Application of Neural Network Modeling in Deformation Measurement of Thermal Expansion Bolts

The application of control and decision neural network modeling in the measurement of thermal expansion bolt deformation is enhanced. Yuan Zhuyu's algorithm is used to optimize the search learning rate by using the order interpolation method. The æ–© method has a fast convergence speed and a good acceptance accuracy. After serving, the vocal practice of the Fa sounds and meridians has been ruined, and the block is smashed, and the result of the smoldering heat shape is åž , æ˜Ž , = = = = = = = = = = = = = = 55 55 55 55 55 55 55 55 55 Introduction Neural network has great potential in the field of control engineering. 2. The neural network has good approximation ability to nonlinear functions, making it a very important technology and wide-ranging method in nonlinear system modeling identification and control. The weighting of the multi-feedforward 1 network before use is slow. The search speed is slow based on the gradient descent direction. It is often necessary to obtain many steps in order to obtain satisfactory approximation to the system, and it is even difficult to achieve the desired accuracy.

Received date 1 less 1243 repair date 2000 also 28 fund project country 86308 applied basic research fund project 8635445410; Tianjin natural science fund project 9836020; Ministry of Education backbone teacher program project island people, professor, doctoral tutor, engaged in adaptive control Research on intelligent control and other aspects.

Through the learning of network weights, the paradigm method of sub-interpolation is used to search the learning step size 0, and the variable scale order is proposed. The algorithm has faster convergence speed and higher approximation accuracy.

Using this method, a nonlinear mathematical model of the relationship between the deformation of the thermal expansion bolt and the heating time in the film stretching line of Pan is established. The results show that the method converges quickly and the established model has high precision. This model has been used in the thickness control system of the production line and has achieved good results.

2 The feedforward neural network assumes that the input and output are obtained by a nonlinear dynamic system, xKYj=vj., then these data can be used to learn the three-layer feedforward neural network. The approximation model is the weight vector in which the network is taken. In the learning algorithm of the network, the neural network model is output by adjustment, and the actual output of the near system is used to identify the input layer nodes of the 3-layer feedforward neural network. One, one hidden layer node, and one output layer node. When will be the first. The original data heart of the transcript, when 1 input network, the corresponding hidden unit net input is recorded as =2, the hidden unit output status is recorded as open = 2, B, the final output of the network, that is, the output of the output unit is recorded as; =, 2, 14 nodes to the hidden node's connection right 1 =, upper B + 1; claw. =2 is the connection weight of the hidden layer node to the output layer node, where the guest is a function of 8 and it is generally poor to be +=e.

Ask the most efficient algorithm. The weight learning of multi-layer feedforward neural network is also an unconstrained extremum problem, which provides a theoretical basis for the variable scale method applied to the weight learning of neural networks. Let lf change the weight of the network to enable the function 7 to be as small as possible, set the gradient of the pair to çŽ¢ç–‹, and use the variable-scale order fast learning algorithm to learn the weight calculation steps as follows: The decision to allow the error 4 is the weight of the optimization, pity iteration; otherwise. Turn to less.

In the household direction, the dimension search is performed to determine the optimal step size, so that the weight of the next group is taken as 1 and the optimized weight is obtained. Stop the iteration. Otherwise, press F to make the dimension search in the corpse. Jia Buchang can get the following set of weights and scales. The fast learning method is a valid method for solving unconstrained extremum problems. Since it avoids the calculation of the order derivative matrix and its inversion 557, +1 satisfies the tip requirement, iterative learning is stopped; if it is, then 5 rushes to 4 until the tip meets the requirement. If iterative, the dimension 1 whose weight is 1 is still not converged, then iteratively re-enters the hall wheel with 1 as the starting point.

After the interception, the brothersâ€™ method of attacking and arranging the special bombs, and the mysterious Wei search for the rabbits in the south, Yang has remarkable superiority. It is therefore considered to solve the unconstrained extremum method. The specific method is as follows. Take the first step as an example to take (1) + (4) and assume that the corpse (4) is a unit length vector. Calculate 1 if å— å¯¹ , å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£« å£«A number of bolts in the film stretching line are used to adjust the thickness distribution of the polyester base by expanding the heat at the lips of the production line. However, the deformation of each bolt of the corpse is difficult to measure online in real time, so the thermal time ratio of each bolt in each control interval can be controlled by the solid state relay, that is, the duty cycle. This requires a mathematical model of the relationship between bolt deformation and heating time, which is a nonlinear mathematical model.

We use the feedforward network to build a shuttle. The variable scale order fast learning algorithm is used for identification. The drawing interval is to study the relationship between the heating time and the bolt deformation in this interval. Adopt 241 network structure, the input is 1=.17, the small heating time unit 5 output garment expansion deformation unit 1.

In the off-line case, 60 samples were taken with a special instrument. The learning algorithm of this paper is used for training. The initial value of the network is taken as a random number between the two, and the initial value of the step is taken, which is 0.98. The model obtained after identification is the model with high precision. Curve 1 in the model output fitting is the output of the trained network model; the curve is the corresponding actual measured data.

This mathematical model has been applied to the thickness control system of the film, and the thickness of the slab at a plurality of points in the lateral direction is measured to form a transverse profile. In order to make the profile reach the ideal profile, the deformation displacement of each thermal expansion bolt is calculated by a fixed control algorithm, and then the heating time of each thermal expansion bolt is calculated by the mathematical model, and the relay is driven to heat the bolt. .

The conclusion can be seen from the application research, the variable-scale fast learning algorithm has great advantages in neural network identification. The first is that the convergence speed is fast. It takes only a few dozen steps to achieve a satisfactory result, so the network can be trained in a small sample. Secondly, there is no need to manually select the step size 1 learning effect base tree + the sample sample of the learning sample wood and the convergence precision is high. In the system with high precision requirements, the calculation amount of each step is more than the calculation amount of the algorithm. Large, but the time required to achieve satisfactory accuracy is still much less than 8 savings. This method has broad application prospects in the field of nonlinear system identification.

Lin Maoqiong, Chen Zengqiang, He Jiangfeng, et al. Neural network self-correcting step predictive controller based on damped least multiplication. Control and Decision, 1999, Wang Qunxian, Chen Zeng, Yuan Zhuyu. The nonlinear system based on wavelet network is built in love. Operations research. Beijing Tsinghua University Press, 199 0 he 1. Introduction to nonlinear optimization. Ma Zhengwu, translated by Wang Peiling. Beijing China Outlook Press, and dare, the embarrassing effect of the rules and regulations, Liu, postal painting, 1.

Completely pre-assembled, no assembly needed
Frame made of powder-coated aluminum
High quality fiberglass fabric
UV-resistant, washable, light and air permeable
Durable, stable structure, ready made, easy installing, EN13561-2015 certification.
Resistance to wind loads: Class 1
Operating effort: Class 2
Mechanical Endurance: Class 1
Resistance to corrosion: Class 2
As the professional Insect Screen supplier, we cooperated with consumer from all the world to settle the insect problem.

Sliding Insect Screen

Sliding Insect Screen,Sliding Screen,Insect Window,Mosquito Net

Huanghua Techo Building Material Co., Ltd. , http://www.insectsscreen.com