CN102394017A - Knowledge-based welding defect prediction method - Google Patents

Abstract

The invention relates to a knowledge-based welding defect prediction method which is suitable for the field of virtual welding operation training. Access serves as a database; a typical structure serves as a structure of an expert system; the structure of the expert system comprises a man-machine interface module, a knowledge base module, an inference engine module, a knowledge base management module, an interpreter module, and the like; and a prediction process sequentially comprises data acquisition, real-time welding defect prediction and welding defect diagnostic analysis. According to the knowledge-based welding defect prediction method provided by the invention, common defects during a welding operation process can be predicted, diagnosed and analyzed, and without a direction of a teacher, a welding operator can independently operate and study so as to promote welding operating skill and theoretical level.

Description

A kind of weld defects Forecasting Methodology based on knowledge

Technical field

The present invention relates to a kind of virtual welding operation Training Methodology, be specifically related to a kind of weld defects Forecasting Methodology based on knowledge.

Background technology

Solder technology is one of gordian technique in the machine building industry; It is the important component part of modern advanced manufacturing technique; Along with the continuous development of science and technology, the continual renovation of the new material of welding, new technology, new equipment, welder's rejuvenation, more educated in addition; This just needs us must strengthen welder's technical merit and theoretical level, to adapt to the requirements at the higher level of social development to welder's technical ability.

Traditional welding operation training exists the training cost height; Length consuming time, problems such as environmental pollution, the continuous development of Along with computer technology; Computer technology is introduced the welding operation training has become possibility, and virtual welding operation training technique is with its safety, environmental protection, energy-conservation receiving much concern.Because the research of virtual welding operation training apparatus is in the starting stage; Most studies mechanism has been placed on its research emphasis improves on the fidelity and accuracy of detection of welding analog scene; Make every effort to offer the welder and be tending towards real welding surroundings; And it is comprehensive inadequately in the evaluation to the welding operational quality; The operation evaluation of existing both at home and abroad virtual welding training system only rests on the warning level mostly, and is single often for the warning of parameter, independently, and the evaluation of postwelding also is confined to welding current, voltage is isoparametric redraws in the reproduction.In the welding training of reality; Because elementary welder is not because theoretical foundation is not firm, because improper operation is easy to cause the generation of weld defects, and existing virtual welding training equipment is because Function of Evaluation is perfect inadequately when carrying out welding operation; The weld defects that not possibly not cause student's improper operation carries out real-time prompting; Even teacher's guidance is in person arranged, because teacher's level irregular, also be difficult to accomplish the standardization of training; Standardization, these all make the efficient of virtual welding operation training and result have a greatly reduced quality.

Expert system is the computer intelligence programming system that possesses special knowledge and experience, and the representation of knowledge in the employing artificial intelligence and knowledge reasoning technology realize usually can only be by the challenge of domain expert's solution.So far, expert system has number of applications in fields such as medical diagnosis, chemical engineering, Flame Image Process, geologic prospectings, has produced huge economic benefit and social influence.Because the significant complicacy of welding field knowledge, empirical; Welding is considered to be suitable for most one of field of developing expert system, and units such as Tsing-Hua University, Harbin Institute of Technology, University Of Tianjin, Shanghai Communications University, Gansu Polytechnical Univ have all successively carried out the research and development of welding expert system.

Summary of the invention

The objective of the invention is to the welding operation analog training system; A kind of simulation welding operation information through collecting in real time is provided; Weld defects to causing is reported to the police, and analyzes the weld defects Forecasting Methodology based on knowledge that weld defects produces reason and its corresponding solution is provided.

For realizing above-mentioned purpose, the technical scheme of employing of the present invention is:

1) foundation of knowledge base: knowledge base representes that mode adopts the production expression, and its representation is defined as: IF A ₁(α ₁) AND A ₂(α ₂) AND ... A _n(α _n) THEN B

A wherein ₁, A ₂..., A _nBe the prerequisite or the condition of rule, α ₁, α ₂, α _nBe the weighting coefficient of each influence factor in every rule, B is the A that satisfies condition ₁, A ₂..., A _nThe time conclusion of drawing, the significance level of welding defective effect factor is carried out assignment, guarantee ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1];

2) foundation of inference mechanism: inference mechanism is that the forward reasoning control strategy is eliminated the control strategy that strategy combines with conflict, according to the simulation welding real time data that collects, in knowledge base, carries out the prerequisite coupling in order, selects knowledge through Strategy of Conflict Resolution:

Introduce y _nRepresent real-time detected True Data A _n' for condition true value A _nDegree of membership, the real time data that collects of definition is at 0.8A _n-1.2A _nIn the time of in the scope, calculate the degree of membership of these data:

$y_n=\left\{\begin{array}{cc}0& ,{A_n}^{\&prime;}<0.8A_n\\ ((0.4\&times;A_n-\mathrm{abs}({A_n}^{\&prime;}-A_n))/(0.4\&times;A_n)& ,0.8A_n\&le;{A_n}^{\&prime;}\\ 0& ,{A_n}^{\&prime;}>1.2A_n\end{array}\&le;1.2A_n\right.$

For following rule:

IF?A ₁(α ₁)AND?A ₂(α ₂)AND...A _n(α _n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁, A ₂..., A _nThe degree of membership value be respectively y ₁, y ₂..., y _n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is:

When two rules lead to the same conclusion:

R ₁：IF?A ₁₁(α ₁₁)ANDA ₁₂(α ₁₂)AND…AND?A _1N(α _1n)THEN?B

R ₂：IF?A ₂₁(α ₂₁)ANDA ₂₂(α ₂₂)AND…AND?A _2n(α _2n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁₁, A ₁₂..., A _1nAnd A ₂₁, A ₂₂..., A _2nThe degree of membership value be respectively y ₁₁, y ₁₂..., y _1nAnd y ₂₁, y ₂₂..., y _2n, y _1n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is: $B=\mathrm{Max}(\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{1i}\&times;{\mathrm{\&alpha;}}_{1i},\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{2i}\&times;{\mathrm{\&alpha;}}_{2i})$

Thereby draw weld defects degree of confidence numerical value between [0,1];

3) in the welding operation process; Collection in real time can the mirror operation process status information be real-time welding data; And successively the rule condition in the welding defect knowledge base is inquired about; Seek and real-time welding data matching rules; If welding data is in the prerequisite scope of weld defect occurring in real time; Then calculate its subjection degree to rule condition by membership function, and in conjunction with the difference of different welding parameters to welding defect influence weight, the confidence level of the weld defect that the calculating reasoning draws;

4) when matching rules not only the time; Relatively each bar rule the confidence level of corresponding conclusion; Get that confidence level maximum and preset threshold compare in the defective conclusion; When surpassing this threshold value; Think that promptly the corresponding defective conclusion of this rule is credible, judgement this weld defect of preferential generation is also pointed out;

5) after welding operation is accomplished, produce defects count and type in the statistics welding process, and reasoning weld defects generation reason, provide the welding operation recommendation on improvement.

The present invention is based on the simulative training device for manual arc welding operation of Xi'an Communications University's research and development; 200610042953.X; The data-signal that hardware unit through welding operation analog training systems such as single spot position detecting device, electro-conductive glass simulation test plate (panel), data collecting cards collects; Situations such as the positional information of welding rod, velocity information, angle information and striking blow-out in the mirror operation process, thus the data foundation is provided for weld defects prediction.

The present invention has the following advantages:

(1) the present invention is directed to simulation welding operation training real-time reminding is carried out in student's improper welding operation; And the reason that weld defects produces analyzed; Rational suggestion for operation is proposed; Guiding operation personnel reduce the extent of injury of weld defects timely and effectively, when improving welding operation, improve the soldering theory level.

(2) the present invention has set up the model of inexact reasoning, according to this model, can infer the uncertainty of conclusion by true and regular uncertainty, can judge the contingent probability of weld defect based on conclusion confidence level size.

Description of drawings

Fig. 1 is a structure composition diagram of the present invention;

Fig. 2 is a degree of membership distribution plan of the present invention;

Fig. 3 is a reasoning process flow diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing the present invention is done further explain.

The present invention is based on the necessary requirement of real time of weld defects prediction of virtual welding operation training system, just can give operating personnel and instruct timely and effectively, reach the effect that instructs welding.For this requirement, the present invention is decomposed into two task layers with the failure prediction process, in real time failure prediction layer and postwelding defect diagonsis analysis layer.

Real-time estimate module major function is to realize real-time weld defects prediction; Rule in data message that can reflect welder's mode of operation that collects and the expert system knowledge base is mated one by one; The inference strategy of machine is analyzed by inference; Dope issuable weld defects, and carry out alarm.

The postwelding defect diagonsis has been realized the task that the real-time diagnosis part can not be accomplished; Promptly add up the number of weld defects in the welding operation process; Position and life period that each weld defects occurs are analyzed the generation reason of each weld defects, and corresponding solution are provided.Through further statistics and analysis, can realize improving the purpose of welder's soldering theory level.

Referring to Fig. 1, system of the present invention is made up of knowledge base, inference machine, integrated data base, KBM, interpreter and man-machine interface.

(1) knowledge base

Knowledge base is in order to deposit the professional knowledge that the welding field expert provides, and knowledge base is made up of the rule of a rule, therefore is also referred to as rule base, and the weld defects prediction is exactly through the rule in the traversal knowledge base, and the object of searching coupling is realized.

(2) inference machine

Under certain control strategy in the database and current information; Identification and choose and in the knowledge base current problem is found the solution useful knowledge and carry out reasoning; In the present invention; Because the knowledge in the knowledge base is often not exclusively with coarse, thereby reasoning process of the present invention adopts inexact reasoning.

(3) integrated data base

Be used to deposit the primary data about problem solving, find the solution state, intermediate result is supposed, target and final solving result.

(4) interpretive routine

According to user's enquirement, to the conclusion that system provides, the current state of finding the solution of solution procedure and system furnishes an explanation, and is convenient to the problem solving of user's understanding system.

(5) KBM

KBM can let operating personnel, expert, knowledge engineer easily knowledge base added, revise and delete, and for virtual welding operation training system, knowledge-base management and maintenance only limit to specified permission, accomplishes like teacher or keeper.

(6) man-machine interface

Expert or user's input information is translated as the acceptable internal form of system, becomes human understandable external form to system to the information translation of expert or user's output.Bookkeepings such as the interpolation of knowledge base, modification are accomplished through man-machine interface.

Referring to Fig. 3,1) foundation of knowledge base

The knowledge base of expert system representes that mode adopts the production expression; Because the influence factor of weld defects is complicated; A kind of weld defects is often relevant with a plurality of welding parameters, and therefore the condition in every rule of the present invention is not single generally, and; Different welding parameters generally are different to the influence degree that produces weld defects, and being reflected in every rule is the weighting coefficient α of each influence factor _nBe different, therefore the representation with knowledge is defined as:

IF?A ₁(α ₁)AND?A ₂(α ₂)AND...A _n(α _n)THEN?B

A wherein ₁, A ₂..., A _nBe the prerequisite or the condition of rule, α ₁, α ₂, α _nBe the weighting coefficient of each influence factor in every rule, B is the A that satisfies condition ₁, A ₂..., A _nThe time conclusion of drawing, the significance level of welding defective effect factor is carried out assignment, guarantee ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1];

2) inference mechanism

The inference strategy that the present invention used is that the forward reasoning control strategy is eliminated the control strategy that strategy combines with conflict.Based on the simulation welding real time data that collects, in knowledge base, carry out the prerequisite coupling in order, select knowledge through Strategy of Conflict Resolution.

Because more than one of the precondition in the rule; The data probability that the welding operation data that collect accurately meet each precondition is very little; Therefore, in reasoning process, the present invention adopts following inexact reasoning technology to come probabilistic propagation of processing rule.

Introduce y _nRepresent real-time detected True Data A _n' for condition true value A _nDegree of membership, in the present invention, the real time data that collects of definition is at 0.8A _n-1.2A _nIn the time of in the scope, calculate the degree of membership of these data.Degree of membership definite as shown in Figure 2:

The computing formula of degree of membership is:

$y_n=\left\{\begin{array}{cc}0& ,{A_n}^{\&prime;}<0.8A_n\\ ((0.4\&times;A_n-\mathrm{abs}({A_n}^{\&prime;}-A_n))/(0.4\&times;A_n)& ,0.8A_n\&le;{A_n}^{\&prime;}\\ 0& ,{A_n}^{\&prime;}>1.2A_n\end{array}\&le;1.2A_n\right.$

For following rule:

IF?A ₁(α ₁)AND?A ₂(α ₂)AND...A _n(α _n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁, A ₂..., A _nThe degree of membership value be respectively y ₁, y ₂..., y _n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is:

When two rules lead to the same conclusion:

R ₁：IF?A ₁₁(α ₁₁)ANDA ₁₂(α ₁₂)AND…AND?A _1N(α _1n)THEN?B

R ₂：IF?A ₂₁(α ₂₁)ANDA ₂₂(α ₂₂)AND…AND?A _2n(α _2n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁₁, A ₁₂..., A _1nAnd A ₂₁, A ₂₂..., A _2nThe degree of membership value be respectively y ₁₁, y ₁₂..., y _1nAnd y ₂₁, y ₂₂..., y _2n, y _1n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is: $B=\mathrm{Max}(\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{1i}\&times;{\mathrm{\&alpha;}}_{1i},\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{2i}\&times;{\mathrm{\&alpha;}}_{2i})$

The weld defects degree of confidence numerical value that draws according to this reasoning algorithm is [0; 1] between,, avoids less the predicting the outcome of weld defects probability in order to improve the weld defects prediction accuracy; The present invention is provided with alarm threshold value, is used for judging whether the weld defects that dopes being carried out alarm.

The reasoning flow process is as shown in Figure 3:

1) in the welding operation process, the information that collection in real time can the mirror operation process status is real-time welding data, and successively the rule condition in the welding defect knowledge base is inquired about, and seeks and welding data matching rules in real time;

2) if welding data is in the prerequisite scope of weld defects occurring in real time; Then calculate its subjection degree to rule condition through membership function; And combine different welding parameters to welding the difference of defect influence weight, the degree of confidence of the weld defects that the calculating reasoning draws;

3) next bar rule is mated, process is with 2);

4) when matching rules not only the time, need each bar rule relatively the degree of confidence of corresponding conclusion.Get that degree of confidence maximal value and preset threshold compare in the defective conclusion; When surpassing this threshold value; Think that promptly the corresponding defective conclusion of this rule is credible, judgement this weld defects of preferential generation is also pointed out, when providing the weld defects conclusion; Also provide the confidence value of conclusion, and stop this time matching process.Otherwise, when the maximum confidence value of conclusion less than threshold value, it is not pointed out, and removes the The reasoning results of keep in the integrated data base, continuation matching process next time.

5) after welding operation is accomplished, produce defects count and type in the statistics welding process, and reasoning weld defects generation reason, and provide the welding operation recommendation on improvement.

Claims (1)

1. weld defects Forecasting Methodology based on knowledge is characterized in that:

1) foundation of knowledge base: knowledge base representes that mode adopts the production expression, and its representation is defined as: IF A ₁(α ₁) AND A ₂(α ₂) AND ... A _n(α _n) THEN B

A wherein ₁, A ₂..., A _nBe the prerequisite or the condition of rule, α ₁, α ₂, α _nBe the weighting coefficient of each influence factor in every rule, B is the A that satisfies condition ₁, A ₂..., A _nThe time conclusion of drawing, the significance level of welding defective effect factor is carried out assignment, guarantee ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1];

2) foundation of inference mechanism: inference mechanism is that the forward reasoning control strategy is eliminated the control strategy that strategy combines with conflict, according to the simulation welding real time data that collects, in knowledge base, carries out the prerequisite coupling in order, selects knowledge through Strategy of Conflict Resolution:

Introduce y _nRepresent real-time detected True Data A _n' for condition true value A _nDegree of membership, the real time data that collects of definition is at 0.8A _n-1.2A _nIn the time of in the scope, calculate the degree of membership of these data:

$y_n=\left\{\begin{array}{cc}0& ,{A_n}^{\&prime;}<0.8A_n\\ ((0.4\&times;A_n-\mathrm{abs}({A_n}^{\&prime;}-A_n))/(0.4\&times;A_n)& ,0.8A_n\&le;{A_n}^{\&prime;}\\ 0& ,{A_n}^{\&prime;}>1.2A_n\end{array}\&le;1.2A_n\right.$

For following rule:

IF?A ₁(α ₁)AND?A ₂(α ₂)AND...A _n(α _n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁, A ₂..., A _nThe degree of membership value be respectively y ₁, y ₂..., y _n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is:

When two rules lead to the same conclusion:

R ₁：IF?A ₁₁(α ₁₁)ANDA ₁₂(α ₁₂)AND…AND?A _1N(α _1n)THEN?B

R ₂：IF?A ₂₁(α ₂₁)ANDA ₂₂(α ₂₂)AND…AND?A _2n(α _2n)THEN?B

Wherein, ∑ α _i=1, i=1,2 ..., n; α _i∈ [0,1]

The true value of condition is for A ₁₁, A ₁₂..., A _1nAnd A ₂₁, A ₂₂..., A _2nThe degree of membership value be respectively y ₁₁, y ₁₂..., y _1nAnd y ₂₁, y ₂₂..., y _2n, y _1n, y _i∈ [0,1], 1≤i≤n, the degree of confidence that produces conclusion B so is: $B=\mathrm{Max}(\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{1i}\&times;{\mathrm{\&alpha;}}_{1i},\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_{2i}\&times;{\mathrm{\&alpha;}}_{2i})$

Thereby draw weld defects degree of confidence numerical value between [0,1];

3) in the welding operation process, the information that collection in real time can the mirror operation process status is real-time welding data, and successively the rule condition in the welding defect knowledge base is inquired about, seek and welding data matching rules in real time,

If welding data is in the prerequisite scope of weld defects occurring in real time; Then calculate its subjection degree to rule condition through membership function; And combine different welding parameters to welding the difference of defect influence weight, the degree of confidence of the weld defects that the calculating reasoning draws;

4) when matching rules not only the time; Relatively each bar rule the confidence level of corresponding conclusion; Get that confidence level maximum and preset threshold compare in the defective conclusion; When surpassing this threshold value; Think that promptly the corresponding defective conclusion of this rule is credible, judgement this weld defect of preferential generation is also pointed out;

5) after welding operation is accomplished, produce defects count and type in the statistics welding process, and reasoning weld defects generation reason, provide the welding operation recommendation on improvement.

Images