Nerve network makes the application in machining craft solely in pulse electrochemistry

  • Time:
  • Click:132
  • source:SHERBURNE CNC Machining
Summary: The approximate maths model between the exterior quality after using BP nerve network to build pulse electrochemistry to make treatment solely and treatment craft parameter, need not consider each craft parameter surely between complex underlying concern, be opposite according to the mathematical model between them pulse electrochemistry rectifies the worse technology parameter in treatment to undertake forecasting solely, forecast result and experimental data to have better consistency, can make the reference that treatment studies theoretically and machines actually solely as pulse electrochemistry. Keyword: Pulse electrochemistry; Light is whole; BP nerve network; Build a model. Pulse electrochemistry rectifies treatment to make processing technique as a kind of new-style light solely, have a lot of distinct advantages, it and union of photograph of numerical control technology form the pulse electrochemistry that can accuse to make processing technique solely, there will be broader applied perspective in mechanical production domain. The pulse electrochemistry that can accuse makes processing technique solely is the craft parameter in controlling treatment process in real time with the computer, thereby the metallic purify speed of surface of adjusting control treatment, reasonable match craft parameter, the nicety that realizes high accuracy and fast face quality machines a method. There is more distinct advantage in the smooth full treatment of a few different curved surface. To the treatment requirement of workpiece surface, but through adjusting pulse electrochemistry rectifies each craft parameter in machining a process solely, make its are achieved reasonable match, in order to acquire the exterior quality of the requirement. In pulse electrochemistry light makes treatment in craft parameter, the clearance between positive pole of chroma of current density, electrolyte, shade and stream distributing, the parameter such as frequency of electrolyte velocity of flow, pulse and handling time is right the exterior quality of treatment is affected bigger. The change of each parameter can cause the change of smooth full effect, and the mutual influence between each parameter also makes treatment process more complex, these craft concern daedally each other between parameter, will build pulse electrochemistry to make treatment solely through classic transfer function and modern state function the academic maths model between craft parameter and exterior quality is very difficult, and the mathematical model between the exterior quality after often needing to have treatment again however in treatment and treatment craft parameter comes seasonable and accurate report the correlation between them. Accordingly, build the demand that can estimate well and truly according to machining exterior quality and purify to be designed accurately reasonable the experience maths model of the craft parameter that match is a difficult problem that faces at present. As a result of artificial nerve network with its height is nonlinear map, review from organization, self-study and associate the function such as memory, can be opposite complex actual nonlinear the system builds a model, pass training study, achieve taller precision. In the in studying a process proof of software, BP nerve network can realize pulse electrochemistry to rectify those who machine a model to build solely, and precision is taller. The thought that BP network algorithm and network of BP of its maths model solve a problem is: To a certain number of the output example with input study foregone and corresponding sample, the error of the actual output that uses a network and expectation target vector will amend power worth, make actual output is close to the network as far as possible expectation target vector, the sum of squared error that makes the network outputs a layer is achieved the smallest, drop to worth of advantageous position of network of the computation on direction and deviation change and approach an end gradually at error function slope relatively through be in continuously; Every time authority is worth and the change of deviation becomes direct ratio with the influence of network error, deliver each with the means of retrorse transmission. Make the error between network output and example output is achieved thereby the smallest. But at present to nonlinear optimize a problem, from theoretic make sure convergence arrives overall situation is best the algorithm of solution is not much, and it is very difficult that implementation rises, this makes a network easy in learn and training be immersed in local minimum to nod, and the network is intricate, computational amount is large, take time. Had appeared in computation of BP nerve network at present a few improvement algorithm, if heuristic study,BP network of the method and numerical value optimize BP network of the method, convergent rate raises while the algorithm that these improvement can assure precision. The article plans to choose numerical value to optimize the Quasi-Newton algorithm in the method to have consideration. Study algorithm is as follows: Quasi-Newton algorithm: Spread out error function E Taylor is progressional: (1.

In 1) type, e -- computational value and the error that are worth actually; W -- nerve yuan power worth. To type (1.

1) is begged guide, make (because be in,seek extreme value) , can get: (1.

2) (1.

3) type (1.


The algorithm that 3) comprises calls Newton method, the Hessian that H calls E blast. Because be in Newton method, the iteration that to each authority is worth should be calculated the inverse matrix of matrix of N of × of a N, want calculation of N3 amount form in all, and need consideration 2 rank derivative. In addition Hessian blast often clammy. Be aimed at such circumstance, use Quasi-Newton algorithm, with matrix G approximate H-1. In type, η -- study is led (1.

4) among them: (1.

5) (1.

6) pulse electrochemistry rectifies the BP network of treatment to build modular pulse electrochemistry to make treatment solely solely is to use pulse electricity, use nonlinear electrolyte () of solution of neuter and inorganic brine, the light of a kind of surface that lesser treatment clearance holds between tool cathode and workpiece positive pole rectifies treatment method. In because be rectified solely in pulse electrochemistry,be being machined, the academic maths model to treatment surface quality establishs each craft parameter very hard. So we keep away from through BP network interior of academic maths model concerns daedally between each parameter, seek a N dimension space directly the map to M dimension space, the function form with internal system drafts in certain level upper mould, what rectify each craft parameter in treatment solely to beg pulse electrochemistry instead is reasonable match bring huge to go to the lavatory. BP network develop attacks if knot composition of a picture pursues 2.

1 is shown, include an input layer, one outputs a layer, one among implicit layer. The structure of affirmatory BP network of layer of network input output includes an input layer, one outputs a layer, one among implicit layer. Input a layer to be comprised by each experimental parameter, output layer by experimental result namely Ra, Rz and exterior reflectivity comprise exterior surface roughness. Train example choose as follows, input quantity P chooses velocity of flow of clearance, frequency, current density, electrolyte and electrode relatively floating rate (handling time) as travel vector, bag of each travel vector is contained a certain number of the level regards a row as vector; Output layer is exterior surface roughness respectively Ra, Rz and exterior reflectivity; Among implicit the layer chooses 1, 10 nerve yuan. Among them, choose place of as above section algorithmicly to narrate, choice Quasi-Newton algorithm will calculate exterior surface roughness Ra, Rz and exterior reflectivity. Surface roughness of attainable after be calculated respectively to different output layer surface the nerve network imitate of reflectivity of Ra, Rz, surface. The experiment designs 1, experimental data: Tool material is brass or stainless steel material; Light rectifies 1Cr18Ni9Ti of treatment workpiece material. 2, experimental craft parameter: Choice quality per cent the neuter salt electrolyte of 20% ; The voltage of job of pulse power source of the choice 0 ~ 20V, current density 10 ~ 50A/cm2; Process clearance 0.

2 ~ 0.

5mm; Electrolyte velocity of flow is 1.

4 ~ 14m/s, pulse frequency 100 ~ 2200Hz, pulse width 500 μ of 100 ~ S, electrode is relatively mobile speed (handling time) 0.

0058 ~ 0.

0425 M/s. 3, measure an instrument: (1) surface surface roughness: Use the exterior contourgraph Surfcorder SE-3H that company of Kasaha Laborotacy Ltd produces; (2) reflectivity: Use laser of neon of the HJ-1B that factory of Nanjing laser instrument produces helium and support features. Norms is: Output power 1.


5W, frequency 814Hz, phase angle 40.

7 ° , input voltage 120mV, output voltage 71mV. 4, experimental principle: If pursue,pulse electrochemistry makes treatment test principle and experimental installation solely 2.

2 are shown. Workpiece receives positive pole of pulse power source, tool cathode receives the cathode of pulse power source. Polar the clearance that keeps certain. Electrolyte is mixed at workpiece by pump supply between tool cathode. A reciprocate exists in process of smooth full treatment, namely the reciprocate of tool cathode. Because the input, early days processing that outputs a data is in in BP network computation, input quantity chooses velocity of flow of clearance, frequency, current density, electrolyte and floating rate (handling time) , sendout is exterior surface roughness Ra, Rz and exterior reflectivity, the data limits of these craft parameter and experimental result is different. Then, in the network input output layer and algorithm decide hind, the network inputs a layer to output a data to need to undertake early days normalization is handled, namely interval [A, the numerical value on B] according to type (2.

1) changes interval [- 1, on 1] . (2.

The X in 1) type -- input, output number of plies seizes actual plant, [A, b] -- the limits extraction a cost that is data X correspondence, y -- the data after be normalization. Model training and test according to above the formula that gives out, undertake data normalization is handled, after the numerical value that wins experiment place uses afore-mentioned formulary processing, take 2/3 test result to regard training as example, build BP network model; The 1/3 test result that leave regards a test as example, of model of network of BP of test and verify correct, it is to be below different craft parameter situation, the BP network consideration that has checking gain arriving with built BP network model is worth and be worth comparative result actually. Can see from inside the comparative result of the watch, result of computation of BP nerve network and real value are very adjacent, can use at actual production is forecasted mediumly, provide a help to rectify what treatment parameter optimizes to choose solely. Conclusion (1) pulse electrochemistry makes treatment solely can substantially the exterior quality that improves a spare parts, use method of BP nerve network to forecast the quality of spare parts surface after pulse electrochemistry makes treatment solely, its forecast result and actual condition to be close to very much, precision is very tall, show this method is used building pulse electrochemistry to make treatment model solely is feasible, effective. (2) is forecasted through BP nerve network, can provide data of important craft parameter reference to the following smooth full treatment not only, and still can undertake an analysis to the data that has gotten, purify among them unreasonable data. (3) forecasts BP nerve network the method to rectify those who machine technical properties to forecast a method solely as pulse electrochemistry, as groovy as other method together complementary use, it is a kind of new exploration. In the meantime, still need to study its are algorithmic further in order to improve the study efficiency of BP network, be united in wedlock with the knowledge base photograph of expert system. CNC Milling CNC Machining