TWI776606B - A method for parameter optimization and through-hole detection of an edm hole drilling - Google Patents

Abstract

The present invention provides a method for parameter optimization and through-hole detection of an EDM Hole Drilling. The steps include: making the EDM Hole Drilling carry out a through-hole process on a workpiece; capturing the discharge voltage and discharge current while processing a through-hole on a workpiece before/after the hole formed by the EDM Hole Drilling; establishing a detection model of a one-dimensional Convolutional Neural Network, making the said discharge voltage and the said discharge current as the input layer of the one-dimensional Convolutional Neural Network; furthermore, defining the output layer of the one-dimensional Convolutional Neural Network whether the EDM Hole Drilling has processed a through hole on the workpiece, so as to train the detection model. By means of it, the present invention can automatically detect and control whether the EDM Hole Drilling carry out a through-hole process or not, preventing the insufficient electrode decline from unsuccessful through-hole, or effecting the underneath object due to the excessive decline, which can improve the accuracy and efficiency of the EDM Hole Drilling.

Description

Parameter optimization of deep hole EDM and detection method of through hole

The present invention provides a method for optimizing parameters of a deep-hole electrical discharge machine and detecting a through hole, especially a method for automatically detecting whether a deep-hole electrical discharge machine has a through hole through a detection model of a one-dimensional convolutional neural network. .

According to the current trend of high-tech products with complex structures, special materials, and thin thicknesses, compared with ordinary drilling tools, deep-hole electrical discharge machining can melt metal through spark discharge, which is not affected by material strength, toughness and hardness. It has the advantages of impact and can be processed on the curved inclined surface that is difficult to be processed by traditional drilling.

The main purpose of the deep hole electric discharge machine is through holes. Because the electric discharge machining parameters, electrode materials and workpiece materials of the deep hole electric discharge machine all have a certain effect on the drilling processing time, drilling area, drilling circularity (Circularity) and electrode consumption. The influence of drilling will also affect the judgment of the voltage and current characteristics of the through hole, and aerospace processing has high requirements on the performance indicators of the drilling. If the performance indicators after processing can meet expectations, the relationship between the research parameters and the performance results must be It becomes very important; among them, in the processing of blind holes, since the drilling only needs to penetrate the first few layers and does not need to penetrate the entire workpiece, backlash occurs. However, in electric discharge machining, the processed electrodes will follow EDM parameters, materials, and electrode length changes affect the performance of the electrode. Consumption makes the electrodes consumed for each processing different, and the height of the feed shaft falling is also different. As a result, the current manufacturers rely on processing experience to set the fixed height of the feed shaft. However, the method often fails to penetrate the hole or penetrate to the bottom. The situation of the first layer occurs; therefore, the automatic through hole inspection of deep hole electric discharge machining through automation is an important issue to improve the through hole machining accuracy.

In view of this, our inventors have devoted themselves to further research on automatic through-hole inspection of deep-hole electrical discharge machining, and started to develop and improve, hoping to solve the above problems with a better invention, and after continuous testing and modification, there is a problem. The advent of the present invention.

In other words, the purpose of the present invention is to solve the aforementioned problems. In order to achieve the above purpose, our inventors provide a method for optimizing parameters of a deep-hole EDM machine and detecting a through hole, the steps of which include: making a deep-hole EDM The device performs a through-hole process on a workpiece; when the deep-hole electrical discharge machining device is in the through-hole process, the discharge voltage and discharge current before and after the workpiece is penetrated into a hole; and a one-dimensional convolution neural network is established for detection model, and set the discharge voltage and the discharge current as the input layer of the one-dimensional convolutional neural network, and define the output layer of the one-dimensional convolutional neural network as whether the deep hole electrical discharge machining device has been Through holes are generated in the workpiece to train the detection model.

According to the above-mentioned method for optimizing parameters of a deep-hole electric discharge machine and detecting a through hole, the steps further include: defining a repeated trigger index; and continuously judging from the detection model that the deep-hole electric discharge machining device has produced a through hole on the workpiece. When the number of holes is greater than the repeated triggering index, it is defined that the deep hole electrical discharge machining device has indeed produced through holes in the workpiece.

According to the above-mentioned parameter optimization of deep-hole electric discharge machine and through-hole detection method, the repeated trigger index is defined as 1 to 25.

According to the above-mentioned parameter optimization of deep hole electric discharge machining machine and through hole detection method, wherein, during the through hole process of the deep hole electric discharge machining device, during a first period before the through hole is generated in the workpiece, and in a second period after the moment when the through hole is generated, the discharge voltage and the discharge current are respectively captured at an interval, and input into the detection model.

According to the above-mentioned parameter optimization of deep-hole electric discharge machine and through-hole detection method, the interval time is 0.025 seconds to 0.1 seconds.

According to the above-mentioned parameter optimization of the deep-hole electric discharge machine and the detection method of the through hole, the sampling rate for capturing the discharge voltage and the discharge current is between 50 kHz and 200 kHz.

According to the above-mentioned method for parameter optimization of deep-hole electric discharge machine and detection method for through-hole, the steps further include: inputting the discharge voltage and the discharge current of the through-hole process into the detection model after training, so that The detection model judges whether the deep hole electrical discharge machining device has produced through holes in the workpiece.

From the above description and settings, it is obvious that the present invention mainly has the following advantages and effects, which are described in detail as follows:

1. In the present invention, through the establishment and training of a one-dimensional convolutional neural network detection model, it can accurately detect whether the through-hole of the deep-hole electrical discharge machining device is not, so that the deep-hole discharge can be set without visual inspection or relying on experience. The descending height of the processing electrode can improve the efficiency of the through hole, and when processing blind holes, it can also ensure the through holes of a specific layer, and it is not easy to generate unrealized through holes, and the electrode is excessively lowered and causes the object The problem of backflushing caused by wrong through holes can effectively improve the accuracy and efficiency of blind hole processing.

2. The present invention can prevent the detection model from erroneously judging as a through hole before the through hole through the definition of the repeated trigger index, and the detection model continuously judges that the deep hole electric discharge machining device has produced through holes in the workpiece more times than the workpiece. When the indicator is repeatedly triggered, it is determined that the deep hole electrical discharge machining device has indeed Through holes are formed in the workpiece, so as to improve the overall detection accuracy of the present invention.

1: Electrode

2: Workpiece

S001~S003: Steps

Figure 1 is a flow chart of the present invention.

Fig. 2 is a schematic diagram of the moment when the electrode penetrates the workpiece in the through-hole process of the present invention.

Fig. 3 is a schematic diagram of the electrode completely penetrating the workpiece in the through-hole process of the present invention.

FIG. 4 is a schematic diagram of the configuration of the multi-channel data acquisition device of the present invention.

FIG. 5 is an experimental diagram of the discharge voltage and discharge current of the deep-hole electrical discharge machining device actually captured by the present invention.

FIG. 6 is an architectural diagram of a one-dimensional convolutional neural network of the present invention.

FIG. 7 is a schematic diagram of the detection and training structure of the present invention.

FIG. 8 is a waveform diagram of discharge voltage and discharge current captured by the electrical discharge machining of the present invention.

Fig. 9 is a confusion matrix diagram of the present invention.

Fig. 10 is the confusion matrix of the training set with a sampling rate of 200 kHz according to the present invention.

Fig. 11 is the confusion matrix of the validation set with a sampling rate of 200 kHz according to the present invention.

Fig. 12 is the confusion matrix of the test set with a sampling rate of 200 kHz according to the present invention.

Fig. 13a is a signal detection diagram of the present invention before the repeated trigger indicator is applied.

Fig. 13b is a signal detection diagram of the present invention applying the repeated trigger indicator.

Figure 14a is a schematic diagram of the present invention with an interval of 0.1 seconds.

Figure 14b is a schematic diagram of the present invention with an interval time of 0.05 seconds.

Fig. 15 is the confusion matrix of the training set with a sampling rate of 100 kHz according to the present invention.

Fig. 16 is the confusion matrix of the validation set with a sampling rate of 100 kHz according to the present invention.

Fig. 17 is the confusion matrix of the test set with a sampling rate of 100 kHz according to the present invention.

Fig. 18 is the confusion matrix of the training set with a sampling rate of 50 kHz according to the present invention.

Fig. 19 is the confusion matrix of the validation set with a sampling rate of 50 kHz according to the present invention.

Fig. 20 is the confusion matrix of the test set with a sampling rate of 50 kHz according to the present invention.

FIG. 21 is a comparison diagram of each sampling rate and repeated trigger index of the present invention.

Regarding the technical means of our inventors, several preferred embodiments are described in detail below together with the drawings, so as to provide the readers with an in-depth understanding and approval of the present invention.

Please refer to Fig. 1 first, the present invention is a method for parameter optimization and through-hole detection of a deep-hole electric discharge machine. The steps include:

Step S001: Make a deep hole electric discharge machining device, through its electrode 1, perform a through-hole process on a workpiece 2; The test is not limited by this; and the through-hole process of the deep-hole electric discharge machining device is mainly by placing the electrode 1 configured by itself, and the workpiece 2 to be processed on the positive and negative electrodes, respectively, and dielectric The liquid is continuously washed, or placed in the dielectric liquid, and then a DC power supply is applied, and the amplitude of the shaking of the electrode 1 is fixed through the eye mold, and when the positive and negative electrodes are close to each other, the dielectric liquid will be ionized and sparks will be generated Discharge, by which the workpiece 2 is melted and evaporated at high temperature, and the molten metal residue is removed by the gasification and expansion of the dielectric liquid caused by the dielectric liquid sprayed in the electrode tube and processing, and then the insulation state is restored. Maintaining the spark discharge state makes the energy concentrated, so that the material of the workpiece 2 can be accurately removed for drilling processing. The operation method and principle belong to the conventional technology in the field of deep hole electrical discharge machining, so they will not be repeated here.

Step S002 : capturing the discharge voltage and discharge current of the deep hole electrical discharge machining device before and after the hole is formed through the workpiece 2 during the through hole process, wherein the deep hole electrical discharge machining device performs the through hole process on the workpiece 2 , as follows As shown in Figures 2 and 3, when the electrode 1 penetrates the workpiece 2, there will be obvious sparks ejected from the bottom of the workpiece 2, but at this time, the electrode 1 is in the shape of a cone due to the machining relationship, as shown in the second As shown in the figure, the drilling at this time is smaller than that when the through hole is completed, and the through hole is not completed until the electrode 1 protrudes from the bottom of the workpiece 2 by a length, as shown in Figure 3.

For the extraction of discharge voltage and discharge current in the through-hole process of the deep-hole EDM device, due to the characteristics of deep-hole EDM, when the voltage is open, the voltage should be in an open state, so the discharge current at this time should be However, in fact, there is still a current of about 18 to 40 amperes. Therefore, the extraction of the discharge voltage and the discharge current is not the extraction of the voltage and current input to the deep-hole electrical discharge machining device. Therefore, in one embodiment, The through-hole discharge voltage and discharge current can be collected through a multi-channel data acquisition device (DAQ). The configuration is shown in Figure 4. The discharge circuit of the deep-hole EDM device is connected in parallel through a differential probe. , and its current probe is sheathed in the discharge loop, so that the differential probe is connected to the multi-channel data acquisition device to measure the discharge voltage, and the current probe is connected to the voltage attenuator to measure the discharge current. In the example, the multi-channel data acquisition device is NI-USB-6361, which uses a sampling rate of 200kHz for two channels, and the input voltage ranges from +10V to -10V; the software DAQ-EXPRESS is used to record the data through the computer, and the differential probe The rod is PINTEK DP-25 (attenuation rate 20), the current probe is Tektronix TCP-303 with amplifier TCPA-300 (5A/V), plus voltage attenuator PICO TA-050 (use attenuator 3db in series with 10db ), and use the computer to capture the 10Hz coordinate information from the controller through the network route, and record the FPS60 experiment through the camera.

Therefore, the actual captured discharge voltage and discharge current of the deep-hole EDM device are shown in Fig. 5, where the blue line is the voltage, and the orange line is the current.

Step S003: Establish a detection model of a one-dimensional convolutional neural network, the structure of which is as shown in Fig. 6, and use the discharge voltage and the discharge current as the input of the one-dimensional convolutional neural network layer, and the output layer that defines the one-dimensional convolutional neural network is whether the deep-hole electrical discharge machining device has produced through-holes in the workpiece 2, so as to train the detection model. The detection structure is shown in Figure 7. ; Thereby, it can be set that when the detection model judges that through-holes are generated, the deep-hole electric discharge machining device stops processing; in this embodiment, the used data training set is 97200, the verification set is 48600, and the test set 15,000 pieces of data were processed for parameters that have not been trained at all, and the one-dimensional convolutional neural network architecture was trained using 200kHz data as shown in Table 1 below:

Regarding the processing of the discharge voltage and discharge current data, in one embodiment, a voltage threshold is set. In this embodiment, the voltage is set to be greater than 19, so that the normal discharge data can be changed from one record to one including discharge and undischarged. cut out from the data, and then extract the data by comparing the coordinates of 10Hz (that is, transparent Through the electrode consumption, the coordinates of the through-hole can be reversed) and the processing image record of 60FPS, and the moment when the blind-hole through-hole is obtained and the time when the electrode through-hole touches the next layer (distance about 5mm), the marking error is about 0.2 seconds. Due to the disparity in the number ratio of data before and after the perforation, in the first period before the perforation, the average length of the data is 78 seconds, and in the second period after the instant of the perforation, the average time is 12 seconds, which are balanced training data. It takes a period of time to see the change with the naked eye after the through hole, so only the last 80% of the through hole data is taken; and for the sampling interval of the discharge voltage and the discharge current, in one embodiment, each A total of 100 strokes are taken every 0.1 seconds in the first 10 seconds of processing the through hole, and a total of 100 strokes are marked as non-through holes, and a total of 100 strokes are taken every 10 seconds after the through hole to the last 80 seconds of the through hole, and a total of 100 strokes are marked as through holes. If the non-through-hole or through-hole data of the pen data is less than ten seconds, change the reduction interval to obtain 100 pieces of data within this interval. As shown in Figure 8, 486 pieces of processing are trained and 243 pieces of processing are used for verification. Then use 75 data as the test, take the voltage and current data every 20,000 data points as the input data, and convert it to half-precision (float16) data, which can reduce the data size to a quarter, which is not suitable for training. No difference, but can reduce the amount of memory required.

The results of its training with 200kHz data are shown in Table 2 below:

The confusion matrix is shown in Figure 9, where the sensitivity (Sensitivity) represents the ratio of judging the pre-via signal as a non-via, which is TP/(TP+FN) shown in Figure 9, and if the sensitivity If the degree is low, it is easy to stop the processing before the through hole, so that the through hole process cannot be completed smoothly; ), since the electrode 1 only needs to penetrate the first layer of the workpiece 2 in the blind hole processing, without drilling to the second layer, because there is a buffer space with a certain distance from the first layer to the second layer, a certain height is required. The specificity, therefore, preferably, the sensitivity should be as high as possible, so that it will not stop before the through hole, and has a large specificity; and the training set and the verification set with a sampling rate of 200kHz in this example The confusion matrix with the test set is shown in Figures 10, 11, and 12; it can be seen that the training results of the one-dimensional convolutional neural network detection model are very good, with 97% on the validation set with the same parameters The accuracy rate is high, which shows that the present invention can indeed detect whether the deep-hole electrical discharge machining device is indeed a through hole according to the characteristics obtained from the data of the discharge voltage and the discharge current.

Therefore, after the above-mentioned training, the discharge voltage and the discharge current of the through-hole process can be input into the trained detection model, so that the detection model can judge whether the deep-hole electric discharge machining device has produced the workpiece 2 . Through hole.

Since the sensitivity is 97%, it is still possible to judge the non-through hole state as a through hole and stop the processing. It is necessary to add multiple consecutive detections to the through hole judgment before it is judged as a through hole and the processing is stopped. Therefore, , in a preferred embodiment, it is possible to define a repeated trigger index; and when the detection model continuously judges that the number of times that the deep hole electrical discharge machining device has generated through holes in the workpiece 2 is greater than the repeated trigger index, define the repeated trigger index. The deep hole electrical discharge machining device has indeed produced through holes in the workpiece 2 .

As mentioned above, it is the result of the training of 1000 pieces of data at most 10 seconds before and after the input of the through hole. In one embodiment, for the input of the discharge voltage and discharge current signals, the deep hole electric discharge machining device can be considered for the discharge in advance. The five factors of duration, discharge current, discharge gap, servo feed and electrode length are used to organize training of processing parameters to optimize the performance of deep-hole electrical discharge machining equipment and improve its processing accuracy. Each group was carried out three times, two of which were used as training data and one was used as verification data, with a total of 729 pieces of training data, and 75 pieces of data from the untrained processing parameter group were added for comparison, so as to compare the actual 729 and 75 pieces of complete data at every interval. The discharge voltage and discharge current signals with a length of 20,000 are captured in 0.1 second and input into the trained one-dimensional convolutional neural network. On average, the through-hole can be detected 0.5 seconds after the occurrence of the through-hole; while the non-through-hole is detected as Through-holes are more likely to occur in parameters with slow processing time, and the probability of non-through-holes being detected as through-holes is 0.1121. Therefore, by setting the repeated trigger index N _b , the probability of occurrence can be reduced; as shown in Figures 13a and 13b As shown, the orange part represents the time axis before the through-hole, while the blue represents the time axis after the through-hole, the black point represents whether the detection model judges it as a through-hole within the unit time, and the black line is the touch to the first. The time on the second floor, Figure 13a is the signal detection diagram before the repeated trigger indicator has been applied, and Figure 13b is the signal detection diagram with the repeated trigger indicator applied. The correct number of strokes detected before the through hole represents the time axis on the orange line There are no black dots on the data, and the data detected only after touching the second layer means that the black dots appear after the black line; the results of the 729 processing data are shown in Table 3 below, and the processing parameters are not trained for 75 data. The test results of the complete processing data of the combination are shown in Table 4 below:

From Table 3 and Table 4, it can be seen that the increase of the repeated trigger index does improve the sensitivity and accuracy, but it will reduce the specificity, and cause the delay of the judgment time of the through hole, which will increase the time when the second layer is encountered and cause backflushing. To judge the probability of stopping the processing of the through hole, it is necessary to choose the repeated trigger index carefully. Preferably, the repeated trigger index is defined between 1 and 25, so that the sensitivity is high enough, and the processing will not be stopped early before the through hole. And after the through hole, it can be quickly determined that it is a through hole and the processing can be stopped.

In the aforementioned, through-hole detection is performed by inputting a signal with a length of 0.1 second to the detection model every 0.1 seconds. In one embodiment, the interval time for judgment is reduced. As shown in Figure 14a, the horizontal axis represents the input signal. On the time axis, the yellow circle represents the data of the input detection model for through-hole detection. At this time, the judgment is performed every 0.1 seconds. In Figure 14b, the judgment interval is reduced to 0.05 seconds. It can be seen that the signal length of the input model is still 0.1 seconds but the interval It is only 0.05 seconds; therefore, the interval of 0.1 seconds is changed to 0.05 seconds for judgment, and the repeated trigger indicator is added. The results of the complete processing data of 729 processed processing parameter groups are shown in Table 5 below, and 75 are not trained. The processing data test results of the over-processing parameter group are shown in Table 6 below:

The judgment interval was further reduced to 0.025 seconds for through-hole detection and repeated trigger indicators were added. The results of 729 complete processing data with trained processing parameter groups are shown in Table 7 below, and the complete data of 75 untrained processing parameter groups. As shown in Table 8 below:

It can be seen from the above that the interval time is better, it can be set between 0.025 seconds and 0.1 seconds, and when the judgment interval is reduced to 0.025 seconds, the parameters that have obvious misjudgments in the through hole detection can be eliminated, so as to improve the correct detection. Therefore, it is better to use a detection model with a judgment interval of 0.025 and a repeated trigger index N _b =21, which can detect the through hole within the delay time of 0.6 seconds after the through hole. If the time prediction model is used, it will improve the its accuracy.

Due to the high price of the 200kHz current sensor and the large files captured, the training model requires a huge memory. In order to reduce the cost, in this embodiment, the sampling rate is reduced to reduce the cost of the current sensor and the acquisition of data. Take the data size to test the accuracy of judging through holes.

In one embodiment, the 100kHz data training set is 194,000, the validation set is 97,200, and the test set is 30,000. The one-dimensional convolutional neural network architecture is shown in Table 9 below:

The training results are shown in Table 10 below:

The confusion matrix is shown in Figures 15, 16, and 17, where 0 is no through hole, 1 is through hole, the vertical axis is the marked data, and the horizontal axis is the prediction data.

Next, the results of judging using 729 complete data of the 100kHz trained processing parameter group are shown in Table 11 below, and the results of judging 75 complete data of the untrained processing parameter group are shown in Table 12:

Then, the sampling rate is reduced again, using 50kHz data for the training set of 388,000, the validation set of 194,400, and the test set of 60,000. Its one-dimensional convolutional neural network architecture is shown in Table 13 below:

The training results are shown in Table 14 below:

The confusion matrix is shown in Figures 18, 19, and 20. Similarly, 0 means no through hole, 1 means through hole, the vertical axis is the marked data, and the horizontal axis is the prediction data.

The results of judging using 729 complete data of the 50 kHz trained processing parameter group are shown in Table 15 below, and the results of judging 75 complete data of the untrained processing parameter group are shown in Table 16 below.

Comparing the different sampling rates with the repeated trigger indicators, the comparison chart is shown in Figure 21. It can be observed that the data sensitivity and specificity at the 50kHz sampling rate are significantly lower than those at 200kHz and 100kHz, and the sampling rate is 50kHz. If the data of 100kHz is divided into the data of 200kHz, the amount of data is doubled, so the accuracy in the training set may be better than the data of 200kHz. It is B. The sampling rate for capturing the discharge voltage and the discharge current is It can be between 50kHz and 200kHz, but the sampling rate of 50kHz has a significant drop in the accuracy of the test set, and the proportion of data detected after touching the second layer has increased significantly, improving the accuracy of judgment and not in the through hole. On the premise that the pre-detection has not been completed, it is preferable to use a signal of 100 kHz and the repeated trigger index of 5 to perform the through-hole detection.

Since the surfaces of the electrode 1 and the workpiece 2 are usually not perpendicular to each other when deep hole EDM is applied in aerospace, they have different angles. In the present invention, the included angles between the workpiece 2 and the electrode 1 are 30 degrees, 45 degrees, and 60 degrees. The experimental results are shown in Table 17 below:

It can be seen from the experimental results that at various angles, the through-hole can be detected to stop processing after the through-hole. It is obvious that the present invention can indeed be applied to various angles. For the through-hole process of the workpiece 2, the through-hole can be automatically detected. or not.

To sum up, the technical means disclosed in the present invention can indeed effectively solve the problems of conventional knowledge, and achieve the expected purpose and effect, and it has not been published in publications before the application, has not been used publicly, and has long-term progress. The invention referred to in the Patent Law is correct, and the application is filed in accordance with the law.

However, the above are only several preferred embodiments of the present invention, which should not limit the scope of the present invention. It should still fall within the scope of the patent of the present invention.

S001~S003: Steps

Claims (6)

A method for optimizing parameters of a deep-hole electric discharge machine and detecting a through hole, the steps comprising: making a deep-hole electric discharge machining device perform a through-hole process on a workpiece; The discharge voltage and discharge current before and after the workpiece is penetrated into a hole; a detection model of a one-dimensional convolutional neural network is established, and the discharge voltage and the discharge current are input to the one-dimensional convolutional neural network layer, and the output layer that defines the one-dimensional convolutional neural network is whether the deep-hole electrical discharge machining device has produced through holes in the workpiece, so as to train the detection model; and define a repeated trigger index; and in the When the detection model continuously judges that the number of times that the deep hole electrical discharge machining device has produced through holes in the workpiece is greater than the repeated trigger index, it is defined that the deep hole electrical discharge machining device has indeed produced through holes in the workpiece. The parameter optimization of a deep-hole electric discharge machine and the detection method of a through hole according to claim 1, wherein the repeated trigger index is defined as one of 1 to 25. The parameter optimization of a deep hole electric discharge machine and the method for detecting a through hole according to claim 1 or 2, wherein the deep hole electric discharge machining device is a first step before generating a through hole in the workpiece during the through hole process. In a period, and in a second period after the moment when the through hole is generated, the discharge voltage and the discharge current are respectively captured at an interval time and input into the detection model. The parameter optimization of a deep-hole electric discharge machine and the detection method of a through hole according to claim 3, wherein the interval time is 0.025 seconds to 0.1 seconds. The parameter optimization of a deep-hole electric discharge machine and the detection method of a through hole according to claim 3, wherein the sampling rate for capturing the discharge voltage and the discharge current is between 50 kHz and 200 kHz. As claimed in claim 1 or 2, the parameter optimization of a deep-hole electric discharge machine and the method for detecting a through hole further comprise: inputting the discharge voltage and the through hole process into the trained detection model. The discharge current enables the detection model to determine whether the deep hole electrical discharge machining device has produced a through hole in the workpiece.

Images