CN105547717A - Diesel engine lubricating system fault diagnosis method based on Bayes network - Google Patents

Abstract

The invention relates to a diesel engine lubricating system fault diagnosis method based on the Bayes network. According to the method, fault types and external symptoms of a lubricating system are abstracted into fault layer nodes and symptom layer nodes, and a diesel engine lubricating system Bayes network model is established; performance parameters of the diesel engine lubricating system are detected by utilizing a data acquisition system, a linear proportion transformation method is employed to carry out classification processing on the performance parameters, and the actual work state information of the lubricating system is acquired; a Hugin combined tree algorithm is employed to convert a corrected lubricating system Bayes network model into a combined tree. Before inference diagnosis, on the basis of the actual work state of the lubricating system, through resetting the prior probability of the fault layer nodes, adaptability correction on the Bayes network model is carried out, so the actual work state of the lubricating system can be accurately described through the model, nondeterminacy of model inference is reduced, and thereby fault diagnosis accuracy is improved.

Description

Based on the lubricating system of diesel oil engine method for diagnosing faults of Bayesian network

Technical field

What the present invention relates to is lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network.

Background technology

Diesel engine plays an important role in the every field of national economy.But engine block is complicated, many parts be in high temperature, high pressure, high load capacity mal-condition work, make system failure rate higher, maintenance expense is very large.Statistics shows, in every cost of use of diesel engine, the expenditure of maintenance aspect reaches 15%-30%.Separately there is statistics display, when carrying out equipment control maintenance, determining that the fault time used accounts for the 70%-90% of T.T..As can be seen here, inefficient Diagnosis Method of Diesel Fault wastes a large amount of human and material resources resources, brings great inconvenience to commercial production.

Diesel Fault Diagnosis is the effective means realizing fault early prediction and preventive maintenance, for the harm of reduction accident, guarantees that the safe operation of diesel engine has vital role.The key issue that first application of this technology will solve is exactly that mapping between fault signature and the source of trouble is non-linear.Bayesian network is a directed acyclic graph, node on behalf stochastic variable wherein, and internodal directed edge represents the incidence relation between stochastic variable, and characterizes the size of correlation degree between stochastic variable with the form of prior probability.Bayesian network is expressed in multistate logic and uncertain reasoning is had unique advantage.Bayesian network is utilized to need to be undertaken by relevant reasoning algorithm to the diagnosis of equipment failure.Hugin Junction tree is a kind of conventional Bayesian network Accurate Reasoning algorithm.First Bayesian network is converted into a primary structure by this algorithm---and combining tree, then by being defined in the message process on associating tree, probabilistic causal reasoning being carried out to object event.In recent years, there is scholar that Bayesian network is applied to lubricating system of diesel oil engine fault diagnosis field, achieve certain achievement.But, in existing research, the Bayesian network model form of lubricating system is fixed, can not according to the actual working state correction institute established model of system, make model to the dynamic change bad adaptability of lubricating system, the problems such as it is lower that diagnostic result exists accuracy rate, and referential is not strong, seriously constrain the further application of this technology.Invention one can according to device physical status, accommodation model structure, the security that the lubricating system of diesel oil engine method for diagnosing faults identifying fault type quickly and accurately runs for raising equipment, realizes having great importance to the condition maintenarnce of diesel engine.

Through finding the literature search of prior art, open file " naval vessel diesel main engine oil system Bayesian Network Inference method for diagnosing faults " (Sichuan war industry's journal, 2015) a kind of lubricating system of diesel oil engine method for diagnosing faults proposed, the readme of the disclosure file is: " with the diesel main engine lube pipe system in ship's powerplant for research object, for oil system troubleshooting issue, analyze most common failure mechanism, establish the fault tree synthesis of oil system most common failure, construct the Bayesian network model for malfunction reasoning on this basis, analyze the Bayes's State reasoning process under oil system typical fault state, rapid failure diagnosis for oil system provides a kind of new method ".Its weak point is: the method is built Bayesian network model form and fixed, accommodation can not be carried out according to the dynamic change of lubricating system, cannot the virtual condition of accurate description lubricating system, therefore cause model reasoning uncertain large, diagnostic accuracy is lower; And the fault diagnosis of the method to lubricating system is a kind of static reasoning, its process according to lubricating system of diesel oil engine actual motion information, really cannot not realize the diagnosis to lubricating system of diesel oil engine fault, be difficult to guiding work personnel and carry out specific aim maintenance to equipment.

Summary of the invention

The object of the invention is to for the deficiencies in the prior art, a kind of lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network is provided.

The object of the present invention is achieved like this:

(1) fault type of lubricating system and outward sign are distinguished abstract to be fault node layer and sign node layer, to set up lubricating system of diesel oil engine Bayesian network model;

(2) utilize data acquisition system (DAS) to detect the performance parameter of lubricating system of diesel oil engine, adopt linear scale transform's method to carry out classification process to performance parameter, ω represents the actual performance parameter of lubricating system, ω ^*represent the standard value of performance parameter, represent the performance parameter after conversion, and then extensive lubricating system performance parameter, obtain the actual working state information e of lubricating system;

$\widetilde{\mathrm{\ω}}=\frac{\mathrm{\ω}}{{\mathrm{\ω}}^{*}}$

(3) according to the lubricating system actual working state information e obtained, to Bayesian network model sign node layer v _jstate π (v _j) carry out binary value, (ω _j) represent performance parameter ω _jextensive, represent and meet performance parameter ω _jextensive corresponding outward sign describes; The lubricating system Bayesian network model that adaptability correction is set up;

$\mathrm{\π}\left(v_j\right)=\left\{\begin{array}{cc}1,& \begin{array}{cc}if& v_j~{\tilde{v}}_j=\arg Q\left({\mathrm{\ω}}_j\right)\end{array}\\ 0,& else\end{array}\right.$

(4) adopt Hugin Junction tree that revised lubricating system Bayesian network model is converted into associating tree, and then using lubricating system actual working state information e as reasoning evidence, by calculating fault node layer s _imarginalisation conditional probability p (s _i| e), the current failure type of lubricating system is diagnosed.

Described fault type specifically comprises: piston ring packing inefficacy S ₁, add low on fuel S ₂, Cooler Fault S ₃, exceed S in serviceable life ₄, the improper S of lubricating oil oil product ₅, in lubricating oil containing bubble S ₆, pipeline leakage of oil S ₇, line clogging S ₈.

Described outward sign specifically comprises: the too low V of lubricating oil liquid level ₁, the too high V of oil temperature ₂, enter the too low V of machine lubricating oil pressure ₃, go out the too low V of machine lubricating oil pressure ₄, go out the too low V of machine oil flow ₅.

The concrete grammar of the lubricating system Bayesian network model that described adaptability correction is set up is: according to sign node layer v _jstate π (v _j), reset fault node layer pa (v in lubricating system Bayesian network model _j) prior probability P (pa (v _j)); Pa (v _j) be sign node layer v _jfather node;

$P\left(pa\right(v_j\left)\right)=\left\{\begin{array}{ccc}1,& if& \mathrm{\π}\left(v_j\right)=1\\ 0,& if& \mathrm{\π}\left(v_j\right)=0\end{array}\right..$

Compared with prior art, beneficial effect of the present invention is: the present invention is before enforcement reasoning diagnosis, according to the actual working state of lubricating system, by resetting the prior probability of fault node layer, adaptability correction has been carried out to Bayesian network model, make model can the actual working state of accurate description lubricating system, thus reduce the uncertainty of model reasoning, improve the accuracy rate of fault diagnosis; In addition, the present invention implements the actual working state information of the diagnostic reasoning of lubricating system fault type according to lubricating system, and therefore diagnostic result can reflect lubricating system actual performance truly, has stronger practical guided significance.

Accompanying drawing explanation

Fig. 1 is the lubricating system of diesel oil engine method for diagnosing faults process flow diagram that the present invention is based on Bayesian network.

Fig. 2 is certain type four-cylinder diesel engine lubricating system Bayesian Network Topology Structures figure.

Embodiment

Below in conjunction with accompanying drawing, embodiments of the invention are elaborated: the present embodiment is implemented under premised on technical solution of the present invention, give detailed embodiment, but protection scope of the present invention is not limited to following embodiment.

The present invention relates to a kind of lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network, belong to Diesel Fault Diagnosis field.First by lubricating system fault type and outward sign abstract be network node, set up lubricating system Bayesian network model; Secondly, detect lubricating system performance parameter, obtain lubricating system actual working state information; Again, according to the lubricating system actual working state information obtained, by resetting the prior probability of fault node layer, adaptability correction is carried out to lubricating system Bayesian network model; Finally, using the actual working state information of lubricating system as reasoning evidence, Hugin Junction tree is utilized to carry out probabilistic diagnosis to lubricating system fault.The Bayesian network model that the present invention can be set up by the correction of knot modification probabilistic information, make model reflect lubricating system current operating state more accurately, improve the accuracy of fault diagnosis, diagnostic result has higher practical guided significance.

First the Bayesian network model of lubricating system of diesel oil engine is set up, then, lubricating system actual working state information according to obtaining carries out adaptability correction to Bayesian network model, to enable the actual working state of model accurate description lubricating system, improve the accuracy rate of fault diagnosis, finally, using lubricating system actual working state information as reasoning evidence, Hugin Junction tree is utilized to carry out diagnostic reasoning to fault type, maintainer is instructed to implement specific aim maintenance to lubricating system of diesel oil engine accordingly, support equipment safety, reduces maintenance management cost.

Method of the present invention specifically comprises the following steps:

1, by the fault type S of lubricating system _iwith outward sign V _jabstract is respectively fault node layer s _iwith sign node layer v _j, set up lubricating system of diesel oil engine Bayesian network model;

2, data acquisition system (DAS) is utilized to detect the performance parameter ω of lubricating system of diesel oil engine _i, adopt linear scale transform's method to performance parameter ω _icarry out classification process, and then the performance parameter of extensive lubricating system, obtain the actual working state information e of lubricating system;

3, according to the lubricating system actual working state information e obtained in step 2, to sign node layer v in Bayesian network model _jstate π (v _j) carry out binary value, on this basis, reset the prior probability P (pa (v of fault node layer _j)), to revise the lubricating system Bayesian network model of foundation;

4, adopt Hugin Junction tree that revised lubricating system Bayesian network model is converted into associating tree, and then using lubricating system actual working state information e as reasoning evidence, by calculating fault node layer s _imarginalisation conditional probability p (s _i| e), the current failure type of lubricating system is diagnosed.

As shown in Figure 1, the present invention includes following steps: the adaptability correction of the foundation of lubricating system Bayesian network model, lubricating system actual working state acquisition of information, Bayesian network model and lubricating system fault combine tree diagnosis.Specific as follows:

1, the foundation of described lubricating system Bayesian network model is by the fault type S of lubricating system _iwith outward sign V _jrespectively as fault node layer s _iwith sign node layer v _j, set up the Bayesian network model of lubricating system of diesel oil engine.Bayesian network model can utilize two tuple B<G, and P> represents, wherein G is the topological structure of Bayesian network, and P is the probabilistic information of node.Further, the probabilistic information P of node specifically comprises: prior probability p (s _i) and conditional probability p (v _j| pa (v _j)), wherein pa (v _j) represent and sign node layer v _jthere is the fault node layer (father node) because of, really relation.At Bayesian network model B<G, in P>, the joint probability distribution between all outward signs represents by formula (1).

$p(v_1,\mathrm{...},v_n)=\underset{j=1}{\overset{n}{\&Pi;}}p\left(v_j\right|pa\left(v_j\right))---\left(1\right)$

Described fault type specifically comprises: piston ring packing inefficacy S ₁, add low on fuel S ₂, Cooler Fault S ₃, exceed S in serviceable life ₄, the improper S of lubricating oil oil product ₅, in lubricating oil containing bubble S ₆, pipeline leakage of oil S ₇, line clogging S ₈.

Described outward sign refers to the operation situation of the performance parameter of lubricating system of diesel oil engine, specifically comprises: the too low V of lubricating oil liquid level ₁, the too high V of oil temperature ₂, enter the too low V of machine lubricating oil pressure ₃, go out the too low V of machine lubricating oil pressure ₄, go out the too low V of machine oil flow ₅.

Further, described performance parameter ω _ispecifically comprise: lubricating oil liquid level ω ₁, oil temperature ω ₂, enter machine lubricating oil pressure ω ₃, go out machine lubricating oil pressure ω ₄, go out machine oil flow ω ₅.

2, described lubricating system actual working state acquisition of information is the performance parameter ω utilizing sensor and data collecting card to detect lubricating system of diesel oil engine current time _i, and by linear scale transform's method, as shown in formula (2), make performance parameter ω _ito map between given zone ([0,1), [1,1], (1 ,+∞)), to eliminate the unit restriction of performance parameter, be translated into nondimensional pure values.On this basis, according to the performance parameter of the extensive acquisition of rule of agreement in definition 1, and it can be used as the actual working state information e of lubricating system.

$\widetilde{\mathrm{\&omega;}}=\frac{\mathrm{\&omega;}}{{\mathrm{\&omega;}}^{*}}---\left(2\right)$

In above formula, ω represents the actual performance parameter of lubricating system; ω ^*represent the standard value of performance parameter; represent the performance parameter after conversion.

Provide key definition as follows:

Definition 1:(defines the abstraction rule of performance parameter) linear scale transform is carried out to the performance parameter collected after, high-level concept " too low ", " normally ", " too high " is utilized to substitute performance parameter respectively interval [0,1), [1,1], (1, + ∞) in value, to be described the operation situation of performance parameter by fuzzy division.Claim this fuzzy division method to be the extensive of performance parameter, be denoted as Q ().

3, the adaptability correction of described Bayesian network model refers to, according to lubricating system actual working state, utilizes the method shown in formula (3), to Bayesian network model sign node layer v _jstate π (v _j) carry out binary value.In formula (3), Q (ω _j) represent performance parameter ω _jextensive, represent and meet performance parameter ω _jextensive corresponding outward sign describes.On this basis, according to sign node layer state value, formula (4) is adopted to reset the dependent failure node layer pa (v of Bayesian network model _j) prior probability P (pa (v _j)), adaptability correction is carried out to built lubricating system Bayesian network model, with the actual performance feature enabling model reflect lubricating system exactly.

$\mathrm{\&pi;}\left(v_j\right)=\left\{\begin{array}{cc}1,& \begin{array}{cc}if& v_j~{\tilde{v}}_j=\arg Q\left({\mathrm{\&omega;}}_j\right)\end{array}\\ 0,& else\end{array}\right.---\left(3\right)$

$P\left(pa\right(v_j\left)\right)=\left\{\begin{array}{ccc}1,& if& \mathrm{\&pi;}\left(v_j\right)=1\\ 0,& if& \mathrm{\&pi;}\left(v_j\right)=0\end{array}\right.---\left(4\right)$

4, the tree diagnosis of combining of described lubricating system fault refers to, Hugin Junction tree is utilized to transform Bayesian network model B<G, P> for combining tree, and using lubricating system actual working state information e as evidence, reasoning and calculation fault node layer s _imarginalisation conditional probability p (s _i| e).P (s _i| e) be under the actual working state information e condition of lubricating system, fault node layer s _ithe probability that the fault type characterized occurs.Now, according to maximum likelihood principle, as formula (10), the fault node layer s that marginalisation conditional probability is maximum is chosen ^*, the current performance state of diagnosis lubricating system is s ^*the fault type S characterized ^*, accordingly plan, targetedly condition maintenarnce are implemented to diesel engine.

Utilize the detailed process of Hugin Junction tree edge calculation conditional probability as follows:

The first step: Bayesian network model B<G, P> are converted into associating tree according to set converting algorithm.

Second step: the potential function of uniting some X is combined in tree in initialization be 1, and upgrade potential function according to the conditional probability in Bayesian network model update method adopts formula (5), wherein, and v _j, pa (v _j) ∈ X (all meeting this condition in following steps).

3rd step: utilize the method shown in formula (6) using the actual working state information e of collection, extensive lubricating system as reasoning evidence, input associating tree.

4th step: by uniting the message process between point, as formula (7), upgrades all potential functions of uniting point.

In formula (7), X represents the unity point sending message; Y represents the unity point accepting message; S is point cut set between X, Y; be respectively the former potential function of uniting some Y, minute cut set S.

5th step: utilize the method shown in formula (8) to the potential function of uniting some X at fault node layer s _ion do marginalisation process.

6th step: recycling formula (9) calculates fault node layer s _imarginalisation conditional probability p (s _i| e), obtain the possibility that all fault types occur.

$p\left(s_i\right|e)=\frac{p(s_i,e)}{p\left(e\right)}=\frac{p(s_i,e)}{{\&Sigma;}_{s_i}p(s_i,e)}---\left(9\right)$

s ^*＝argmaxp(s _i|e)(10)

Fig. 2 is the Bayesian Network Topology Structures figure of the invention process type four-cylinder diesel engine lubricating system.

Claims (4)

1., based on the lubricating system of diesel oil engine method for diagnosing faults of Bayesian network, it is characterized in that, comprise the steps:

(1) fault type of lubricating system and outward sign are distinguished abstract to be fault node layer and sign node layer, to set up lubricating system of diesel oil engine Bayesian network model;

(2) utilize data acquisition system (DAS) to detect the performance parameter of lubricating system of diesel oil engine, adopt linear scale transform's method to carry out classification process to performance parameter, ω represents the actual performance parameter of lubricating system, ω ^*represent the standard value of performance parameter, represent the performance parameter after conversion, and then extensive lubricating system performance parameter, obtain the actual working state information e of lubricating system;

$\widetilde{\mathrm{\&omega;}}=\frac{\mathrm{\&omega;}}{{\mathrm{\&omega;}}^{*}}$

(3) according to the lubricating system actual working state information e obtained, to Bayesian network model sign node layer v _jstate π (v _j) carry out binary value, (ω _j) represent performance parameter ω _jextensive, represent and meet performance parameter ω _jextensive corresponding outward sign describes; The lubricating system Bayesian network model that adaptability correction is set up;

$\mathrm{\&pi;}\left(v_j\right)=\left\{\begin{array}{cc}1,& if{v}_j~{\tilde{v}}_j=\arg Q\left({\mathrm{\&omega;}}_j\right)\\ 0,& else\end{array}\right.$

(4) adopt Hugin Junction tree that revised lubricating system Bayesian network model is converted into associating tree, and then using lubricating system actual working state information e as reasoning evidence, by calculating fault node layer s _imarginalisation conditional probability p (s _i| e), the current failure type of lubricating system is diagnosed.

2. the lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network according to claim 1, is characterized in that: described fault type specifically comprises: piston ring packing inefficacy S ₁, add low on fuel S ₂, Cooler Fault S ₃, exceed S in serviceable life ₄, the improper S of lubricating oil oil product ₅, in lubricating oil containing bubble S ₆, pipeline leakage of oil S ₇, line clogging S ₈.

3. the lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network according to claim 1, is characterized in that: described outward sign specifically comprises: the too low V of lubricating oil liquid level ₁, the too high V of oil temperature ₂, enter the too low V of machine lubricating oil pressure ₃, go out the too low V of machine lubricating oil pressure ₄, go out the too low V of machine oil flow ₅.

4. the lubricating system of diesel oil engine method for diagnosing faults based on Bayesian network according to claim 1, is characterized in that: the concrete grammar of the lubricating system Bayesian network model that described adaptability correction is set up is: according to sign node layer v _jstate π (v _j), reset fault node layer pa (v in lubricating system Bayesian network model _j) prior probability P (pa (v _j)); Pa (v _j) be sign node layer v _jfather node;

$P\left(pa\left(v_j\right)\right)=\left\{\begin{array}{cc}1,& \begin{array}{cc}if& \mathrm{\&pi;}\left(v_j\right)=1\end{array}\\ 0,& \begin{array}{cc}if& \mathrm{\&pi;}\left(v_j\right)=0\end{array}\end{array}\right..$