A numerical control machine tool process parameter self-adaptive regulation and control method considering tool state

Abstract

The application discloses a kind of numerical control machine tool process parameter self-adaptive regulation and control methods considering tool state, belongs to intelligent manufacturing technical field.First, variable cutting depth milling machining test is carried out on machine tool, spindle motor power signal is collected in real time during processing, and signal is handled with missing value completion and normalization;Second, tool state prediction and spindle power prediction model are respectively constructed;Then, based on the predicted value of tool state and spindle power, a process parameter self-adaptive regulation and control model considering tool state is established;Finally, according to the regulation and control model, the processing process parameters are adjusted in real time.This method considers the influence of tool state on processing efficiency in self-adaptive regulation and control, maximizes the processing efficiency under the premise of avoiding abnormal damage of tool and machine tool due to parameter regulation, and meets the high-quality, high-efficiency processing requirements of parts.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent manufacturing, and relates to a numerical control machine tool process parameter self-adaptive regulation method considering tool state. BACKGROUND

[0002] The principle of numerical control machine tool process parameter self-adaptive regulation is to maximize the processing efficiency under the premise of avoiding tool and machine damage and ensuring part processing quality, so as to meet the processing demand of high quality and high efficiency. In the traditional numerical control machining process, the process parameters are determined in the programming stage before machining, and these parameters are often conservatively set according to the experience of operators or the processing manual, which are not the optimal cutting parameters and are not suitable for the dynamic cutting process. Therefore, the high-quality and high-efficiency processing technology capable of self-adaptive regulation of machine tool process parameters has become an urgent demand in the current mechanical processing field.

[0003] In terms of process parameter self-adaptive regulation, scholars have carried out related research. In the patent "Machine Tool Self-Adaptive Control Method Based on GA-BP Neural Network Algorithm" (Application No. CN201910732917.3), the neural network algorithm is used to optimize and self-adaptively adjust the feed speed and spindle speed in real time, which can effectively improve the processing efficiency and processing quality; in the patent "Numerical Control Machining Parameter Self-Adaptive Fuzzy Control Rule Optimization Method" (Application No. CN201310081486.1), the power bond graph method is used to optimize the fuzzy control rule of numerical control machining self-adaptive control, which improves the control performance and processing stability; in the paper "On-line Monitoring of Surface Roughness and Adaptive Optimization of Drilling Parameters for Composite Materials" (Journal of Mechanical Engineering, 2020, 56(02): 27-34+42.), an on-line monitoring model of hole wall roughness based on support vector regression is established, and the simulated annealing algorithm is used to optimize the drilling parameters under the current monitoring state to ensure the drilling quality; in the paper "Research on Constant Power Constraint Adaptive Machining Method Fusing Chatter Control" (Mechanical Manufacturing and Automation, 2018, 47(05): 41-44+52.), based on fuzzy theory and variable speed suppression cutting chatter theory, the adjustment principle of machine tool spindle speed and feed speed is formulated with processing efficiency as the target, and constant power constraint adaptive machining fusing chatter control is realized.

[0004] Through the analysis of the adaptive regulation technology of machine tools, it is found that: (1) the existing adaptive regulation model does not consider the tool state, and the fixed reference input constraint of the model is not suitable for the dynamic change of the tool state in the machining process, resulting in that the machining efficiency is not significantly improved; (2) at present, the adaptive regulation is mainly carried out by using neural network or fuzzy theory alone, and there are problems such as difficulty in obtaining training data, delay in response of the control system, and the like, which easily lead to poor model generalization and abnormal damage of the tool. In view of the above problems, the present application proposes a fuzzy control model with time-varying input constraint, and organically combines neural network and fuzzy theory, so that the machining efficiency of the machine tool can be significantly improved. SUMMARY

[0005] In order to solve the problems that the common adaptive regulation method of machine tools does not consider the influence of the tool state on the control effect, the abnormal damage of the tool caused by the delay of the control system, and the like, the present application proposes a numerical control machine tool process parameter adaptive regulation method considering the tool state.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A numerical control machine tool process parameter adaptive regulation method considering the tool state, first, a variable cutting depth milling test is carried out on the machine tool, the spindle power signals under different working conditions are obtained, and the obtained signals are preprocessed; then, the processed power signals are used to train a radial basis function neural network (RBF neural network) and a recurrent neural network (RNN network) respectively, to obtain a tool state prediction model and a cutting power prediction model; then, the input constraint of the fuzzy control is updated according to the tool state, at the same time, the larger value between the power data predicted by the RNN network and the actual monitored power data is selected as the feedback input of the fuzzy control, and a machining parameter adaptive regulation model with time-varying input constraint is established; finally, in the actual machining process, the spindle power signal of the machine tool is monitored in real time, and the machining parameter adaptive regulation model is used to calculate the regulation value of the feed speed, and the feed speed is regulated in real time. The method comprises the following steps:

[0008] First step, variable cutting depth processing information acquisition and processing;

[0009] A stepped sample is prepared, the same milling tool is selected, and cutting is carried out under the condition of different spindle speeds and feed speed combinations. The power signal of each cutting process is collected by using a power sensor. In view of the abnormal situation of signal loss in the power signal collection process, the average value of the two values before and after the missing value is used as the new value after compensation, and the compensated power P is calculated according to formula (1) i :

[0010]

[0011] P = (P + P) / 2 (1) wherein, Pi-1 P is the signal value at the time point before the signal is lost i+1 P is the signal value at the time point after the signal is lost

[0012] The collected power data is normalized by a linear function, and the normalized value P of the power sensor collected value is calculated according to formula (2) x

[0013]

[0014] P is the original collected power value, P is the maximum value of the original collected power value, and P is the minimum value of the original collected power value. o max min

[0015] Second, the tool state prediction model and the cutting power prediction model are constructed.

[0016] The main part is divided into two parts of the tool state prediction model and the cutting power prediction model.

[0017] (1) The first part is the construction of the tool state prediction model, an RBF neural network is established, and the tool state is predicted according to the collected power value in the machining process, the input is the processed power value P x , and the output is the state T i of the tool.

[0018] The first layer of the RBF neural network is the input layer, which is composed of the power value P x , the second layer is the hidden layer, and the activation function p(P x , C j ) of the neurons in the hidden layer is a non-negative linear function which is radially symmetric and decays to the center point C j , and the activation function of the jth neuron in the hidden layer is represented by formula (3):

[0019] p(P x , C j ) = exp(-β j ||P x -C j || 2 ) (3)

[0020] Where β j represents the input received by the jth neuron, and C j represents the jth center point.

[0021] The third layer is the output layer, and the output layer adjusts the linear weight, and its output T i is represented by formula (4):

[0022] ​​​​

[0023] where n is the number of nodes of the hidden layer; w ki is the weight of the hidden layer to the output layer.

[0024] (2) The second part is the construction of the cutting power prediction model. The RNN network is established to predict the next moment power by using the real-time collected power value. The input is the processed power value P x , and the output is the next moment normalized power value P p . The recurrent neural network output P p is represented by formula (5):

[0025] P p = g(V·f(U·P x +W·s t-1 +b1)+b2) (5)

[0026] where U is the weight matrix of the input layer to the hidden layer, V is the weight matrix of the hidden layer to the output layer, W is the weight matrix of the previous moment hidden layer to the current moment hidden layer, s t-1 is the value of the previous moment hidden layer, b1 and b2 are the bias of the output layer and the hidden layer, g(·) and f(·) are the activation functions of the output layer and the hidden layer.

[0027] Thirdly, the machining parameter adaptive control model considering the tool state is constructed.

[0028] The double-input single-output fuzzy logic controller is selected as the adaptive control model, and the input of the fuzzy logic controller is the power deviation e and the power deviation change rate ec, and the output is the control value f c of the feed speed.

[0029] Firstly, according to the change characteristics of the spindle power when the part machining allowance is uneven, the feedback input P f of the fuzzy logic control is calculated according to formula (6):

[0030] P f =max{P p , P x} (6)

[0031] where P p is the normalized power value predicted by the RNN network, and P x is the normalized value of the power sensor collected value.

[0032] Then, the normalized power constraint value of the fuzzy logic controller is given as P ref , and the time-varying input constraint value P′ ref is calculated according to formula (7):

[0033] P′ref = a i P ref (7)

[0034] wherein a i is the tool state T i corresponding coefficient.

[0035] The power deviation e in the fuzzy logic controller input is calculated according to equation (8), and the power deviation change rate ec is calculated according to equation (9):

[0036]

[0037]

[0038] Next, the power deviation e, the power deviation change rate ec and the feed speed control value f c are described using the negative big NB, the negative medium NM, the negative small NS, the zero ZO, the positive small PS, the positive medium PM and the positive big PB linguistic variables c , the fuzzy domain of e, ec and f c is set as [-n, n], the fuzzy set thereof is represented as {NB, NM, NS, ZO, PS, PM, PB}, and a triangular function is used as the membership function of e, ec and f

[0039] Finally, a fuzzy control rule table is formulated, the traditional fuzzy reasoning method (Mamdani fuzzy reasoning method) is used to calculate the fuzzy membership output, the area barycenter method is used for defuzzification calculation, and the feed speed control value f c is obtained.

[0040] Fourth step, real-time control of the feed speed;

[0041] The feed speed multiple control value M of the machine tool is calculated according to equation (10):

[0042]

[0043] wherein f c is the feed speed control value, and f o is the feed speed set by the program.

[0044] The machine tool and the industrial computer executing the above adaptive control system are in real-time communication; the machine tool dynamic link library is called, the feed speed multiple control value M is sent to the machine tool, and adaptive control of the machining parameters is realized.

[0045] Further, in the third step, n in the fuzzy domain [-n, n] of e, ec and f c is a positive integer not greater than 7.

[0046] The innovation of the present application is as follows: the existing machining parameter adaptive adjustment method does not consider the continuous change of the tool state in the machining process, the regulation and control model cannot be dynamically adjusted according to the tool state, resulting in insignificant improvement of machining efficiency; when adjusting the process parameters, the abnormal impact received by the tool is not considered, which easily causes tool damage. In the technical scheme of the present application, the influence of tool state change on machining efficiency improvement and the influence of process parameter regulation and control on tool state are fully considered, the parameters of the regulation and control model are updated in real time and the load condition of the spindle is predicted by using the data collected in the machining process, which can significantly improve the machining efficiency and reduce the abnormal damage of the tool caused by process parameter regulation and control.

[0047] The beneficial effects of the present application are as follows:

[0048] (1) The part machining efficiency is improved, and the production time cost is saved.

[0049] (2) A control method with time-varying input constraint is proposed, which considers the change state of the tool in the cutting process during the adaptive regulation and control of the process parameters, further improves the machining efficiency of the part, and improves the intelligent level of the machine tool.

[0050] (3) The impact force on the tool caused by the adjustment of the machining parameters is reduced, the abnormal damage of the tool is reduced, and the service life of the tool is prolonged. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a schematic diagram of power sensor arrangement.

[0052] Figure 2 It is an RBF neural network structure diagram for predicting tool state.

[0053] Figure 3 It is an RNN network structure diagram for predicting cutting power.

[0054] Figure 4 It is a membership function diagram of power deviation e, power deviation change rate ec and control value f of feed speed in fuzzy control module. c

[0055] Figure 5 It is a control value f of feed speed changing with time. c

[0056] Figure 6 It is a control value M of feed speed ratio changing with time.

[0057] Figure 7 It is a flow chart of the adaptive regulation and control method of the numerical control machine tool process parameters.

[0058] In the figure: 1 is a power sensor; 2 is a spindle servo motor; 3 is a vertical numerical control milling machine; 4 is a workbench. ​​Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings.

[0060] The first step is the collection and processing of information for variable shear deep processing;

[0061] like Figure 1 As shown, a spindle servo motor 2 is installed on the vertical CNC milling machine 3, which is located above the worktable 4. The power sensor 1 is arranged on the spindle servo motor 2, and its three coils are respectively wrapped around the three power lines U, V and W of the spindle motor.

[0062] A stepped cast iron prototype was fabricated using a domestically produced end mill with a diameter of 10 mm. A total of 121 cutting operations were performed using different machining parameters. The spindle speed ranged from 1000 N / min to 4000 N / min, with each group consisting of 300 N / min intervals. The feed rate ranged from 50 mm / min to 150 mm / min, with each group consisting of 10 mm / min intervals. A power sensor was used to record the power signal of each cutting process.

[0063] To address the anomaly of signal loss during power signal acquisition, the average of the two values ​​before and after the missing value is used as the new value after compensation. The compensated power value P is calculated using equation (1). i .

[0064] The collected power data is normalized to [0, 1] using a linear function, and the normalized value P of the power sensor is calculated using equation (2). x .

[0065] The second step is to construct the tool condition prediction model and the cutting power prediction model; this stage is mainly divided into the establishment of the tool condition prediction model and the establishment of the cutting power prediction model.

[0066] The first part is the establishment of the tool condition prediction model.

[0067] An RBF neural network is established to predict the tool state using real-time acquired power values. The input is the processed power value P. x The output is the tool status T. i The processed power value is input into the RBF neural network, and the tool state is calculated using equations (3) and (4). When the tool state is normal, the model output is T1; when the tool state is slightly worn, the model output is T2; when the tool state is moderately worn, the model output is T3; and when the tool state is heavily worn, the model output is T4. The network topology is as follows: Figure 2 As shown.

[0068] The second part is the establishment of the cutting power prediction model.

[0069] An RNN network is established to predict the cutting power at the next time step using real-time collected power values. The input is the processed power value P. x The output is the normalized power value P at the next time step. p The processed power value is input into the RNN network, and the normalized power value P is calculated using equation (5). p The network activation function is the sigmoid function, and the network is trained using the time-series backpropagation algorithm. The network topology is as follows: Figure 3 As shown.

[0070] The third step is to construct an adaptive control model for machining parameters that takes into account the tool condition;

[0071] A dual-input, single-output fuzzy logic controller is selected as the adaptive control model. The inputs to the fuzzy logic control module are the power deviation e and the rate of change of power deviation ec, and the output is the control value f of the feed rate. c .

[0072] First, the feedback input P of the fuzzy logic control is calculated using equation (6). f The time-varying input constraint value P′ of the fuzzy control is calculated using equation (7). ref When the tool state model output is T1, α1 is 1.0; when the tool state model output is T2, α2 is 1.1; when the tool state model output is T3, α3 is 1.2; and when the tool state model output is T4, α4 is 1.3. Next, the fuzzy control input power deviation e and power deviation change rate ec are calculated using equations (8) and (9).

[0073] Use negative large, negative medium, negative small, zero, positive small, positive medium, and positive large linguistic variables to describe the power deviation e, the rate of change of power deviation ec, and the control value f of the feed rate. c e, ec and f c The fuzzy universe of discourse is set to [-6, 6], and its fuzzy set is represented as {NB, NM, NS, ZO, PS, PM, PB}. Trigonometric functions are used as e, ec, and f. c Membership function, such as Figure 4 As shown in Table 1, a fuzzy control rule table is then developed based on expert experience. The Mamdani fuzzy inference method is used to calculate the fuzzy membership output. Finally, the area centroid method is used for defuzzification calculation to obtain the feed rate control value f. c In this embodiment, with a spindle speed of 1000 rpm and a preset feed rate of 50 mm / min, the feed rate control value f is obtained. c like Figure 5 As shown.

[0074] Table 1 Rule chart of fuzzy control module

[0075]

[0076] Fourth step, real-time control of feed speed;

[0077] The feed speed ratio control value M of the machine tool is calculated by using formula (10), when the spindle speed of the embodiment is 1000 n / min and the preset feed speed is 50 mm / min, the obtained feed speed ratio control value M is as shown in the table. Figure 6

[0078] The machine tool and the industrial computer executing the adaptive control system are connected through a network cable to realize real-time communication; the machine tool dynamic link library is called to send the feed speed ratio control instruction to the machine tool, and the adaptive control of the machining parameters is realized through the built-in PLC module of the machine tool. Figure 7

[0079] When the spindle speed of the embodiment is 1000 n / min and the preset feed speed is 50 mm / min, the part machining efficiency is improved by 34%, the production time cost is saved, the intelligent level of the machine tool is improved, and the tool service life is improved by 23%.

[0080] It should be noted that the above specific embodiments of the present application are only used to illustrate the principles and processes of the present application, and do not constitute a limitation on the present application. Therefore, any modifications and equivalent replacements made without departing from the spirit and scope of the present application shall be included in the protection scope of the present application.​​

Claims

1. A numerical control machine tool process parameter self-adaptive regulation method considering tool state, characterized in that, The numerical control machine tool process parameter self-adaptive regulation method comprises the following steps: Firstly, the spindle power signal under different working conditions is obtained through variable cutting depth milling test on the machine tool, and the obtained signal is pretreated; then, the processed power signal is used to train RBF neural network and RNN network respectively, to obtain a tool state prediction model and a cutting power prediction model; then, the input constraint of fuzzy control is updated according to the tool state, and the larger value between the power data predicted by the RNN network and the actual monitored power data is selected as the feedback input of the fuzzy control, to establish a machining parameter self-adaptive regulation model with time-varying input constraint; finally, the spindle power signal of the machine tool is monitored in real time during actual machining, and the machining parameter self-adaptive regulation model is used to calculate the regulation value of the feed speed, to regulate the feed speed in real time; The specific steps are as follows: First step, variable cutting depth processing information acquisition and processing; A stepped sample is trial-manufactured, the same milling cutter is selected, cutting is carried out under the condition of different spindle speed and feed speed combination, and the power signal of each group of cutting process is collected by using a power sensor; in view of the abnormal situation of signal loss in the power signal collection process, the average value of the two values before and after the missing value is used as the new value after compensation, and the compensated power P i is calculated according to formula (1) wherein P i-1 is the signal value at the time instant preceding the signal loss, P i+1 is the signal value at the time instant following the signal loss; The collected power data is normalized by a linear function to obtain normalized power value P x ; Second step, construction of tool state prediction model and cutting power prediction model; It is divided into two parts of tool state prediction model and cutting power prediction model; (1) The first part is the construction of the tool state prediction model. The RBF neural network is established to predict the tool state according to the power value collected in the machining process. The input is the processed power value P x , and the output is the tool state T i ; The first layer of the RBF neural network is an input layer, which is composed of power values P x The second layer is a hidden layer, and the activation function p(P x , C j ) of neurons in the hidden layer is a non-negative linear function that is radially symmetric and decays around the center point C j The activation function of the jth neuron in the hidden layer is represented by equation (3): ρ(P x ,C j )=exp(-β j ||P x -C j || 2 ) (3) where β j represents the input received by the jth neuron, C j represents the jth center point; The third layer is an output layer that adjusts the linear weights, and outputs T i is represented by Equation (4): where n is the number of nodes of the hidden layer; w ki is the weight of the hidden layer to the output layer; (2) The second part is the construction of the cutting power prediction model. The RNN network is established to predict the power at the next moment by using the real-time collected power value. The input is the processed power value P x , and the output is the normalized power value P p at the next moment; the output of the recurrent neural network P p is represented by formula (5): P p = g(V · f(U · P x + W · s t-1 + b1) + b2) (5) where U is a weight matrix from the input layer to the hidden layer, V is a weight matrix from the hidden layer to the output layer, W is a weight matrix from the hidden layer at the previous time to the hidden layer at this time, s t-1 is the value of the hidden layer at the previous time, b1and b2are the biases of the output layer and the hidden layer, g(·) and f(·) are the activation functions of the output layer and the hidden layer; Third step, construction of machining parameter self-adaptive regulation model considering tool state; A double-input single-output fuzzy logic controller is selected as the adaptive control model, the input of the fuzzy logic controller is power deviation e and power deviation change rate ec, and the output is the control value f of the feeding speed c ; First, according to the variation characteristics of the spindle power when the machining allowance of the parts is uneven, the feedback input P of the fuzzy logic control is calculated according to formula (6) f : P f = max{P p , P x} (6) wherein P p is the normalized power value predicted by the RNN network, P x is the normalized power value collected by the power sensor; Next, the normalized power constraint value P for the fuzzy logic controller is given ref The time-varying input constraint value P' is calculated from equation (7) ref : P' ref = a i P ref (7) wherein a i is the tool state T i corresponding coefficient; The power deviation e in the input of the fuzzy logic controller is calculated according to formula (8), and the power deviation change rate ec is calculated according to formula (9): e = P f - P' ref (8) Next, the power deviation e, the power deviation change rate ec, and the regulated value f of the feed speed are described using negative large NB, negative medium NM, negative small NS, zero ZO, positive small PS, positive medium PM, and positive large PB linguistic variables c The fuzzy domains of e, ec, and f c are set to [-n, n], and their fuzzy sets are represented as {NB, NM, NS, ZO, PS, PM, PB}, using a triangular function as the membership function of e, ec, and f c . Finally, the fuzzy control rule table is made, the fuzzy membership degree output is calculated by using Mamdani fuzzy inference method, the de-fuzzification calculation is made by using area barycenter method, and the regulation value f of the feeding speed is obtained c ; Fourth step, real-time regulation of feed speed; The feed speed multiple regulation value M of the machine tool is calculated according to formula (10): wherein f c is the regulated value of the feed speed, f o is the programmed feed speed; Real-time communication is carried out between the machine tool and the industrial computer for executing the self-adaptive regulation model; the machine tool dynamic link library is called, the feed speed multiple regulation value M is sent to the machine tool, and the self-adaptive regulation of the machining parameter is realized.

2. The method according to claim 1, wherein, In the first step, the power value P normalized by the power sensor acquisition value is calculated according to formula (2) x : where P o is the original collected power value, P max is the maximum value of the original collected power value, P min is the minimum value of the original collected power value.

3. The method according to claim 1, characterized in that, In the third step, e, ec and f c n in the fuzzy domain [-n, n] of f is a positive integer not greater than 7.

Images