TWI773483B - Sensing data processing method - Google Patents

Abstract

A sensing data processing method includes a data reduction step and a reduction point data curve fitting step. The data reduction step is to find the period of the noise bandwidth in the original data volume by Fourier analysis, then perform a moving average, and then use the difference point to acquire the effective reduction point data, wherein the Y-axis is used as the reference for segmentation capture to obtain reduction point data. The reduction point data curve fitting step is to perform curve fitting on the reduction point data. The effect of the present invention is that the method of capturing reduction point data is based on the Y-axis as the basis for segmented capture, which can visualize the turning points and characteristic points of the data, and preserve the characteristic value of the curve, so that the combined curve can be closer to the original data curve.

Description

Sensing data processing method

The present invention relates to a sensing data processing method, in particular to a sensing data processing method for reducing and fitting the huge amount of data obtained in the monitoring injection molding process.

During the injection molding process, the mold will be closed, and the filling, pressure holding, and cooling stages are all invisible. Therefore, traditional plastic injection personnel are based on experience or trial and error. After completion, a machine parameter setting is given, but in fact, it is easily affected by factors such as environment, material batch, melt state, machine control, etc., which makes plastic products prone to defective products during production. Therefore, in injection molding, sensors are added in the mold to measure physical quantities such as pressure, temperature, and speed of the melt in the mold during the molding process, in order to understand the state change of the melt during the injection process.

In response to the development of today's big data and cloud technology, many industries collect and analyze data from machines. However, the excessive amount of data will cause troubles in storage and analysis. If the required information is to be extracted from the large amount of data, the data curve will oscillate due to factors such as machine error, human error, and wire frequency problems. Phenomenon. Therefore, in the judgment of the curve, it is impossible to judge whether a certain data belongs to noise, and it is impossible to judge the accurate data in the judgment of the eigenvalue.

Therefore, the purpose of the present invention is to provide a processing method for reducing the amount of sensing data, which is used to process the huge amount of data obtained in the injection molding process, and can greatly reduce the amount of data of the sensing data while retaining the curve characteristics to the greatest extent. .

The sensing data processing method of the present invention includes a data reduction step and a reduction point data curve fitting step. The data reduction step is to find out the period of the noise bandwidth in the original data volume by Fourier analysis, and then perform a moving average, and then use the difference value point to extract the effective reduction point data, wherein the difference value point is based on the Y-axis. The sensing value is set as a reference value and a Y-axis difference value for segmental capture. When the comparison result between the reference value and the subsequent Y-axis sensing value is less than the Y-axis difference value, the subsequent Y-axis sensing value is obtained. The measured value is regarded as noise and ignored. When the comparison result between the reference value and the subsequent Y-axis sensing value is greater than the Y-axis difference value, the subsequent Y-axis sensing value is regarded as a valid decrement point data. The curve fitting step of the decrement point data is to perform curve fitting and restoration on the decrement point data.

Another technical means of the present invention is that the difference value selection point refers to whether the difference value of the Y-axis point data is greater than a specific value, and if it is greater than the specific value, it is selected as valid decrement point data.

Another technical means of the present invention is that the Y-axis difference is set to the melt pressure in the injection mold of 1Mpa.

Another technical means of the present invention is that in the step of fitting the curve of the abatement point, the curve must pass through all the valid abatement point data to join all the valid abatement point data.

Another technical means of the present invention is that when the valid decrement point data is joined, it is judged whether the next point data has an interval extreme value, if not, the joining is continued, and if so, the interval extreme value is examined.

Another technical means of the present invention is that the discussion of the interval extreme value is to determine (a) whether there is an inflection point between the interval extreme value and the previous point of the interval extreme value. Extend, and then fit the elliptic arc curve; if so, then judge (b) whether the slope of the connecting line between the interval extreme value and the previous point of the interval extreme value is greater than 1; if the slope is not greater than 1, the interval extreme value and the interval extreme value The previous point of the interval is fitted with two equal-diameter arcs; if the slope is greater than 1, then (c) the interval extreme value and the previous point of the interval extreme value are fitted with two unequal-diameter arcs and the two unequal-diameter circles The common tangent of the arc is fitted.

The effect of the present invention lies in that the method of capturing the data of the reduction point is based on the Y-axis for segmental capture, which can display the regional extreme values and characteristic points of the original data curve, better preserve the characteristics of the curve, and then obtain the reduction point. The data curve after data fitting can be closer to the original data curve. This patent can use the Y-axis as the reference to capture the reduction point data to greatly reduce the data volume of the sensing data, and then restore the sensing data curve through the reduction point data curve fitting step, which can preserve the curve characteristics to the greatest extent. The amount of data is greatly reduced.

11: Data optimization steps

12: Curve fitting steps

FIG. 1 is a flow chart, which is a preferred embodiment of the sensing data processing method of the present invention; FIG. 2 is a graph illustrating the raw data obtained by the data capture card; Part of the data magnification; Fig. 4 is a graph, which is to find out the period of the noise bandwidth by Fourier analysis, and then carry out the curve after moving average; Fig. 5 is a graph, which is a part of the data magnification of Fig. 4; FIG. 6 is a graph illustrating a reduction graph with a data difference greater than 1; FIG. 7 is a graph illustrating an enlarged view of a decrement with a data difference greater than 1; FIG. 8 is a graph illustrating a decrement with a data difference greater than 2 Fig. 9 is a graph, illustrating that the data difference is greater than 2, the reduction magnification; Fig. 10 is a graph, which is a curve fitting graph without considering the interval extreme value; Fig. 11 is a graph, which is a simple curve Fitting diagram; Figure 12 is a curve and decrement point data diagram, which is the original data curve and the decrement point data diagram considering the interval extreme value; Figure 13 is a one-point data diagram, which is the original point data and the decrement point data considering the interval extreme value Enlarged figure; Fig. 14 is a graph, which is a superimposed graph of the original data curve and the fitting curve of the reduction point data; Fig. 15 is a graph, which is an enlarged graph of the curve fitting cavity pressure of 15.5 to 17.5 seconds in Fig. 14; Fig. 16 is a graph showing an enlarged view of the cavity pressure of the curve fitting from 16.6 to 18.2 seconds in Fig. 14; Fig. 17 is a graph showing an enlarged view of the cavity pressure of the curve fitting from 17.5 to 19.5 seconds in Fig. 14; FIG. 18 is a graph showing an enlarged view of the cavity pressure of the curve fitting from 21 to 27 seconds in FIG. 14 ; and FIG. 19 is a graph showing a schematic diagram of an extreme value fitting control point in the curve interval.

The features and technical contents of the relevant patent applications of the present invention will be clearly presented in the following detailed description of the preferred embodiments with reference to the drawings. Before the detailed description, it should be noted that similar elements are designated by the same reference numerals.

The invention develops a mathematical model for reducing the data of the sensing point and fitting and restoring the data curve, and takes the mold cavity pressure curve as the benchmark to reduce and adjust the data captured by injection molding. As a matter of fact, injection molding collects a large amount of data every time it operates, including noise from electronic equipment, so there will be oscillations on the curve, and the characteristic data cannot be confirmed. The invention reduces the point data through the function model, and eliminates the noise without destroying the characteristic of the curve. It is expected to solve the problem that the current common fitting technology is likely to cause the curve features to be filtered out, thereby achieving the purpose of data reduction and reducing the storage space of the hardware.

First of all, it should be explained that the present invention uses the injection molding machine of FANUC roboshots-2000i100b, and connects the DAQ Card (data acquisition card) with wires, which can capture the original data of the injection molding process for further analysis. In the present invention, the data captured by the DAQ Card is the data of the cavity pressure sensors installed near the gate and the far gate. The captured data volume will be input into MATLAB software for analysis using mathematical software. Among them, the injection molding machine, the DAQ Card, and the cavity pressure sensor can be understood by those with ordinary knowledge in the field, and they are not the focus of this case, so they will not be repeated in detail.

Referring to FIG. 1, it is a preferred embodiment of the sensing data processing method of the present invention, which is used to process the huge amount of raw data obtained in the injection molding process, including a data reduction step 11 and a reduction point data curve fitting step 12. .

Since the DAQ Card will capture data every millisecond, 20,000 data will be captured in just 20 seconds. The more point data, the more accurate the curve, but due to the influence of noise, it will be unfavorable for the future. Because it is impossible to know whether the point data is a valid value or a noise, it is impossible to know the regional extreme value characteristics during the analysis process. As shown in Figure 3, if the curve 10.28 to 10.34 seconds of the original data map of Figure 2 is partially enlarged, it will be found that there are too many eigenvalues of the regional extreme value. Because the time is captured every millisecond, most of the features are that the noise cannot calculate the slope integral and the regional extreme value. Values and other features need to be filtered out, otherwise it will affect the analysis curve feature results, and it is impossible to determine whether the error on the curve is effective curve. It should be noted that, in the acquisition of sensing data, the X axis is the time axis, the Y axis is the sensing value, the X value of the coordinates of the sensing point on the XY plane is the time, and the Y value is the current sensing at the time. value.

Therefore, the data reduction step 11 is to find out the period of the noise bandwidth in the original data volume by Fourier analysis, and then perform a moving average, and then use the difference point to extract the effective reduction point data. Fragmented capture to obtain decrement point data.

More specifically, in this embodiment, the period of the noise bandwidth is found by Fourier analysis, and then a moving average is performed to reduce the noise to make the curve smoother, and eliminate the spike value points where the noise sharply rises and falls. As shown in the moving average chart in Figure 4. In addition, referring to Figure 5, the red dots from 10.28 to 10.34 seconds are the original data volume, and the blue dots are the results of moving average. It can be seen that the curve is smoother, and many spurious value points with sharp rises and falls are relatively filtered out. To avoid the value of the noise surge being taken by mistake.

Next, take the difference points for data simplification. As shown in Figure 6, it is a decrement graph with the Y-axis difference set to 1Mpa. The green points are the effective decrement point data captured. The acquisition conditions take the first point as the reference value, and the data of the next point and the first point When the difference is greater than 1Mpa, it is the effective reduction point data, and then this point is used as the next reference value, and so on, when the data difference is larger than 1Mpa, the next bit of data is captured to achieve the effect of data reduction. It should be noted that, in this embodiment, the method of capturing point data is different from the traditional way of taking the X-axis as the benchmark, and taking the Y-axis as the benchmark to capture the point data, which can save the eigenvalues of the curve, as shown in Figure 7 As shown in the figure, it is an enlarged view of the decrement with the Y-axis difference set to 1Mpa. It can be seen that when the point data is captured with the Y-axis as the reference, the noise of the Y-axis difference whose data difference is less than 1Mpa will not be detected. Select. This method is an effect that cannot be achieved by the traditional X-axis reference.

In this embodiment, the Y-axis difference of the data difference is greater than 2Mpa to test the capture effect. Figure 8 is a reduction graph with the Y-axis difference set to 2Mpa, and the green points are the captured effective reduction points. According to the data, the first point is used as the reference value for the conditions to be retrieved. When the data difference between the next point and the first point is greater than 2Mpa, it is the valid reduction point data, and then this point is used as the next reference value, and so on. If the difference is greater than 2Mpa, the effective reduction point data will be captured. Figure 9 is an enlarged view of the decrement when the Y-axis difference is set to 2Mpa. It can be seen that when the data difference is greater than the Y-axis difference of 2Mpa to select the next point data, many eigenvalues will be sacrificed and not selected (for example, Around the turning points of 17.9 seconds, 18.2 seconds, and 18.5 seconds), such a curve is easily distorted because sufficient eigenvalues are not captured. Therefore, the Y-axis difference of the points should be selected according to the curve pattern and the effective difference value of actual production to select the appropriate Y-axis difference as the standard for simplifying the data. Small easy to incorporate noise fluctuations into features. In this embodiment, an appropriate Y-axis difference value of 1Mpa is selected as the standard for simplifying the data according to the shape of the curve, which can retain sufficient eigenvalues.

Next, the curve fitting step 12 is performed, wherein the curve joining needs to be performed first. The next step in capturing each valid point is to join the curve. The point of joining is that the fitting curve must pass through each valid point, so the first curve is used for joining. The formula of the linear curve function is formula (1): y =a× x + b (1)

Substitute the first point and the first point of the valid point data to form formula (2) and formula (3): y ₁ =a× x ₁ + b (2)

y ₂ =a × x ₂ + b (3)

By solving the simultaneous equations of formula (2) and formula (3), the unknown values of a and b can be obtained, and then data fitting is performed. After analyzing the fitting curve, it is found that the left and right derivative functions are not equal, resulting in The fitting curve will be somewhat jagged, and at the extreme value of the interval, it will be a shape with sharp corners, which has a certain gap with the arc of the curve, as shown in the first curve fitting diagram in Figure 10.

Since the fitting curve will appear jagged when using the primary curve, the mathematical model is changed to use the exponential function. The advantage of the exponential function is that its waveform path is fixed. Among them, the function index formula is formula (4): y =a× e ^{b * x} + c (4)

Substitute into the points P ₁ and P ₂ of Figure 11, and the slope m of the tangent to P ₁ :

Using the three formulas (5), (6) and (7) to solve the three unknown functions a, b, and c, and then the mathematical model can be used for curve fitting. For the fitted curve, when a certain point is an extreme value, the exponential function cannot be operated, so the extreme value needs to be further calculated.

As shown in Figure 12, the red point in the figure is the extracted interval extreme point, the interval extreme value. From the enlarged view of Figure 13, it can be seen that from 16.3 to 16.85 seconds, the extracted point is at the turning point of the curve. value or minimum value so that the curve characteristics are preserved. The disadvantage of the exponential function is that it cannot be used for functions with a slope equal to zero. Therefore, this embodiment uses three mathematical models to solve the problem of interval extremes. , and (3) the geometric relationship between the slope of the tangent line at the point before the extreme value is divided into three cases, as follows.

Please refer to the appendix together. For the case of (1) extreme value, the formation condition of Figure 1 is that there will be no inflection point between the interval extreme value and the point before the interval extreme value. When making a curved point, we extend the tangent and use the elliptic curve for fitting, as shown in Figure 2, as shown in Figure 2.

Fig. 3 and Fig. 5 are formed on the condition that an inflection point will appear between the interval extreme value and the point before the interval extreme value. The difference is that the slope of the line connecting the two points in Fig. 3 is less than 1, and the slope of the line connecting the two points in Fig. 5 Greater than 1, based on these two conditions, when there is an inflection point between two points, it is divided into two cases. When the slope of the line connecting the extreme value and the point before the extreme value is less than 1, two equal diameters are used. The arc is used for curve fitting, as shown in Figure 4 with two equal diameter circular arcs for curve fitting. When the slope of the line connecting the extreme value and the point before the extreme value is greater than 1, the curve fitting is performed with the common tangent between the two circular arcs and the circular arc, as shown in Figure 6, where the inner common tangent between the two circular arcs is used for curve fitting. .

When the extremum and the previous point of the extremum are completed, it is necessary to consider that the extremum point and the exponential function should be connected, so after the extremum, a binary quadratic equation will be connected to join, such as formula (8).

y = a* x ² + b * x + c (8)

Substitute into the P ₂ and P ₃ points of the binary quadratic fitting diagram of Figure 7 to form formulas (9) and (10)

The slope value of the extreme value is zero, which is substituted into the differential of formula (8) to form formula (11): 2 a * x ₂ + b =0 (11)

Combining formulas (9), (10), and (11) into a simultaneous equation to solve the three unknown values of a, b, and c, and substituting them into the mathematical model, a quadratic curve can be fitted to join the next exponential curve.

Figures 13 and 14 zoom in on the original point data and the decrement point data considering the interval extreme value. It can be seen that the captured decrement data point plus the interval extreme value point can fully reflect the original data point The change.

Figures 15 to 18 are the magnified graphs of the curve fitting cavity pressure. It can be seen that the curve is more smooth. The original frequency vibration is fitted to a smooth curve, and the extreme value difference is not large, which can explain the fitting at the extreme value. is a fit that can use an ellipse. The extension of the data from the up and down vibrations to the fitting of a curve means that the fitted curve has a fixed value, and the next correlation index can be calculated mathematically.

The optimized fitting curve is more convenient in the research and analysis of injection molding. Thousands of curve features can be made with the fitting curve, and each feature will have its physical meaning.

Figure 19 The curve fitting graph of the extreme value index in the interval is the index of the maximum and minimum value. It can be seen that the injection molding is at the cavity pressure in the second few seconds. The maximum cavity pressure can also be made into a table, such as Table 1 shows the interval extreme value index table. Table 2. The interval slope and interval integral index table is the index of interval slope and interval integral. The slope can be seen by whether the curve of the interval is rising or falling, and the integral can be seen by the pressure value spent in this interval.

Table 2

The injection molding machine places multiple sensors on the machine, and captures relevant data during injection, and analyzes whether the machine error, human error, environment... and other uncertain factors affect the injection during the injection molding process. molding quality. One of the factors that will affect the quality of the finished product in injection molding is the cavity pressure. When the cavity pressure is too large, the mold will be stretched, resulting in burrs on the finished product, which will affect the smoothness of the finished product. If the cavity pressure is too small, the finished product will have short shots. Affects the size of the finished product.

The present invention takes the data captured by the cavity pressure as the main curve, and filters out the noise and fits the curve, so that the fitting curve can be used to analyze the cavity pressure more precisely, and has the following effects:

1. The curve is more smooth, the original frequency vibration is fitted into a smooth curve, and the extreme value difference is not large, and the characteristic value is retained.

2. The index of the maximum value and the minimum value can be used to know the maximum cavity pressure in a few seconds when the injection molding is in the cavity pressure, and which range of the pressure distribution is observed when comparing the finished product, which affects the quality of the finished product.

3. The fitting curve can be used to analyze other curves such as large column, injection pressure...etc. The curve can be further analyzed.

4. The data curve of this case reduces the data of several hundred points from the data of 20,000 points, saving a lot of storage space.

5. The fitted curve will pass through each of the captured data and will not be skewed compared to the original data.

To sum up, the contribution of the present invention is data reduction and curve optimization. At present, in the development of big data and cloud, there will be a lot of data waiting to be sorted and analyzed. In the process of analysis, the more data, the better the accuracy. Close to correct, but a lot of memory waste. In this study, the original data volume of 3754KB was reduced to 30KB, and the curve characteristics were retained, and the mathematical model was used to calculate the data to calculate the relevant characteristics of injection molding. The characteristics can be used for further big data analysis files. When the analysis files can be more, they can be put into mechanized learning, which is the preliminary model of intelligence.

However, the above are only preferred embodiments of the present invention, and should not limit the scope of the present invention, that is, any simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the description of the invention, All still fall within the scope of the patent of the present invention.

11: Data reduction steps

12: Steps of fitting the curve of the decrement point data

Claims (3)

A sensing data processing method is suitable for using a computer to process a huge amount of raw data obtained by a sensor in a manufacturing process, and includes the following steps: a data reduction step, using Fourier analysis to find out the noise frequency in the raw data amount A moving average is then carried out for a wide period, and then the effective decrement point data is captured by the difference value points, wherein the difference value points are based on the Y-axis sensing value to set a reference value and a Y-axis difference value for segmentation Capture, when the comparison result of the reference value and the subsequent Y-axis sensing value is less than the Y-axis difference value, the subsequent Y-axis sensing value is regarded as noise and ignored, when the reference value and the subsequent Y-axis sensing value are ignored. When the comparison result of the sensed value is greater than the Y-axis difference, the subsequent Y-axis sensed value is used as a valid decrement point data; and a decrement point data curve fitting step is performed to restore the decrement point data by curve fitting , the curve fitting and restoration must pass all the valid decrement point data to join all the valid decrement point data. When the valid decrement point data is joined, it will be judged whether there is an interval extreme value with the next point data, if If not, continue to join, if so, carry out the exploration of the extreme value of the interval. The sensing data processing method according to claim 1, wherein the Y-axis difference is set to a melt pressure of 1 Mpa in the injection mold. The sensing data processing method according to claim 1, wherein the discussion of the interval extreme value is to judge (a) whether there is an inflection point between the interval extreme value and the previous point of the interval extreme value; The tangent line of the point before the value is extended, and then fitted with an elliptic arc curve; if so, then determine the difference between the (b) interval extreme value and the previous point of the interval extreme value. Whether the slope of the connection line is greater than 1; if the slope is not greater than 1, fit the interval extreme value and the previous point of the interval extreme value with two arcs of equal diameter; if the slope is greater than 1, then (c) fit the interval extreme value to the The previous point of the interval extreme value is fitted by two unequal diameter arcs and the common tangent of the two unequal diameter arcs.

Images