In order to survive and develop, domestic CNC machine tool manufacturers have put forward urgent requirements for reliability prediction. The methods of reliability prediction of CNC machine tools can be divided into two categories: one is based on experience-based prediction methods, which is based on the investigation, sampling, etc. by experienced professionals, combined with the actual situation at the time to make predictions. Another method is based on historical data and empirical data, first establish the corresponding mathematical model, and then use the mathematical model to predict. The first forecasting method needs to accumulate a lot of experience, so it is not easy to learn, the inheritance is very poor, and the accuracy of prediction is relatively low. The prediction process has no solid theoretical basis and the persuasive power is relatively poor. The second method is predicting CNC machine tools. The application of reliability is extensive, the prediction accuracy is high, and the prediction process is based on mathematical theory knowledge, and the persuasive power is relatively strong. At present, the methods used for system reliability prediction mainly include mathematical model method, conditional probability method, and minimum cut set approximation. However, traditional mathematical methods, such as mathematical model methods, have poor adaptability. In solving the problem of predicting the reliability of CNC machine tools, the accuracy is poor, and some prediction problems cannot even be solved. Based on the in-depth study of neural networks, this paper proposes a three-layer BP neural network model and algorithm for reliability prediction of CNC machine tools, and compares it with traditional data model prediction methods to prove that the neural network prediction model proposed in this paper is more traditional than the traditional one. Mathematical prediction methods are more accurate and feasible.
2 Prediction network model The main idea of ​​using neural network to predict the reliability of CNC machine tools is to use the reliability parameter x of each subsystem that constitutes the CNC machine tool as the information input of the neural network, and the reliability index y of the CNC machine tool as the network output. Establish a three-layer BP neural network. The M reliability data pattern pair (x) of some types of CNC machine tools is used as a training sample set to train the network. Finally, the reliability index of the CNC machine tool of the predicted model is calculated by using the trained network.
The BP neural network model for reliability prediction of CNC machine tools is shown. It consists of three layers of input layer, hidden layer and output layer. x is the network input value, y is the output value of the jth neuron of the hidden layer, y is the final output value of the network, w is the jth neuron of the hidden layer and the i th neuron of the input layer The connection weight between them, w is the connection weight of the i-th neuron of the output layer and the hidden layer, and is the value of the i-th neuron of the hidden layer, which is the value of the output layer, and f is the function function. In this study, the Sigmo id function is taken as the function function, that is, f(x) = 1 /( 1+ exp( - x ) ).
3 algorithm principle set a certain input sample x
The target value is dy is the output value of the hidden layer, and y is the output value of the network. Then there is: the design study analyzes the error function as: E = find the gradient of E to the weight: a unified formula, let w denote the first layer The connection weight of the jth neuron to the i-th neuron of the l-1st layer; w represents the connection weight of the jth neuron of the 1st layer and the kth neuron of the l+1 layer; Q Representing the input value of the i-th neuron of the previous layer, the gradient formula can be unified as:
When unit j is an output unit: when unit j is an implicit unit: the weight changes along the negative gradient direction of the error function, so the amount of change of the weight is: where is the learning factor, so the weight correction formula can Uniformity: In addition to considering the convergence of the learning process, the larger the learning factor is, the larger the value is. The more intense the change of each weight, the more likely the oscillation will occur during the learning process. Therefore, in order to make the learning factor large enough and no oscillation, add a potential term in the weight correction formula: the same can be used to derive the correction formula: the weight correction and the interpretation correction are in the error Completed layer by layer in the process of propagation. The following is a flow chart of BP neural network numerical control machine tool reliability prediction algorithm.
4 parameter discussion 4. 1 Pre-processing of input value and output value For the parts of CNC machine tools, the failure rate is generally between 0 1 . If it is not within this range, you can make the input of the network have a more stable range by adjusting its value to fall between 0 1 . This is advantageous for the selection of parameters. For the output value, if the input value is limited to 0 1 , the overall failure rate of the CNC machine tool will generally be greater than 1. However, for the action function Sigmo id function used in this algorithm, the output value should fall between 0 1 . Therefore, the output value should be processed, which should be handled for different practical problems. The output value y is generally processed as follows:
This will ensure that the value falls within the range of 0 1 .
4. 2 Determination of input data linear regression and x
The goodness test proves that the life distribution of the CNC machine tool can be approximated to obey the exponential distribution, so that it can be determined which data is used as the input data. The algorithm will use the failure rate as the input data. In addition, in order to ensure the performance of the network and the complexity of control calculation, the number of input data should not be too much, preferably within 10, but the structure of the CNC machine tool is generally more complicated, only the direct subsystem is more than 10, To solve this problem, this paper proposes a secondary input parameter merging method. Because for most CNC machine tools, faults are generally concentrated on several subsystems, such as a certain type of CNC lathe, which occurs in the CNC system, turret tool holder, electrical system, main drive system, hydraulic and pneumatic system, and feed system. The failure of the system accounted for 74.34 of all faults. In this way, the reliability data of each major subsystem is taken as an input data, and then the secondary subsystems are combined as one or several input data according to the functional relationship between them, so that the input data can be controlled at Within 10, it can guarantee the performance of the network and meet the requirements of calculation accuracy.
4. 3 Determination of environmental factors In addition to the reliability data of each component unit, the input value of the algorithm adds an environmental factor. Because the reliability level of CNC machine tools is determined not only by the reliability level of each component and the interaction between them, but also by the environment in which they are used. The environment referred to here is a broad concept including the use environment, operational proficiency, and processing conditions.
Here is a rough way to determine the environmental factors. The level of the environmental factor is as shown, and the final environmental factor's h is calculated as:
4. 3. 1 working environment h
The working environment refers to the operating environment of the machine tool. Excellent working environment can be modern machinery in the second phase of 2007: the ambient temperature is relatively constant, kept at 10 30, indoor cleaning, less dust, good fixation or shock absorption measures, which is more common in advanced automated production lines. A good working environment can be: comfortable temperature, clean indoors, occasional noise and vibration. The general working environment is: moderate temperature, clean environment, a small amount of noise and vibration. The poor working environment is: indoor temperature fluctuations are relatively large, indoor noise is large, and workers will feel bored after working for a long time. The harsh working environment can be: the ambient temperature fluctuates greatly, and even the wind is exposed to the sun; the indoor sanitary conditions are poor, the air is more dusty; the fixing is not complete or can not be fixed well, and the workers are not easy to work in this environment. .
4. 3. 2 operational proficiency h
Operational proficiency refers to the proficiency of the operator. For skilled and skilled operators, you can use the optional level. Apprentices choose apprenticeships when they use them. The working years are less than 0.5 years. The selection is less skilled. The working years are 1.1.5 years. The mechanics can choose the general level. The working years are 1. 5 2 years. The skilled workers can choose more skilled and have a working life of more than 2 years. The choice is skilled.
4. 3. 3 processing conditions h
Processing conditions are related to a variety of factors, such as work strength, cutting amount, and processing materials. The processing conditions can be divided into five grades according to these conditions, and the medium load can be selected in general.
4. The initialization weights and threshold initialization of the weights and thresholds have a certain impact on the convergence of the network. The selection is good and the convergence is fast. If the selection is not good, the network may be in a saturated region, it is difficult to converge, or the network Limited to local minimum. However, there is currently no optimal initialization method, but it is best not to have the same value, generally between -1 1 . The algorithm will use the random number between -10 or 01 as the initial value of the weight according to the actual situation.
4. 5 learning factor and potential factor determination The learning factor is large, the convergence speed is fast, and vice versa is slow. If it is too large, it is easy to cause network oscillation and divergence. The role of the potential factor is to make the learning factor large enough without turbulence. Reasonable learning factors and potential factors can improve the efficiency of network learning. Usually 0< < < 1. To shorten the convergence time of the algorithm, you can use the variable learning factor. The learning factor value is very large at the beginning, with the number of iterations. The increase can be appropriately lowered, and this can also improve the graduality of the algorithm.
4. 6 Iterative Calculation Endpoint Judgment After the network converges, iterative calculation can be stopped. At this time, calculations such as simulation and prediction can be performed. There are generally three methods for determining the end point of the iteration. One is that the total error is less than the specified error, the other is that the number of iterations reaches a predetermined number of times, and the third is that the error of the monitoring sample set increases. In this paper, the first method is adopted to specify the iteration precision in advance, and its size should be selected according to the actual situation.
5 Case Study The example in this paper is two types of CNC machine tools (A, B) in an automobile manufacturing center. Their reliability data are obtained based on the previous operation statistics, including the overall reliability data of the machine tool, turret, and electrical system. , chip removal system, card loading system, X-feed system and environmental factors. The traditional prediction results of the mathematical model method and the neural network method of this paper are shown.
It can be seen that the neural network prediction method can significantly reduce the prediction error compared with the traditional mathematical model method. Using the trained network to predict the two types of CNC machine tools, the average error is 2.76, which is much lower than the average error rate of 11.29 predicted by the mathematical model. It can be seen that in terms of prediction accuracy, the neural network method has a greater improvement than the mathematical model prediction method.
When it is assumed that the reliability data of the X-feed system is unknown, the conventional mathematical model method and the prediction result data of the neural network method of this paper are shown. 0. 274 970 cannot predict unpredictable 0. 273 940 cannot predict unpredictable prediction neural network
It can be seen that the neural network method has better adaptability than the traditional mathematical model method. When the reliability data of the X-feed system is unknown, the mathematical model method cannot predict. This is because the mathematical model method requires higher data integrity. The neural network method can not only predict, but also predict accuracy. It is still relatively high, although the average error is changed from 2.76 to 3.23, but it is still much lower than the error predicted by the mathematical model method when data is complete (11.29).
6 Conclusion This paper combines the characteristics of CNC machine tool reliability prediction, and establishes a three-layer BP neural network model and algorithm for reliability prediction of CNC machine tools. The principle, steps and processing methods of the problems encountered in the algorithm are systematically expounded. Compared with the traditional mathematical model prediction method, the reliability prediction method proposed in this paper is more accurate and reliable than the traditional mathematical model prediction method.
(Finish)
Battery Paper Packaging Machine
Automatic Wrapping Machine,Battery Packaging Automatic Production Line,Automatic Battery Packaging Machine,Battery Paper Packaging Machine
Dongguan City Li you machinery and Equipment Co., Ltd. , https://www.gdliyou.com