Research On Many Measurement Points Flatness Error Intelligent Evaluation And Uncertainty Analytical Method

There are many measurement points for large scale and high precision workpieces,therefore, the traditional form error evaluation method in this case does not converge orcalculates slowly, and it is difficult to meet the needs of the development of product testing,so that the research of high accuracy, high reliability, high efficiency of modern evaluationmethod has important practical significance. The reaseach work received financial supportfrom Ministry of Education Support Program for New Century ExcellentTalent(NCET-08-2011), Guangdong Province Institutes of Higher Education High-qualifiedTalent Project(Yue Jiao Shi Han[2010]79).Beginning with discussion of flatness definition, difficulties for many measurementpoints evaluation, evaluation performance indexes, this research progress at home and abroadof the flatness evaluation based on numerical algorithm, computational geometry, andintelligent algorithm were reviewed generally to determine the research content of this paper.The main research work in this paper is as follows:Many measurement points intelligent evaluation optimization algorithm and thesimulation method for measurement points were researched. It was analyzed thatcharacteristic measurement points are distributed nearby the minimum zone planes, acharacteristic measurement points extraction method Least Square Residual-LSR wasproposed to reduce computational complexity in the evaluation. Two extreme cases ofminimum zone were calculated according to coordinate extremums of measurement points inx,y,z–axis, then minimum zone normal vector range was determined, which intelligentalgorithm individuals initialization was depending on. Fitness function and terminationcondition were researched. Based on the above, several bionic intelligent flatness evaluationmethods based on genetic algorithm, particle swarm optimization, and artificial bee colonywere proposed. According to analyze the affecting factors in the planar process, the planeprocessing simulation model was constructed, which consisted of a polynomial function,trigonometric function, random error function and satisfied minimum zone criterion. Themodel was used to verify intelligent evaluation methods, and the results showed that whenextracting threshold factor kP=0.6, initial individuals amount sINIT=20, evaluation timeTC≤0.0508s, evaluation error δt≤0.8745×10~-5mm, which proved that the intelligent evaluationmethods were effective for many points evaluation.The assessment method for intelligent evaluation uncertainty based on β distributionuniform expression method was researched. Taking particle swarm optimization (PSO) evaluation method as an example, through analyzing enveloping zone distribution feature ofthe new particle, it was obtained that the probability density function of intelligent evaluationresults f (t_e)is bounded and right skewed. It was pointed that β distribution, if h> g>1,coincides with f (t_e)'s distribution feature, therefore β distribution was used for fitting f (t e)probability distribution. The best evaluation value and the left margin of β distribution werechosen for flatness error interval estimation, while percentile Qpintercept method wasproposed to reserve better evaluation samples, the maximum entropy method(MEM) wasused to calculate the important parameters of the left marginmte,steto improve thereliability of the interval estimation. The simulation experiments results showed that βdistribution uniform expression method has a good fitting for intelligent evaluation results,while the length of flatness estimate sectionDt Thas a short span. As the number ofintelligent evaluation samplesN S=100, intercept percentile Q p=20,Dt T [10~-7,10~-5]mm,whileDt Tincluded flatness error value. This method could effectively present intelligentevaluation uncertainty.Flatness error evaluation method based support vector machine (SVM) was researched.Through analyzing the mathematic models of support vector machine regression (SVR) andflatness error minimum zone, it was pointed that SVR optimization goals-optimal hyper planeand the minimum zone criterion results in exactly the same geometric mechanism.Consequently, the evaluation method based on ε-SVR was proposed. Aiming at ε was hard tofix, the evaluation method based on seperated support vector machine classification (SSVC)was also proposed, which uses SMO algorithm to decomposite the even model intocalculating two samples' Langarange factors to improve computational efficiency.Measurement uncertainty assessment methods using Monte Carlo Method was proposed. Alsosimulation measurement data evaluation experiments were performed. The simulationexperiments results showed that SSVC evaluation error was lower, and evaluation timeTC≤0.0508s, which proved that SSVC was also effective for many points evaluation. Throughmapping F, minimum zone center circle and center cylindrical surface are transformed intohyper plane in Hilbert space, SVM could be used in roundness error and cylinderity errorevaluation, which provided a new idea for other form error evaluation.Granite surface plates, track plates flatness error, bearing ring roundness errorintelligent evaluation experiments were performed; the results verified that the flatness errorand roundness error intelligent evaluation and uncertainty assessment in this paper is of practicality and effectiveness in practical measurement.