CN105782071A - Water isolation pulp pump fault diagnosis method based on probabilistic neural network - Google Patents

Abstract

The invention discloses a water isolation pulp pump fault diagnosis method based on a probabilistic neural network. The method comprises the following steps: firstly, fault categories of a water isolation pulp pump are divided into m classes; secondly, the Ni set of training sample data in the i class are collected; thirdly, the training sample data are subject to normalization processing, and an input vector X is obtained; fourthly, the X is connected to a mode layer, and output of the mode layer is obtained; fifthly, variables, sent by the mode layer, in the same class are subject to accumulation and summation through a summation layer, and the i class of probability density value is obtained; sixthly, a trained probabilistic neural network model is obtained; and seventhly, the currently-collected data are input to the probabilistic neural network model, and the current fault category of the water isolation pulp pump is obtained and output. According to the water isolation pulp pump fault diagnosis method, the structure is simple, the convergence rate is high, the training time is short, convergence to local is not easily achieved, stability is high, and the sample adding capacity is high.

Description

A kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network

Technical field

The invention belongs to the fault diagnosis field of water-isolation slurry pump, particularly to a kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network.

Background technology

At present, various mine resources are in remote mountain areas, and Ore is transported to smeltery needs long-distance transportation.Ores lifting is the technique that one in ore dressing process is important, is generally adopted water-isolation slurry pump and carries.

The conveying principle of water-isolation slurry pump is, solid matter (Ore) is mutually mixed into ore pulp with liquid (clear water), adopts clarified water pump 11 to provide power, carries in slurry transportation pipeline.This method has the advantage on cost relative to railway transportation and highway transportation, has the feature of energy-conserving and environment-protective simultaneously.

The structure of water-isolation slurry pump is as shown in Figure 1.Water-isolation slurry pump includes equipped with the clear water reserviors 12 of clear water, clarified water pump 11, water intaking valve 10, back-water valve (BWV) 9, sealing chamber the 5, first stop valve the 3, second stop valve 4 and slurry transportation pipeline 2；Wherein sealing chamber 5 is internal is provided with ball float 6 moving up and down, and sealing chamber 5 is separated into superposed first cavity 7 and is positioned at the second cavity 8 of bottom by described ball float 6, and the first cavity 7 and the second cavity 8 are not connected；The outlet of clear water reserviors 12 passes sequentially through clarified water pump 11 and water intaking valve 10 and communicates with the first cavity 7, and the entrance of clear water reserviors 12 is communicated with the first cavity 7 by back-water valve (BWV) 9；The outlet of the first stop valve 3 and the entrance of the second stop valve 4 all communicate with the second cavity 8, and the entrance of the first stop valve 3 and the outlet of the second stop valve 4 all communicate with slurry transportation pipeline 2；One end of slurry transportation pipeline 2 communicates with the Slurry Bin 1 equipped with ore pulp, and the other end leads to the destination of ores lifting.

During conveying Ore, provide power by clarified water pump 11, make clear water and ore pulp isolation by sealing chamber 5.When sucking ore pulp, closing water intaking valve 10, open back-water valve (BWV) 9, ore pulp under gravity, is pressed into sealing chamber 5 from the first stop valve 3, and ball float 6 rises, and by back-water valve (BWV) 9, the clear water in first cavity 7 is pressed into clear water reserviors 12.When discharging ore pulp, closing back-water valve (BWV) 9, open water intaking valve 10, the clear water in clear water reserviors 12 is clamp-oned the first cavity 7 with certain pressure by clarified water pump 11, promote ball float 6 that by the second stop valve 4, ore pulp is pressed into slurry transportation pipeline 2, be then delivered to destination by slurry transportation pipeline 2.In whole work process, by the opening and closing of programme-control water intaking valve 10 and back-water valve (BWV) 9, it is achieved the process sucking ore pulp and discharge ore pulp of water-isolation slurry pump.

Water-isolation slurry pump is the visual plant in ore slurry pipeline course of conveying, can cause that when it breaks down ore pulp precipitates in ore pulp delivery duct, can cause that ore pulp delivery duct blocks time serious.Water-isolation slurry pump is remotely monitored, diagnoses its running status, be conducive to staff can make remedial measure early when breaking down, reduce the probability that this type of major accident of line clogging occurs.

Water-isolation slurry pump is a complicated nonlinear system, is difficult to set up accurate mathematical model, therefore invents a kind of accuracy method for diagnosing faults high, energy quick diagnosis water-isolation slurry pump significant.

The method for diagnosing faults of current water-isolated slurry pump is the diagnostic method based on BP neutral net, owing to BP neural network algorithm convergence rate is slow, it is necessary to the longer training time；Easily converge to local minimum point, cause failure to train；Learning and memory has unstability, and the dependency of sample is strong, if adding learning sample, the network trained is accomplished by the training that starts anew, and does not remember for former weights and threshold value.

Probabilistic neural network is the neutral net of a kind of pattern classification, is a kind of feedforward neural network.Compared with BP neutral net, probabilistic neural network simple in construction, fast convergence rate, the training time is short, it is not easy to converge to local, and stability is high, and sample supplemental capabilities is strong.The present invention, just from the advantage of probabilistic neural network, obtains a kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network.

It is illustrated in figure 2 the basic structure of probabilistic neural network.Probabilistic neural network is made up of input layer I, mode layer II, summation layer III and output layer IV.Input vector delivers to mode layer II after treatment at input layer I.The neutral net number of mode layer II is equal to the number of training sum of each classification, wherein g_ijJth for the i-th class of mode layer II exports, i=1, and 2 ..., m；J=1,2 ... N_i, i is fault category number, and m is fault category total number, N_iIt it is total number of the training sample of the i-th class.The neutral net number of summation layer III is equal to fault category number, and each class probability density of summation layer III self mode layer II in the future delivers to output layer IV after being separately summed and averaging.Output layer IV receives all kinds of probability density of summation layer III output and judges to obtain output state classification.

Summary of the invention

Owing to existing water-isolation slurry pump method for diagnosing faults is all based on BP neural network algorithm, and BP neural network algorithm convergence rate is slow, and the training time is long；Easily converge to local minimum point, cause failure to train；Learning and memory has unstability, and the dependency of sample is strong.It is an object of the invention to, for above-mentioned the deficiencies in the prior art, it is provided that a kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network.

For solving above-mentioned technical problem, the technical solution adopted in the present invention is:

A kind of water-isolation slurry pump method for diagnosing faults based on probabilistic neural network, comprises the following steps:

The fault category of water-isolation slurry pump is divided into m class, m >=2 by step (1), and a class of described m apoplexy due to endogenous wind represents normal condition, all the other class representing fault states；

Step (2) gathers the N of the i-th class_iGroup training sample data, wherein i is fault category, i=1,2 ..., m；

Step (3) is to described N_iGroup training sample data are normalized, and obtain input vector X；

X is connected to mode layer by step (4), obtains the output of mode layer

$g_{ij}\left(X\right)=\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp (-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2}),i=1,2,\mathrm{...},m,j=1,2,\mathrm{...}{N}_i$

Wherein g_ij(X) exporting for the jth of the i-th class of mode layer, p is the dimension of input vector, and σ is smoothing parameter, X_ijIt is the i-th class jth sample vector weights in a network；

The same class variable that mode layer is sent here by step (5) summation layer carries out cumulative and sues for peace, and the probability density value obtaining the i-th class is

$f_i\left(X\right)=\frac{1}{N_i}\underset{j=1}{\overset{N_i}{\Σ}}\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp \left(-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2}\right);$

Step (6) output layer receives the probability density value of summation layer, and by formula max (f_i(X) the sample fault category output theoretical value of water-isolation slurry pump) is obtained；If sample fault category output theoretical value is inconsistent with sample fault category output actual value, then change X_ijSize, until sample fault category output theoretical value and sample fault category output actual value is consistent；Storage X_ijValue, obtain the probabilistic neural network model trained；

Step (7) will currently gather data and input described probabilistic neural network model, obtain the current failure classification output of water-isolation slurry pump.

As a kind of optimal way, the fault category in described step (1) includes normal condition, outer tube blocked state and water pump state of wear 3 class.

Probabilistic neural network is a kind of neutral net for pattern classification, model fault can classified it is only applicable to during for fault diagnosis field, the present invention is through substantial amounts of experiment, the fault category of water-isolation slurry pump is divided into normal condition, outer tube blocked state and water pump state of wear 3 class, achieves good fault diagnosis effect.

As a kind of optimal way, the normalization processing method in described step (3) is Z-score standardized method.

The pressure of clarified water pump current of electric, the plasma discharge amount of slurry transportation pipeline, slurry transportation pipeline is included as a kind of optimal way, described training sample data and the current data that gather.

As a kind of optimal way, σ=0.1.

Compared with prior art, present configuration is simple, fast convergence rate, and the training time is short, it is not easy to converge to local, and stability is high, and sample supplemental capabilities is strong.

Accompanying drawing explanation

Fig. 1 is the structural representation of water-isolation slurry pump.

Fig. 2 is the basic structure of probabilistic neural network.

Fig. 3 is the diagnostic result of training data.

Fig. 4 is the diagnostic result of checking data.

Wherein, 1 is Slurry Bin, and 2 is slurry transportation pipeline, 3 is the first stop valve, and 4 is the second stop valve, and 5 is sealing chamber, 6 is ball float, and 7 is the first cavity, and 8 is the second cavity, 9 is back-water valve (BWV), and 10 is water intaking valve, 11 clarified water pumps, 12 is clear water reserviors, and I is input layer, and II is mode layer, III is summation layer, and IV is output layer.

Detailed description of the invention

One embodiment of the present invention comprises the following steps:

The fault category of water-isolation slurry pump is divided into m class, m >=2 by step (1), and a class of described m apoplexy due to endogenous wind represents normal condition, all the other class representing fault states；

Step (2) gathers the N of the i-th class_iGroup training sample data, wherein i is fault category, i=1,2 ..., m；

Step (3) is to described N_iGroup training sample data carry out Z-score standardization normalized, obtain input vector X；

X is connected to mode layer by step (4), obtains the output of mode layer

$g_{ij}\left(X\right)=\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp (-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2}),i=1,2,\mathrm{...},m,j=1,2,\mathrm{...}{N}_i$

Wherein g_ij(X) exporting for the jth of the i-th class of mode layer, p is the dimension of input vector, and σ is smoothing parameter, σ=0.1, X_ijIt is the i-th class jth sample vector weights in a network；

The same class variable that mode layer is sent here by step (5) summation layer carries out cumulative and sues for peace, and the probability density value obtaining the i-th class is

$f_i\left(X\right)=\frac{1}{N_i}\underset{j=1}{\overset{N_i}{\Σ}}\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp \left(-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2}\right);$

Step (6) output layer receives the probability density value of summation layer, and by formula max (f_i(X) the sample fault category output theoretical value of water-isolation slurry pump) is obtained；If sample fault category output theoretical value is inconsistent with sample fault category output actual value, then change X_ijSize, until sample fault category output theoretical value and sample fault category output actual value is consistent；Storage X_ijValue, obtain the probabilistic neural network model trained；

Step (7) will currently gather data and input described probabilistic neural network model, obtain the current failure classification output of water-isolation slurry pump.

Described training sample data and the current data that gather include the pressure of clarified water pump current of electric, the plasma discharge amount of slurry transportation pipeline, slurry transportation pipeline.

In order to verify the feasibility of the present invention, utilize in test the present invention that the running status of water-isolation slurry pump is diagnosed.In test, the fault category of water-isolation slurry pump is divided into normal condition, outer tube blocked state and water pump state of wear 3 class, wherein normal condition is numbered state 1, outer tube blocked state is numbered state 2, water pump state of wear is numbered state 3, if increasing the status number of correspondence when having other state successively.

Gathering the data of 5 groups of normal operating conditions, the data of 5 groups of outer tube blocked states, the data of 5 groups of water pump state of wear, data value is as shown in table 1.

The training of table 1 water-isolated slurry pump and checking data

Often group state takes 4 groups of training carrying out the probabilistic neural network diagnostic cast of water-isolation slurry pump, carries out modelling verification by the remaining one group of data of each running status.The diagnostic result of training data such as Fig. 3, diagnostic result such as Fig. 4 of checking data, as can be seen from Figure, utilize the present invention, and the accuracy of water-isolation slurry pump fault diagnosis reaches 100%.

Claims (5)

1. the water-isolation slurry pump method for diagnosing faults based on probabilistic neural network, it is characterised in that comprise the following steps:

The fault category of water-isolation slurry pump is divided into m class, m >=2 by step (1), and a class of described m apoplexy due to endogenous wind represents normal condition, all the other class representing fault states；

Step (2) gathers the N of the i-th class_iGroup training sample data, wherein i is fault category, i=1,2 ..., m；

Step (3) is to described N_iGroup training sample data are normalized, and obtain input vector X；

X is connected to mode layer by step (4), obtains the output of mode layer

$g_{ij}\left(X\right)=\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp (-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2}),i=1,2,...,m,j=1,2,...N_i$

Wherein g_ij(X) exporting for the jth of the i-th class of mode layer, p is the dimension of input vector, and σ is smoothing parameter, X_ijIt is the i-th class jth sample vector weights in a network；

The same class variable that mode layer is sent here by step (5) summation layer carries out cumulative and sues for peace, and the probability density value obtaining the i-th class is

$f_i\left(X\right)=\frac{1}{N_i}\underset{j=1}{\overset{N_i}{\Σ}}\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{p}{2}}{\mathrm{\σ}}^p}\exp (-\frac{\left|\right|X-X_{ij}|{|}^2}{2{\mathrm{\σ}}^2});$

Step (6) output layer receives the probability density value of summation layer, and by formula max (f_i(X) the sample fault category output theoretical value of water-isolation slurry pump) is obtained；If sample fault category output theoretical value is inconsistent with sample fault category output actual value, then change X_ijSize, until sample fault category output theoretical value and sample fault category output actual value is consistent；Storage X_ijValue, obtain the probabilistic neural network model trained；

Step (7) will currently gather data and input described probabilistic neural network model, obtain the current failure classification output of water-isolation slurry pump.

2. the water-isolation slurry pump method for diagnosing faults based on probabilistic neural network as claimed in claim 1, it is characterised in that the fault category in described step (1) includes normal condition, outer tube blocked state and water pump state of wear 3 class.

3. the water-isolation slurry pump method for diagnosing faults based on probabilistic neural network as claimed in claim 1, it is characterised in that the normalization processing method in described step (3) is Z-score standardized method.

4. the water-isolation slurry pump method for diagnosing faults based on probabilistic neural network as claimed in claim 1, it is characterized in that, described training sample data and the current data that gather include the pressure of clarified water pump current of electric, the plasma discharge amount of slurry transportation pipeline, slurry transportation pipeline.

5. the water-isolation slurry pump method for diagnosing faults based on probabilistic neural network as claimed in claim 1, it is characterised in that σ=0.1.