METHOD FOR DIAGNOSING HEALTH OF CNC MACHINE TOOL

Abstract

The present invention belongs to the field of automation equipment predictive maintenance, and relates to a method for diagnosing health of a CNC machine tool. Parameters such as vibration and temperature are detected by Zigbee nodes adopted in the present invention and uploaded to a Zigbee gateway. The detected data is transmitted by the Zigbee gateway to a cloud server for digital signal processing, data mining and analysis processing. After a training test of the above data is carried out by the cloud server, a health degree model is obtained. The health degree model is transmitted by the cloud server to the Zigbee gateway. A clock circuit can be used for synchronizing system time and recording time as data is recorded. For the on-line predictive maintenance of a CNC machine tool, the most rapid and accurate predictive maintenance information is provided.

Claims

1. A method for diagnosing health of a CNC machine tool, comprising the following steps: step 1: using a cloud server to carry out FFT transformation processing on vibration data collected by a vibration sensor node, and obtaining spectrum averages of 800-1800 Hz, 1800-2800 Hz and 2800-3800 Hz; obtaining a regression coefficient of relationship among the spectrum average of a relevant frequency band, temperature and operating state of equipment according to corresponding historical data of the spectrum average of the frequency band and temperature; 1.1 creating a data vector
x=(x.sup.(1),x.sup.(2),x.sup.(3),x.sup.(4)) wherein x.sup.(1) is the spectrum average of 800-1800 Hz, x.sup.(2) is the spectrum average of 1800-2800 Hz, x.sup.(3) is the spectrum average of 2800-3800 Hz, and x.sup.(4) is a temperature average of equipment; 1.2 creating a coefficient vector w=(w.sup.(1), w.sup.(2), w.sup.(3), w.sup.(4)), wherein w.sup.(1) is a coefficient of the spectrum average of 800-1800 Hz; w.sup.(2) is a coefficient of the spectrum average of 1800-2800 Hz; w.sup.(3) is a coefficient of the spectrum average of 2800-3800 Hz; and w.sup.(4) is a temperature coefficient; 1.3 x.sub.i is the i.sup.th trained data vector, and y.sub.i is a class marker of x.sub.i; when y.sub.i is −1, it means that the equipment fails; when y.sub.i is +1, it means that the equipment is normal; and N is a quantity of trained data; finding a separating hyperplane with a maximum geometric margin, and a problem is expressed as the following constrained optimization problem: $\underset{w,b,\xi }{\min }\frac{1}{2}{.Math.W.Math.}^2+C\underset{i=1}{\overset{N}{.Math.}}{\xi }_i$ $S.t{y}_i\left(w.x_i+b\right)\ge 1-{\xi }_i{\xi }_i\ge 0i=1,2,\mathrm{.Math.}N$ C is a penalty coefficient, and ξ.sub.i is a slack variable; assuming that the solutions of the constrained optimization problem are * and b*; transforming the original problem into a dual problem, and using KKT conditions to obtain an optimal solution of the dual problem, then: $w^{*}=\underset{i=1}{\overset{N}{.Math.}}{\alpha }_i^{*}{y}_i{x}_i$ $b^{*}=-\frac{{\max }_{i=y_i=-1}{w}^{*}{x}_i+{\min }_{i=y_i=-1}{w}^{*}{x}_i}{2}$ wherein α* is a solution of the dual problem in a Lagrange multiplier vector; step 2: 2.1 health degree of a CNC machine tool: health degree J of a CNC machine tool is a number between 0 and 1; the greater or the closer to 1 the value is, the more healthy the operating state is; the smaller or the closer to 0 the value is, the more unhealthy the operating state is and the greater the possibility of failure is; the health degree J of the CNC machine tool can be calculated by the following steps: $\mathrm{Lc}=\{\begin{array}{cc}\frac{w^{*}{x}_0+b^{*}}{.Math.w^{*}.Math.}& w^{*}{x}_c\ge w^{*}{x}_0\\ \frac{w^{*}{x}_c+b^{*}}{.Math.w^{*}.Math.}& w^{*}{x}_c<w^{*}{x}_0\end{array};L_0=\frac{w^{*}{x}_0+b^{*}}{0.95.Math.w^{*}.Math.};J=\frac{Lc}{L_0}$ wherein x.sub.0 is a data vector of a normally operating CNC machine tool calibrated by experts, and x.sub.c is a data vector of a current CNC machine tool obtained by on-line monitoring; L.sub.0 is an equivalent distance of the health degree of the calibrated CNC machine tool, and L.sub.c is an equivalent distance of the health degree of the current CNC machine tool; 2.2 determination of predictive maintenance alarm threshold a failure possibility index p means the possibility of failure; $p=\frac{2e^{-J}}{e^J+e^{-J}}$ when p>r, alarm is given to prompt that the CNC machine tool needs maintenance, wherein r is a predictive maintenance alarm threshold.

2. The method for diagnosing health of a CNC machine tool of claim 1, wherein in step 1, vibration signal AD conversion sampling frequency f.sub.c of the vibration sensor node is set to different values according to the change of the health degree: $f_c=\frac{f_0}{N};$ $N=\{\begin{array}{cc}{N}_0& J\ge 0.5\\ 2N_0& J<0.5\end{array};$ wherein f.sub.c: vibration signal AD conversion sampling frequency; f.sub.0: crystal oscillator frequency; N: frequency dividing ratio of variable frequency divider; and N.sub.0: reference value of frequency dividing ratio.

3. The method for diagnosing health of a CNC machine tool of claim 1, wherein the value of the predictive maintenance alarm threshold r is set to be 0.7-0.9.

4. The method for diagnosing health of a CNC machine tool of claim 1, wherein in step 1.3, the penalty coefficient C=0.35.

5. The method for diagnosing health of a CNC machine tool of claim 3, wherein in step 1.3, the penalty coefficient C=0.35.

6. A system adopted in the method for diagnosing health of a CNC machine tool of claim 1, comprising a Zigbee gateway, a cloud server and sensor nodes; detected data is transmitted by the Zigbee gateway to the cloud server for digital signal processing, data mining and analysis processing; after a training test of the above data is carried out by the cloud server, a health degree model is obtained; the health degree model is transmitted by the cloud server to the Zigbee gateway; a clock circuit can be used for synchronizing system time and recording time as data is recorded; and the health degree of the CNC machine tool is predicted in real time by the Zigbee gateway according to the vibration and temperature data monitored by the sensor nodes and in combination with the model obtained by training.

7. The system of claim 6, wherein the sensor nodes include a vibration sensor node; the vibration sensor node is composed of a DSP TMS320C6748 processor of Texas Instruments (TI), a CC2530 chip of TI, an A/D converter, a vibration sensor, an SD card storage module, a communication unit and a power supply module, and is used for collecting and storing mechanical vibration signals and sending the collected data to the Zigbee gateway in a wireless transmission mode.

Description

DESCRIPTION OF DRAWINGS

[0036] FIG. 1 is a structural schematic diagram of a system of the present invention.

[0037] FIG. 2 is a circuit structural schematic diagram of a vibration sensor node of the present invention.

DETAILED DESCRIPTION

[0038] To make the summary of the present invention easier to understand clearly, the present invention is further described below in details according to specific embodiments and in combination with drawings.

[0039] A method for diagnosing health of a CNC machine tool, comprising the following steps:

[0040] Step 1: using a cloud server to carry out FFT transformation processing on vibration data collected by a vibration sensor node, and obtaining spectrum averages of 800-1800 Hz, 1800-2800 Hz and 2800-3800 Hz;

[0041] Vibration signal AD conversion sampling frequency f.sub.c of the vibration sensor node is set to different values according to the change of the health degree:

[00006]$f_c=\frac{f_0}{N};$$N=\{\begin{array}{cc}{N}_0& J\ge 0.5\\ 2N_0& J<0.5\end{array};$

Wherein f.sub.c: vibration signal AD conversion sampling frequency;

[0042] f.sub.0: crystal oscillator frequency;

[0043] N: frequency dividing ratio of variable frequency divider; and

[0044] N.sub.0: reference value of frequency dividing ratio.

[0045] Obtaining a regression coefficient of relationship among the spectrum average of a relevant frequency band, temperature and operating state of equipment according to corresponding historical data of the spectrum average of the frequency band and temperature;

[0046] 1.1 Creating a data vector

x=(x.sup.(1),x.sup.(2),x.sup.(3),x.sup.(4))

[0047] Wherein x.sup.(1) is the spectrum average of 800-1800 Hz, x.sup.(2) is the spectrum average of 1800-2800 Hz, x.sup.(3) is the spectrum average of 2800-3800 Hz, and x.sup.(4) is a temperature average of equipment;

[0048] 1.2 Creating a coefficient vector w=(w.sup.(1), w.sup.(2), w.sup.(3), w.sup.(4)), wherein w.sup.(1) is a coefficient of the spectrum average of 800-1800 Hz; w.sup.(2) is a coefficient of the spectrum average of 1800-2800 Hz; w.sup.(3) is a coefficient of the spectrum average of 2800-3800 Hz; and w.sup.(4) is a temperature coefficient;

[0049] 1.3 x.sub.i is the i.sup.th trained data vector, and y.sub.i is a class marker of x.sub.i; When y.sub.i is −1, it means that the equipment fails; when y.sub.i is +1, it means that the equipment is normal; and N is a quantity of trained data;

[0050] Finding a separating hyperplane with a maximum geometric margin, and a problem is expressed as the following constrained optimization problem:

[00007]$\underset{w,b,\xi }{\min }\frac{1}{2}{.Math.W.Math.}^2+C\underset{i=1}{\overset{N}{.Math.}}{\xi }_i$$S.t{y}_i\left(w.x_i+b\right)\ge 1-{\xi }_i{\xi }_i\ge 0i=1,2,\mathrm{.Math.}N$

[0051] C is a penalty coefficient, C=0.35, and ξ.sub.i is a slack variable;

[0052] Assuming that the solutions of the constrained optimization problem are * and b*;

[0053] Transforming the original problem into a dual problem, and using KKT conditions to obtain an optimal solution of the dual problem, then:

[00008]$w^{*}=\underset{i=1}{\overset{N}{.Math.}}{\alpha }_i^{*}{y}_i{x}_i$$b^{*}=-\frac{{\max }_{i=y_i=-1}{w}^{*}{x}_i+{\min }_{i=y_i=-1}{w}^{*}{x}_i}{2}$

[0054] Wherein α* is a solution of the dual problem in a Lagrange multiplier vector;

[0055] Step 2:

[0056] 2.1 Health degree of a CNC machine tool:

[0057] Health degree J of a CNC machine tool is a number between 0 and 1; the greater or the closer to 1 the value is, the more healthy the operating state is; the smaller or the closer to 0 the value is, the more unhealthy the operating state is and the greater the possibility of failure is;

The health degree J of the CNC machine tool can be calculated by the following steps:

[00009]$\mathrm{Lc}=\{\begin{array}{cc}\frac{w^{*}{x}_0+b^{*}}{.Math.w^{*}.Math.}& w^{*}{x}_c\ge w^{*}{x}_0\\ \frac{w^{*}{x}_c+b^{*}}{.Math.w^{*}.Math.}& w^{*}{x}_c<w^{*}{x}_0\end{array};L_0=\frac{w^{*}{x}_0+b^{*}}{0.95.Math.w^{*}.Math.};J=\frac{Lc}{L_0}$

[0058] Wherein x.sub.0 is a data vector of a normally operating CNC machine tool calibrated by experts, and x.sub.c is a data vector of a current CNC machine tool obtained by on-line monitoring;

[0059] L.sub.0 is an equivalent distance of the health degree of the calibrated CNC machine tool, and

[0060] L.sub.c is an equivalent distance of the health degree of the current CNC machine tool;

[0061] 2.2 Determination of predictive maintenance alarm threshold

[0062] A failure possibility index p means the possibility of failure;

[00010]$p=\frac{2e^{-J}}{e^J+e^{-J}}$

[0063] When p>r, alarm is given to prompt that the CNC machine tool needs maintenance, wherein r is a predictive maintenance alarm threshold. The value of the predictive maintenance alarm threshold r is set to be 0.7-0.9.