CNC machine tools are advanced manufacturing equipment integrating machine, electricity and liquid. Their faults are complex and require high maintenance personnel. It is difficult to achieve fast and accurate positioning faults in traditional maintenance methods. It is an irresistible trend to realize intelligent diagnosis. At present, the expert system based on symbolic reasoning has made great progress in the fault diagnosis of CNC machine tools, but still encounters some difficulties, mainly as the “bottleneck†of knowledge acquisition and the “combination explosion†problem of logical reasoning, ie Inferred to be inefficient, poorly adaptable, etc. The neural network can effectively avoid the above problems with its unique learning ability, association ability and knowledge acquisition ability, but there are also problems such as the inability to explain its own reasoning method.
This research combines neural network with expert system to develop a neural network fault diagnosis expert system.
1 Neural network expert system structure According to the different ways of combining expert system and neural network, the expert system based on neural network can be divided into three types: serial, parallel and hybrid. This scheme uses neural network in front and expert system in After the serial mode. The “Fault Sign Acquisition and Processing Module†is responsible for the acquisition of fault symptom signals in the machine tool and is used as the input signal of the neural network after proper processing. The neural network uses multiple layers of parallel structure to solve and deduct the types of faults in CNC machine tools. It accepts the normalized symptom signal input, gives the processed result, and then uses the expert system to verify and interpret the diagnosis result.
2. BP neural network structure and improved algorithm 2. 1BP neural network topology The numerical control machine has complex composition and various faults. It has both electrical faults and mechanical faults. There are both strong electrical faults and weak electrical faults, which require many measuring points. Therefore, designing a single network is likely to cause a huge network structure, an increase in training samples, making network training difficult; and causing low classification accuracy and unreliable diagnosis results. In the actual application, the layered and modular design method is adopted. According to the composition of the CNC machine tool, it is divided into four diagnostic modules: servo system, PLC and electronic control system, CNC system and other systems. Diagnose speed and accuracy, and diagnose multiple faults simultaneously.
The neural network adopts the BP network which is relatively mature in application. It is a multi-layer forward network with one-way propagation. The network has one or more hidden layer nodes in addition to the input/output nodes. The theory has proved that under the premise that the number of hidden layer nodes can be freely set according to actual needs, the three-layer BP neural network can realize the function of approximating any continuous function with arbitrary precision. Therefore, this study uses a standard three-layer network topology.
In the figure, the input layer corresponds to the fault phenomenon, and the output layer corresponds to the fault cause. Adjacent layers in the network are connected in a fully interconnected manner. There is no connection between the neurons in the same layer, and there is no direct connection between the output layer and the input layer.
2. 2 Neural network improvement algorithm For the traditional BP network algorithm, there are problems such as low operation speed and easy to fall into local minimum points. The learning algorithm of the neural network model uses BP algorithm with impulse term. The training process of the network is as follows: 1) Set the initial values ​​ωj(0) and θj(0) of the respective weights and thresholds to be small non-zero random numbers.
(2) Input learning samples: input vector X p ( p = 1, 2, ..., P) and target output T p ( p = 1, 2, ..., P). o pj = fj(∑ωi oi -θj)(1)(3) Calculate the actual output of the network and the state of the implicit unit: o pj = fj(∑ωi oi -θj) where the excitation function f is the Sigmoid function, That is, f ( x) = 1 / (1 + exp( - x) ).
(4) Calculate the training error: output layer: δpj = o pj(1 - o pj) ( t pj - o pj) (2) hidden layer: δpj = o pj(1 - o pj) ∑kδpkωk(3)( 5) Modify the weight and threshold: ωi( t + 1) = ωi(t) + ηδj o pj + α(ωi( t) - ωi( t - 1) ) (4) θi( t + 1) = θi( t) + ηδj + α(θi( t) - θi( t - 1) ) (5) where η? learning step size, α? potential state term.
(6) When p experiences 1~P, judge whether the index meets the accuracy requirement E, here E 3 diagnosis example 3.1 Fault mode and fault analysis This study takes SINUMER IK802C servo drive system as an example, and compares 13 typical servo faults. The mode and corresponding cause of failure analysis are as shown. Other diagnostic model methods are similar.
The input vector represents the failure mode X, where "1" indicates failure and "0" indicates normal. In the table, the fault points corresponding to the 13 monitoring points have the following meanings: X 1? Servo drive power failure; X 2? Drive not ready; X 3? CNC machine overtravel alarm; X 4? CNC machine can not find reference point; X 5? Feedback loop fault; X 6? Position error too large alarm; X 7? Two-axis linkage roundness error; X 8? Servo system overload alarm; X 9? Servo system over-voltage alarm; X 10? Fault; X 11? Machine crawling and vibration; X 12? Servo motor speed is abnormal; X 13? Servo motor does not turn fault.
The output vector represents the fault cause analysis result Y, and the corresponding meanings are as follows: Y 1? Check the AC power supply (); Y 2? Check the DC power supply; Y 3? Check the fuses, relays and air switches, etc.; Y 4? Check the terminals and connections Y; Check the chip load; Y 6? Check the parameters set by the CNC system; Y 7? Check the CNC machine limit switch and zero switch; Y 8? Check the servo system parameters; Y 9? Check the servo system hardware; Y 10? Check the feedback encoder hardware; Y 11? Check the servo motor; Y 12? Check the mechanical drive chain components.
3. 2 Simulation analysis According to the fault sample table, it can be determined that the input layer number of the BP network is 13, the number of output layers is 12, the number of hidden layers is 16 according to experience, and the excitation function of the hidden layer and the output layer neurons selects the Sigmoid function. The network training function selects the traingdm function, and uses the MATLAB neural network toolbox for simulation training, and selects the learning rate lr =0. 5, the momentum factor mc = 0.7, the error e = 0. 000 1. The error curve, after 6 070 training sessions to meet the requirements. At the same time, the weights are recorded for the hardware implementation of the neural network.
For the trained network structure, the input fault mode vector X = <1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 > for verification, the actual output fault analysis vector Y = < 1 1 1 1 0 1 1 0 0 0 0 0 0 0 0 >, the result is completely correct. After several experiments, the model was verified to meet the diagnostic requirements.
4 Conclusion The neural network is combined with the expert system and applied to the fault diagnosis of CNC machine tools. It is an intelligent diagnostic system with complementary advantages. According to the characteristics of the neural network using a huge amount of parallel distributed information processing structure, a large-scale field programmable gate array (FPGA) integrated circuit can also be used to fabricate a diagnostic module, and it can be embedded into the numerical control device as a dedicated circuit, thereby realizing Real-time diagnosis of faults in CNC machine tools.
It is believed that this integrated fault diagnosis system based on the integration of expert system and neural network will be a trend of intelligent development of CNC machine tool fault diagnosis. But with the development of CNC technology and machine tools, the faults will become more complicated and diverse. How to adopt a hierarchical structure, further refine the diagnostic subnet, and collect enough learning samples to ensure the accuracy of diagnosis will be the subject of future research.
(Finish)
Water Tank Blow Molding Machine
Water Tank Moulding Machine Price,Plastic Water Tank Manufacturing Machine Price,Water Tank Blow Molding Machine
Weifang Uyoung Iknow plastic machinery co,.ltd, , https://www.uyoungiknow.com