TWI872895B - Quality assessment method for a stamping process - Google Patents

Abstract

A quality assessment method for a stamping process includes a training phase and an online phase. The training phase includes: collecting a plurality of stamping signals by a strain sensor when a plurality of stamping processes is performed by a stamping device; calculating a plurality of stamping energies according to the plurality of stamping signals by a computing device; filtering an outlier in the plurality of stamping energies to establish a stamping sample dataset by the computing device; and calculating at least one feature reference value according to the stamping sample dataset by the computing device. The online phase includes: collecting at least one current stamping signal by the strain sensor when at least one current stamping process is performed by the stamping device; and generating a stamping quality estimation according to the at least one current stamping signal and the at least one feature reference value by the computing device.

Description

Stamping process quality evaluation method

The present invention relates to a stamping process, and in particular to a stamping process quality evaluation method.

Quality control is critical in the stamping process of metal forming. Various factors in the stamping process, such as raw material selection, mold design and adjustment of molding parameters, will affect the quality of the final product. A poor stamping process may cause product defects, inaccurate dimensions, and even reduce material strength, further affecting product performance and reliability.

At present, most companies still use sampling inspection to control quality. However, this method can only inspect a limited number of samples and cannot fully understand the quality of the entire production batch. If progressive stamping is used, it will increase the difficulty of visual inspection. In addition, when the stamping process is running, the existing technology only provides a simple display and alarm about the stamping force (such as strain and load). When the problem of the stamping process is discovered, a large number of defective products have been produced. In general, there is currently a lack of an effective solution for managing the status of processes and products, which makes it difficult for companies to grasp or predict abnormal production status in real time.

In view of this, the present invention proposes a method for evaluating the quality of a stamping process. By collecting strain signals during the processing and applying the analysis method proposed by the present invention, the shortcomings of traditional quality sampling methods, such as measurement variability, measurement repeatability, and the inability to monitor production status abnormalities in real time, can be compensated. In addition, through real-time quality monitoring, the generation of excessive waste can be reduced, thereby improving product quality and efficiency.

A method for evaluating the quality of a stamping process according to an embodiment of the present invention includes a training phase and an online phase. The training phase includes: a strain gauge collects a plurality of stamping signals of a stamping device performing a plurality of stamping processes, a computing device calculates a plurality of stamping energies according to the plurality of stamping signals, the computing device selects outliers in the stamping energies to establish a stamping sample data set, and the computing device calculates at least one characteristic benchmark value according to the stamping sample data set. The on-line stage includes: the strain gauge collects at least one current forward pressing signal when the pressing device performs at least one current forward pressing signal, and the computing device generates a pressing quality evaluation result according to the at least one current forward pressing signal and the at least one characteristic reference value.

In summary, the present invention proposes a method for evaluating the quality of a stamping process. The method collects strain signals of multiple single stamping processes, calculates at least one characteristic benchmark value as a reference for process quality monitoring, and uses the characteristic benchmark value to analyze the quality of the finished product of the stamping process. The present invention not only solves the problem of insufficient accuracy of manual sampling, but also solves the problem of being unable to evaluate abnormal production status in real time.

The above description of the disclosed content and the following description of the implementation methods are used to demonstrate and explain the spirit and principle of the present invention, and provide a further explanation of the scope of the patent application of the present invention.

The detailed features and advantages of the present invention are described in detail in the following embodiments, and the contents are sufficient to enable anyone familiar with the relevant technology to understand the technical content of the present invention and implement it accordingly. According to the contents disclosed in this specification, the scope of the patent application and the drawings, anyone familiar with the relevant technology can easily understand the relevant purposes and advantages of the present invention. The following embodiments are to further illustrate the viewpoints of the present invention, but they are not to limit the scope of the present invention by any viewpoint.

Fig. 1 is a flow chart of a stamping process quality evaluation method according to an embodiment of the present invention. As shown in Fig. 1, the method includes a training phase (steps S1 to S4) and an online phase (steps T1 to T2).

In step S1, a strain gauge collects a plurality of stamping signals when a stamping device, such as a crankshaft stamping press, performs a plurality of stamping processes.

The strain gauge is suitable for supporting parts (not shown) arranged around the punch bed. The strain gauge starts collecting the punch signal when receiving the first trigger signal, and stops collecting the punch signal when receiving the second trigger signal. The generation of the first and second trigger signals depends on the motor shaft encoder, crankshaft angle, or slider height in the punching equipment. Please refer to Figure 2 for details. The upper part of Figure 2 is a segmented example of a single punching process, the middle part is the slider height change of the single punching process (unit not marked), and the lower part is the punch signal corresponding to the single punching process.

Generally speaking, the stamping equipment includes a motor M1, a crankshaft M2 and a slider M3. The rotation of the motor M1 can drive the crankshaft M2 to rotate, and the height of the slider M3 connected to the crankshaft M2 drops from the top dead center TDC to the bottom dead center BDC, and returns to the top dead center TDC after stamping the material.

The stamping device can output a first trigger signal when the crankshaft M2 rotates to a first angle or when the slider M3 is at a first height, to notify the strain gauge to start collecting the stamping signal; then when the crankshaft M2 rotates to a second angle or when the slider M3 is at a second height, it can output a second trigger signal to notify the strain gauge to stop collecting the stamping signal. The first angle is different from the second angle, and the first height is different from the second height. Taking Figure 2 as an example, the first angle is 130°, the second angle is 270°, the first height is H1, and the second height is H2, but the present invention is not limited to these examples. The above-mentioned angle information can be obtained through an encoder arranged on the motor shaft, and the height information can be obtained through a laser displacement meter or a mold height indicator arranged on the stamping device.

If it is an old sample stamping process, it means that the stamping signal has been obtained in the past, and the angle or height information can be determined in advance. If it is a new sample, the first and second angles (heights) can be set as boundary values, and then the stamping process can be performed once to obtain the stamping signal, and then the first and second angles (heights) can be reset according to the stamping signal.

The present invention uses a single punching signal as the minimum unit of signal analysis. In step S1, the starting and ending ranges of punching signal collection can be determined by angle information, and the collected punching signals are aligned with this angle information. If the punch press feed signal is used as the trigger signal for starting collection, the unprocessed parts of the front and back sections of the punching process are collected, resulting in the need to re-capture the signal range during subsequent analysis, resulting in a waste of data storage space. Therefore, the present invention uses the first and second trigger signals to accurately capture the starting and ending points of each punching process.

In step S2, the computing device calculates a plurality of impulse energies according to a plurality of impulse signals.

In one embodiment, the stamping signal collected from a single stamping process includes multiple stamping forces, and the stamping energy is calculated as follows:

E= (Formula 1)

Where E represents the impact energy, X ₁ , X ₂ , …, X _M represent the impact pressure values, and M represents the amount of impact pressure data collected in a single impact signal.

The computing device is communicatively connected to the stamping device and the strain gauge to obtain information such as the stamping signal, the first and second triggering signals, and the angle (height). In one embodiment, the computing device may be at least one of the following examples: a computing device such as a personal computer, a network server, a microcontroller (MCU), an application processor (AP), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system-on-a-chip (SOC), a deep learning accelerator, or any electronic device with similar functions. The present invention does not limit the hardware type of the computing device.

In step S3, the computing device selects outliers from the plurality of impulse energies to establish an impulse sample data set.

In order to improve the quality of collected data, the data cleaning method proposed in step S3 is as follows: the computing device sorts multiple impulse energies and deletes the parts less than Q1-W×IQR and greater than Q3+W×IQR, where Q1 represents the first quartile, Q3 represents the third quartile, IQR represents the interquartile range (IQR), and W is the weight (e.g., 1.5). After step S3 is completed, the multiple impulse signals that are not deleted constitute the impulse sample data set.

In step S4, the computing device calculates at least one characteristic benchmark value according to the impact sample data set. The specific details will be described later.

In step T1, the strain gauge collects at least one forward stamping signal when the stamping equipment performs at least one forward stamping process. Step T1 is basically the same as step S1, except that the online stage uses the information obtained in the training stage to perform real-time monitoring and status feedback of the stamping process, as shown in step T2.

In step T2, the computing device calculates at least one impulse quality evaluation result according to the at least one current impulse signal and at least one characteristic reference value. The at least one characteristic reference value mentioned in step S4 includes: at least one of a dynamic envelope, a stability index, and an impulse health status index, which are described as follows:

Dynamic Envelope:

Figures 3 and 4 are schematic diagrams of the current impulse signal and dynamic envelope in the on-line stage. In the training stage, the computing device calculates the upper envelope L1 and the lower envelope L2 based on multiple impulse signals in the impulse sample data set. The upper envelope L1 is drawn by the computing device based on the maximum impulse value of each angle, and the lower envelope L2 is drawn by the computing device based on the minimum impulse value of each angle. The upper envelope L1 and the lower envelope L2 constitute an envelope interval.

Figure 3 is a schematic diagram of a normal stamping signal and a dynamic envelope, where L0 represents multiple current stamping signals in the online stage. It can be seen that these current stamping signals L0 are all within the envelope range. During the subsequent spot check, it was confirmed that the finished product of the current stamping process had no defects (the average burr value of 0.01 was less than the average burr value of 0.03 during the training stage).

FIG4 is a schematic diagram of an abnormal current stamping signal and a dynamic envelope, where L4 represents multiple current stamping signals in the on-line stage. It can be seen that between angles 230 and 240, the current stamping signal L4 exceeds the upper envelope L1. During subsequent spot checks, it was also found that the finished product of the current stamping process had defects (its average burr value of 0.05 was greater than the average burr value of 0.03 during the training stage). As can be seen from FIG3 and FIG4, the dynamic envelope proposed by the present invention can indeed reflect the abnormality of the current stamping process in real time during the on-line stage, avoiding the production of a large number of defective products.

The dynamic envelope can be updated manually or automatically. Manual mode: When the user observes that the current stamping signal is outside the envelope range, the user determines whether there is a problem with the current stamping process based on experience. If there is no problem, the computing device is instructed to update the envelope range. Automatic mode: As long as the stamping equipment has cumulatively executed the current stamping process M times, the computing device will generate new upper and lower envelopes based on the latest collected M current stamping signals and the N stamping signals collected during the training phase. The dynamic envelope calculation only requires N impulse signals collected during the training phase, and does not require manual signal marking. The calculation speed is fast during the online phase, so real-time monitoring of the impulse process can be achieved.

Stability indicators:

In the training phase, the computing device calculates an average impulse signal based on multiple impulse signals in the impulse sample data set. In the online phase, every time the computing device receives a current impulse signal, it calculates the time series distance between the current impulse signal and the average impulse signal through a dynamic time warping algorithm as a stability indicator of the current impulse process.

Figure 5 is an example of the average impulse signal t1 and two current impulse signals t2 and t3 of the impulse sample data set. As shown in Figure 5, the stability index S(t1, t2) of the current impulse signal t2 is 704, and the stability index S(t1, t3) of the current impulse signal t3 is 237. The larger the value of the stability index used in the present invention, the more unstable it is. As can be seen from FIG. 5 , the trends of the current impulse signal t3 and the average impulse signal t1 of the impulse sample data set are closer than the current impulse signal t2, and the shape of the current impulse signal at an angle of 250 to 270 is obviously different from the shape of the average impulse signal t1 of the impulse sample data set.

Figure 6 is an example of a trend chart of the stability index at the online stage. It can be seen from Figure 6 that the stability index rises significantly at about 18,000 times. During the subsequent sampling inspection, it was also found that after the current stamping process was executed 18,025 times, due to the wear of the stamping head, the finished product began to have defects (the average burr value was greater than the allowable value of 0.05). This means that the stability index proposed by the present invention can indeed reflect the abnormality of the current stamping process in real time at the online stage, avoiding the production of a large number of defective products.

It should also be noted that the sampling error on the hardware may cause the number of data points collected for a single stamping process to be different during the training phase and the online phase. In other words, the time series lengths of the average stamping signal and the current stamping signal are different. Therefore, the present invention adopts a dynamic time correction algorithm to avoid the above problem. If the difference in time series length can be ensured to be within the allowable range, the Euclidean distance can also be used as a stability indicator.

The update of stability index can refer to the update of dynamic envelope, and the impulse sample data set can be redefined automatically or manually. Stability index can fill the blind spot when the peak value or dynamic envelope is monitored online, and can help users find the abnormal part of the contour in the curve corresponding to the current impulse signal. Stability index is particularly suitable for applications such as continuous mold and servo punch press.

Stress health status indicators:

In order to balance the training data, the present invention applies the synthetic minority oversampling technique (SMOTE) to the outliers deleted in step S3 (e.g., 10 data) and the impact sample data set (e.g., 90 data) established in step S3 to generate a sufficient number (e.g., 80) of abnormal data. Then the computing device trains a machine learning model based on the abnormal data generated after the oversampling technique, the original outlier data, and the impact sample data set. In the online stage, the computing device inputs each current impact signal into the machine learning model to generate an impact health status indicator. Figure 7 is an example of a trend chart of the pressure health indicator in the online stage. It can be seen from Figure 7 that the pressure health indicator rises significantly at about 18,000 times, which is consistent with the example shown in Figure 6.

In one embodiment, the present invention adopts a random forest model. This is because the processing parameters of the stamping process are complex and the process is fast, so it is not suitable to use a model with too long training time. The random forest model includes multiple decision trees, is not easy to overfit, does not require normalization of features, and has a short training time and the function of ranking feature importance. However, the present invention does not limit the type of machine learning model. For example, in other embodiments, a support vector machine (Support Vector Machine) or a convolutional neural network (Convolutional Neural Network, CNN) may also be used. The present invention can output the stamping health status indicator in real time through the computing device when each current stamping process is completed in the online stage, thereby achieving the effect of real-time monitoring of the mold status of the stamping equipment. In addition, the random forest model can output the feature importance ranking as a reference for mold repair or troubleshooting.

In one embodiment, the stamping device performs a plurality of current stamping processes. In other words, the number of the at least one current stamping process, the at least one current stamping signal, and the at least one stability indicator in step T1 is multiple.

The computing device collects multiple current impulse signals, obtains the maximum impulse pressure of each current impulse signal, and thereby obtains multiple maximum impulse pressure values corresponding to these current impulse signals. In addition, the computing device also calculates multiple stability indicators corresponding to the multiple current impulse signals. The computing device uses a moving average method to smooth the multiple maximum impulse pressure values and the multiple stability indicators respectively, and then inputs the smoothed multiple stability indicators into an AutoRegressive Integrated Moving Average (ARIMA) model to generate an impulse quality prediction result related to the stability indicator. The computing device inputs the smoothed multiple maximum impact pressure values into the ARIMA model to generate the impact quality prediction results related to the maximum impact pressure values. The ARIMA model can predict the time series data. The present invention uses this to evaluate the future impact quality trend, and can be used as a warning reference for quality deterioration in conjunction with the threshold setting.

In summary, the present invention proposes a method for evaluating the quality of a stamping process. The method collects the strain signal of a single stamping process, and calculates characteristics such as a dynamic envelope, a stability index, and a stamping health status index as a reference for process quality monitoring, and can also be used to analyze the quality of finished products of the stamping process. In addition, the method also uses information such as historical peak values, historical stability indicators, etc., in conjunction with a time series analysis model to predict future stamping process quality trends. The present invention not only solves the problem of insufficient accuracy of manual sampling, but also solves the problem of being unable to instantly evaluate abnormal production status.

Although the present invention is disclosed as above with the aforementioned embodiments, it is not intended to limit the present invention. Any changes and modifications made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention. Please refer to the attached patent application for the scope of protection defined by the present invention.

S1-S4, T1-T2: Steps M1: Motor M2: Crankshaft M3: Slider TDC: Top Dead Center BDC: Bottom Dead Center Ɵ: Angle H1: First Height H2: Second Height L1: Upper Envelope L2: Lower Envelope L0, L4, t2, t3: Current impulse signal t1: Average impulse signal of impulse sample data set S(t1, t2), S(t1, t3): Stability index

FIG1 is a flow chart of a method for evaluating the quality of a stamping process according to an embodiment of the present invention; FIG2 is a schematic diagram of a segmented stamping process and an example of a stamping signal; FIG3 is a schematic diagram of a current stamping signal and a dynamic envelope in the on-line stage (normal); FIG4 is a schematic diagram of a current stamping signal and a dynamic envelope in the on-line stage (abnormal); FIG5 is an example of an average stamping signal and a current stamping signal of a stamping sample data set; FIG6 is an example of a trend chart of a stability indicator in the on-line stage; and Figure 7 is an example of a trend chart of the impact health status indicator during the launch phase.

S1-S4, T1-T2: Steps

Claims (11)

A method for evaluating the quality of a stamping process includes a training phase and an online phase, wherein: The training phase includes: Collecting multiple stamping signals when a stamping device performs multiple stamping processes using a strain gauge; Calculating multiple stamping energies based on the stamping signals using a computing device; Screening outliers from the stamping energies using the computing device to establish a stamping sample data set; and Calculating at least one characteristic benchmark value based on the stamping sample data set using the computing device; The online phase includes: The strain gauge is used to collect at least one current stamping signal when the stamping equipment performs at least one current stamping process; and the computing device is used to generate a stamping quality evaluation result based on the at least one current stamping signal and the at least one characteristic reference value. The method for evaluating the quality of a stamping process as described in claim 1, wherein the stamping signals collected by the strain gauge when the stamping equipment performs the stamping processes include: When a first trigger signal is received, the strain gauge starts to collect one of the stamping signals; and When a second trigger signal is received, the strain gauge stops collecting one of the stamping signals. A method for evaluating the quality of a stamping process as described in claim 2, wherein the stamping device includes a crankshaft, and the method further includes: When the crankshaft rotates to a first angle, the stamping device outputs the first trigger signal; and When the crankshaft rotates to a second angle different from the first angle, the stamping device outputs the second trigger signal. A method for evaluating the quality of a stamping process as described in claim 2, wherein the stamping device includes a slider, and the method further includes: When the slider is at a first height, the stamping device outputs the first trigger signal; and When the slider is at a second height different from the first height, the stamping device outputs the second trigger signal. A method for evaluating the quality of a stamping process as described in claim 1, wherein each of the stamping signals includes a plurality of stamping forces, and the computing device calculates the stamping energies based on the stamping signals, including: Calculating the sum of the squares of the stamping forces by the computing device; and Calculating the ratio of the sum of the squares to the number of the stamping forces by the computing device; and Calculating the square root of the ratio by the computing device as one of the stamping energies. The method for evaluating the quality of a stamping process as described in claim 1, wherein the outliers among the stamping energies are screened by the computing device, including: Deleting at least one of the stamping energies less than Q1-1.5×IQR by the computing device; and Deleting at least one of the stamping energies greater than Q3+1.5×IQR by the computing device; Wherein Q1 is the first quartile, Q3 is the third quartile, and IQR is the interquartile range. The method for evaluating the quality of a stamping process as described in claim 1, wherein the at least one characteristic reference value includes an upper envelope formed according to the maximum values of the stamping signals and a lower envelope formed according to the minimum values of the stamping signals; and the upper envelope stage further includes: when the current stamping signal is outside the interval formed by the upper envelope and the lower envelope, the computing device sends an alarm signal. The method for evaluating the quality of a stamping process as described in claim 1, wherein the at least one characteristic reference value includes an average stamping signal calculated based on the stamping signals, and generating a stamping quality evaluation result based on the at least one current stamping signal and the at least one characteristic reference value includes: Inputting each of the average stamping signal and the at least one current stamping signal into a dynamic time correction algorithm to generate a plurality of stability indicators as the stamping quality evaluation result. The stamping process quality assessment method as described in claim 8 further includes: Smoothing the stability indicators by moving average; and Inputting the smoothed stability indicators into an Autoregressive Integrated Moving Average Model (ARIMA) model to generate a stamping quality prediction result. The method for evaluating the quality of a stamping process as described in claim 1, wherein the at least one current stamping process and the at least one current stamping signal are multiple, and the method further comprises: Obtaining multiple maximum stamping pressure values corresponding to the at least one current stamping pressure signal; Smoothing the stamping pressure maximum values by a moving average method; and Inputting the smoothed stamping pressure maximum values into an integrated moving average self-regression model to generate a stamping quality prediction result. The method for evaluating the quality of a stamping process as described in claim 1, wherein the training phase further includes: generating an abnormal data based on the stamping sample data set and the outlier using a synthetic minority oversampling technique (SMOTE); and training a machine learning model based on the abnormal data, the outlier and the stamping sample data set using the computing device; the online phase further includes: inputting the current stamping signal into the machine learning model using the computing device to output a stamping health status indicator.

Images