A multi-axis current signal-based online fault monitoring method for numerical control machine tools

Abstract

The application discloses a kind of numerical control machine tool online fault monitoring methods based on multi-axis current signal: including the following steps: the current and vibration data of machine tool multi-axis drive motor are collected;Extract current and vibration signal characteristics, and the characteristics are screened;According to the characteristics of multi-axis data, establish multiple machine tool motor fault diagnosis models, integrate multiple diagnostic models into a more robust comprehensive evaluation model, and realize online monitoring of machine tool through the model obtained from historical data.The application fully considers the frequent working condition change of machine tool processing, multi-axis linkage processing problem.According to actual demand, the problem of insufficient robustness of single vibration signal under variable working condition is solved by introducing current signal and multi-axis signal regression, and the model can be quickly retrained during actual use, thereby continuously improving the expansibility and accuracy of machine tool fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of online monitoring of mechanical faults, and in particular to the problems of rapid changes in processing conditions and the complex structure and operation of CNC machine tools. It achieves accurate online monitoring of machine tools by introducing multi-axis motion information and current signals related to changes in working conditions. Background Technology

[0002] Online monitoring technology can track the operating status of machine tools in real time, identify potential problems, and take timely measures to avoid downtime. Effective monitoring can reduce malfunctions and errors in production, improve machine tool utilization, and ultimately increase overall production efficiency.

[0003] Vibration signals are particularly useful for rotating machinery, such as motors, fans, and pumps. These devices typically exhibit mechanical problems through vibration. Furthermore, vibration signals demonstrate high sensitivity to various faults in mechanical systems, such as bearing failure, gear failure, and imbalance. Abnormal vibration patterns can often indicate specific types of faults, and online monitoring systems frequently utilize vibration signals.

[0004] Due to the complex coupling relationships between the machine tool's mechanical structure, CNC system, and various control components, and the fact that machine tools involve multi-axis linkage and frequent speed changes during machining, accurate monitoring of the vibration signals of the mechanical system itself may be interfered with, making it difficult to monitor the machine tool's condition using traditional single vibration signals.

[0005] The effective value of a current signal can reflect changes in load during machining to some extent, while the rotational frequency can reflect changes in rotational speed. In the absence of clear machining conditions, current signals can approximate changes in operating conditions, making them ideal for machine tools, which experience significant variations in these conditions. Furthermore, machine tools have complex coupling relationships; a single machining process may involve multi-axis combined motions and single-axis motions. Therefore, considering the operating states of multiple axes for joint analysis can achieve higher monitoring accuracy. Summary of the Invention

[0006] The problem that this invention aims to solve is: In view of the above-mentioned problems, this invention combines multi-axis current signals for real-time online monitoring of machine tools, which can achieve more accurate monitoring results.

[0007] The technical solution adopted in this invention is as follows: an online fault monitoring method for CNC machine tools based on multi-axis current signals, the method comprising the following steps:

[0008] (1) Using a vertical four-axis CNC milling machining center as the analysis object, vibration and current signals of each motor were collected;

[0009] (2) The spectrum of the current signal is obtained by fast Fourier transform (FFT), the spectral features of the current signal are extracted, and the time-domain features of the current signal and the vibration signal are extracted at the same time.

[0010] (3) Determine if the data is in a shutdown state and delete the data in a shutdown state;

[0011] (4) Based on the current characteristics of all monitored motors in the historical data as input to multiple models, and the vibration signal characteristics of one shaft as the model output, the models are solved.

[0012] (5) Use the data from the training set to obtain the appropriate anomaly detection threshold for each model;

[0013] (6) After obtaining features from the real-time data in step (2), the data is input into the model in step (4) to obtain the predicted value.

[0014] (7) Compare the error between the predicted vibration value and the actual vibration value in all models. When all models exceed the threshold of step (5), an alarm is triggered, thereby achieving robust online monitoring of CNC machine tools.

[0015] (8) Update the model when abnormal data is detected and the abnormal data is in a normal state.

[0016] Furthermore, the vibration data acquisition parameters in step (1) are: the sampling unit is m / s. 2 The sampling rate is 25.6Hz. Current signal acquisition parameters are: sampling unit is A, sampling rate is 25.6Hz.

[0017] Furthermore, the spectral characteristics of the current signal extracted in step (2) are the frequencies corresponding to the maximum peak values ​​in the spectrum, and the time-domain characteristics are the effective values ​​(RMS). The time-domain characteristics of the vibration signal are also the effective values ​​(RMS).

[0018] Furthermore, in step (3), in order to reduce data redundancy, all data in which the effective value (RMS) of the shaft current is less than 0.005 are considered to be in a shutdown state and are deleted.

[0019] Furthermore, in step (4), random forest, decision tree, K nearest neighbor algorithm and linear regression model are used to solve the relationship between current and vibration signal. Through historical data, the current signal features of multiple axes are mapped to the vibration signal features including the main axis, X axis, Y axis and Z axis to obtain the corresponding axis model.

[0020] Further, in step (5), the error between the predicted effective value and the true effective value obtained from the training set is calculated using relative error. Based on the error in the training set, the anomaly detection threshold for each model is determined. The threshold in the training set error that satisfies the condition that five consecutive points do not exceed a certain threshold is used as the hard threshold. To increase robustness, the threshold is increased by 20% to determine the final anomaly detection threshold.

[0021] Further, in step (6), features of real-time data are extracted and the real-time data is judged. If the data is in a processing state, the predicted value of the effective vibration value is obtained by solving multiple models; otherwise, no calculation is performed.

[0022] Furthermore, in step (7), an alarm is triggered if the real-time data error exceeds the alarm threshold for five consecutive time periods.

[0023] Furthermore, if the model alarms in step (8), the machine needs to be stopped to check the machine tool status. If no fault is found, the data of the five time periods are saved in the historical database, and the model is retrained and updated.

[0024] The beneficial effects of this invention are: it fully considers the frequent changes in working conditions and the problem of multi-axis linkage machining during machine tool processing. Based on actual needs, it addresses the insufficient robustness of a single vibration signal under varying working conditions by introducing current signals and multi-axis signal regression. Furthermore, it allows for rapid retraining of the model during actual use, thereby continuously improving the scalability and accuracy of machine tool fault diagnosis. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the present invention;

[0026] Figure 2 This is a flowchart of the online monitoring model training process for this discovery;

[0027] Figure 3 This is a flowchart of the online monitoring system for this discovery;

[0028] Figure 4 The features extracted in this invention;

[0029] Figure 5 This is a flowchart of the abnormal threshold judgment process of the present invention;

[0030] Figure 6 The results are from the random forest model of this invention.

[0031] Figure 7 This is the regression result of the decision tree model in this invention;

[0032] Figure 8 This is the regression result of the K-nearest neighbor algorithm model of this invention;

[0033] Figure 9 The regression results are from the linear regression model of this invention; Detailed Implementation

[0034] The invention will be further illustrated below with an example of a spindle:

[0035] 1. This example uses a vertical four-axis CNC milling machining center as the analysis object to collect vibration and current signals from each motor. The sampling method is interval sampling, with each sample lasting 1 second and the sampling interval being 1 second. The vibration data acquisition parameters are: sampling unit is m / s². 2 The sampling rate is 25.6Hz. Current signal acquisition parameters are: sampling unit is A, sampling rate is 25.6Hz.

[0036] 2. The overall process of this method is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0037] Step 1: Extract features from the vibration and current data collected from each motor, such as... Figure 4 The features extracted in this invention;

[0038] The feature extraction process is as follows:

[0039]

[0040]

[0041] Where x is the signal, rms and maxfrequency are the extracted features, fs is the sampling frequency, and n is the signal length.

[0042] Step 2: Delete all data with effective values ​​of shaft current less than 0.005 to obtain the filtered features.

[0043] Step 3: Use random forest, decision tree, K-nearest neighbor algorithm and linear regression model to solve the relationship between current and vibration signal. Through historical data, map the current signal features of multiple axes to the vibration signal features including the principal axis, X axis, Y axis and Z axis to obtain the corresponding axis model.

[0044] Taking the principal axis as an example, the linear regression model is shown in the following equation:

[0045] VR m =A*ER m +B*ERF m +C*(ER m *ERF m )+D*ER x +E*ERF x +F

[0046] *(ERx *ERF x )+G*ER y +H*ERF y +I*(ER y *ERF y )+J*ER z +K

[0047] *ERFz z +L*(ER z *ERF z )

[0048] Among them, VR m This represents the effective value of the spindle vibration after regression; ER m Indicates the effective value of the current in the main axis; ERF m Indicates the spindle current frequency; subscripts indicate individual axes.

[0049] Step 4: Obtain the threshold for identifying abnormal data through the calculated model. For example... Figure 5 This is a flowchart of the abnormal threshold judgment process of the present invention;

[0050] (1) Calculate the relative error between the predicted effective value and the true effective value. The formula for calculating the relative error is:

[0051]

[0052] Where E i Let RMS be the error of the i-th model. pi Let RMS be the predicted value of real-time data in the i-th model. r This represents the true and valid values ​​of the real-time data. i = 1, 2, 3, 4, representing the random forest, decision tree, K-nearest neighbor algorithm, and linear regression model, respectively. Figures 6-9 The regression results are for random forest, decision tree, K-nearest neighbor algorithm, and linear regression model, respectively.

[0053] (2) According to the appendix Figure 5 Based on the calculation steps and the regression error results of the above four models, the threshold of each model is calculated, and the threshold is increased by 20% as the judgment threshold for real-time data.

[0054] Step 5: Extract the features of the real-time data and judge the real-time data. If the data is in a processing state, the predicted value of the effective vibration value is obtained by solving multiple models; otherwise, no calculation is performed.

[0055] Step 6: Issue an alarm based on the error obtained from real-time data. If all four models exceed the alarm threshold for five consecutive time periods, an alarm will be triggered. Otherwise, no alarm will be triggered if all four models do not exceed the threshold simultaneously.

[0056] Step 7: If the model triggers an alarm, the machine needs to be stopped to check its status. If no fault is found, the data from the five time periods are saved in the historical database, and the model is retrained and updated.

[0057] 3. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for online fault monitoring of CNC machine tools based on multi-axis current signals, characterized in that... Includes the following steps: (1) Using a vertical four-axis CNC milling machining center as the analysis object, vibration and current signals of each motor were collected; (2) The spectrum of the current signal is obtained by fast Fourier transform (FFT), the spectral features of the current signal are extracted, and the time-domain features of the current signal and the vibration signal are extracted at the same time. The spectral characteristics of the current signal extracted in this step are the frequencies corresponding to the maximum peak values ​​in the spectrum, and the time-domain characteristics are the effective values ​​(RMS). The time-domain characteristics of the vibration signal are also the effective values ​​(RMS). (3) Determine if the data is in a shutdown state and delete the data in a shutdown state; In this step, to reduce data redundancy, all data where the effective value (RMS) of the shaft current is less than 0.005, i.e., the data is considered to be in a shutdown state, are deleted. (4) Based on the current characteristics of all monitored motors in the historical data as input to multiple models, and the vibration signal characteristics of one shaft as the model output, solve all models. In this step, random forest, decision tree, K-nearest neighbor algorithm and linear regression model are used to solve the relationship between current and vibration signal. Through historical data, the current signal features of multiple axes are mapped to the vibration signal features including the principal axis, X axis, Y axis and Z axis to obtain the corresponding axis model. Taking the principal axis as an example, the linear regression model is shown in the following equation: ； in, This represents the effective value of the spindle vibration after regression. Indicates the effective value of the spindle current; Indicates the spindle current frequency; subscripts indicate individual axes; (5) Use the training set data to obtain the appropriate anomaly detection threshold for each model; In this step, the error between the predicted effective value and the true effective value obtained from the training set is calculated using relative error. Based on the error of the training set, the anomaly judgment threshold for each model is determined. The threshold in the training set error that satisfies the condition that no more than 5 consecutive points is used as the hard threshold. In order to increase robustness, the threshold is increased by 20% to determine the final anomaly judgment threshold. The formula for calculating the relative error is as follows: ； in Let be the error of the i-th model. Let i be the predicted value of the real-time data in the i-th model. This represents the true and valid value of the real-time data; (6) After obtaining features from the real-time data in step (2), the data is input into the model in step (4) to obtain the predicted value. This step extracts features from real-time data and judges the real-time data. If the data is in a processing state, the predicted value of the effective vibration value is obtained by solving multiple models; otherwise, no calculation is performed. (7) Compare the error between the predicted vibration value and the actual vibration value in all models. When all models exceed the threshold in step (5), an alarm is triggered, thereby achieving robust online monitoring of CNC machine tools. In this step, an alarm is triggered if the real-time data error exceeds the alarm threshold for five consecutive time periods. (8) Update the model when abnormal data is detected and the abnormal data is in a normal state; If the model triggers an alarm during this step, the machine needs to be stopped to check its status. If no fault is found, the data from the five time periods are saved in the historical database, and the model is retrained and updated.

2. The online fault monitoring method for CNC machine tools based on multi-axis current signals according to claim 1, characterized in that, The vibration data acquisition parameters in step (1) are: the sampling unit is m / s. 2 The sampling rate is 25.6Hz, and the current signal acquisition parameters are: the sampling unit is A, and the sampling rate is 25.6Hz.

Images