CN111331211B

CN111331211B - On-line penetration detection method and system for electric spark small hole machining

Info

Publication number: CN111331211B
Application number: CN201811566495.9A
Authority: CN
Inventors: 张亚欧; 夏蔚文; 赵万生
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2018-12-19
Filing date: 2018-12-19
Publication date: 2021-06-15
Anticipated expiration: 2038-12-19
Also published as: CN111331211A

Abstract

An on-line penetration detection method for electric spark small hole machining and a penetration detection system thereof collect original signals in the process of perforation machining, extract characteristic signals for preprocessing and generate a machining state diagram. When in a modeling mode, processing the processing state diagram which is manually classified by using a mode identification method to obtain a penetration judgment model, namely a classification model; and when detecting the mode, reading the penetration judgment model and classifying the machining state diagram obtained in real time by using a mode identification method so as to obtain a penetration judgment result. The characteristic that the processing state changes suddenly before and after the penetration phenomenon occurs is utilized, the processing process characteristic signals are processed in real time, the extracted characteristic signal change trend curve set in a certain time window is regarded as an image representing the processing state, and the image is classified on line by adopting a mode recognition algorithm to realize penetration detection.

Description

On-line penetration detection method and system for electric spark small hole machining

Technical Field

The invention relates to a technology in the field of machining, in particular to an electric spark small hole machining online penetration detection method and a penetration detection system thereof.

Background

In the electric spark small hole machining, the phenomenon that the electrode penetrates out of the other side or the inner cavity of the workpiece is called penetration, and generally, after the penetration, a section of the electrode needs to be fed continuously to ensure that the machining is finished. Since the electrode wear is large and the wear amount is variable during machining, the control system cannot determine whether or not penetration occurs from the feed amount.

At present, the common method is to set the processing depth, and automatically finish the processing after reaching the depth, but the complete penetration is difficult to ensure, and the phenomenon that the workpiece with the cavity is damaged and the back injury is generated due to overfeeding is avoided. The existing penetration detection method mainly sets a threshold value, compares signals such as discharge current, voltage and the like with the threshold value, and judges whether to penetrate according to the conditions of excess, deficiency or fluctuation. Such methods are highly dependent on the effectiveness of the threshold and the specific processing conditions.

Disclosure of Invention

The invention provides an online penetration detection method for electric spark small hole machining and a penetration detection system thereof aiming at the defects of the existing method, which utilize the characteristic that the machining state changes suddenly before and after the penetration phenomenon occurs to process the characteristic signal in real time in the machining process, take the extracted characteristic signal change trend curve set in a certain time window as an image representing the machining state, and adopt a mode recognition algorithm to classify the image online to realize penetration detection.

The invention is realized by the following technical scheme:

the invention relates to an on-line penetration detection method for electric spark small hole machining. When in a modeling mode, processing the processing state diagram which is manually classified by using a mode identification method to obtain a penetration judgment model, namely a classification model; and when detecting the mode, reading the penetration judgment model and classifying the machining state diagram obtained in real time by using a mode identification method so as to obtain a penetration judgment result.

The characteristic signals in the processing process include but are not limited to: (1) directly collectable signals: interelectrode current, voltage, feed depth, etc., (2) obtaining secondary signals after processing: effective discharge rate, short circuit rate, feed rate, etc.

The processing state diagram is as follows: the processing state diagram can comprehensively represent the change trend of the processing state within a period of time, is slightly influenced by the fluctuation of the processing state, external interference, measurement noise and the like, and the change of the processing state trend can not be changed by changing the processing condition. The application of the pattern recognition algorithm enables the classification modeling aiming at the processing state diagram to be completed by a computer program, so that the human intervention is reduced; the classification result is accurate, quick, stable and reliable; after a classification model is established by using a group of sample data, the method can be used for new and different data sample classifications, and has strong generalization capability. After the classification model is established according to typical processing conditions, the classification model can be directly applied to other various processing conditions without calibrating or adjusting parameters for each condition.

The pattern recognition is that:

wherein: f represents a pattern recognition process; y is the recognition result, expressed by an integer, and is the output of the pattern recognition process;

is a feature signal vector, contains a group of feature signals, and is input in the pattern recognition process; and m is a classification model which is used as a basis for pattern recognition and is input in a pattern recognition process.

The classification model is as follows:

wherein: y is a classification result, is expressed by an integer and is the output of the classification model;

is a feature signal vector, contains a set of feature signals, and is the input of the classification model.

Technical effects

Compared with the prior art, the penetration detection method provided by the invention has high accuracy; the processing parameters are adjusted within a reasonable range, detection parameters do not need to be changed, and the generalization capability is strong. Compared with other existing penetration detection methods, the method has remarkable advantages.

Drawings

FIG. 1 is a schematic diagram of an embodiment;

FIG. 2 is a schematic diagram of the internal modules of the control module;

FIG. 3 is a state diagram of the embodiment;

FIG. 4 is a schematic diagram illustrating penetration determination according to an embodiment;

in the figure: the device comprises a workbench 1, a processing workpiece 2, a lifting motor 3, a screw rod 4, a hollow tubular electrode 5, an electrode chuck 6, a grating ruler 7, a pulse power supply 8, a discharge gap 9, high-voltage working liquid 10, a current detection module 11, a voltage detection module 12 and a penetration detection system 13.

Detailed Description

As shown in fig. 1, an electric spark small hole machining apparatus according to the present embodiment includes a worktable 1, a workpiece 2, a lifting motor 3, a screw rod 4, a hollow tubular electrode 5, an electrode chuck 6, a grating ruler 7, a pulse power supply 8, a discharge gap 9, a high-voltage working fluid 10, a current detection module 11, a voltage detection module 12, and a penetration detection system 13, wherein: a processing workpiece 2 is placed on the workbench 1, and the lifting motor 3 is arranged on the screw rod 4 and is connected with the hollow tubular electrode 5 through the electrode chuck 6 to form a lifting mechanism which can control the lifting of the electrode 5. The lifting mechanism also comprises a grating ruler 7 for tracking the position of the current lifting mechanism, a pulse power supply 8 adds high-frequency voltage between an electrode 5 and a workpiece 2, discharge is generated in a discharge gap 9, workpiece material removal is realized, high-pressure working fluid 10 is sprayed into the discharge gap 9 from the inside of the electrode to play roles in chip removal and cooling, the continuous and stable proceeding of the machining process is realized, a current detection module 11 for collecting the current value passing through the discharge gap in real time is connected with the pulse power supply 8, a voltage detection module 12 for collecting the voltage values at two ends of the discharge gap in real time is connected with the electrode 5, a penetration detection system 13 is respectively connected with the current detection module 11, the voltage detection module 12, the grating ruler 7 and the lifting motor 3, and is used for receiving the lifting mechanism position signal from the grating ruler 7, the interelectrode current signal from the current detection module 11, and the intere, The inter-pole voltage signal from the voltage detection module 12 sets the sampling time window width and the sampling period.

The penetration detection system 13 includes: the device comprises a signal acquisition unit, a signal processing unit, a state diagram generation unit, a mode identification unit and a classification model unit, which are shown in figure 2. Wherein: the signal acquisition unit is respectively connected with the grating ruler 7, the current detection module 11 and the voltage detection module 12 to acquire distance signals, current information and voltage information, the signal processing unit is respectively connected with the signal acquisition unit and the state diagram generation unit to perform signal preprocessing, the state diagram generation unit is connected with the signal processing unit to acquire processed signal data, generate a state diagram sample and output the state diagram sample to the mode identification unit for establishing a classification model, namely, the mode identification unit which penetrates through the judgment model and implements classification outputs a classification result to the classification model unit according to the state diagram sample, and the classification model unit is used for storing and providing model data.

The specific function implementation and execution process in this embodiment are as follows:

the sampling period in this embodiment is 10 ms, the signal window width is 100 sampling periods, the original signal sampling rate is 100 khz, and the order of the low-pass filter is 30.

In each sampling period, the signal acquisition unit acquires three original signals of current, voltage and electrode position, and stores the three original signals in a buffer area, as shown in fig. 2. The amount of raw signal data within a single sampling period is related to the raw signal sampling rate. And after sampling in each period is finished, outputting the original signal data to the signal processing unit, emptying a buffer area, and starting data acquisition and storage in the next period.

In each sampling period, the signal processing unit firstly obtains original data from the signal acquisition unit, and then extracts a secondary signal from the original signal data, wherein the obtained secondary signal comprises: electrode feed rate, effective discharge rate and short circuit rate, normalizing the secondary signal to the [0,1] interval and using the secondary signal as a characteristic signal of the machining process. And filtering the characteristic signal by using a digital Finite Impulse Response (FIR) low-pass filter, and removing high-frequency interference and noise to obtain a smooth signal. The smoothed signal is predicted using a discrete-time Moving Average (MA) model, the length of the predicted period being equal to the order of the low-pass filter, the prediction being appended to the smoothed signal. As shown in fig. 3.

In each sampling period, the state diagram generating unit obtains the processed characteristic signal from the signal processing unit, the processed characteristic signal comprises the smooth signal and a prediction part thereof, namely a characteristic signal change trend curve, the change trend curves of the characteristic signal are combined to generate a processing state diagram, the length of the processing state diagram is the sum of the width of a signal window and the prediction length, and the height of the processing state diagram is 1.

The operation mode of the mode identification unit comprises the following steps: a modeling mode for obtaining a classification model by using the method and a detection mode for carrying out penetration detection by using the method in a machining process, wherein: in a modeling mode, in each sampling period, manually classifying the state diagram samples generated by the state diagram generating unit, recording each sample and a corresponding manual classification result thereof, and inputting the state diagram samples and the manual classification results into the mode recognition unit to generate a classification model and outputting the classification model to the classification model unit to be stored for use in a detection mode after obtaining a proper number of state diagram samples and manual classification results thereof; the pattern recognition unit obtains classification model data from the classification model unit during the detection mode, and obtains a processing state diagram sample from the state diagram generation unit, and the classification result is the output of the penetration detection system 13 as the basis for the automatic control of the processing process.

Each of the process state diagram samples may be classified by pattern recognition methods as pre-penetration and post-penetration, as shown in fig. 4, wherein the state diagram sample at penetration may be considered post-penetration at the time of classification.

The manual classification is as follows: after the processing is started, an operator marks the state pattern which is obtained in real time as a mark before the penetration; and (3) after the operator manually judges that the penetration occurs by observing the processing phenomenon, marking the state diagram sample obtained in real time as the penetration. The marking is realized by an operator through a change-over switch, and the switch is only required to be switched once in the whole machining process.

In this embodiment, each of the processing state diagrams is composed of 390 data, which can be regarded as a point in 390-dimensional space

And (4) showing. A sample set containing N processing state maps, with N points in 390-dimensional space, has been manually classified into two categories, labeled y-1 and y-1, representing pre-and post-penetration, respectively.

In this embodiment, the mode identification unit performs classification by using a support vector machine algorithm, specifically: finding an optimal hyperplane in 390-dimensional space, separating the two types of points, and maximizing the shortest distance between the two types of points and the hyperplane, wherein the hyperplane is

Wherein:

is a 390-dimensional vector, i.e., the normal vector of this hyperplane; b is a constant;

is a 390-dimensional vector, representing a point; k is a kernel function representing a vector point multiplication in a 390-dimensional or higher dimensional space.

The step of finding the optimal hyperplane is as follows: in all samples, minimize:

obtained by

And b is the classification model data.

The pattern recognition unit periodically obtains a sample after reading the classification model data in the detection mode

Carry over into the calculation of classification model to obtain

Samples are classified into two categories according to the negativity and the positivity of y.

In other conventional penetration detection methods, a current time value of one or more machining signals is generally used as a feature signal, and a threshold value is set for discrimination. Compared with the state diagram, the signal current time value is easily influenced by processing state fluctuation, external interference, measurement noise and the like, and has low accuracy and reliability on the representation of the processing state; and the recent state change trend can not be expressed, and the real processing state is difficult to reflect. The threshold is a predetermined single or several signal value criteria for comparison. Comparing the signal value obtained by each period with a threshold value, and obtaining a judgment result according to the comparison result; the threshold is set according to the experience of an operator or a system developer, and needs to be adjusted by repeated trial and error manually, so that the operability is poor, and the accuracy and the reliability are low; the threshold value is generally set according to a single processing condition, and changing the processing condition has a large influence on the threshold value and needs to be reset.

If the state diagram is used for representing the machining state and the threshold value is set for judgment, the threshold value is set for each datum with great difficulty due to more data in the state diagram, and an effective and reliable method is not available. If the machining state is represented by the signal value at the current moment and the mode identification method is used for judging, the accuracy of the judgment is low because the accuracy and the reliability of the representation of the machining state are low. The generation of the machine state diagram and the classification of the pattern recognition algorithm are therefore a core part of the invention.

Under the same processing conditions, the detection accuracy of the method of the invention and the threshold detection method is compared. The detection accuracy refers to the proportion of the number of correct samples in the total number of samples. Samples are the signal data obtained in each sampling period. The detection accuracy of the embodiment is more than 99%, and the detection accuracy of the threshold detection method is more than 70%.

The generalization ability of the method of the present invention was compared with that of the conventional detection method. The generalization ability refers to an ability to ensure the accuracy of determination without changing a threshold or a classification model after changing the processing conditions within a reasonable range that ensures normal processing. In the embodiment, the processing conditions are changed without modifying the classification model; the threshold detection method needs to adjust the threshold, and the number of the thresholds needing to be adjusted is related to specific processing conditions.

Comparison of the detection of the method of the invention with the conventional detection method takes time under the same processing conditions. The detection time consumption means the time spent in detecting the penetration phenomenon in the processing process, and the condition that the penetration is not accurately detected does not take into account statistical data. The detection time of the present embodiment is stabilized within 1 second. The detection time of the traditional detection method is also within 1 second. Both meet the real-time requirements of online detection.

As can be seen from the description of the embodiments, the method of the present invention has significant advantages in terms of accuracy, operability, and generalization ability compared to the threshold detection method.

The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Claims

1. An electric spark small hole machining on-line penetration detection method is characterized in that an original signal in a perforation machining process is collected, a characteristic signal is extracted for preprocessing, and a machining state diagram is generated;

when in a modeling mode, processing the processing state diagram which is manually classified by using a mode identification method to obtain a penetration judgment model, namely a classification model; when detecting the mode, reading the penetration judgment model and classifying the machining state diagram obtained in real time by using a mode identification method so as to obtain a penetration judgment result;

the characteristic signals in the processing process comprise: interelectrode current, voltage, feed depth, effective discharge rate, short circuit rate, feed rate;

the processing state diagram is as follows: selecting each characteristic signal in a preset window, preprocessing and drawing a corresponding change trend curve to form a processing state diagram;

the manual classification is as follows: in each sampling period, classifying the state diagram samples manually, and recording each sample and a corresponding manual classification result thereof;

the pattern recognition is that:

is a feature signal vector comprising a set ofA signature signal, which is an input to the pattern recognition process; m is a classification model which is used as a basis for pattern recognition and is input in a pattern recognition process;

the classification model is as follows:

2. A penetration detection system for implementing the method of claim 1, comprising: signal acquisition unit, signal processing unit, state diagram generation unit, mode identification unit and classification model unit, wherein: the signal acquisition unit is respectively connected with the grating ruler, the current detection module and the voltage detection module to acquire distance signals, current information and voltage information, the signal processing unit is respectively connected with the signal acquisition unit and the state diagram generation unit to perform signal preprocessing, the state diagram generation unit is connected with the signal processing unit to acquire processed signal data, generate a state diagram sample and output the state diagram sample to the pattern recognition unit, the pattern recognition unit outputs a classification result to the classification model unit according to the state diagram sample, and the classification model unit is used for storing and providing model data.

3. The system of claim 2, wherein in each sampling period, the signal acquisition unit acquires three original signals of current, voltage and electrode position and stores the three original signals in a buffer area; and outputting the original signal data to the signal processing unit after sampling of each period is finished, emptying a buffer area and starting data acquisition and storage of the next period.

4. The system of claim 3, wherein the signal processing unit obtains the raw data from the signal acquisition unit and extracts secondary signals including electrode feed rate, effective discharge rate, and short circuit rate during each sampling period, normalizes the signals to obtain process characteristic signals, filters the characteristic signals to obtain smoothed signals, predicts the smoothed signals by using a discrete time moving average model, and appends the prediction results to the smoothed signals.

5. The system according to claim 4, wherein the state diagram generating unit obtains the processed characteristic signals including the smoothed signal and the predicted part thereof, i.e. characteristic signal variation trend curves, from the signal processing unit in each sampling period, and combines the characteristic signal variation trend curves to generate the processing state diagram.

6. The system of claim 5, wherein the mode of operation of the pattern recognition unit comprises: a modeling mode for obtaining a classification model by using the method and a detection mode for carrying out penetration detection by using the method in a machining process, wherein: in a modeling mode, in each sampling period, manually classifying the state diagram samples generated by the state diagram generating unit, recording each sample and a corresponding manual classification result thereof, and inputting the state diagram samples and the manual classification results into the mode recognition unit to generate a classification model and outputting the classification model to the classification model unit to be stored for use in a detection mode after obtaining a proper number of state diagram samples and manual classification results thereof; the mode recognition unit obtains the classification model data from the classification model unit when detecting the mode, obtains a processing state diagram sample from the state diagram generation unit, and the classification result is the output of the control module and is used as the basis of the automatic control of the processing process.

7. The system of claim 6, wherein the pattern recognition unit is classified using a support vector machine algorithm, specifically: in a processing state diagram consisting of 390 data, an optimal hyperplane is found in a corresponding 390-dimensional space, two types of points are separated, and the shortest distance between the two types of points and the hyperplane is maximized, wherein the hyperplane is

Wherein:

8. The system of claim 7, wherein the pattern recognition unit periodically obtains a sample after reading the classification model data in the detection mode

Carry over into the calculation of classification model to obtain