CN101540009B - Method for predicting facility and equipment failure - Google Patents

Abstract

The invention provides a method for predicting facility and equipment failure by using a fuzzy neural network. The method predicts the service life of facility and equipment parts by the importance that historical data automatically adjusts factors influencing the service life through a learning mode of the fuzzy neural network.

Description

The method of prediction facility and equipment failure

Technical field

The present invention is about a kind of method of predicting facility and equipment failure, and particularly uses the method for fuzzy nerve network prediction facility and equipment part fault about a kind of what computer system.

Background technology

If diminish the consumption part, just must repair or replace at set intervals in facility and the equipment, yet the serviceable life of part is different because of environment and properties of materials.In general, be subjected to the influence of some service factors the serviceable life of part, and under different situations, the importance of each influence factor can change to some extent, for the part of avoiding facility and equipment damages before maintaining or exhausts, therefore influence its due function, if can further initiatively predict the time that the part of facilities and equipment may damage or exhaust, can take precautions against in possible trouble.

Summary of the invention

The purpose of this invention is to provide a kind of method of predicting facility and equipment failure, to estimate the serviceable life of facilities and equipment part exactly.

Comprising according to the method for prediction facility of the present invention and equipment failure and to select one to specify part, is to select a part in the facilities and equipment as specifying part.After having selected the appointment part in the facilities and equipment, then select to influence the factor in the appointment part serviceable life in the facilities and equipment, and with each influence factor as input variable.

Then according to the influence factor that is adopted, the ambiguity in definition rule, in order to the serviceable life of the appointment part of prediction facilities and equipment, fuzzy rule is defined as:

IFx ₁ISμ _1j(x ₁)ANDx ₂ISμ ₂(x ₂)AND......ANDx _nISμ _n(x _n)THENyISc _j

X wherein ₁... .x _nBe the parameter of setting in the fuzzy nerve network, i.e. the part factor in serviceable life is specified in influence, is x with parameter setting ₁, it is appointment part service time, x ₂For specifying the frequency of utilization of part, x ₃Be the interval time of specifying part to use, x ₄Be temperature, x ₅Be humidity.μ _1j(x ₁) be subordinate function, y is c for according to the result that every influence factor produced _jThe serviceable life of the part of predicting.

According to fuzzy rule, set up a fuzzy nerve network and revise this fuzzy rule again.Wherein, the method for building up of fuzzy nerve network comprises and determines an input layer, determines an obscuring layer, uses a contact bed and use an output layer.Input layer comprises a plurality of neurons, the number of the corresponding input variable of neuronic number.Obscuring layer comprises a plurality of groups, and each group comprises a plurality of neurons, and each neuron in each group is represented a fuzzy membership functions.Contact bed comprises a plurality of neurons, and it is the back output of multiplying each other of output valve with obscuring layer, the output valve of output layer reception interface layer and with its back output of multiplying each other.

Then,, fuzzy nerve network is trained, to set up the life forecast pattern of specifying part according to the historical service data in the database.Then, the fuzzy nerve network that training is finished can be in order to predict the possible serviceable life of appointment part in the facilities and equipment.

By the scheme of the invention described above as can be known, before maintaining, promptly damage or exhaust but use the present invention's containment facility equipment, and cause facilities and equipment to operate, cause regular works to stop the loss that is produced.

Description of drawings

For above and other objects of the present invention, feature, advantage can be become apparent, below conjunction with figs. is elaborated to preferred embodiment of the present invention, wherein:

Fig. 1 is a kind of process flow diagram of predicting the method for facility and equipment failure according to a preferred embodiment of the present invention.

Fig. 2 is the arrangement plan of the fuzzy nerve network among Fig. 1.

Fig. 3 is the synoptic diagram of the employed triangle subordinate function of the obscuring layer of the fuzzy nerve network among Fig. 2.

Embodiment

In the present embodiment, so-called facility and equipment comprise facility, equipment and furniture three parts, and facility can refer to a building or local so that specific service or specific industrial use to be provided; Equipment can be a job or serves needed instrument; Furniture can be the used utensil of family, as tables and chairs, cabinet etc.

Please refer to Fig. 1, it illustrates a kind of process flow diagram of predicting the method for facility and equipment failure according to a preferred embodiment of the present invention.The method 200 of prediction facility and equipment failure comprises selects one to specify part, as step 210.The part factor in serviceable life is specified in selected influence, as step 220.The part fuzzy rule in serviceable life is specified in definition one prediction, as step 230.Set up a fuzzy nerve network, as step 240.Train this fuzzy nerve network, as step 250.Estimate the possible serviceable life of appointment part in the facilities and equipment, as step 260.

Selecting one to specify part, as step 210, is to select part in the facilities and equipment as specifying part.Because facilities and equipment is made of many parts, therefore, need select the prediction of which part being carried out serviceable life at the beginning.To have used a nitrogen cylinder to be example in the facilities and equipment, under general normal use,, can determine when nitrogen can exhaust along with the time length of using, the factors such as number of times of use as part.In order to predict when nitrogen can exhaust in this nitrogen cylinder, therefore select this nitrogen cylinder as specifying part, come it is carried out serviceable life the prediction of (promptly when need change).

The part factor in serviceable life is specified in selected influence, as step 220.After having selected the appointment part in the facilities and equipment, must select can influence the factor in appointment part serviceable life in the facilities and equipment, and with each influence factor as an input variable.In the present embodiment, selected and specified part service time, specified the frequency of utilization of part, interval time, temperature and the humidity etc. of specifying part to use, as the influence factor in the appointment part serviceable life in the facilities and equipment.

According to the influence factor that is adopted, the part fuzzy rule in serviceable life is specified in definition one prediction, as step 230.Present embodiment uses fuzzy nerve network as Forecasting Methodology, by fuzzy neuroid mode of learning, automatically adjust the importance of each influence factor by historical data, and then the lifting prediction accuracy, the fuzzy rule definition that part (being the facilities and equipment part) serviceable life is specified in prediction is suc as formula (1):

IFx ₁ISμ _1j(x ₁)ANDx ₂ISμ ₂(x ₂)AND......ANDx _nISμ _n(x _n)THENyISc _j (1)

X wherein ₁... .x _nBe the parameter of setting in the fuzzy nerve network, i.e. the part factor in serviceable life is specified in influence, is x with parameter setting ₁For specifying part service time, x ₂For specifying the frequency of utilization of part, x ₃Be the interval time of specifying part to use, x ₄Be temperature, x ₅Be humidity.μ _1j(x ₁) be subordinate function, y is c for according to the result that every influence factor produced _jThe serviceable life of the part of predicting.Subordinate function adopts the triangle subordinate function in the present embodiment.

With reference to Fig. 2, it is the arrangement plan of the fuzzy nerve network among Fig. 1.Fuzzy nerve network is the artificial network in conjunction with simulating human thinking pattern and human nerve's operation principles, and this network has human thinking and the characteristic of learning simultaneously.Set up a fuzzy nerve network according to formula (1) and revise these fuzzy rules.In the present embodiment, fuzzy nerve network comprises four stratum, these four stratum are respectively: input layer (Input Layer) 310, obscuring layer (Fuzzify Layer) 320, contact bed (Intermediate Layer) 330 and output layer (OutputLayer) 340, and it is configured to a fuzzy nerve network that contains J bar fuzzy rule, it is further specified as follows:

Input layer 310 comprises n neuron 311, and each neuron 311 cooperates an input variable.Neuron 311 after receiving the signal of input variable, the neuron of other that is directly passed to down in one deck (being obscuring layer 320) that neuron 311 therewith is connected.Therefore, i neuronic output valve is in the input layer 310 It can be represented suc as formula (2):

$o_i^{\left(1\right)}=p_i,\mathrm{for}1\≤i\≤n---\left(2\right)$

The neuron number of input layer 310 is the input variable number, that is in input layer 310 directly with input value p _iBe sent in the obscuring layer 320, do not make any change.

Total J group 321 in the obscuring layer 320, each group 321 comprises n neuron 322, and each neuron 322 is represented a fuzzy membership functions, can be represented with formula (3), formula (4), formula (5) respectively.Neuron 322 in the obscuring layer 320 is being played the part of the role who the numerical value of input is converted to degree of membership.I in the group 321 of j neuronic output valve is

It is

Pairing subordinate function.The obscuring layer neuron number is the long-pending of the input variable and the fuzzy meaning of one's words.

Please refer to Fig. 3, it is the synoptic diagram of the triangle subordinate function that obscuring layer made of the fuzzy nerve network among Fig. 2.Employed in the present embodiment subordinate function defines as shown in Figure 3, and it is made up of three fuzzy membership functions respectively, is expressed as follows with formula (3), formula (4), formula (5) respectively.

$o_{1\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="S2008100868663E00021.png,S2008100868663E00031.png"\; state="\{\{state\}\}">j</figure-callout>}}^2=\left\{\begin{array}{cc}1& \mathrm{when\; x}\&le;2\\ \frac{(4-x)}{2}& \mathrm{when}2<x\&le;4\\ 0& \mathrm{when\; x}>4\end{array}\right.---\left(3\right)$

$o_{2\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="S2008100868663E00021.png,S2008100868663E00031.png"\; state="\{\{state\}\}">j</figure-callout>}}^2=\left\{\begin{array}{cc}1& \mathrm{when\; x}\&le;3\\ \frac{(x-3)}{2}& \mathrm{when}3<x\&le;5\\ \frac{(7-x)}{2}& \mathrm{when}5<x\&le;7\\ 0& \mathrm{when\; x}>7\end{array}\right.---\left(4\right)$

$o_{3\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="S2008100868663E00021.png,S2008100868663E00031.png"\; state="\{\{state\}\}">j</figure-callout>}}^2=\left\{\begin{array}{cc}0& \mathrm{when\; x}\&le;6\\ \frac{(x-6)}{2}& \mathrm{when}6<x\&le;8\\ 1& \mathrm{when\; x}>8\end{array}\right.---\left(5\right)$

Please refer to Fig. 2, contact bed 330 comprises J neuron 331, and each neuron 331 is all represented each rule, and neuron 331 numbers are identical with regular number, and links with corresponding neuron in the obscuring layer 320.This one deck is equivalent to the prerequisite (if in the rule ...) partly, neuron 331 can calculate input variable and regular matching degree (Matching degree).The output valve of each neuron 331

For all input values are multiplied each other, it can represent an accepted way of doing sth (6):

$o_j^{\left(3\right)}={\mathrm{\&Pi;}}_{i=1}^n{o}_{\mathrm{ij}}^{\left(2\right)}---\left(6\right)$

Contact bed 330 neuron numbers adopt utilization rule decision the most widely, calculate as the formula (7).

$N_i=\sqrt{(N_f+N_o)}---\left(7\right)$

Wherein, N _iBe contact bed 310 neuron numbers, N _fBe obscuring layer 320 neuron numbers, N _oBe output layer 340 neuron numbers.

Output layer 340 only comprises a neuron 341, its corresponding output variable, the i.e. time of possibility fault.Input value multiplied each other is the output layer result:

$o_j^{\left(4\right)}={\mathrm{\&Pi;}}_{i=1}^n{o}_{\mathrm{ij}}^{\left(3\right)}---\left(8\right)$

The neuron of this layer is responsible for output valve is carried out ambiguity solution, to obtain clear and definite (Crisp) output valve.

Then, fuzzy nerve network being trained, is the historical data of safeguarding according to the past in the database, and pass-algorithm (Error back propagation) training fuzzy nerve network is fallen in execution error, and sets up part and estimate pattern serviceable life.It is to adopt steepest gradient method to train that mistake is fallen pass-algorithm, so that energy function (error function) can arrive minimizes.

The fuzzy nerve network that training is finished can be in order to estimate the possible serviceable life of appointment part in the facilities and equipment.In the present embodiment, the method 200 of prediction facility and equipment failure can be created as a prediction malfunctioning module, and be disposed in the information computer system.The prediction malfunctioning module comprises four sub-function choosing-items, is respectively management special project function choosing-item, provides the newly-increased or editor's special project of user; Set the subordinate function function choosing-item, provide the user and set Membership Function Distribution; Training special project function choosing-item provides the user and specifies special project to carry out the fuzzy nerve network training; And prediction fault function choosing-item, after providing the user and selecting special project, estimate the life-span of facilities and equipment part.

Wherein, management special project function choosing-item comprises two subfunctions, is respectively new project function choosing-item and editor's special project function choosing-item.The new project function choosing-item provides the user to import new project filename and parameters, as: prediction part name, predictor and maximin scope, contact bed neuronal quantity etc.Editor's special project function choosing-item provides the user already present special project is edited or revised, and revises existing special project content.

When operation prediction malfunctioning module, by using the new project function choosing-item, provide the user to select specific component and save as create name, and can select to influence the part life factor and set the contact bed neuronal quantity.When new project, computer system can produce Membership Function Distribution automatically, and in addition, the user also can use and set subordinate function function modification Membership Function Distribution.

Training special project function choosing-item is according to the past maintain historical data, provides the user to use fuzzy nerve network to set up part life and estimates pattern.Prediction fault function choosing-item provides the special project that the user uses training to finish, and estimates and specifies the possible life-span of part, and the reference as maintenance management is provided.

The mode of operation of prediction malfunctioning module can be set up new special project at the part of specific facilities equipment for the user, or according to demand modification past special project, special project newly-built or revise after can train fuzzy nerve network according to historical data, training result can be used for estimating the possible life-span of facilities and equipment part.

Though the present invention discloses as above with a preferred embodiment; yet it is not in order to limit the present invention; any person skilled in the art; without departing from the spirit and scope of the present invention; when can making various changes that are equal to or replacement, so protection scope of the present invention is when looking accompanying being as the criterion that the application's claim scope defined.

Claims (7)

1. method of predicting facility and equipment failure is applied to comprise in the computer system:

One user selects one to specify part by this computer system, is to select the part in the facilities and equipment to be this appointment part;

This user specifies the influence factor in the serviceable life of part by selected this of this computer system, and wherein each influence factor is an input variable;

According to these influence factors, define a fuzzy rule, in order to predict this appointment part serviceable life;

According to this fuzzy rule, in this computer system, set up a fuzzy nerve network, this this fuzzy rule of fuzzy nerve network correction wherein, the method for building up of this fuzzy nerve network comprises:

Determine an input layer, wherein this input layer comprises a plurality of neurons, the number of corresponding these input variables of these neuronic numbers;

Determine an obscuring layer, wherein this obscuring layer comprises a plurality of groups, and each group comprises a plurality of neurons, and each neuron in each group is represented a fuzzy membership functions, and this obscuring layer also produces an output valve;

Use a contact bed, with the output valve of this obscuring layer back output one output valve that multiplies each other, wherein this contact bed comprises a plurality of neurons; And

Use an output layer, receive the output valve of this contact bed and its back output of multiplying each other;

Training this fuzzy nerve network, is in the database according to this computer system, in the historical service data of this computer system that the past is imported, trains this fuzzy nerve network about this appointment part; And

The fuzzy nerve network that uses this training to finish is predicted serviceable life of this appointment part by this computer system.

2. the method for prediction facility according to claim 1 and equipment failure is characterized in that this fuzzy rule is defined as:

IFx ₁ISμ _1j(x ₁)ANDx ₂ISμ ₂(x ₂)AND......ANDx _nISμ _n(x _n)THENyISc _j

It is characterized in that x ₁... .x _nBe these input variables, x ₁For this specifies the service time of part, x ₂For this specifies the frequency of utilization of part, x ₃Be this interval time of specifying part to use, x ₄Be a temperature, x ₅Be a humidity, μ _1j(x ₁) be a subordinate function, y is c for according to the result that these input variables produced _jThe serviceable life of this appointment part of predicting.

3. the method for prediction facility according to claim 2 and equipment failure is characterized in that this subordinate function is a triangle subordinate function.

4. the method for prediction facility according to claim 3 and equipment failure is characterized in that this triangle subordinate function comprises three fuzzy membership functions and is respectively:

$o_{1j}^2=\left\{\begin{array}{cc}1& \mathrm{when\; x}\&le;2\\ \frac{(4-x)}{2}& \mathrm{when}2<x\&le;4\\ 0& \mathrm{when\; x}>4\end{array}\right.$

$o_{2j}^2=\left\{\begin{array}{cc}1& \mathrm{when\; x}\&le;3\\ \frac{(x-3)}{2}& \mathrm{when}3<x\&le;5\\ \frac{(7-x)}{2}& \mathrm{when}5<x\&le;7\\ 0& \mathrm{when\; x}>7\end{array}\right.$

$o_{3j}^2=\left\{\begin{array}{cc}0& \mathrm{when\; x}\&le;6\\ \frac{(x-6)}{2}& \mathrm{when}6<x\&le;8\\ 1& \mathrm{when\; x}>8\end{array}\right..$

5. the method for prediction facility according to claim 4 and equipment failure, it is characterized in that this step of training this fuzzy nerve network is to adopt wrong pass-algorithm to train this fuzzy nerve network, to set up the life forecast pattern of this appointment part.

6. the method for prediction facility according to claim 5 and equipment failure is characterized in that the fuzzy nerve network that this uses this training to finish, and the step of predicting the serviceable life of this appointment part comprises:

New project provides a user and selects a specific component as this appointment part and save as a create name, and this user also selects influence the neuronal quantity of this appointments part life factor and this contact bed of setting;

The training special project is the historical service data according to this appointment part, provides this user to use this fuzzy nerve network to set up the life prediction pattern of this appointment part; And

The prediction fault provides the special project that the user uses training to finish, and predicts the life-span of this appointment part.

7. the method for prediction facility according to claim 5 and equipment failure is characterized in that the fuzzy nerve network that this uses this training to finish, and the step of predicting the serviceable life of this appointment part comprises:

Editor's special project provides and revises an already present special project content, revises the back special project to produce one, and wherein the special project content comprises influences this appointment part life factor and the neuronal quantity of setting this contact bed;

The training special project is according to the historical service data of this appointment part in this modification back special project content, uses this fuzzy nerve network to set up the life-span preset mode of this appointment part; And

The prediction fault is to use special project after the modification that utilization training finishes, and predicts the life-span of this appointment part.

Images