An Arbitrary Angle Measurement System Based On Genetic And Neural Network Real-time Error Correction

A more systemic and in-depth research of error correction on grating dynamic measurement system is performed through analyzing the characteristics of dynamic error, the characteristics of angle and circular division measurement, the dynamic error statistical characteristic of measurement system and the error transfer characteristics.The inscapes and token of dynamic measurement error is depicted. The concept of relative movement and the describing method of relative movement rate of dynamic measurement system are shown. A dynamic error neural network topology structure corresponding to the established system is designed. A cognizing matrix of error source and the transfer, classifying model of dynamic error source are put forward. On the base of the relativity and the data self-sealing characteristics of circular division measurement, an arbitrary measurement system on circular division and dihedral is built, and the processing method of the dynamic data zero average and the balancing based on self-sealing nature reference are also built.Moreover, the fuzzy clustering discriminate analysis method for distinguishing and eliminating the gross error of the measurement sample is established. The gross error of the practical measured data is distinguished by use of the method prove the established the gross error distinguishing model practical to the measurement system. After this, the data processing method and concepts about the stationarity and ergodicity of measurement system in the limited distance are proposed. This experimental system is proven to be limited stationarity and ergodicity in short distance sample space.This paper also proposes that the comparability among prediction data can be described using sample distance space norm, and the comparability 8 function is defined synchronously. Using the function, the relations between dynamic error data comparability and data correlation function are deduced, and the evaluating method that assesses the model's prediction error using correlation function's relative error is built. The effective prediction space concept is established, On the base of these, this paper deduces two representations' evaluating equation, one evaluates the prediction error and the other evaluates the prediction error in limited space. First, the error transfer characteristic among subsystems at different space locations is analyzed, and the direct transfer characteristic from discrete standard measure space to the workpiece measure space under measured in measure system is proven. Second, the error reconstruction condition and method of mapping from discrete standard measurement system to continuous standard measurespace are analyzed. Based on the measurement sample stationarity in limited distance,the prediction model's limited astringency and mensurability to the dynamic measuring error and the prediction error respectively are proven. More,the feasibility and rationality performing prediction error correction on work piece to be measured using discrete standard quantity system are proven.A high-precise arbitrary dihedral and circular division measurement system is exploited at the first time; the measuring and system controlling software based on MS Windows are also exploited. The dual standard quantity (the work piece and the discrete standard quantity) mutual measuring and model verification methods are also proposed,which perfects the whole modifying process from data measuring,error separation,model establishment to real correction.After researching the discrete standard quantity system dynamic error separation technique,two error correction methods based on genetic algorithm and neural network mixed modeling technique are established. The two methods are the discrete standard quantity dynamic error direct/synchronous correction and prediction model correction; the model's parameters and model's exercising method are also confirmed. After analyzing the multi-time prediction characteristic of the prediction model,the equality between multi-time prediction an...