金属带锯床设计【9张PDF图纸+CAD制图+文档】
收藏
资源目录
压缩包内文档预览：
编号：122510134
类型：共享资源
大小：3.40MB
格式：ZIP
上传时间：2021-04-20
上传人：好资料QQ****51605
认证信息
个人认证
孙**（实名认证）
江苏
IP属地：江苏
45
积分
- 关 键 词：
-
金属
锯床
设计
PDF
图纸
CAD
制图
文档
- 资源描述：
-
喜欢这套资料就充值下载吧。。。资源目录里展示的都可预览。。。下载后都有，，请放心下载，，文件全都包含在内，，【下载后加 QQ:1064457796或1304139763,免费领CAD格式图纸】
- 内容简介：
-
journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313journal homepage: /locate/jmatprotecIntelligent adaptive control and monitoring of bandsawing using a neural-fuzzy systemIlhan Asilt urka, AliUn uvarbaFaculty of Technical Education, Selcuk University, Konya 42250, TurkeybFaculty of Mechanical Engineering, Selcuk University, Konya 42250, Turkeya r t i c l ei n f oArticle history:Received 27 June 2007Received in revised form11 March 2008Accepted 14 May 2008Keywords:Intelligent manufacturingFuzzy controlNeural-fuzzy controllerAdaptive control of band sawingBand sawinga b s t r a c tIn bandsaw machines, it is desired to feed the bandsaw blade into the workpiece with anappropriate feeding force in order to perform an efficient cutting operation. This can beaccomplishedbycontrollingthefeedrateandthrustforcebyaccuratelydetectingthecuttingresistance against the bandsaw blade during cutting operation. In this study, a neural-fuzzy-based force model for controlling band sawing process was established. Cutting parameterswerecontinuouslyupdatedbyasecondaryneuralnetwork,tocompensatetheeffectofenvi-ronmental disturbances. Required feed rate and cutting speed were adjusted by developedfuzzy logic controller. Results of cutting experiments using several steel specimens showthat the developed neural-fuzzy system performs well in real time in controlling cuttingspeed and feed rate during band sawing. A material identification system was developed byusing the measured cutting forces. Materials were identified at the beginning of the cuttingoperation and cutting force model was updated by using the detected material type. Con-sequently, cutting speed and feed rate were adjusted by using the updated model. The newmethodology is found to be easily integrable to existing production systems. 2008 Elsevier B.V. All rights reserved.1.IntroductionIn band sawing, the power rating of the machine limits thethickness and hardness of the metal to be cut. In band sawingprocess, metal removal is accomplished by forcing a multi-toothed tool against the workpiece. The depth of cut in sawingcannotbepresetlikeothermetalcuttingprocessesandcontrolcan only be exercised over the thrust load applied between theblade and workpiece material. The amount of metal removedby each tooth is dependent primarily on how well the bladetransmits the applied pressure to the workpiece and alsoon the penetration ability of the cutting teeth. Machiningforces generated during sawing process are therefore foundto have greater significance than in other chip removal pro-cesses. It has been found that thrust and cutting loads pertooth per unit thickness reduced with an increase in cuttingCorresponding author. Tel.: +90 332 223 3344; fax: +90 332 241 2179.E-mail address: iasilturk.tr (I. Asilt urk).speed. A reduction in the thrust force will cause a reductionin the depth of cut taken by the engaged teeth. An increasein the feed rate causes a substantial increase in both cut-ting and thrust loads per tooth. Geometry of the workpiecedoes also have a considerable influence on cutting perfor-mance. In band sawing, the thrust load is normally constantalongtheworkpiecebreadth.Whensawingroundsectionsthewidth of the workpiece changes within the cut, the cutwidthincreasesastheblademovestowardsthecentreanddecreasesas the cut is being finished. Band saw machines that operateon a pressure feed principle maintain a constant chip loadper tooth as described while the blade saws through varyingsections.Artificial intelligence (AI) methods are widely used in solu-tion of complex engineering problems. Some of the mostcommonlyusedAItechniquesareneuralnetworks(NN),fuzzy0924-0136/$ see front matter 2008 Elsevier B.V. All rights reserved.doi:10.1016/j.jmatprotec.2008.05.031journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 230223132303NomenclatureDdiametereoutput errorffeed ratefiinstantaneous feed rateFoutprocess outputFmeasmeasured forceFrefreference forceGmaterial group noHhardnessIintegral of errorIAM 1intelligent adaptive moduleLinstantaneous heightMRRmaterial removal ratePIperformance indexTWRtool wear rateyprocplant outputyrefreference model outputviinstantaneous cutting speedlogics (FL), expert systems (ES) and models using hybrids ofthese.Artificial intelligence methods are used in every stage ofmanufacturing. Machining is one of the basic manufacturingmethods used in the industry. Manufacturers must minimizecostandprocesstime,andadditionallytheproductmustcom-ply with the required dimensions and quality criteria for abetter competition.Increasing the productivity of metal cutting machine toolsis a principal concern for manufacturing industry. In tra-ditional machining systems, cutting parameters are usuallyselected prior to machining according to machining hand-books or the users experience. The selected machiningparameters are usually conservative to avoid machining fail-ure. To ensure the quality of machining products, to reducethe machining costs and increase the machining efficiency,it is necessary to adjust the machining parameters in real-timeandtooptimizemachiningprocessatthattime.Adaptivecontrol of the machining process is preferable to solve aboveproblems.Since band sawing process is non-linear and time-varying,it is difficult for traditional identification methods to providean accurate model. Adaptive control methods provide on-lineadjustment of the operating conditions. Therefore, parame-ter adaptive control techniques for machining processes weredeveloped to adjust the feed rate automatically to maintain aconstant cutting force. Applications of these techniques suc-cessfully increased both the metal removal rate and tool life.In this paper, an intelligent neural-fuzzy adaptive controlscheme is proposed for band sawing process. The proposedadaptive control system can be applied effectively in variouscutting situations.2.Literature surveyTherearealotofworksexistingintheliteratureonmonitoringand controlling of the machining operation.Groover pointed out that conventional control theory couldbe inefficient and unstable due to disturbing variations in themachining conditions. It is stated that fixed cutting forceswouldbeausefulapproachforincreasingtoollifeandmaterialremoval rate (Groover, 1987).The conventional PID feedback control system has beenusedincontrollingmachiningprocessesbynumerousresearchers (Masory and Koren, 1980, 1985; Lauderbaugh andUlsoy, 1989; Koren, 1988). The main problem with the fixedgain Adaptive Control Constraint (ACC) system is the one thatproduce poor performance and may become unstable duringthe time-varying machining process. The use of various formsof adaptive control in an ACC system has been examined byadjusting the gain of the controller.Model reference adaptive control-based ACC systems(MRAC)havebeendevelopedbysomeresearchers(MasoryandKoren, 1980, 1985; Lauderbaugh and Ulsoy, 1989). These stud-ies found that MRAC perform control duties better than fixedgain controllers. A typical MRAC incorporates the parameterestimation of the cutting process.Recently, many studies have been devoted to the theoryof fuzzy control and its application to machining processes.Tarng et al. developed a fuzzy logic-based controller (FLC) foradaptive control of turning operations. The developed FLC canadjust feed rate on-line so as to reduce machining time andmaintain constant force (Tarng and Cheng, 1993; Tarng andWang, 1993).IntheexperimentalstudiesofZhangandKhanchustambham (1993), it is shown that process opti-mization is possible by online monitoring and controllingof the machining process. This eliminates the effect ofdisturbances caused by operator.An online monitoring system was designed by Ordonezet al. (1997) by using artificial intelligence based on sensors.Signals which were taken from sensors are used in AI deci-sion making during the cutting process. The real time signalsobtained through force transducers and estimated cuttingforces obtained by using NN were compared. Consequently,estimated model was implemented to surface roughness,tool wear and geometric tolerances. Feed forward and backpropagation algorithms were used as architecture and train-ing algorithm of NN model, respectively. Direct and indirectadaptive fuzzy techniques and simulations of conventionalcontrols were compared.An adaptive control approach was suggested by Rodolfo etal. (1998) for maintaining the cutting force constant, in themilling process. The constant force feed rate was investigatedwithout delay time.Tsai et al. (1999) observed that, surface roughness canexperimentally be determined by one or more quantita-tive measurements. Estimated surface roughness model wasbased on relative vibrations between the tool and the workpiece. Estimated surface roughness was improved by usingsignals that are taken from vibration and proximity sensors.System accuracy was observed as 9699%.An adaptive controller with optimization was designedbased on two kinds of NN by Liu and Wang (1999) formilling process. A modified back propagation NN was pro-posed adjusting its learning rate and adding dynamic factorin the learning process, and was used for the online modeling2304journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313of the milling system. A modified Augmented Lagrange Multi-plier (ALM) neural network model was proposed adjusting itsiteration step, and was used for the real time optimal controlof the milling process. In this study, the simulation and exper-imental results show that not only does the milling systemwith the designed controller have high robustness and globalstability, but also the machining efficiency of the milling sys-tem with the adaptive controller is much higher than for thetraditional CNC milling system.Adaptive control constraint is one of the methods usedin controlling machining processes. Force control algorithmshave been developed and evaluated by numerous researchers.Among the most common is the fixed gain proportional inte-gral (PI) controller, originally proposed for milling by Kim etal. (1999). The gain of the controller is adjusted in response tovariations in cutting conditions in the proposed controller.The essential aim of the neural network-based controller istoconstructareversefunctionforthemachiningsystemusingtheNNsothattheoutputofthemachiningsystemapproachesto the desired output. Machining process can usually be con-trolled by adjusting the feed rate or spindle speed. The neuralnetwork-based ACC system has been applied to machiningprocess control by (Liu et al., 1999; Hang and Chiou, 1996).In a study by Liu et al. (2001), the major adaptive controlconstraintsystemswerediscussedbasedonthefeedbackcon-trol, parameter adaptive control/self-tuning control, modelreference adaptive control, variable structure control/slidingmodecontrol,neuralnetworkcontrol,andfuzzycontrol.Theirtypical applications to constant cutting force control systemare also described, and some recent experiments results werepresented.Online method of achieving optimal settings of a fuzzy-neural network has been developed by Sandak and Tanaka.Results of the cutting experiments using several wood speciesshow that the fuzzy-neural system developed performs wellin online feed rate optimization during band sawing, whilemaintainingsawdeviationwithinspecifiedlimits(SandakandTanaka, 2003).Zuperl et al. (2005) discussed the application of fuzzy adap-tive control strategy to the problem of cutting force control inhighendmillingoperations.IntheirACsystem,thefeedrateisadjusted on-line in order to maintain a constant cutting forcein spite of variations in cutting conditions. They developed asimplefuzzycontrolstrategyintheintelligentsystemandcar-riedoutsomeexperimentalsimulationswiththefuzzycontrolstrategy.The effect of cutting speed, feed rate and work piece geom-etry in band sawing were investigated by Ahmad et al. (1987).In the experimental studies, reduction in the thrust force andcutting force per teeth for unit thickness were observed, as thecutting speed increased.Ko and Kim observed that, in order to create a mechanis-tic model of cutting force, specific cutting pressure should beobtained through cutting experiments. The band sawing pro-cess is similar to milling in that it involves multi-point cutting,so it is not an easy matter to evaluate specific cutting pres-sure. The cutting force is predicted by analyzing the geometricshape of a saw tooth. They stated that the predicted cuttingforce coincided well with those measured in validation exper-iment. Therefore, the predicted cutting forces in band sawingcan be used for the adaptive control of saw-engaging feed ratein band sawing (Ko and Kim, 1999).Anderson et al. (2001) pointed out a mechanical cuttingforce model for band sawing. The model describes the vari-ation in cutting force between individual teeth and relates itto initial positional errors, tool dynamics and edge wear. Bandsawing is a multi-tooth cutting process, and the terminologyof the cutting action is discussed and compared with othercutting processes.3.Adaptive control of machining processesIntelligent machining system applications include monitor-ing and control technologies. This system also improves themachining operations. In this system, process related datais acquired, and then process is controlled. Manufacturingindustries are affected by computer technologies. Recently,automation works were made on the material handling,quality monitoring, motion control, source planning, pro-cess control, etc. The sensing of the machining process ismuch more comprehensive and complex. Many papers havebeen prepared about monitoring and control of the machinetool.Researchers and industrialists concerned with tool mon-itoring and adaptive control. One of the most importantfunctions of the intelligent control is the provision of requiredaction in the unknown or indefinite ambient processes.Machine tool and cutting tools are protected by the moni-toring system. Tool changing cost, scrap rate and productioncost is reduced by real time tool wear measurement. Thus, fullcapacities of the machine tools are maintained.In the machining processes, feed rate is continuouslyadjusted for keeping on the process with constant referenceforce in the adaptive control systems. Thanks to adaptivecontrol, since it aims to minimize the production time toadjust the feed rate to optimal values in the high capacityworking conditions. Consequently, tool life increases with therestricted load application.3.1.Necessities of the adaptive controls in themachining operationIn the case of depth or width of cut, feed rate are usuallyadjusted to compensate for the variability. This type of vari-ability is often encountered in profile milling or contouringoperations (Groover, 1987).When hard spots or the areas of difficulty to machine areencountered in the workpiece, either speed or feed is reducedto avoid premature failure of the tool.If the work piece deflects as a result of insufficient rigidityinthesetup,thefeedratemustbereducedinordertomaintainaccuracy in the process.As the tool begins to dull, the cutting forces increase. Theadaptive controller will typically respond to tool dulling byreducing the feed rate.The workpiece geometry may contain shaped sectionswhere no machining needs to be performed. If the tool were tocontinue feeding through these so-called air gaps at the samerate, time would be lost. Accordingly, the typical procedure isjournal of materials processing technology 2 0 9 ( 2 0 0 9 ) 230223132305Fig. 1 Adaptive control optimization in milling system.to increase the feed rate, by a factor of two or three, when airgaps are encountered.These sources of variability present themselves as time-varying and, for the most part, unpredictable changes in themachining process. It should be examined how adaptive con-trol can be used to compensate for these changes.3.2.Adaptive control methods in the machiningprocessThe methods that are mentioned below are used in machiningoperations. These are namely Adaptive Control with Opti-mization, Geometric Adaptive Control and Adaptive ControlConstraint.3.2.1.Adaptive control optimization (ACO)In this type of adaptive control, a performance index is spec-ified for the system. This performance index is a measureof overall process performance, such as production rate orcost per volume of metal removed. The objective is to opti-mize the index of performance by controlling speeds and/orfeeds.A system with the adaptive controller for machining pro-cess can be constructed based on NNs as shown in Fig. 1. Thesystem is modeled on-line by the modified BP learning algo-rithm. The feed rate is adjusted and the process is optimizedin real-time by the modified ALM NN. In the process the differ-ence between measured cutting force and estimated cuttingforce is (e), which is used as back propagation NN and is toadjust the weights of the NN. Feed rate is adjusted in the senseof object function constraints (Liu et al., 1999).3.2.2.Geometric adaptive control (GAC)Geometricadaptivecontrolisusuallyusedinfinishmachiningoperations, where the objective is to achieve a desired surfacequality and/or accurate part dimensions despite tool wear ortool deflection (Liang et al., 2004). Owing to the relationshipbetween feed rate and surface quality, surface roughness ordimensional accuracy of the part is continuously measuredby the sensor by means of feed-back.Therefore in most GAC systems, the cutting speed is con-stant and the machining feed is manipulated to achieve thedesired surface quality (Masory and Koren, 1980).The dimensional precision in turning is usually achievedby measuring the part diameter at various points after themachining. Ultrasonic sensors were used in turning operationfortheestimationandthecontrollingofthesurfaceroughness(Coker and Shin, 1996). Offset distance is manually adjustedto compensate for inaccuracy.3.2.3.Adaptive control constraint (ACC)The objective in this method is to manipulate speeds and/orfeeds to maintain the measured variables below their con-straint limit values. A typical configuration of the adaptivecontrol is illustrated in Fig. 2 for machining process. Adaptivecontrol constraint is one of the effective methods of solvingthe above problems. ACC controls the machining parame-ters to maintain the maximum working conditions during theFig. 2 The integral ACC system of the turning process.2306journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313Fig. 3 Self-tuning control-based ACC system.time-varying machining process. Cutting force, power, surfacequality, etc. are constant parameters.Where Foutis the process output, Fmeasis the measuredforce, Frefis the reference force, e is the output error, f is feedrate.The performance index is generally economic function,andalsomaximizessubjecttoprocessandsystemconstraints.In this form of adaptive control, constraint limits are imposedon the measured process variables. Performance index ismaterial removal rate (MRR) to tool wear rate (TWR) as in Eq.(1). This ratio must be maximized.Performance index:PI =MRRTWR(1)where PI=performance index; MRR=material removal rate;TWR=tool wear rate.The determination of the performance index as real timeis rather difficult with nowadays technologies. Because of thisreason TWR is not a measurable value as real time. ConstraintAdaptive Control was classified by Liu et al. (2001) as below,- A feedback controller-based ACC system,- Self-tuning control-based ACC system,- A model reference adaptive control-based ACC system,- A variable structure system-based ACC system,- A neural network-based ACC system,- A fuzzy control-based ACC system.Frequently used methods mentioned above are explainedbelow. Self-tuning control-based ACC system.The early ver-sions of parameter adaptive control-based ACC systems weredeveloped using a simple on-line estimator for the processgain and an integral strategy to adjust the gain of an integralcontroller.Thedefectsofthisstrategyarethatthedynamicsofthecuttingprocesswereneglectedandnotheoreticaladaptivedesign technique was used. The adaptive controller consistsof two functions:1. On-line estimation of the parameters of the cutting pro-cess.2. Real-time control.The servo loop is modeled by a third-order system as fol-lows:The regulator is designed for setting of required per-formance parameters. The block diagram of self-tuningcontrol-based ACC system is shown in Fig. 3.This design is based on the output response of the system.Maindifficultyofthisdesignisstabilityproblemofthesystem.This problem is solved by Root-Locus Plot. The gain of controlalgorithm and the adjustment of time constants are investi-gated by this method. Other processes are generally based onZiegler and Nichols methods. Self-tuning control-based ACCis generally applied to the time delay, non-linear and multi-control cycle systems. The stability of this type of systems isnot clearly defined. Self-tuning control shows diversity in theregulation of the control organs parameters. A model reference adaptive control-based ACC sys-tems.Required system behavior is determined by referencemodel. The performance of this system can be expressedas adaptive servo system. It has a controller and conven-tional feed-back system. Controller parameters are changedby “e” error-signals. The general idea behind Model Refer-ence Adaptive Control is to create a closed loop controllerwith parameters that can be updated to change the responseof the system. The output of the system is compared to adesired response obtained from a reference model. The con-trol parameters are updated based on this error. The goal forthe parameters is to converge to ideal values that cause theplant response to match the response of the reference model.Using MRAC, you could choose a reference model that couldrespond quickly to a step input with a short settling time.Again, the idea behind Model Reference Adaptive Control isto create a closed loop controller with parameters that canbe updated to change the response of the system to match adesired model. There are many different methods for design-ing such a controller. The other type of design is using theMIT rule in continuous time. When designing an MRAC usingthe MIT rule, the designer chooses: the reference model, thecontroller structure and the tuning gains for the adjustmentmechanism.MRAC begins by defining the tracking error, e. This is sim-ply the difference between the plant output and the referencemodel output:e = yproc yref(2)journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 230223132307Fig. 4 The block diagram of the MRAC.The block diagram of the MRAC is shown in Fig. 4.The realization of the MRAC system is easily made in high-speed adaptation. Since dynamic performance determinationis unnecessary. The aim of this system, output of the systemis approached as asymptotic to the reference model.4.Intelligent adaptive control system of theband sawing (IACS)In band sawing, in order to provide stable cutting and to selectthe appropriate machining parameters according to materialcharacteristics, it is necessary to optimize and control themachiningprocessonline.Itisknownthatperformanceoftra-ditional adaptive controller depends greatly on the accuracyof the model of the machining process. Since the machiningprocess is a non-linear and time-varying process with randomdisturbances, it is difficult for the traditional adaptive con-troller to realize effective control in the machining process.Adaptive control constraint is one of the effective methodsof solving the above problems. ACC controls the machiningparameters so as to maintain the maximum working condi-tionsduringthetime-varyingmachiningprocess.Inthiswork,to avoid aforementioned difficulties, a neural-fuzzy-basedadaptive controller with constraints for band sawing wasestablished and model outputs were compared with cuttingforces measured in real time. Cutting parameters were contin-uouslyupdatedbyasecondaryneuralnetwork,tocompensatefor the effect of environmental disturbances. Required feedrate and cutting speed were continuously adjusted by fuzzylogic controller. A new control scheme which is called adap-tive neural-fuzzy control is developed by using NNs with backpropagation algorithm and fuzzy logic controller. The objec-tive of this control is keeping the metal removal rate as highaspossibleandmaintainingcuttingandtrustforcesascloseaspossible to a given reference value. Furthermore, the amountof computational task and time can be reduced as comparedto classical or modern control methods.The training and testing of NNs and also the operationof machining system require a set of experimental data. Forthis reason, an experimental scheme for band sawing wascreated.In order to establish 162 different cutting states in bandsawing, specimens with 18 different types in three differentkinds of material groups were prepared.Specimens were machined to the required dimensionsby turning and milling operations. And then normalizationwas made at 880C for the homogeneity of samples. Thehardness of the samples was measured as HB. Data acquisi-tion was accomplished by using developed software called as“ilhan daq v01”. A 16-channel real-time data acquisition wasaccomplished by using this software.The constants and cutting parameters were entered tothe interface. Outputs were measured as 80 samples/s andtheir average values were recorded as one datum. Circularand square cross-sectional samples which have three differ-ent hardness and three different diameters, were machined atthree different cutting speeds, and three different feed rates.Consequently, tests were performed 162 experimental runs(Asilt urk, 2007). The block diagram of the experiment set isshown in Fig. 5.Filtered data have been combined and data files have beencomposed for NN. Training and test processes were made withthis data and suitable NN model were determined. A super-vised multi-layer back propagation network algorithm wasused as NN.In this adaptive control system, recorded data are evalu-ated by NN and FL and then these data are transformed intoadaptive control decision. Consequently, the operation of theband sawing machine is improved to operate in required cut-ting conditions.The model is composed of three intelligent models whichare IAM 1, IAM 2, and IAM 3. Where IAM 1 is used forestimated cutting force in process variations. IAM 2 is FLalgorithms, which produce control outputs by the differencebetween estimated cutting force and measured cutting force.2308journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313Fig. 5 Block diagram of the experimental setup.IAM 3 is a real time NN algorithm which is used to upgradethe cutting force model.Intelligent adaptive control system is composed of threeintelligentmoduleswhichareIAM 1,IAM 2,andIAM 3.WhereIAM 1 is used for the estimated cutting forces in process vari-ations. IAM 2 is used for FL algorithms, which produce controloutputs with respect to the difference between the estimatedcutting force and the measured cutting force. IAM 3 is for areal time NN algorithm which is used to upgrade the cuttingforces.4.1.Artificial intelligence module (AIM 1)Cutting forces (Fxand Fz) is accepted as measurable outputparameters with respect to input parameters of our systemconsisting of cutting speed (v), feed rate (f), diameter (D), hard-ness (H), material group no (G), instantaneous feed rate (fi).Data consisting of 4482 lines and 8 columns extracted fromthe experiments is source of input data to AIM 1. Data isnormalized between the ranges 01 and network is chosenas multi-layered feed forward network and transfer functionis either tangent sigmoid or logarithmic sigmoid. Moreovergradient descent optimization method is chosen as trainingmethod function and weights and bias values are updatedaccordingly.For AIM 1, a matrix with a dimension of (30006) is usedas input and a matrix with a dimension of (30002) is used asoutput. To test the system, a data matrix with a dimension of(14826) is used as input and a data matrix with a dimensionof (14822) is used as output matrix. Both two matrices arethe remaining data which is not used for training. Modeling,training and test of the AIM 1 are realized by using MATLABTMNeural Network Toolbox.To determine the appropriate network structure, perfor-mance of the network consisting of constant number of inputand output layer neuron is measured with respect to changingnumber of hidden layer neurons which are 2, 3, 4, 5, 6, 10, 15,20, 25, 30, 35, 40, 50. The network with 6302 neurons perlayer which has Mean Squared Error (MSE) of 0.0020407 at the6000th epoch is chosen. The graphic of the data which is partof the test process by using aforementioned network struc-ture are given below. The comparison of measured forces andestimated forces are given in Table 1.4.2.Artificial intelligence module 2 (AIM 2)AIM 2 is a multi-input multi-output fuzzy logic algorithmwhichfunctionsasacontrollerwhichproducesoutputparam-eters for the process network. AIM 2 modeling is realized byusing a MATLAB Fuzzy Logic Toolbox.The module is a controller structure by using fuzzy logic.For this reason, it tries to minimize the difference between thereal time measured values and estimated values produced byjournal of materials processing technology 2 0 9 ( 2 0 0 9 ) 230223132309Table 1 Experimental conditionsExplanationsUnitValueCutting speedm/min40; 63; 100Feed ratemm/min35; 66; 125Cooling liquid (Wateroil) emulsion)Boron oil 20%Work piece material SAE 1015mm30; 46; 70Work piece material SAE 1040mm30; 46; 70Work piece material SAE 4140mm30; 46; 70Saw blade dimensionsmm36700.927Saw blade Eberle M42TPI6Tensile stress of the saw bladeBar85Band saw machineIMAS 280AIM 1. Expert opinions and experiences are used to build themodel and to determine the model parameters. Error valuesand their integrations are used as inputs, and feed rate andcutting speed driving the system are used as output.Since the model used here is non-linear, Mamdani fuzzyprocessing type is preferred. The number of input variables is4, the number of output variables is 2, the degree of fuzzinessis 3 and the number of the rules is 36 for the fuzzy controller.The center of gravity are also used as inference method anddefuzzification methods, respectively.4.3.Artificial intelligence module 3 (AIM 3)AIM 3 module drives the system to the optimum state againstthe instantaneous changes in the process or outside distort-ing variables. The measured inputs of the system are cuttingspeed, feed rate, instantaneous feed rate, speed, diameter,and coordinate/diameter values. Material hardness is the onlyoutput. The experiments to gather data for training and testpurposes result with 5487 lines by 7 columns data matrixwhich is normalized to the values ranging from 0 to 1. For thetraining phase, a data matrix having a dimension of (38006)is used as input and a data matrix having a dimension of(38001) is used as output. To test the system, a data matrixwith a dimension of (16866) is used as input and a datamatrix with a dimension of (16862) is used as output matrix.Both of the matrices are the remaining data which is notused for training. Modeling, training and test of the AIM 1are realized by using MATLABTMNeural Network Toolbox. Todetermine the appropriate number of layers and neurons perlayers, the following network structures are experimented.Performance of the network consists of constant number ofinput and output layer neuron is measured with respect tochanging number of two hidden layer neurons which are 3, 6,10, 15, 20, 25, 30, 40. The network structure with 630201neurons per layer which has MSE of 0.0133115 at the 5000thepoch is chosen.5.Experimental work5.1.Experimental conditionsMinimum and maximum cutting speed and feed rate val-ues used at experimental works were determined by usingthe mean values obtained from the band manufacturershandbooks, and median values were determined by usingfull factorial experiment calculations. A semi-automatic typebandsaw machine which has pulling cut (IMAS 280) was used.To control the machine by a computer, hydraulic, electrical,electronics and mechanic revisions are made on the machine.An experiment set which is able to measure the cutting forces,current, vibration, AE, speed and feed rate from the processwas set up. Experimental conditions are given in Table 2.5.2.Experiment set elementsAnalog data was collected for the different cutting param-eters and materials by using sensors. Force measurementswere realized by using Kistler 9257B dynamometer and Kistler5019B Charge Amplifier. Acceleration measurements weremade by using Kistler 8792A500. Acoustic emission measure-ments were made by using Kistler 8152B111, the valve whichadjust the feed rate is proportional valve Rexroth 2FRE 6 B-2X/K4RV. Cutting speed is adjusted by using Telemecaniqueinventor, and National Instrument 6221 is used as DAQ card.By using the data collection card, analog data is convertedinto digital values for different cutting parameters and mate-rials and stored in the computer. When the data collectionexperiment results are investigated; it can be seen that cuttingforces decreases as cutting speed decreases, and increases asthe feed rate increases. These results are considered carefullywhen the rule table for control structure is generated. Thisdatabase is used both for training and test process.6.Experimental results and discussionWhen the IACS is applied, operator just clamps the materialand enters the diameter value. Material hardness, and mate-rial group info are classified and proper reference model isdetermined as soon as cutting is started. Process adjusts thecutting speed and feed rate based on the principle of con-stant cutting force. When the situations like chipping andwork piece local hardness arise, system updates itself, refer-ence forces changes instantaneously, and process is regulatedaccording to the parameters accommodating the new condi-tions (Figs. 6 and 7).Determined network parameters and results of the con-structed network structure are compared. ANN model of6302 which has the least MSE chosen as the mostappropriatemodel.NNestimatedthereferenceforceswith97.13%and97.05%ratiosfortrainingandtesting, respectively.First ANN system continuously produces new referenceforce by using the feedback from the second ANN regardingto hardness change and then sends it to the fuzzy controller.New cutting parameters produced by fuzzy controller accord-ing to the new situation depending on the reference cuttingforces are produced.To test the system, the work piece having a roundcross-section was used. After the band saw contacts withthe work piece, the cutting forces increase as the cuttinglength increases. As the cutting forces increases, feed rate isdecreasedbyANNFuzzySystem.Duringtheprocess,withfeedrate and cutting speed changing, second ANN (AIM 3) pro-duces output and the hardness changes and this parameter2310journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313Table 2 The comparison of measured cutting forces and cutting forces estimated by NNResults of NN trainResults of NN testNN predict outputsMeasured outputsNN predict outputsMeasured outputsFxFzFxFzFxFzFxFz442.2057763.703445.1773733.1902442.2057763.703445.1773733.1902134.4875195.475133.4109206.7233134.4875195.475133.4109206.7233152.6661284.764150.5178263.7395152.6661284.764150.5178263.7395138.7918211.7589128.3311171.0689138.7918211.7589128.3311171.0689134.818196.2869140.0077184.7306134.818196.2869140.0077184.730659.9776586.8937762.572785.9499459.9776586.8937762.572785.94994351.7195656.4338349.2264652.6025351.7195656.4338349.2264652.6025145.1744258.2446148.8639248.5279145.1744258.2446148.8639248.5279355.8429705.9138349.5964772.7656355.8429705.9138349.5964772.7656307.5868559.0349320.4081508.9903307.5868559.0349320.4081508.9903188.2508298.9447179.2608329.5293188.2508298.9447179.2608329.5293117.264196.0613120.6795193.0908117.264196.0613120.6795193.0908212.4309397.4335210.2907387.5213212.4309397.4335210.2907387.5213412.7616825.1123407.7722835.1263412.7616825.1123407.7722835.126389.07383148.22487.10368169.337489.07383148.22487.10368169.3374189.9915288.5916187.649258.3875189.9915288.5916187.649258.3875204.8489337.9783197.0533349.8911204.8489337.9783197.0533349.8911227.509406.8032229.5201350.7272227.509406.8032229.5201350.7272152.7418284.685156.4969265.2147152.7418284.685156.4969265.214791.92212133.641383.54261135.7691.92212133.641383.54261135.76170.8712308.3833168.707276.0291170.8712308.3833168.707276.0291352.3812658.866346.9379773.7592352.3812658.866346.9379773.7592390.5508787.7399387.5276860.4332390.5508787.7399387.5276860.4332310.7366601.6176310.1643584.0798310.7366601.6176310.1643584.0798198.3021393.7602165.3134384.4596198.3021393.7602165.3134384.4596432.447809.2537445.68761.7104432.447809.2537445.68761.7104251.7293535.0718249.4053493.9936251.7293535.0718249.4053493.9936134.4486199.1893126.4773173.8702134.4486199.1893126.4773173.8702196.7886388.9811198.5428383.57196.7886388.9811198.5428383.57310.1094590.5273312.3921559.5582310.1094590.5273312.3921559.5582352.6021659.6783346.5218710.4463352.6021659.6783346.5218710.4463124.7921206.7643126.0709206.9399124.7921206.7643126.0709206.9399389.3444782.8961361.9368669.3829389.3444782.8961361.9368669.3829117.3195196.0454119.2968191.3657117.3195196.0454119.2968191.3657128.5573182.7823124.0504174.7956128.5573182.7823124.0504174.7956212.0144397.5657230.2204359.4101212.0144397.5657230.2204359.4101102.1894181.3769102.0702175.8783102.1894181.3769102.0702175.8783380.3919746.0669381.2197621.1733380.3919746.0669381.2197621.1733100.4572151.381599.10651146.5054100.4572151.381599.10651146.5054313.5743595.185318.602587.9306313.5743595.185318.602587.9306353.0786661.4317349.5462765.8742353.0786661.4317349.5462765.8742102.9637183.3941102.1868167.9919102.9637183.3941102.1868167.9919126.0354209.9449120.7443205.7517126.0354209.9449120.7443205.7517127.9189214.7463125.1926202.6325127.9189214.7463125.1926202.6325319.3382684.4524316.1491688.0594319.3382684.4524316.1491688.0594112.1071159.0852103.7548172.2993112.1071159.0852103.7548172.2993is feed back to the ANN (AIM 1), as a result of this processit is produced a new reference force value by the first ANN(AIM 1). Consequently, the system adjusts (adapts) itself tothe new situation and cuts the material by using new param-eters. Graphics of change in cutting forces vs. cutting time isgiveninFigs.811,bothforANNfuzzyadaptivecontrolsystemand uncontrolled system. At the start of the cutting, cuttingparameters are entered according to the known material typeand work piece diameter. Cutting operation starts by usingentered parameters, as the cutting operation progresses, thecontroller determines the values of cutting parameters. Thecontroller tries to maintain the cutting forces steady at deter-mined reference values by changing the feed rate and cuttingspeed. Hardness information is extracted from cutting forcefeed back and then reference force model is updated. Cuttingtime is recorded as 62.2s with IACS. It would take 141s forcutting the same material if was cut by using manufacturersrecommendations.Results of measurements on a workpiece of material SAE4140anddiameter70mmforadaptiveIACScuttingconditionsare given in Figs. 810.As seen in Figs. 8 and 9, at the starting of the cuttingprocess, to maintain the chosen reference cutting force, feedrate is decreased and cutting speed is increased. The systemjournal of materials processing technology 2 0 9 ( 2 0 0 9 ) 230223132311Fig. 6 Block diagram of IACS.Fig. 7 Graphic to show the comparison measured andestimated NN output cutting force.Fig. 8 Cutting forces during the cutting of the material(SAE 4140 ?70mm) by IACS.acquires real time material hardness value based on cuttingforces and cutting force model is updated by using this infor-mation. Any local force increase during the cutting sessionupdates the reference force model. The system reacts byFig. 9 Adaptive change of feed rate and cutting speedwhile the material (SAE 4140 ?70mm) cutting by IACS.Fig. 10 Recognition of the material hardness of thematerial (SAE 4140 ?70mm) by IACS during the cuttingprocess.Fig. 11 Cutting forces during the cutting of the material(SAE 4140 ?70mm) by conventional band saw machine.increasing cutting speed and decreasing the feed rate in thecase of increasing cutting forces.Amaterialhardnessrecognitionsystemisdevelopedbasedon cutting forces by means of AIM 3 which is in the structureof IACS. Hardness of material SAE 4140 ?70mm workpiece isobtained in the real time as shown in Fig. 10.In Fig. 11, the change of cutting forces during band sawingof SAE 4140 ?70mm workpiece by using a conventional bandsaw machine is shown. As seen in the graphic, at the time ofstart the cutting forces are low according to the number of theteeth which are in touch with the material. As the band sawprogresses towards the center, the cutting forces increase as2312journal of materials processing technology 2 0 9 ( 2 0 0 9 ) 23022313Fig. 12 Constant feed rate in cutting SAE 4140 ?70mmspecimen in conventional band saw machine.the cutting length increases. In cutting circular workpieces byusing a band saw with constant feed rate, the forces increaseas the cross-sectional length increases.In Fig. 12, it can be seen that feed rate is constant in bandsaw cutting of SAE 4140 ?70mm workpiece by using a con-ventional band saw machine.Proposed system estimates the material hardness con-sistently with real cutting forces and produces cuttingparameters accordingly. Proposed system estimates the realcutting forces consistently with the material hardness andproduces cutting parameters accordingly. IACS provides quickresponse to drive the system to appropriate state and resultswith regular updates in cutting parameters and referencemodel with respect to conventional systems.The experimental results show that the hybrid ANN-fuzzyadaptive controller decreases the tool costs by making it pos-sible to run the band saw machine under appropriate feed rateand cutting speed and provides less cutting time compared tothe conventional controllers.7.ResultsThe present study shows that hybrid artificial intelligencemethods can be used successfully in adaptive control ofmachine tools. Metal removal process which is very hard tomodel mathematically is solved easily by using artificial intel-ligence methods.Modeling and control of the band saw cutting processcannot be achieved effectively by using conventional meth-ods, since there are many parameters affecting the process.Consequently, the system is modeled by using an artificialintelligence model. Adaptive control is used in controllingthe process by regulating decreasingly the affects of environ-mental disturbances and adjusting itself accordingly. Effectof operator intervention is set to minimum by the help ofdeveloped monitoring and control system. The
- 温馨提示:
1: 本站所有资源如无特殊说明，都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
2: 本站的文档不包含任何第三方提供的附件图纸等，如果需要附件，请联系上传者。文件的所有权益归上传用户所有。
3.本站RAR压缩包中若带图纸，网页内容里面会有图纸预览，若没有图纸预览就没有图纸。
4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
5. 人人文库网仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对用户上传分享的文档内容本身不做任何修改或编辑，并不能对任何下载内容负责。
6. 下载文件中如有侵权或不适当内容，请与我们联系，我们立即纠正。
7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
人人文库网所有资源均是用户自行上传分享，仅供网友学习交流，未经上传用户书面授权，请勿作他用。
川公网安备: 51019002004831号