Product quality analysis assistance system
The product quality analysis support system optimizes waveform data classification by minimizing group overlap and distributing classification tasks, ensuring accurate and efficient analysis of continuously manufactured products.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- TMEIC CORP
- Filing Date
- 2024-12-25
- Publication Date
- 2026-07-02
AI Technical Summary
Existing methods for classifying waveform data in steel and non-ferrous metal plants face challenges in maintaining accuracy and efficiency, particularly when handling large volumes of continuously manufactured products, leading to issues with overlapping groups and increased classification time.
A product quality analysis support system that includes a data processing device to classify waveform data by creating groups with similar shapes, determining representative data, and adjusting representative data based on specific conditions to minimize overlap and optimize classification efficiency.
The system achieves accurate and efficient classification of waveform data by minimizing group overlap and distributing classification work, reducing time requirements and improving the efficiency of factor analysis.
Smart Images
Figure JP2024046028_02072026_PF_FP_ABST
Abstract
Description
Product Quality Analysis Support System
[0001] The present disclosure relates to a product quality analysis support system for supporting analysis operations to maintain the quality of products rolled on a rolling line. In particular, the present disclosure relates to a product quality analysis support system for supporting the classification operation of waveform data in order to grasp product quality defects.
[0002] In the operation of steel plants and non-ferrous metal plants, it is required to grasp product quality defects, identify the causes of quality defects, and implement countermeasures against quality defects to promptly restore the quality. Identification of such causes is generally performed by a series of operations of collecting and storing data related to quality, displaying the stored data as waveforms, and analyzing them. Since rolled products are continuously manufactured in the plant, the above series of operations are continuously executed.
[0003] When the number of products to be analyzed is large, in order to grasp typical characteristics of quality defects, the above series of operations are made more efficient by classifying waveform data related to quality defects according to shape characteristics. The shape characteristics of waveform data are linked to product information related to defects that occurred in the past and the countermeasures implemented, and are recorded as a history. Therefore, it is desirable that the waveform data classification method be invariant.
[0004] As a method for classifying waveform data, numerous clustering methods are known. By using a clustering method, starting from a certain analysis time (hereinafter referred to as "time T1"), waveform data is classified (belongs) into groups for past data. At an analysis time after time T1 (hereinafter referred to as "time T2"), new waveform data for the period from time T1 to time T2 is obtained. The following problems occur regarding the handling of waveform data of products manufactured during the period from time T1 to time T2. That is, as one method, when clustering the waveform data obtained during the period from time T1 to time T2 using the same method, editing of the two clustering results performed at time T1 and time T2 is required, and problems such as which groups of the two clustering results should be combined occur.
[0005] Alternatively, if waveform data obtained up to time T1 and waveform data obtained from time T1 to time T2 are clustered together, waveform data belonging to different groups may appear compared to the clustering results performed at time T1. If the factors that do not require quality have already been analyzed using the clustering results at time T1, it is necessary to check the consistency with the results of that factor analysis.
[0006] One possible approach is to ignore the results of past factor analyses and perform clustering using all waveform data up to the analysis time each time, then perform factor analysis. However, this method is not practical because the time required for clustering becomes enormous as the number of waveform data increases.
[0007] The analysis support system disclosed in Patent Document 1 below classifies similar waveform patterns (waveform data) into groups and automatically registers defective waveform data. A representative waveform pattern (representative waveform data) is determined for each group, and it is determined by some method whether the waveform data to be classified is similar to the representative waveform data. If similar representative waveform data exists, it is added to the group to which that representative waveform data belongs. If similar representative waveform data does not exist, the waveform data is registered as the representative waveform data for a new group.
[0008] Japanese Patent No. 5868784
[0009] By the way, in the above-mentioned Patent Document 1, the waveform data that is most similar to any of the waveform data belonging to a group is determined as the representative waveform data. As a result, the representative waveform data is located in the center of the group of waveform data belonging to the group. This leads to a large overlap between adjacent groups, which reduces the accuracy of waveform data classification.
[0010] This disclosure was made to solve the problems described above. The purpose of this disclosure is to provide a product quality analysis support system that can classify continuously acquired waveform data accurately and efficiently.
[0011] The first aspect of this disclosure relates to a product quality analysis support system that assists in analytical work to maintain the quality of steel sheets manufactured on a rolling line. The product quality analysis support system includes a data storage device that stores actual data representing changes in physical quantities related to the above quality, measured by sensors placed on the rolling line, over a fixed time period or a fixed long period, and a data processing device that processes the actual data. The data processing device is configured to create waveform data of physical quantities for each predetermined length by editing the actual data into data for each predetermined length of steel sheet, classify the waveform data into groups so that waveform data with similar shapes belong to the same group, determine representative waveform data from among the waveform data belonging to each group, and determine the destination of newly created waveform data. Determining the destination of new waveform data includes calculating the data distance between the new waveform data and the representative waveform data of each group, creating a new group and assigning the new waveform data to the new group if there are no groups with a data distance smaller than a predetermined value, and assigning the new waveform data to the existing group with the largest number of waveform data belonging to it if there are multiple existing groups with a data distance smaller than a predetermined value.
[0012] The second perspective, in addition to the first perspective, has the following further characteristics: The data processing device is configured to perform the following actions: assign new waveform data to an existing group, then temporarily set the new waveform data as representative waveform data; confirm whether the first condition is met, which is that all waveform data belonging to the existing group are contained within a first virtual circle of a certain radius centered on the temporarily set representative waveform data; confirm whether the second condition is met, which is that the number of waveform data belonging to adjacent groups contained within the first virtual circle is less than the number of waveform data belonging to adjacent groups contained within a second virtual circle of a certain radius centered on the original representative waveform data; and, upon confirming that both the first and second conditions are met, change the original representative waveform data to the temporarily set representative waveform data.
[0013] The third perspective, in addition to the second perspective, has the following further characteristics: The data processing device is configured to further perform the following actions: assigning new waveform data to a new group, determining the new waveform data as representative waveform data, and recording adjacent groups adjacent to the new group.
[0014] The fourth aspect, in addition to the third aspect, has the following further characteristics: The data processing device is configured to further perform the following: after changing the original representative waveform data to a provisionally set representative waveform data, it reviews the records of adjacent groups.
[0015] The fifth aspect has the following additional features in addition to any one of the first to fourth aspects: The data processing device is configured to further perform a normalization process to equalize the number of points in the waveform data to a predetermined number in order to calculate the distance between data points. The normalization process includes dividing the waveform data into a leading edge, a central edge, and a trailing edge, and setting a number of points individually for each of the leading edge, central edge, and trailing edge.
[0016] According to this disclosure, waveform data obtained continuously can be classified without changing the group to which the classified waveform data belongs. Moreover, since the waveform classification is performed in such a way that the overlapping portions between adjacent groups are minimized, the waveform data can be classified with high accuracy. Furthermore, by distributing the classification work to each waveform data set rather than classifying multiple waveform data sets together, the time required for classification work is reduced, and classification can be performed efficiently. As a result, the efficiency of factor analysis work where quality is not a concern can be improved.
[0017] This is a schematic diagram showing a steel plant to which the product quality analysis support system according to the embodiment is applied, and the flow of data used in the system. Figure 2(a) shows actual data, and Figure 2(b) shows signals passing through the sensor. Figure 3(a) shows waveform data along with the transport distance, and Figure 3(b) shows the length of each part and the number of points for normalization processing when the waveform data is divided into three parts. This is a diagram showing the configuration of the data processing device. This is a diagram showing the calculation flow by the waveform data classification processing unit. This is a diagram showing specific examples of the processes in steps S9 and S11 shown in Figure 5. This is a diagram showing the calculation flow of adjacent group recording performed in step S10 shown in Figure 5. This is a diagram showing the calculation flow of representative waveform data review performed in step S12 shown in Figure 5. This is a diagram showing a specific example of representative waveform data review shown in Figure 8. This is a diagram showing the calculation flow of adjacent group recording review performed in step S13 shown in Figure 5. This is a diagram showing an example of the hardware configuration of the data processing device of the product quality analysis support system.
[0018] The product quality analysis support system according to the embodiments of this disclosure will be described below with reference to the drawings. In each drawing, the same or corresponding parts are denoted by the same reference numerals, and redundant explanations will be simplified or omitted as appropriate.
[0019] Figure 1 is a schematic diagram showing a steel plant to which the product quality analysis support system according to the embodiment is applied, as well as the flow of data used in the system. The steel plant comprises a rolling line 2 for rolling rolled material 1, and a control system 3 for controlling the rolling line 2.
[0020] The rolling line 2 is equipped with multiple pieces of equipment. Examples of the main equipment include a heating furnace, a roughing mill, a finishing mill, a cooling device, and a winding machine. As the rolled material 1 passes through these pieces of equipment, heating, processing, cooling, and winding are performed sequentially, ultimately producing a finished product. Each piece of equipment in the rolling line 2 is controlled by command data d1 transmitted from the control system 3.
[0021] Multiple sensors are installed on the rolling line 2. The actual data measured by these multiple sensors can be broadly divided into two types. One is actual data d2 used by the control system 3. Actual data d2 represents changes in the physical quantities of the equipment and the rolled material 1, and is collected at regular time intervals. The control system 3 generates command data d1 based on the actual data d2. The other is actual data d4, which is not used by the control system 3 but is used for the analysis and evaluation of product quality.
[0022] The product quality analysis support system 4 comprises a data storage device 5, a data processing device 6, and an information display device 7. These devices 5, 6, and 7 may each be devices with independent hardware, or they may be computer functions realized by the execution of corresponding programs on a processor.
[0023] The data storage device 5 is a device that stores data used for product quality analysis, that is, actual data representing changes in physical quantities related to quality measured by sensors placed on the rolling line 2 over a fixed time period or a fixed long period. The data storage device 5 includes, for example, magnetic disks such as HDDs, optical disks such as DVDs, and flash memory storage devices such as SSDs. The data storage device 5 receives and stores data set d3 and actual data d4. Data set d3 is a collection of data that combines the actual data d2 used in the control system 3 with intermediate calculation data and calculation result data within the control system 3. The data storage device 5 sends data d5, which is a combination of data set d3 and actual data d4, to the data processing device 6.
[0024] The data processing device 6 processes the data d5 received from the data storage device 5 and performs calculations to support the analysis work, in particular, calculations to classify waveform data into groups. Details of the configuration and calculation flow of the data processing device 6 will be described later. The calculation results from the data processing device 6 are sent to the information display device 7. The information display device 7 edits the calculation results from the data processing device 6 and displays them on the monitor 8, which serves as the display screen for the information display device 7. The user 100 can perform the analysis work while viewing the support information displayed on the monitor 8.
[0025] The data primarily targeted in product quality analysis includes physical quantities of equipment and rolled material measured by sensors, such as speed, opening, and dimensions. This historical data represents changes at fixed sampling intervals (also referred to as "fixed-time sampling data").
[0026] Figure 2(a) shows the actual data, and Figure 2(b) shows the signal during sensor passage. When a graph is drawn with time on the X axis and the physical quantity on the Y axis for one data item of the actual data (time-sampling data), that is, for a physical quantity measured by one sensor, it looks like Figure 2(a). Furthermore, when a graph is drawn with time on the X axis and the signal during sensor passage on the Y axis, it looks like Figure 2(b). While the rolled material 1 is passing through the sensor, the value of the signal during sensor passage is "1". The signal during sensor passage is generated by the control system 3 and is included in the data set d3 sent to the data storage device 5. The signal during sensor passage is stored in the data storage device 5.
[0027] Figure 3(a) shows waveform data along with the transport distance, and Figure 3(b) shows the length of each part and the number of points for normalization processing when the waveform data is divided into three parts. When a graph is drawn for one data item of fixed-time sampling data with the X axis as time and the Y axis as a physical quantity, it looks like Figure 3(a). In the graph of Figure 3(a), a graph of the distance the rolled material 1 has passed through the sensor where the physical quantity was measured (hereinafter referred to as "transport distance") is superimposed along with the change in the target physical quantity. The transport distance is included in the data set d3 which is generated by the control system 3 and sent to the data storage device 5. The transport distance is stored in the data storage device 5.
[0028] The data processing device 6 re-edits the time-sampling data sent from the data storage device 5 into data of predetermined lengths. The data processing device 6 collects and re-edits data of physical quantities corresponding to points where the transport distance is, for example, every 1 meter, thereby creating waveform data of physical quantities of predetermined lengths (also called "time-sampling data"). If the transport distance value of the waveform data is plotted on the X axis and the physical quantity on the Y axis, a graph like Figure 3(b) is obtained. By processing the actual data related to product quality into time-sampling data in this way and plotting it on a two-dimensional XY plane, the change in the actual data with respect to the distance from the leading edge of the rolled material 1 can be visualized.
[0029] When examining changes in performance data related to product quality over the entire length, it is effective to analyze them by dividing them into the leading edge, middle section, and trailing edge. This is because the frequency of quality defects is low in the middle section where control is stable, and high in the leading edge and trailing edge. Therefore, it is more efficient and accurate to divide the rolled material into leading edge, middle section, and trailing edge sections and perform quality assessments there, rather than judging the quality of the rolled material as a whole.
[0030] Therefore, the data processing device 6 divides the waveform data (fixed-length sampling data) by setting the leading end and trailing end to fixed lengths A [m] and B [m], respectively, and setting the central part to a length L-A-B [m] obtained by subtracting the leading end and trailing end lengths from the total length L [m] of the rolled material 1. The region on the leading end side of the rolled material 1 where quality defects are likely to occur is determined by the structure and characteristics of the rolling line 2 of the rolling plant, and does not depend on the total length L of the rolled material 1. The same applies to the trailing end side of the rolled material 1. By setting the leading end and trailing end to fixed lengths when dividing the waveform data (fixed-length sampling data), it is possible to suppress fluctuations in the accuracy of quality judgment depending on the total length of the rolled material 1.
[0031] Furthermore, as will be described later, a normalization process is performed to equalize the number of points Pc in the waveform data in order to make it easier to calculate the distance between data points in the waveform data. At this time, it is preferable not only to set a total number of points Pc for the waveform data, but also to distribute the total number of points Pc to the leading edge, steady edge, and trailing edge and set individual points. That is, rather than distributing the number of points Pc at equal intervals over the entire length of the waveform data, it is preferable to distribute more points to the leading edge and trailing edge, where characteristic waveform shapes are likely to occur, than if they were distributed at equal intervals. For example, it is possible to individually set the number of points in each part such that the distribution ratio of points Pc1, Pc2, and Pc3 in the leading edge, steady edge, and trailing edge is 2:6:2. By individually setting the number of points in each part in this way, it is possible to classify the waveform data by focusing on the characteristics of the waveform data at the leading edge and trailing edge.
[0032] The data processing device 6 receives data d4 from the data storage device 5, performs calculations for classifying the waveform data, and sends the calculation results to the information display device 7. The information display device 7 displays the calculation results of the data processing device 6 on the monitor 8. The user 100 can perform analysis work while viewing the waveform classification results displayed on the monitor 8.
[0033] Figure 4 shows the configuration of the data processing device 6. The data processing device 6 comprises a waveform data editing unit 61, a waveform data classification processing unit 62, a classification result storage unit 63, and a classification result registration database 64. As will be described in detail later, each of the units 61, 62, and 63 may be a device with its own independent dedicated hardware, or it may be a computer function realized by the execution of a corresponding program on a processor.
[0034] The waveform data editing unit 61 extracts the time spent by the rolled material 1 passing through the target equipment for each product from the actual data collected at regular time intervals stored in the data storage device 5. The waveform data editing unit 61 edits the data into regular long-period actual data in the following way. That is, by extracting the range from the start to the end of the transport of the rolled material 1 based on the transport distance, and collecting and editing data of physical quantities corresponding to points, for example, every 1 meter, regular long-period actual data like that shown in Figure 3(b) is obtained. In Figure 3(b), a graph is drawn with the transport distance on the X axis and the physical quantity on the Y axis. The waveform data editing unit 61 also has the function of editing the data into regular long-period actual data. The actual data edited by the waveform data editing unit 61 is called waveform data. In this way, waveform data is obtained for each product for a data item representing a single physical quantity. The functions of each unit 61, 62, and 63 will be explained in the explanation of the calculation flow by the data processing device 6.
[0035] Figure 5 shows the calculation flow by the waveform data classification processing unit 62. This calculation flow is performed similarly for each data item. If K waveform data have been received up to the current time and the sequential number of the waveform data to be classified is k, then k is initialized to 1 (step S1), and the process proceeds to step S2. In step S2, it is determined whether k ≤ K, that is, whether there is any unprocessed waveform data. If k > K, that is, if there is no unprocessed waveform data, the process waits until the next processing timing. If k ≤ K, the process proceeds to step S3.
[0036] In step S3, the kth waveform data D(k) is read. Next, normalization processing is performed (step S4). In the normalization processing, the number of points Pc of the waveform data D(k) is made into a form that makes it easier to calculate the data distance |D(k) - Dt(g)| described later. Specifically, the number of points Pc is determined according to the points where the passage time of the rolled material 1 is long, and the target actual data is edited so that the number of points of the actual data for all products becomes Pc. As described above, in order to focus on the leading edge and the trailing edge, the number of points for the leading edge:steady edge:trailing edge can be set in the ratio Pc1:Pc2:Pc3, respectively. For example, Pc1:Pc2:Pc3 = 2:6:2.
[0037] Next, it is determined whether k = 1 or not (step S5). If k = 1, the first group gr(1) is created, and the waveform data D(1) is set as the representative waveform data Dt(1) of group gr(1) (step S6). If k > 1, the data distance |D(k) - Dt(g)| between the waveform data D(k) and the representative waveform data Dt(g) is calculated for all existing groups gr(g) (step S7). This data distance |D(k) - Dt(g)| is the distance between a point of waveform data D(k) and a point of representative waveform data Dt(g) in N-dimensional space, such as the Euclidean distance or Manhattan distance. Since this type of data distance is well known, further explanation, including its calculation method, is omitted. Next, when R, which represents a constant radius, is set to a positive constant, it is determined whether or not there is an existing group gr(g) such that |D(k) - Dt(g)| < R (step S8). If no such group exists, the process proceeds to step S9. In step S9, as shown in Figure 6, a new group gr(g0) is created, and the waveform data D(k) is set as the representative waveform data Dt(g0) of the new group gr(g0). Next, it is checked whether or not there is an existing group adjacent to the new group gr(g0) (hereinafter referred to as "adjacent group"), and if there is, the adjacent group is recorded (step S10).
[0038] On the other hand, if there are multiple existing groups gr(g) satisfying |D(k) - Dt(g)| < R, the group gr(g) with the largest number of waveform data among these multiple existing groups gr(g) is determined, and the waveform data D(k) is added to the determined group gr(g) (step S11). In the example shown in Figure 6, the waveform data D(k) is added to the existing group gr(1) with the largest number of waveform data among the multiple existing groups gr(1) and gr(2). Note that in step S11, if there are multiple groups with the largest number of waveform data, the group gr(g) with the smallest data distance |D(k) - Dt(g)| is prioritized for determination. Next, the representative waveform data Dt(g) is reviewed as described below (step S12). Furthermore, the records of adjacent groups are reviewed as described below (step S13). After that, the serial number k of the waveform data to be classified is incremented (step S14), and the process returns to step S2.
[0039] Figure 7 is a diagram showing the calculation flow of adjacent group recording performed in step S10 shown in Figure 5. This calculation is performed when a new group gr(g0) is created and records existing groups adjacent to the new group gr(g0). That is, in step S10, assuming C is a constant of 1 or more, it is checked whether there is an existing group gr(g) that satisfies |Dt(g0) - Dt(g)| < C × R, and if there is, that existing group gr(g) is recorded. The number of adjacent groups to the new group gr(g0) is denoted as ng_near(g0), and the adjacent group to the new group gr(g0) is denoted as near(g0, cg). Here, cg = 1, 2, ..., ng_near(g0). At this time, g is recorded for the adjacent group ng_near(g0). Also, g0 is added to ng_near(g). These are recorded and this information is used when reviewing the representative waveform data described later.
[0040] Figure 8 is a diagram showing the calculation flow for the revision of representative waveform data performed in step S12 shown in Figure 5. This calculation is performed when the kth waveform data D(k) is added to the existing group gr(g), and it checks whether a revision of the representative waveform data Dt(g) of the existing group gr(g) is necessary. If a revision is necessary, the representative waveform data is changed (modified). In other words, it checks whether both the first and second conditions described later are met.
[0041] In step S12 described above, first, the first flag Flag1, which is set to 1 when the first condition is met, and the second flag Flag2, which is set to 1 when the second condition 2 is met, are both set to 0 (step S121).
[0042] Here, the first condition is that, when Dg(j) is the waveform data other than the waveform data D(k) belonging to the existing group gr(g), |D(k) - Dg(j)| < R for all waveform data Dg(j). That is, even if waveform data D(k) is tentatively set as the representative waveform data, it is checked whether all waveform data belonging to the existing group gr(g) are included in the first virtual circle with a certain radius R centered on waveform data D(k) (step S122). If the first condition is met, the first flag Flag1 is set to 1.
[0043] The second condition is that, if the number of adjacent groups ng_near(g) adjacent to the recorded existing group gr(g) satisfies ng_near(g) > 0, then for the waveform data belonging to the adjacent group near(g, cg), Nt and Nk are the total number of waveform data belonging to all adjacent groups near(g, cg) that are contained within the ranges of the respective virtual circles |Dt(g) - x| < R and |D(k) - x| < R, respectively, and Nt > Nk holds true. That is, if waveform data D(k) is tentatively set as the representative waveform data, it is checked whether the total number of waveform data of adjacent groups contained within the first virtual circle with a fixed radius R centered on waveform data D(k) is less than the total number of waveform data of adjacent groups contained within the second virtual circle with a fixed radius R centered on the representative waveform data Dt(g) (step S123). If the second condition is met, the second flag Flag2 is set to 1.
[0044] When both the first condition and the second condition are satisfied, that is, when both flags Flag1 and Flag2 are 1, the representative waveform data of the existing group gr(g) is changed from Dt(g) to D(k). That is, the original representative waveform data Dt(g) of the existing group gr(g) is changed to the temporarily set waveform data D(k). The original representative waveform data Dt(g) is the representative waveform data Dt(g) that exists at the time of temporary setting. In the example shown in FIG. 9, the representative waveform data Dt(4) of the existing group gr(4) is changed to the temporarily set waveform data D(k). By changing the representative waveform data Dt(4) in this way, the representative waveform data Dt(4) can be separated from the representative waveform data Dt(2) and Dt(3) of the adjacent existing groups gr(2) and gr(3). As a result, the difference between the classification groups of the waveform data becomes clearer. As a result, it becomes possible to accurately perform the factor analysis operation of unnecessary quality.
[0045] When the number of waveform data belonging to the existing group gr(g) increases, the probability that both the first condition and the second condition are satisfied decreases, and the calculation time for confirming the first condition also becomes longer. Therefore, if the number of waveform data belonging exceeds a certain number, the calculation load can be reduced by not performing the above review of the representative waveform data.
[0046] FIG. 10 is a diagram showing the calculation flow of the review of the adjacent group record executed in step S13 shown in FIG. 5. This calculation is performed after the above-described review of the representative waveform data. When the representative waveform data is changed, it is confirmed whether it is necessary to review the record of the adjacent group, and if the review of the record is necessary, it is changed. That is, when the other group of the existing group gr(g) is gr(g1), if |D(k) - Dt(g1)| < C × R is satisfied and g does not exist in near(g1, cg), g is added. At this time, if g1 does not exist in near(g, cg), g1 is added. On the other hand, if |D(k) - Dt(g1)| ≥ C × R is satisfied and g exists in the adjacent group near(g1, cg), g is deleted. At this time, if g1 exists in near(g, cg), g1 is deleted.
[0047] The classification result storage unit 63 stores the classification results performed by the waveform data classification processing unit 62 in the classification result registration database 64. In addition to the classification results, the classification result registration database 64 stores the product information, group classification identification information, representative waveform data for each group, and information for adjacent groups. The information display device 7 visualizes the classification results stored in the classification result registration database 64 and displays them on the monitor 8.
[0048] FIG. 11 is a diagram showing an example of the hardware configuration of the data processing device 6 of the product quality analysis support system. Each of the above-described functions of the data processing device 6 can be realized by the processing circuit 60 shown in FIG. 9. This processing circuit 60 may be dedicated hardware 60a. This processing circuit 60 may include a processor 60b and a memory 60c. This processing circuit 60 may be partly formed as dedicated hardware 60a and further include a processor 60b and a memory 60c. The example of FIG. 11 is such that part of the processing circuit 60 is formed as dedicated hardware 60a and the processing circuit 60 also includes a processor 60b and a memory 60c. The processing circuit 60 may be at least one dedicated hardware 60a. In this case, the processing circuit 60 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, an ASIC, an FPGA, or a combination thereof. The processing circuit 60 may include at least one processor 60b and at least one memory 60c. In this case, each function of the data processing device 6 is realized by software, firmware, or a combination of software and firmware. The software and firmware are described as programs and stored in the memory 60c. The processor 60b realizes each function of the data processing device 6 by reading and executing the program stored in the memory 60c. The processor 60b is also called a CPU (Central Processing Unit), a processing device, an arithmetic device, a microprocessor, a microcomputer, or a DSP. The memory 60c corresponds to, for example, a non-volatile or volatile semiconductor memory such as a RAM, a ROM, a flash memory, an EPROM, or an EEPROM. Thus, the processing circuit 60 can realize each function of the data processing device 6 by hardware, software, firmware, or a combination thereof.
[0049] As described above, according to this embodiment, waveform data obtained continuously during the operation of a rolling mill can be classified without changing the group to which the classified waveform data belongs. Moreover, since the waveform classification is performed in such a way that the overlapping portion between adjacent groups is minimized, accurate classification can be achieved. Furthermore, by distributing the classification work to each waveform data rather than classifying multiple waveform data together, the time required for classification work is reduced, and classification can be performed efficiently. As a result, the efficiency of factor analysis work that does not require quality control can be improved.
[0050] While embodiments of this disclosure have been described above, this disclosure is not limited to the embodiments described above and can be implemented in various modified forms without departing from the spirit of this disclosure. When the number of elements, quantities, amounts, ranges, etc., are mentioned in the embodiments described above, this invention is not limited to the number mentioned unless it is specifically stated or clearly defined in principle. Furthermore, the structures, etc., described in the embodiments described above are not necessarily essential to this invention unless they are specifically stated or clearly defined in principle.
[0051] Furthermore, although the above embodiment was described using the application of the product quality analysis support system of this disclosure to a steel plant as an example, it can also be applied to non-ferrous metal plants.
[0052] 1...Rolled material, steel plate, 2...Rolling line, 3...Control system, 4...Product quality analysis support system, 5...Data storage device, 6...Data processing device, 60...Processing circuit, 60a...Dedicated hardware, 60b...Processor, 60c...Memory, 61...Waveform data editing unit, 62...Waveform data classification processing unit, 63...Classification result storage unit, 64...Database, 7...Information display device, 8...Monitor, 100...User
Claims
1. A product quality analysis support system that assists in analytical work to maintain the quality of steel sheets manufactured on a rolling line, comprising: a data storage device that stores actual data representing changes in physical quantities related to quality measured by sensors placed on the rolling line over a fixed time period or a fixed long period; and a data processing device that processes the actual data, wherein the data processing device is configured to: create waveform data of the physical quantities for each predetermined length by editing the actual data into data for each predetermined length of the steel sheet; classify the waveform data into groups so that waveform data having similar shapes belong to the same group; determine representative waveform data from among the waveform data belonging to each group; and determine the destination of newly created waveform data, wherein determining the destination of the new waveform data involves: calculating the data distance between the new waveform data and the representative waveform data of each group; and, upon finding that there are no groups where the data distance is less than a predetermined value, creating a new group and assigning the new waveform data to the new group. A product quality analysis support system that includes, in response to the existence of multiple existing groups where the distance between the data is less than a predetermined value, assigns the new waveform data to the existing group with the largest number of waveform data among the multiple existing groups.
2. A product quality analysis support system according to claim 1, wherein the data processing device is configured to further perform the following: assigning the new waveform data to the existing group, then provisionally setting the new waveform data as representative waveform data; checking whether a first condition is met such that all waveform data belonging to the existing group are included in a first virtual circle of a certain radius centered on the provisionally set representative waveform data; checking whether a second condition is met such that the number of waveform patterns belonging to the adjacent group included in the first virtual circle is less than the number of waveform data belonging to the adjacent group included in a second virtual circle of a certain radius centered on the original representative waveform data; and, upon confirming that both the first and second conditions are met, changing the original representative waveform data to the provisionally set representative waveform data.
3. A product quality analysis support system according to claim 2, wherein the data processing device is configured to further perform the following: assigning the new waveform data to the new group, determining the new waveform data as the representative waveform data, and recording the adjacent group adjacent to the new group.
4. A product quality analysis support system according to claim 3, wherein the data processing device is configured to further perform the following: change the original representative waveform data to the provisionally set representative waveform data, and then review the records of the adjacent groups.
5. A product quality analysis support system according to any one of claims 1 to 4, wherein the data processing device is configured to further perform a normalization process to adjust the number of points in the waveform data to a predetermined number in order to calculate the distance between data, and the normalization process includes dividing the waveform data into a leading edge, a central part, and a trailing edge, and setting points individually for each of the leading edge, central part, and trailing edge.