invention may further comprise the steps:

The first step is determined the vitals and the service state thereof of numerical control equipment to be analyzed.

According to 4 factors such as component function, failure effect, historical failure data statistics and parts monitoring property, adopt expert's point system or other method to calculate the parts importance degree of numerical control equipment.According to the design data of numerical control equipment, determine the service state of each vitals and complete machine.If vitals quantity is m, i represents the sequence number of vitals, i ∈ 1,2 ..., and m}, the service state quantity of i vitals is L _i, then the service state set representations of i vitals is Be parts complete failure.If the complete machine number of states is L, the service state set representations of complete machine is S={s ₁, s ₂..., s _L, s _LBe complete machine complete failure.

Second step was utilized the characteristic signal of sensor acquisition vitals.

Utilize testing tool such as acceleration transducer to gather the vibration signal of vitals in the numerical control equipment process, the group number of sampling interval and each acquired signal can be decided according to enterprise practical conditions.

The characteristic signal of the 3rd step to above-mentioned collection carries out the information fusion processing.

The present invention adopts the linear discriminant analysis method, on the basis of Labview software, the vibration signal of each sensor acquisition is carried out information fusion handle, and obtains the composite character vector of unit status.Concrete steps are as follows:

Step (1): at first the vibration signal that collects is carried out the denoising pre-service in Labview software, and then its wavelet transformation function bag that provides is provided, pretreated vibration signal is carried out feature extraction, obtain the vibration signal characteristics vector, note O _IrBe the vibration signal characteristics vector of r sensor of i vitals, r=1,2 ... N _I0, N _I0The summation of i vitals all the sensors is measured in expression.

Step (2): adopt the linear discriminant analysis method respectively the vibration performance vector of each vitals to be carried out the information fusion of characteristic layer, obtain the service state composite character vector of each vitals.

The present invention adopts the one-factor analysis of variance to give the weight of each sensor characteristics information, and the F test value obtains by the ratio of deviation in deviation between the calculating sensor sets of signals and the average sensor signal group.The F test value expression formula of r sensor measurement signal of i vitals is:

Wherein, n _iThe each sets of signals number of measuring of expression, S _IARepresent sum of squares of deviations between the sensor signal group of i vitals, S _IEThe interior sum of squares of deviations of sensor signal group of representing i vitals, F _IrThe F test value of representing r sensor measurement signal of i vitals.

The weight of r sensor of i vitals is made as w _Ir, computing formula is:

The service state composite character vector of i vitals is made as Y _i, computing formula is:

O wherein _IrIt is the vibration performance vector of r sensor of i vitals.

The 4th step is to the service state identification and the predicting residual useful life of each vitals of numerical control equipment.

HSMM (Hidden Semi-Markov Model, latent-semi-Markov model) is the expansion of HMM (HiddenMarkov Model, hidden Markov model).The state presence time is the limitation of exponential distribution in the hidden Markov model in order to improve, and on the basis of hidden Markov model, latent-semi-Markov model allows to distribute according to the self-defined residence time of practical problems.In order to reduce computation complexity, we adopt single Gaussian distribution as state presence time probability distribution function.

I vitals service state s a certain with it _Iu, u=1,2 ..., L _iThe parameter-definition of corresponding concealing-semi-Markov model is as follows: latent state (being the service state of vitals i) quantity is L _i, the possible observed reading number of each latent state correspondence is

Initial state distribution

State transition probability matrix

Represent that i vitals is from j service state s _IjJump to k service state s _IkProbability, j, k represent the sequence number of service state; The maximum rating residence time

The observed value matrix

The service state of expression vitals i is s _Ij, but the service state of the vitals i that observes is s _IkProbability.Like this, i vitals and the corresponding s of its a certain service state _IuLatent-semi-Markov model also can write:

Numerical control equipment vitals service state identification of the present invention conceals-semi-Markov model based on above-mentioned, and as shown in Figure 2, this process can be divided into two parts: first is a model training, and second portion is service state identification.

Step (a1): model training.At first to the vectorial Y of the composite character of vitals i _iDo vector quantization and handle the service state estimated value s that obtains dispersing _Ij, j=1,2 ..., L _iAdopt the Baum-Welch algorithm to the model training then, promptly solve the parameter estimation problem of model: obtain model parameter Estimated value

Repeating above-mentioned vector quantization then handles, obtain all service state estimated values of vitals i, and successively to each service state of vitals i corresponding latent-semi-Markov model trains, and obtains the estimates of parameters of latent-semi-Markov model of every kind of service state correspondence.

Step (a2): service state identification.Every kind of service state of vitals i corresponding latent-after the semi-Markov model training finishes, to certain composite character vector of at a time newly gathering at vitals i, doing vector quantization handles, obtain this service state estimated value of vitals i constantly, each service state correspondence of the above-mentioned vitals i that trains of substitution conceals-semi-Markov model successively, calculate the log-likelihood function value of each service state, the service state of getting the greatest measure correspondence is the current service state of vitals i.

Method below the present invention has adopted is calculated the remaining life of vitals i:

I vitals and its all service states corresponding latent-parameter-definition of semi-Markov model is as follows: latent number of states is

The possible observed reading number of each latent state correspondence is

Initial state distribution State transition probability matrix

Represent that i vitals is from j service state s _IjJump to k service state s _IkProbability, j, k represent the sequence number of service state; The maximum rating residence time

The observed value matrix

The service state of expression vitals i is s _Ij, but the service state of the vitals i that observes is s _IkProbability.I vitals and its all service states corresponding latent-semi-Markov model is expressed as:

Step (b1):,, set up latent-semi-Markov model at all service states of vitals i as the sample of training data the estimated value of all service states of vitals i that obtain from step a1 And adopt the Baum-Welch algorithm to the model training, and solve the parameter estimation problem, obtain the estimated value of parameter

Step (b2): the service state collection of vitals i is { s _Ij, i=1,2 ... m, j=1,2 ..., L _i, each service state s _IjResidence time D (s _Ij) the single Gaussian distribution N of obedience (μ (s _Ij), σ ²(s _Ij)).The life cycle of vitals i is T, satisfies:

Then to D (s _Ij) expectation value μ (s _Ij) and variances sigma ²(s _Ij) carry out parameter estimation, obtain the estimated value of average and variance

With And definition D (s _Ij) estimated value be:

$\hat{D}\left(s_{\mathrm{ij}}\right)=\widehat{\mathrm{\μ}}\left(s_{\mathrm{ij}}\right)+\mathrm{\ρ}{\widehat{\mathrm{\σ}}}^2\left(s_{\mathrm{ij}}\right)$

Wherein $\mathrm{\ρ}=(T-\underset{j=1}{\overset{L_i}{\mathrm{\Σ}}}\widehat{\mathrm{\μ}}\left(s_{\mathrm{ij}}\right))/\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\widehat{\mathrm{\σ}}}^2\left(s_{\mathrm{ij}}\right).$

Step (b3): the current service state value of establishing the vitals i that step (a2) identifies is s _Ij, i=1,2 ... m, j=1,2 ..., L _i, its remaining life is RUL _Ij, can set up following recurrence equation formula, thereby estimate the remaining life of vitals i.

$\left\{\begin{array}{c}{\mathrm{RUL}}_{\mathrm{ij}}={\hat{a}}_{j,j}^{i,c}\×[\hat{D}\left(s_{\mathrm{ij}}\right)+{\mathrm{RUL}}_{\mathrm{ij}+1}]+\underset{k=j+1}{\overset{L_i-1}{\mathrm{\Σ}}}{\hat{a}}_{j,k}^{i,c}\×{\mathrm{RUL}}_{\mathrm{ik}}\\ {\mathrm{RUL}}_{i{L}_i-1}={\hat{a}}_{L_i-1,L_i-1}^{i,c}\×\hat{D}\left(s_{{\mathrm{iL}}_i-1}\right)\end{array}\right.$

Wherein

With

Be state transition probability:

Expression vitals i keeps j service state s _IjProbability,

Expression vitals i is from j service state s _IjTransfer to k service state s _IkProbability.

The 5th step number control equipment complete machine conceals-semi-Markov modeling and predicting residual useful life.

Set up all vitals latent-semi-Markov model after, the present invention adopts support vector machine (SVM:Support Vector Machine) sorting technique to set up service state contact between numerical control equipment complete machine and the building block.It is as follows to utilize SVM to carry out numerical control equipment complete machine service state identification step:

Step (5.1): set up the support vector machine disaggregated model, the service state data sample capacity of establishing above-mentioned all vitals that obtain and complete machine is n, and n 〉=10.Can be expressed as: D={ (x _f, y _f) | x _f=(x _F1, x _F2..., x _Fm) ∈ S ₁* S ₂* ... * S _m, y _f∈ S, f ∈ 1,2 ..., n}}, wherein, x _fBe m dimension input vector, the vitals service state of expression numerical control equipment, y _fIt is the observed reading of complete machine service state.Utilize above sample data, can train supported vector machine disaggregated model, wherein kernel function adopts radial basis function: K (x _f, x)=exp (‖ x-x _f‖ ²/ σ ²), x represents numerical control equipment vitals service state sequence to be diagnosed, σ is the standard deviation of vitals service state sample data, the desirable σ ∈ (2 of its initial range ^-13, 2 ³⁰), the training of support vector machine has ready-made ripe algorithm, no longer superfluous herein chatting.

Step (5.2): the identification of numerical control equipment complete machine service state, with the service state of certain moment all vitals of numerical control equipment, the above-mentioned supporting vector machine model that trains of substitution obtains this current service state of numerical control equipment complete machine constantly.

Step (5.3): the numerical control equipment complete machine conceals-semi-Markov modeling and remaining life prediction.

With all service states of numerical control equipment complete machine corresponding latent-parameter-definition of semi-Markov model is as follows: latent number of states is L, and the possible observed reading number of each latent state correspondence is M, initial state distribution π=(π ₁..., π _L), state transition probability matrix A=(a _{J, k}) _{L * L}, (a _{J, k}) represent that numerical control equipment is from j service state s _jJump to k service state s _kProbability, j, k represent the sequence number of service state; Maximum rating residence time D, observed value matrix B=(b _{I, k}) _{L * M}, (b _{I, k}) represent that the service state of numerical control equipment complete machine is s _i, but the service state of the numerical control equipment complete machine that observes is s _kProbability.With all service states of numerical control equipment complete machine corresponding latent-semi-Markov model can be expressed as: HSMM _s(λ _s)=(L, M, A, D, B, π).

According to above-mentioned SVM prediction model, choose 10 to 20 moment point successively, and according to each vitals service state constantly, prediction obtains this service state of numerical control equipment complete machine constantly; All moment point complete machine service state predictions finish, and gather the historical service state sequence that becomes complete machine; Adopt Baum-Welch algorithm training complete machine to conceal-semi-Markov model HSMM then _s(λ _s)=(L, M, A, D, B π), obtains the estimated value of model parameter

The process of calculating the complete machine remaining life is as follows:

The service state set representations of numerical control equipment complete machine is S={s ₁, s ₂..., s _L, note s _qThe service state value of representing q complete machine, q=1,2 ..., L.

Step (c1): establish complete machine service state s _qResidence time D (s _q) the single Gaussian distribution N of obedience (μ (s _q), σ ²(s _q)), the life cycle of complete machine is T _s, satisfy:

Then to D (s _q) expectation value μ (s _q) and variances sigma ²(s _q) carry out parameter estimation, obtain the estimated value of average and variance

With

And definition D (s _q) estimated value For:

$\hat{D}\left(s_q\right)=\widehat{\mathrm{\μ}}\left(s_q\right)+{\mathrm{\ρ}}_s{\widehat{\mathrm{\σ}}}^2\left(s_q\right)$

Wherein ${\mathrm{\ρ}}_s=(T_s-\underset{q=1}{\overset{L}{\mathrm{\Σ}}}\widehat{\mathrm{\μ}}\left(s_q\right))/\underset{q=1}{\overset{N}{\mathrm{\Σ}}}{\widehat{\mathrm{\σ}}}^2\left(s_q\right);$

Step (c2): establishing step (5.2) complete machine service state classification prediction result is s _q, q=1,2 ..., L, its remaining life are RUL _q, set up following recurrence equation formula, estimate the remaining life of complete machine;

$\left\{\begin{array}{c}{\mathrm{RUL}}_q={\hat{a}}_{q,q}\×[\hat{D}\left(s_q\right)+{\mathrm{RUL}}_{q+1}]+\underset{t=q+1}{\overset{L-1}{\mathrm{\Σ}}}{\hat{a}}_{q,t}\×{\mathrm{RUL}}_t\\ {\mathrm{RUL}}_{L-1}={\hat{a}}_{L-1,L-1}\×\hat{D}\left(s_{L-1}\right)\end{array}\right.$

Wherein With

Be state transition probability: The expression complete machine keeps q service state s _qProbability,

The expression complete machine is from q service state s _qTransfer to t service state s _tProbability.

Example: the analyzing service state of numerical control DM4600 type vertical milling machine and remaining life prediction.

1:, adopt expert's point system to determine the parts importance degree of numerical control equipment according to 4 factors such as component function, failure effect, historical failure data statistics, parts monitoring property.The result is: six parts such as main shaft, cutter, servo feed system, anchor clamps, worktable and hydraulic system are as vitals and carry out the service state monitoring.Rule of thumb, the service state of main shaft is divided into 5 kinds: I, II, III, IV, V, and wherein I represents " normal fully ", and V represents " complete failure ", and the service state of other 5 kinds of vitals is 4 kinds, and in addition, the service state of complete machine also is divided into 5 kinds.

2: monitoring of vitals service state and service state data aggregation.Adopt the sign amount of vibration signal, arranged a plurality of measuring points at each signal as vitals service state feature.Adopt virtual instrument Labview software as analysis tool, the hardware that whole test system needs only is common cable, data collecting card and some sensors, and extensibility also strengthens greatly.

3: set up the HSMM model of each vitals each service state except that total failure mode and the HSMM model of vitals Life cycle.Virtual instrument software Labview does not provide ready-made HSMM modeling program storehouse.We have developed a cover HSMM modeling program according to the statistical inference process of HSMM model ourselves, are embedded into the virtual instrument software the inside.Just begin training pattern then, the vitals state composite character vector that promptly Labview is obtained is as importing, by the parameter estimation means, train the HSMM model of each service state correspondence of each vitals, use the HSMM model of the historical data training vitals Life cycle of all service states of vitals then.In this example, such vitals HSMM model one has 1 * 6+5 * 5=31.

4: discrete sampling obtains the sample of the service state corresponding relation of complete machine and parts, form such as following table:

Utilize above data, train supported vector machine disaggregated model.We have adopted SVMC (the Support Vector Machine Classifier) instrument of own exploitation in this example, and are integrated among the Labview.

The service state sample of 1 parts and complete machine

5: utilize 4 supporting vector machine models of setting up, according to complete machine active time order, choose 10 to 20 moment point successively, according to this vitals service state constantly, prediction obtains the service state of complete machine, obtain the military service historic state sequence of complete machine at last, by the parameter estimation means, training obtains the HSMM model of complete machine military service life cycle.

By the 3-5 step, finish the structure of HSMM model and SVM model.

6: real-time data acquisition and vitals state identification in the numerical control equipment military service process.Utilize the method in 2, obtain the operation characteristic data of each parts of DM4600 lathe in real time, such as tool parts, 4 branch state HSMM models at cutter, utilize maximum Likelihood, pick out the current service state of cutter, such as being service state II.

7: the prediction of vitals remaining life.Utilize state presence time estimation formulas provided by the invention and remaining life recurrence equation formula, and, calculate the remaining life of vitals according to the current service state of 6 vitals that picked out.In this example, the service state of some time icking tool tool is II, and the remaining life that calculates cutter is 10 hours.

8: the complete machine state identification.Carve at a time,, utilize 4 supporting vector machine models of setting up then, dope the current service state of complete machine by 6 current states that pick out all vitals.In this example, the service state of certain moment main shaft is II, and the service state of cutter is II, the service state of servo feed system is II, and the service state of anchor clamps is I, and the service state of worktable is II, the service state of hydraulic system is I, and the service state that prediction obtains complete machine is III.

9: the prediction of complete machine remaining life.Utilize state presence time estimation formulas provided by the invention and remaining life recurrence equation formula, and, calculate the remaining life of complete machine according to the current service state of 8 complete machines that doped.For complete machine service state in 8 is the situation of III, and the remaining life that calculates complete machine is about 50 hours, and the duration of III level service state is about 40 hours, enter IV level state after, again through more than 10 hour complete machines with complete failure.

The above 6-9 step is finished analyzing service state of numerical control equipment and remaining life prediction.

The present invention not only is confined to above-mentioned embodiment; persons skilled in the art are according to content disclosed by the invention; can adopt other multiple embodiment to implement the present invention; therefore; every employing project organization of the present invention and thinking; do some simple designs that change or change, all fall into the scope of protection of the invention.

Claims (1)

1. A method for analyzing the state of service of numerical control equipment, is characterized in that, the method comprises the following steps: The first step is to determine the important components of the CNC equipment to be analyzed and their service status. Let the number of important components be m, i represents the serial number of important components, i∈{1,2,...,m}, the i-th important component The number of states is L _i , then the state set of the i-th important component is expressed as

Indicates that the i-th important component fails completely; Step 2 Use sensors to collect characteristic signals of important components; Step 3 Carry out information fusion processing on the characteristic signals collected above, and obtain the service state mixed characteristic vector Y _i of each important component; Step 4 Identify the service state and predict the remaining life of each important component of the CNC equipment: The process of service status identification is: Step (a1): Build a model Express the hidden-semi-Markov model of the i-th important component as HSMM _i (λ _i )=(L _i , M _i , A _i , D _i , B _i , π _i ), where the hidden state is the component The number of serving states of i is L _i , the number of possible observations corresponding to each hidden state is M _i , and the initial state distribution

State Transition Probability Matrix

Indicates the probability that the i-th important component jumps from the j-th service state s _ij to the k-th service state s _ik , j, k represents the serial number of the service state; the maximum state residence time D _i , the observation matrix

Indicates the probability that the service state of important component i is s _ij , but the observed service state of important component i is s _ik ; Step (a2): Model training Firstly, do vector quantization on the mixed eigenvector Y _i of service state of important component i to obtain discrete service state s _ij , j=1, 2,..., L _i , and then use Baum-Welch (Baum-Welch) The algorithm trains the model, that is, solves the parameter estimation problem of the model, and obtains the estimated values of the model parameters A _i , D _i , B _i , and π _i

Train all the service states of the important component i in sequence to obtain the hidden-semi-Markov model of each service state; Step (a3) service state identification: After the model training is completed, calculate the logarithmic likelihood function value of each service state for the newly collected service state mixed eigenvector, and take the state corresponding to the maximum value as the service state of the current important component; The remaining life prediction process is: Step (b1): Assume that the residence time D(s _ij ) of an important component in service state s _ij obeys a single Gaussian distribution N(μ(s _ij ), σ ² (s _ij )), and the life cycle of the component is T, which satisfies:

Then estimate the expected value μ(s _ij ) and variance σ ² (s _ij ) of D(s _ij ) to obtain the estimated value of mean and variance

and

and define an estimate of D(s _ij )

for: $\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>$ in $\mathrm{\&rho;}=(T-\underset{j=1}{\overset{L_i}{\mathrm{\&Sigma;}}}\widehat{\mathrm{\&mu;}}\left({\mathrm{the\; s}}_{\mathrm{ij}}\right))/\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&sigma;}}}^2\left({\mathrm{the\; s}}_{\mathrm{ij}}\right);$ Step (b2): Let the j-th service state of the i-th component be {s _ij }, i=1, 2, ... m, j = 1, 2, ..., L _i , and its remaining service life is RUL _ij , Establish the following recurrence equation to estimate the remaining service life of the components; $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; RUL</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>{{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; [</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; RUL</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; RULi</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\\ {\mathrm{<span\; class="notranslate">\; RUL</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; a</span>}}}^{\mathrm{<span\; class="notranslate">\; i</span>}}}_{{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&times;</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>\end{array}\right.$ in

and

is the state transition probability:

Indicates the probability that an important component i maintains the jth service state s _ij ,

Indicates the probability that an important component i is transferred from the jth service state s _ij to the kth service state s _ik ; Step 5 Use the support vector machine classification method to establish the service status relationship between the CNC equipment and its components, and complete the service status identification and remaining life prediction of the CNC equipment; Step 5 includes the following process: Step (5.1): Establish a support vector machine classification model, set the sample size of the service status data of important components and complete machines obtained above as n, and n≥10, expressed as: D={(x _f , y _f )|x _f = (x _f1 , x _f2 ,...,x _fm )∈S ₁ ×S ₂ ×...×S _m , y _f ∈ S, f∈{1,2,...,n}}, where x _f is m dimensional input vector, which represents the service status of important components of CNC equipment, and y _f is the observed value of the service status of the whole machine; using the above sample data, the support vector machine classification model is obtained through training, in which the kernel function adopts the radial basis function: K(x _f , x)=exp(-‖xx _f ‖ ² /σ ² ), x represents the service state sequence of CNC equipment components to be diagnosed, σ is the standard deviation of the sample data of the service state of the components, and its initial range is σ∈(2 ^-13 , 2 ³⁰ ); Step (5.2): Identify the service status of the complete machine of the CNC equipment, substitute the service status of all important components of the CNC equipment into the above-mentioned trained support vector machine model, and obtain the service status of the complete machine of the CNC equipment; Step (5.3): Hidden-semi-Markov modeling and remaining service life prediction of the CNC equipment; According to the above support vector machine prediction model, 10 to 20 time points are selected in turn, and according to the service status of important components at this moment, the service status of the whole machine is predicted, and finally the service history state sequence of the whole machine is obtained; then step 4 The service state sequence of the important components in the machine is replaced by the service state sequence of the CNC equipment machine, and the hidden-semi-Markov model of the service life cycle of the machine is obtained by using the same Baum-Welch training algorithm.

Images