阅读说明：本技术 一种基于智能化技术的复杂曲面数字化闭环精密加工方法 (A kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology ) 是由 郭渊 杜敏 任国栋 陈炜 朱英霞 于 2019-07-29 设计创作，主要内容包括：本发明公开了一种基于智能化技术的复杂曲面数字化闭环精密加工方法,该方法具体为：构建复杂曲面零件加工知识事例库,将待加工曲面零件在事例库中进行匹配,获得相似度A；当A＞0.95时,直接利用检索获得事例提供的加工知识对曲面零件进行数控加工,原位测量后,通过与理想的数字化模型比较获取加工误差,并进行误差补偿；当0.95≥A≥0.80时,利用检索获得事例提供的仿真知识修正曲面零件可能存在的缺陷,然后再进行加工。本发明以智能化方式实现复杂曲面数字化闭环集成制造,从而提高了加工效率和准确性。(The invention discloses a kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology, this method specifically: building complex curved surface parts process knowledge case searching, and curved surface part to be processed is matched in case searching, obtain similarity A；As A > 0.95, the processing knowledge that example provides directly is obtained using retrieval, numerical control processing is carried out to curved surface part, after in situ measurement, by the acquisition mismachining tolerance compared with ideal digital model, and carry out error compensation；As 0.95 >=A >=0.80, the Simulation knowledge that example provides is obtained using retrieval and corrects curved surface part defect that may be present, is then processed again.The present invention realizes complex-curved Digitized Closed Loop Integrated manufacture in a manner of intelligent, to improve processing efficiency and accuracy.)

1. a kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology, which is characterized in that building is complicated Curved surface part processes knowledge case searching, and curved surface part to be processed is matched in case searching, obtains similarity A；As A > When 0.95, numerical control processing directly is carried out to curved surface part, after in situ measurement, is added by being obtained compared with ideal digital model Work error, and carry out error compensation；As 0.95 >=A >=0.80, curved surface part defect that may be present is corrected.

2. the complex-curved Digitized Closed Loop precision machining method according to claim 1 based on intellectualized technology, special Sign is that the case searching is constructed based on integrated-type knowledge representation meta-model, the integrated-type knowledge representation member mould Type includes the description of curved surface part feature, attributive character and parameter setting needed for intelligent simulation module, the manufacture of curved surface part Information, surface reconstruction optimization algorithm.

3. the complex-curved Digitized Closed Loop precision machining method according to claim 1 based on intellectualized technology, special Sign is that the similarity uses the IC similarity algorithm based on concept leaf, and the concept leaf is the institute as derived from common node There is concept.

4. the complex-curved Digitized Closed Loop precision machining method according to claim 1 based on intellectualized technology, special Sign is, when numerical control processing directly obtains part manufacturing information from case searching.

5. the complex-curved Digitized Closed Loop precision machining method according to claim 1 based on intellectualized technology, special Sign is, when amendment curved surface part directly obtains CAD suggestion mode from case searching.

6. the complex-curved Digitized Closed Loop precision machining method according to claim 1 based on intellectualized technology, special Sign is, the data selection and data analysis algorithm, the digital model used of when amendment curved surface part passes through song Reconstruction and optimization algorithm in face obtains, and the data selection and data analysis algorithm, surface reconstruction optimization algorithm pass through data processing Decision Control submodule is selected.

7. the complex-curved Digitized Closed Loop precision machining method according to claim 6 based on intellectualized technology, special Sign is that the data processing Decision Control submodule is knowledge base constructed by Process Based.

Technical field

The present invention relates to the technical fields such as digitlization, closed loop Precision Machining, knowledge engineering (KBE) and ontology, specifically It is related to a kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology.

Background technique

In recent years, with the rapid development of high-quality precision and sophisticated technology, to the accuracy of manufacture and efficiency, more stringent requirements are proposed, and The tradition progress integrated by independent link or locally is no longer satisfied new skill come the method for improving the accuracy of manufacture and efficiency Art needs, and era development an urgent demand will be designed, and be processed, three links of measurement carry out integrated closed loop Integrated manufacture.

But how three design, processing, measurement field organic integration to get up to form closed loop, still lack at present Practicable theory and technology is involved in integrated-type knowledge representation model to the further research of the problem and establishes, learn more The problems such as the cross-cutting knowledge information fusion of section and intelligent data processing, specifically:

(1) from the point of view of integrated strategy and technique study, the research of this respect is also very weak at present.Realize these three necks Domain integrates, and is not easy to realize by the way that the engagement of hardware is integrated, and one side hardware is not easy to merge Knowledge Element by the way that interfacing is integrated Element, and being turned under the background of future development inexorable trend with intelligence, can not also absorb in time artificial intelligence technology it is newest at Fruit.

(2) from the point of view of the research of integration realization technology, current Research of Integration primarily directed to CAX integrate based on The technology of feature, and shortage is compared in design based on knowledge technical research.Based on the technology of feature in information representation ability compared with Weak, expression of results is inaccurate.

(3) traditional data processing method excessively relies on the participation manual decision of expert, far can not adapt to quickly, Timeliness, dynamic, the decision data of magnanimity and processing.

Summary of the invention

In order to solve the problems in the existing technology, the present invention provides a kind of based on the complex-curved of intellectualized technology Digitized Closed Loop precision machining method realizes design, processing, the closed loop Integrated manufacture for measuring three fields.

A kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology, building complex curved surface parts add Work knowledge case searching, curved surface part to be processed is matched in case searching, obtains similarity A；As A > 0.95, directly Numerical control processing is carried out to curved surface part, after in situ measurement, by the acquisition mismachining tolerance compared with ideal digital model, is gone forward side by side Row error compensation；As 0.95 >=A >=0.80, curved surface part defect that may be present is corrected.

Further, the case searching is constructed based on integrated-type knowledge representation meta-model, the integrated-type knowledge Expression meta-model includes the description of curved surface part feature, attributive character and parameter setting needed for intelligent simulation module, curved surface zero Manufacture information, the surface reconstruction optimization algorithm of part.

Further, the similarity uses the IC similarity algorithm based on concept leaf, and the concept leaf is by common node Derivative all concepts.

Further, part manufacturing information directly is obtained from case searching when the numerical control processing.

Further, CAD suggestion mode directly is obtained from case searching when the amendment curved surface part.

Further, data selection used when the amendment curved surface part and data analysis algorithm, the digitlization mould Type is obtained by surface reconstruction optimization algorithm, and the data selection and data analysis algorithm, surface reconstruction optimization algorithm pass through Data processing Decision Control submodule is selected.

Further, the data processing Decision Control submodule is knowledge base constructed by Process Based.

The invention has the benefit that

1, the present invention first mutually ties ontology and KBE (Knowledge Based Engineering, knowledge engineering) technology Build vertical integrated-type knowledge representation meta-model jointly, which can express the implicit sex knowledge in field, fuzzy sex knowledge etc., in knowledge table It has breakthrough on up to mode, establishing has more powerful knowledge, the realization technology of information representation ability；And using this meta-model as base Plinth constructs three field integrated-type case searchings and Precise Semantics understanding type inference machine, using Feature Extraction Technology by three fields Software and hardware element is mapped as the element or parameter of knowledge, and by design, processing, three fields of measurement are integrated realizes Closed loop manufacture, is broken through on integrated strategy and method.

2, the present invention realizes the intelligent processing and Decision Control of data with rule-based knowledge reasoning technology, with intelligence Change mode realizes complex-curved Digitized Closed Loop Integrated manufacture, to improve processing efficiency and accuracy.

3, the present invention uses modularization research strategy, in order to which debugging, the maintenance and expansion of system of processing upgrade.Processing system System generally includes Knowledge Aggregation module, emulation module and in situ measurement-compensation processing module, in addition, in Knowledge Aggregation module Also include data processing Decision Control submodule, for emulating the control decision of data and in situ measurement data processing, and this three A module has the function of can be convenient extension, absorbs new technological achievement in time and constantly enhance system of processing.

Detailed description of the invention

Fig. 1 is that multi-level more granularities based on GHSOM construct case searching flow chart；

Fig. 2 is the semantic understanding type inference machine structural schematic diagram based on IC；

Fig. 3 is control planning schematic diagram of the Knowledge Aggregation module in system of processing；

Fig. 4 is the working mechanism figure of emulation module；

Fig. 5 is in situ measurement-compensation processing module working mechanism figure；

Fig. 6 is NC code building flow chart；

Fig. 7 is processing general flow chart.

Specific embodiment

The specific technical solution of the present invention is further described below in conjunction with attached drawing, it should be noted that this hair Bright protection scope is not limited to this.

As shown in fig. 7, a kind of complex-curved Digitized Closed Loop precision machining method based on intellectualized technology, including it is following Step:

Step 1: the building of Knowledge Aggregation module

Integrated-type knowledge representation meta-model is initially set up, constructs complicated song using GHSOM technology based on this meta-model Surface parts process knowledge case searching, while constructing the data processing Decision Control submodule of Process Based, and data processing is determined Plan control submodule is rule-based knowledge base.Curved surface part to be processed is matched to obtain the higher thing of similarity in case searching Example collection, the similarity of each example in curved surface part to be processed and case set is calculated by the semantic understanding type inference machine based on IC, Highest similarity is best example.

(1) the integrated-type knowledge representation meta-model based on Ontology, meta-model connection design, processing, measurement are established Three fields and become Knowledge Aggregation cell, the integrated-type knowledge representation meta-model based on Ontology includes three parts:

First part is the description of curved surface part feature, and part feature includes design requirement or rule, i.e. integrated-type knowledge thing The problem of example, describes；

Second part is attributive character needed for intelligent simulation module (format, material etc.), parameter setting (size number According to, non-geometry feature etc.)；

Part III includes the manufacture information (cutter path, Processing Strategies, mismachining tolerance compensation etc.) of curved surface part, curved surface Reconstruction and optimization algorithm (B-spline or T spline algorithms).

(2) building of case searching, which uses, is based on GHSOM (Growing Hierarchical Self-Organizing Map, growth type hierarchical ad-hoc mapping) multi-level more granularity knowledge build library strategy

When existing example is after feature extraction and data conversion, the input space vector as GHSOM constructs multilayer The case searching of secondary more granularities, as shown in Figure 1, specific steps are as follows:

1) GHSOM is initialized and is established the 0th layer (virtual level), and the 0th layer of weights initialisation is corresponded to for input space vector The average value of weight calculates the arithmetical mean error m of the 0th layer of weight vectors and input unit weight vectors, wherein input unit Belong to input space vector；

Wherein: n_IFor the number of all input space vectors, ω₀For the 0th layer of weight, x_jFor input unit weight；

2) the 1st layer for establishing GHSOM

SOM (Self-Organizing Map, the Self-organizing Maps) system for initially setting up one 2 × 2 dimension structure, goes forward side by side Row neural metwork training；The weight vectors of this layer of each upper neuron and the error M' of input unit weight vectors are calculated, are sought The maximum neuron of error is looked for, and the neuron and adjacent neurons are compared, similarity is the smallest to be defined as not phase Like neuron.New neuron is added between the neuron and dissimilar neuron, as soon as every addition new neuron, needs The weight vectors of this layer of each upper neuron and the arithmetical mean error M of input unit weight vectors are recalculated, constantly It repeats the above process until the M of this layer is reduced to threshold value λ, training can terminate.

Wherein: ω_iFor i-th layer of weight,For the number of i-th layer of neuron.

3) foundation of subsequent layers

1st layer each neuron is mapped, subsequent layers are established, subsequent layers carry out neural network instruction according to process 2) Practice, until the arithmetical mean error m of subsequent layers (being set as i-th layer)_iLess than a certain threshold gamma, deconditioning.

4) after training, that is, the building of case searching is completed；In the training process, can by change λ and γ size, Control the granularity of classification of each layer of GHSOM.

(3) the semantic understanding type inference machine based on IC is constructed

As shown in Fig. 2, by the description of curved surface part to be processed by forming Feature concept after Feature Extraction Technology, by these Similar example is retrieved in case searching and matched to Feature concept, obtains similar case set, then passes through IC similarity calculation Select best example.

IC similarity algorithm is specific as follows:

Assuming that curved surface part descriptive statement to be processed is likely to form n Feature concept, by these after through feature extraction Feature concept forms a n member Feature concept collection B'=[B₁',B₂'…B_i'…B_n'], then by Feature concept collection B' and case searching In concept matched after, generate another concept set: B=[B₁,B₂…B_i…B_n], the overall similarity of concept set B' and B Use Sim_IC(B', B) is indicated.To calculate the overall similarity of B' and B, it is necessary to it is general first to calculate each in this two groups of concept sets It reads to (B_i',B_i) similarity, use Sim_IC(B_i',B_i) indicate；It is whole to obtain by calculating the standard value of each concept pair Body similarity.The practical height for referring to information registration contained by the two concepts of the size of the concept similarity of two different contents It is low.Assuming that: B_i-It is concept to (B_i',B_i) in the common node of the GHSOM bottom, by node B_i-All concepts being derived It is referred to as common node B_i-Concept leaf L (B_i-)；Each concept of expression more abundant is come by using the definition of concept leaf Pair meaning and the information content, and then differentiation concept definitely is to existing difference.Currently, according to existing theoretical knowledge, IC similarity algorithm is calculated by the height of statistical nature concept frequency of occurrences in case searching, the IC of Feature concept B Similarity value is calculated by following formula:

IC=log^-1P(B) (4)

Wherein: P (B) indicates the frequency that Feature concept B occurs in case searching；

When calculating IC similarity using leaf, have:

IC(B_i-)=log^-1 P(L(B_i-)) (6)

Then there is Sim_IC(B_i',B_i) calculating be shown below:

Standardize similarity:

The size of curved surface part more to be processed and each example similarity, it is best example that it is maximum, which to select similarity,.

As shown in figure 3, setting similarity threshold A, if when A > 0.95, case searching, which connects and controls in situ measurement-compensation, to be added Work module；If when 0.95 >=A >=0.80, case searching connects and control emulation module, original is entered back into after the completion of the work in the stage Position measurement-compensation processing module.

(4) data processing Decision Control submodule is constructed

Data processing Decision Control submodule includes the appropriate data mining algorithm (DMA) of selection, data fitting algorithms (DFA) and selection data sample (Ad), the algorithm can be completed to update and constantly absorbing new results, data processing decision The particular content of control includes:

0.95 then:DMA=GA+SVM of if:A >

DFA=T

Ad=Ad₁=NO.0001

Ad₂=NO.0003

Ad₃=NO.0005

……

Wherein: GA is genetic algorithm, and SVM is support vector machines, and T is T spline algorithms；Ad is using equidistant sampling.

The then:DMA=GA of if:0.95 >=A >=0.80

DFA=B

Ad=Ad₁=NO.0001

Ad₂=NO.0003

Ad₃=NO.0005

……

Wherein: B is B-spline algorithm.

Data processing Decision Control submodule is applied to: 1) intelligent simulation module selection data and data analysis are calculated Method, 2) in situ measurement-compensation processing module selection surface reconstruction optimization algorithm.

Step 2: the building of intelligent simulation module

When similarity is not very high (0.95 >=A >=0.80), need to predict curved surface part by intelligent simulation module Possible defect is corrected: the CAD (Computer Aided Design, CAD) in case searching is suggested Model carries out CAE (Computer Aided Engineering, computer-aided engineering) data simulation by data conversion, leads to It crosses data processing Decision Control submodule and data mining algorithm (DMA), data fitting algorithms (DFA) and data sample is provided Digital model is analyzed and formed to selection to emulation data, and carries out similarity with the part model of curved surface part to be processed It calculates, judges whether there is issuable defect.If entering in situ when similarity A > 0.95 and prediction zero defect generation Otherwise parameter (such as parameter A, B, C, x in formula (6) of digital model are modified in measurement-compensation processing module₀、y₀、z₀With And curved-surface materials etc.) afterwards feedback driving CAD suggestion mode be corrected, continue to execute data conversion, emulation and analysis.Emulate mould Block working mechanism is as shown in Figure 4.

Assuming that digital model are as follows:

Step 3: building in situ measurement-compensation processing module

In situ measurement-compensation processing module working mechanism is as shown in figure 5, call directly curved surface part manufacture letter by case searching Breath forms NC code and carries out numerical control processing, then drives measuring instrument to carry out accurate survey by in-situ measuring method and operation sequence Amount, and the measurement data of acquisition is analyzed, is fitted processing curve, mismachining tolerance is obtained compared with digital model, is commented Valence processing quality.Assuming that assessed value is α, as mismachining tolerance > α, error is further decreased by compensating processing, until processing misses When difference < α, complex-curved numerical control processing is completed.The error assessment index of the module can be with the improve of manufacture measure of precision It constantly expands, algorithm abundant and constantly absorbing newest fruits can update, so that the accuracy of manufacture is continuously improved.Specific building Process are as follows:

(1) part that will be obtained after the part model directly acquired from case searching or completion intelligent simulation module work Model (the CAD suggestion mode after correction) obtains point cloud data by data conversion, then passes through data processing Decision Control Module selects surface reconstruction optimization algorithm (B-spline or T spline algorithms) to carry out curved surface weight according to curved surface part similarity to be processed Structure obtains ideal digital model (ideal surface).

(2) directly NC is completed using the cutter path of curved surface part matching example, Processing Strategies, mismachining tolerance compensation etc. to compile Journey carries out numerical control processing and carries out in situ measurement.

(3) in situ measurement error analysis

1) according to having actual processing test point, use space curved surface data fitting algorithms true to describe for least square method Qualitative complex-curved P, and the complex-curved P of certainty is defined as regression model, specific algorithm is as follows:

The reference representation of the complex-curved P of certainty are as follows:

Wherein: x₀、y₀、z₀, A, B, C be unknown parameter；

Write as general expression are as follows:

x²+ay²+bz²+ cx+dy+ez+f=0 (11)

Wherein:

It is fitted sample data (having practical measuring point) by least square method, six unknown numbers are estimated.

If each sample point is (x_i,y_i,z_i), then the error of each sample point can indicate are as follows:

e_i(a, b, c, d, e, f)=x_i ²+ay_i ²+bz_i ²+cx_i+dy_i+ez_i+f (12)

The then error sum of squares of each sample point are as follows:

It enablesFirst-order partial derivative to a, b, c, d, e, f is 0, then has:

And it enables

Following form can be obtained by bringing above-mentioned average value into partial derivative simplification:

Write as matrix form are as follows:

The inverse matrix of premultiplication coefficient matrix is distinguished on both sides, solves the estimated value of a, b, c, d, e, f.

Then by following various:

Solve x₀、y₀、z₀, A, B, C parameter value.

2) by comparing processing curve and ideal surface, the mismachining tolerance of all sampling test points is calculated, calculation formula is as follows:

Wherein: (x, y, z) indicates the measuring point coordinate on processing curve, (x₁,y₁,z₁) indicate on ideal surface most from measuring point The coordinate of near point；

3) decompose error: mismachining tolerance includes systematic error and random error, using space statistical analysis method In Moran, sI (not blue index statistic law) carries out error decomposition, when not orchid index I level off to 0 when, indicate systematic error and Random error is independently distributed.Use ε_iIndicate actual processing detection location point i to ideal processing curve normal direction deviation,Indicate z The average value of ε at test point.The expression formula of not blue index are as follows:

Wherein: S₀For the set of all space weights, ω_ijIndicate another reality at curved surface actual processing test point i Test point j is processed to the weight coefficient of i space behavior.

The statistic of processing test point may be expressed as:

E (I)=- 1/ (z-1) (17)

Var (I)=E (I²)-E(I)² (18)

Wherein: E (I) indicates that not blue Mean value of index, Var (I) indicate not blue index variance.

Certainty complex surface model P (x, y, z)=Q (x, y, z)+d_s(x, y, z), Q indicate complex-curved, the d of ideal_sTable Show the systematic error in NC Machining Process, and Indicate the average value of e at z test point.

Actual processing curved surface M (x, y, z)=P (x, y, z)+d_μ(x, y, z), d_μIndicate random error, and

It seeks analyzing the regression model of processing curve P by iteration, using one-sided test, takes the significance to be 0.01, the critical value T of the standardized normal distribution of mismachining tolerance_0.01It is 2.33, if T < T_0.01, it is independent that mismachining tolerance obeys space Property distribution, then can regard random error as；If T > T_0.01, need to re-establish processing curve model and test statistics, directly It obeys space to sample point Norma l deviation to be independently distributed, until error is decomposed into systematic error and random error.

(4) error compensation (as mismachining tolerance > α)

1) contact measuring head pretravel compensates: for contact measuring head when carrying out in situ detection, probe needs to generate certain journey The deformation of degree can be such that activation signal triggers.Therefore when the gauge head of trigger signal feedback to numerically-controlled machine tool, gauge head has been produced Certain displacement difference is given birth to.In addition, when trigger signal feedback is ceased and desisted order to numerically-controlled machine tool, numerically-controlled machine tool stop motion this Process gauge head can generate certain displacement difference.To stopping since in situ detection, the displacement difference of gauge head is known as pretravel error.

The actual processing test point of processed curved surface is mapped in standard ball on its normal vector, in standard ball not It is detected with normal vector, realizes the calibration to gauge head pretravel error.Standard ball includes the various different methods in three-dimensional space This can be obtained by comparing the coordinate value in actual processing test point normal vector direction in standard ball and ideal coordinates value in vector The pretravel error in normal vector direction realizes gauge head to obtain the pretravel error of processing curve actual processing test point Pretravel error compensation.

2) contact measuring head radius error compensates: when online in situ measurement detects processing curve, gauge head ball center Coordinate value be in situ measurement instrument detection coordinate value, deviate about radius of a ball of bulb and curved face contact point at a distance from, therefore It needs to carry out error compensation to gauge head radius.Along the normal direction of processing district face actual processing test point when gauge head progress in situ detection Amount direction is moved, therefore the compensation of gauge head radius error is compensated also along the normal vector direction of test point.Compensation formula ForWherein P is gauge head ball's dead center, and R is the radius of the ball,For the normal vector direction of test point i.

(5) NC code is generated based on error compensation

1) mismachining tolerance model generates: after obtaining systematic error, by one distance d of tool offset_sTo compensation position, lead to Overcompensation point generates mismachining tolerance model.The calculation formula of compensation point are as follows:

Wherein: A (x, y, z) is mismachining tolerance compensation point,For mismachining tolerance direction.

2) tool-path planning: the basic demand of complex surface machining is safe and efficient, feasible, the letter of simultaneous processing point Breath will change with curved surface and be changed.Method of section can be used when Machining of Curved Surface, enabled one group of parallel plane intersect with processing curve, obtained One group of intersection curve is obtained, as by tool sharpening path.And intersection curve is discrete for intensive point, data when being programmed so as to NC Reading.

3) biasing of cutter location: for the surface such as plane etc. of rule, center cutter point is just and processing stand differs one The distance of the cutter radius of a ball, actual tool coordinate are exactly processing point coordinates plus radius value.It is complex-curved for carrying out For processing, situation can become more complex, but basic principle is identical.When cutter is processed, it is assumed that processing contact point is certainly By the A point on curved surface, but when actual motion, feed is to be walked according to tool radius, therefore coordinate to be offered is needed in code It should be center cutter B point.

Tool offset formula are as follows:

Wherein: B (x, y, z) is center cutter point, and i (x, y, z) is actual processing test point, r tool radius；Per unit system Vector.

4) generation of NC code

As shown in fig. 6, file header is written first, the cutter location after biasing is then read, and cutter location X, Y, Z data are write Enter in NC code, processing method is controlled by writing Machining Instruction, until terminating the volume of NC code after Path is read It writes.

The above is present pre-ferred embodiments, is merely to illustrate design philosophy and feature of the invention, purpose exists In enabling those skilled in the art's content according to the present invention and implement accordingly, protection scope of the present invention is not limited to State embodiment.So all any modification, equivalent substitution, improvement and etc. done under technical thought of the invention, should be included in this Within the protection scope of invention.