JP5151556B2 - Process analysis apparatus, process analysis method, and process analysis program - Google Patents

Description

  The present invention relates to a technique for modeling a relationship between apparatuses in a production line.

  When an abnormality occurs in a production line such as a factory, the entire line stops and production stops. Therefore, maintenance personnel periodically patrol the production line and inspect whether there is an abnormality or a sign of abnormality. The detection of abnormalities often relied on the five senses of maintenance personnel.

  When an abnormality or a sign of the abnormality is detected, there may be a real cause of the abnormality in the device in the process prior to the device in which the abnormality is detected. Therefore, in order to identify the cause of the true abnormality, it is necessary to understand the relationship between the devices in the production line. In the production line, the number of devices may be enormous (several hundreds or more), so it is difficult to grasp the relationship between the devices, and relies on the experience and intuition of skilled maintenance personnel.

  The relationship between devices is defined by design information, but changes daily due to device deterioration, lot type, parameter adjustment in improved maintenance, and the like. Therefore, unskilled maintenance personnel cannot grasp the relationship between devices only by design information.

  Most of the repair time when an abnormality occurs in the production line is spent investigating the cause of the failure. Moreover, the time required for investigating the cause of the failure varies greatly depending on the skill level of maintenance personnel, and in the case of a non-skilled person, it may take several times as long as the skilled person. In spite of this situation, recently, there are increasing opportunities for unskilled maintenance personnel to maintain alone. Therefore, a technique for making maintenance work easier by using a computer without relying on the experience and intuition of an expert is being researched.

  For example, Patent Document 1 discloses a technique for creating a process-quality model from process state data and quality inspection data. According to this technology, it is possible to detect the occurrence of an abnormal product by monitoring the process state data.

Patent Document 2 also discloses a technique for creating a model representing the relationship between the production process and quality. In Patent Document 2, the correlation strength between the manufacturing history and the product quality history is obtained. A causal structure model of the production line process is created using the correlation strength and the production order information. Using this causal structure model, it is possible to narrow down the fundamental causes of fluctuations that have caused the product quality fluctuations.
JP 2005-197323 A JP 2006-65598 A

  However, in the case of the prior art as described above, the following problems have occurred. That is, an abnormality in the production line may appear not only as a defective product but also as a symptom that the production speed is slow. In the method of Patent Document 1, it is not possible to represent an abnormality of the apparatus that does not appear in the quality with a model.

In the method of Patent Document 2, a causal structure model is created, and the relationship between devices can be grasped. However, in order to create this causal structure model, it is necessary to give order information between processes based on the design information of the production line to the system. Therefore, a correct causal structure model cannot be created when design information cannot be acquired in advance or when the design information and the actual production line operate differently.

  In an actual production line, the operation order changes due to device deterioration, lot type change, improved maintenance, and the like. It is troublesome and unrealistic to correct the design information by acquiring this every time.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to easily model a relationship between apparatuses in a production line.

  In order to achieve the above object, a process analysis apparatus according to the present invention analyzes a relationship between a plurality of signals input and output substantially periodically. Signals to be analyzed include, for example, control signals and response signals that are input and output between the manufacturing apparatus and the control apparatus in a production line that includes a plurality of manufacturing apparatuses and a control apparatus that controls these manufacturing apparatuses. Signal. When the tact production method is performed in such a production line, signals input / output between the plurality of manufacturing apparatuses and the control apparatus are exchanged substantially periodically.

  The process analysis apparatus according to the present invention includes a signal acquisition unit, a feature amount calculation unit, an occurrence order extraction unit, and a causal structure creation unit. The signal acquisition means acquires signals (control signals and response signals) input / output between the plurality of manufacturing apparatuses and the control apparatus. The feature amount calculating means calculates a feature amount from the acquired signal. A feature amount of a certain signal can be calculated as a difference in rising time from an arbitrary reference signal. Note that, as described above, signals are input and output periodically, so that the feature amount is calculated in each cycle.

  The generation order extraction unit extracts the generation order of the acquired signals. Here, the generation order can be determined based on a value obtained by calculating the generation order of signals in each period and averaging them over the entire period.

  The causal structure creation means creates a causal structure between the signals based on the correlation between the feature quantities and the signal generation order. It is preferable that the causal structure creation means creates a causal structure for each signal, using a signal having the highest correlation with the signal among the signals in the order of generation before the signal as the cause of the signal.

  Thus, by paying attention to the relationship between signal generation and correlation, it is possible to create a causal structure (causal structure tree) between signals. Since the causal structure is created by acquiring signals exchanged between the manufacturing equipment and the control equipment, even if the operating order changes, such as equipment deterioration or parameter correction due to improved maintenance, it is automatically performed. In addition, a causal structure is created so that human hands are not bothered. Further, when an abnormality is detected, it is possible to narrow down the range of manufacturing apparatuses to be inspected when searching for the cause of the true abnormality by referring to the causal structure.

  Moreover, it is preferable that the process analysis apparatus according to the present invention includes display means for displaying the causal structure created as described above. More preferably, the display means obtains a name corresponding to the signal from the design information and displays the causal structure using the name.

Moreover, it is preferable to estimate in which part of the causal structure between signals the abnormality has occurred using the process analysis apparatus according to the present invention. Specifically, using the signal data for learning, the storage means for storing the causal structure between signals obtained in advance and the time difference between signals having a causal relationship as learning results, and the signal data for diagnosis It is preferable to further include an abnormal point estimation unit that calculates a time difference between signals having causal relationships by using and estimates whether there is an abnormality in the diagnostic data in comparison with the learning result.

  In this case, the abnormal part estimation means estimates that a part where the difference between the time difference between the causal signals in the diagnostic signal data and the time difference in the learning result is larger than a predetermined threshold is an abnormal part. Is preferred.

  Furthermore, it is preferable to have a display unit that displays the causal structure between signals while emphasizing the estimated abnormal part.

  In this way, using the learning data beforehand, the occurrence time difference between the signals at the normal time is examined, and when the abnormality occurs, the place where the time difference between the signals is greatly deviated from the normal time is regarded as the abnormal place. It can be presented to the user. Therefore, when an abnormality is detected, the user can quickly cope and repair.

  In addition, this invention can be grasped | ascertained as a process analysis apparatus which has at least one part of the said means. The present invention can also be understood as a process analysis method including at least a part of the above processing, or a program for realizing the method. Each of the above means and processes can be combined with each other as much as possible to constitute the present invention.

  For example, the process analysis method as one aspect of the present invention is a process analysis method for analyzing a relationship between a plurality of signals input and output substantially periodically, and the information processing apparatus acquires the plurality of signals. And the step of calculating the feature quantity from the acquired signal, the step of extracting the generation order of the acquired signal, and the causal structure between the signals based on the correlation between the feature quantity and the signal generation order And step to perform

  The process analysis program as one aspect of the present invention is a process analysis program for analyzing a relationship between a plurality of signals input and output substantially periodically, and the information processing apparatus acquires the plurality of signals. And the step of calculating the feature quantity from the acquired signal, the step of extracting the generation order of the acquired signal, and the causal structure between the signals based on the correlation between the feature quantity and the signal generation order And executing a step.

  According to the present invention, it is possible to easily model the relationship between devices in a production line.

  Exemplary embodiments of the present invention will be described in detail below with reference to the drawings.

(First embodiment)
The process analysis apparatus 1 according to the present embodiment transmits signals exchanged between a plurality of manufacturing apparatuses 2 related to a manufacturing line and a programmable logic controller (PLC) 3 that is a control apparatus that controls the manufacturing apparatuses 2. Acquire and create a causal structure between these signals.

  FIG. 1 is a diagram illustrating a functional configuration of the process analysis apparatus. The process analysis apparatus 1 includes a signal measurement collection unit 11, a feature amount calculation unit 12, an occurrence rank extraction unit 13, a ranked correlation coefficient calculation unit 14, a causal structure acquisition unit 15, and a display unit 17. The process analysis apparatus 1 is configured by a computer (information processing apparatus), and each function unit described above is realized by a program stored in a memory being executed by a CPU (central processing unit). In addition, the process analysis apparatus 1 may be comprised by one computer, and may be comprised by the some computer connected by the network.

[Signal Measurement Collection Unit]
PLC3 receives the input of a sensor and a response signal, and sends the control signal for controlling the manufacturing apparatus 2 according to ladder logic. The signal measurement collection unit 11 collects signals exchanged between the manufacturing apparatus 2 and each contact point of the PLC 3. The signal measurement collection unit 11 is connected to the PLC 3 via a bus, and is a unit that reads the I / O memory of the PLC 3 and stores it as a file. The signal measurement collection unit 11 may collect data from the PLC 3 by Ethernet (registered trademark) or serial connection in addition to the bus connection. Moreover, the signal collection measurement part 11 does not need to be directly connected with PLC3, if the signal data of PLC3 can be acquired. For example, the PLC 3 signal data may be temporarily stored in a storage medium such as a CD-ROM, DVD-ROM, or memory card, and the signal data may be acquired from the storage medium.

  FIG. 2 is a flowchart showing an operation flow of the signal measurement collection unit 11. The signal measurement collection unit 11 measures and collects signals exchanged between the manufacturing apparatus 2 and the PLC 3 (S201). FIG. 3 is a diagram illustrating an example of signals collected by the signal measurement collection unit 11. The signal measurement collection unit 11 stores the signal on / off at each time point in the memory as the signal information 101 and stores it (S202). FIG. 4 shows an example of the signal information 101. The signal information 101 stores, for each contact, whether the signal at each time t is on (1) or off (0).

  In a production line composed of the production apparatuses 2, products are produced by a tact production method, and the operation of each production apparatus 2 is synchronized with a certain cycle (tact cycle). Therefore, the signal exchanged between the manufacturing apparatus 2 and the PLC 3 has the periodicity of this tact cycle.

  Here, when an arbitrary signal is defined as a reference signal, all other signals operating in one cycle of the reference signal can be regarded as data relating to behavior in the cycle. FIG. 5 is a diagram illustrating a signal cycle. For example, when the top signal in FIG. 3 is defined as a reference signal, each cycle is divided by the rising edge of the reference signal (in the figure, three cycles of i−1 cycle, i cycle, and i + 1 cycle are depicted. ). The cycle defined in this way defines which cycle the other signals are turned on / off.

  As to which signal is adopted as the reference signal, a method of adopting the signal of the first operating device as the reference signal can be considered.

[Feature amount calculator]
Next, the feature amount calculation unit 12 will be described. The feature amount calculation unit 12 calculates the feature amount of the time series signal for each contact collected by the signal measurement collection unit 11. Although various values can be considered as the feature value of the time series signal, since the present process analysis apparatus 1 focuses on the relationship between the signals, the difference in the state change time between the reference signal and the target signal is calculated. The feature amount of the target signal. More specifically, as shown in FIG. 6A, the difference between the rising time of the reference signal and the rising time of the target signal can be used as the feature amount. In the present embodiment, such a feature amount is adopted. As shown in FIG. 6B, the difference (a) between the rising time of the reference signal and the falling time of the target signal, the falling time of the reference signal and the target signal. The difference between the rise times (b) or the difference between the fall time of the reference signal and the fall time of the target signal (c) may be used as the feature amount.

  Note that the feature amount of the signal is defined for each contact and each cycle.

FIG. 7 is a flowchart showing the flow of the feature amount calculation process performed by the feature amount calculation unit 12. The feature amount calculation unit 12 first sets 1 to j, which is an index representing the target signal (S701). Next, the state change time information 102 is created by extracting the state change time of the signal s _j from the signal information 101 (see FIG. 4) created by the signal measurement collection unit 11 (S702). That is, the signal information 101 is inspected, the time when the signal state changes from on (1) to off (0), or from off (0) to on (1) is extracted and stored as state change time information 102. To do. An example of the state change time information 102 is shown in FIG. The state change time information 102 stores the time at which the signal on / off state is switched for each contact.

  The feature amount calculation unit 12 determines whether or not the state change time extraction has been completed for all N signals (S703). If there is an unprocessed signal (S703-YES), j is incremented (S704). The state change time is extracted for the next signal.

  When the extraction of the state change times for all N signals is completed (S703-NO), processing for calculating the feature amount from the state change time information 102 is performed. First, one signal out of N signals is set as a reference signal (S705). Here, the signal whose state has changed at the earliest time is selected as the reference signal.

Then, 1 is set to j, which is an index representing the target signal, and 1 is set to i, which is an index representing the target cycle (S706). The feature amount calculation unit 12 calculates the feature amount of the signal s _j in the cycle i from the state change time information 102 and stores it as the feature amount data 103 (S707). Specifically, the rising time of the reference signal in cycle i and the rising time of signal s _j are acquired, and the difference is calculated as feature quantity f _j of signal s _j in cycle i. An example of the feature amount data 103 is shown in FIG. The feature value data 103 stores the value of the feature value for each signal (contact point) and for each cycle.

  Next, it is determined whether or not feature amount extraction has been completed for all N signals (S708). If there is an unprocessed signal (S708-YES), j is incremented (S709) and the next signal Extract feature values.

  When feature value extraction has been completed for all N signals in cycle i (S708-NO), it is determined whether feature value extraction has been completed for all M cycles (S710). If there is a cycle in which the feature quantity has not yet been extracted, the cycle index i is incremented, 1 is set to the signal index j (S711), and the feature quantity is extracted for the next cycle.

  When the feature amount extraction is completed for all the cycles (S710—NO), the feature amount calculation process by the feature amount calculation unit 12 ends.

[Occurrence rank extraction unit]
The generation order extraction unit 13 extracts the generation order of the rise of each signal. Here, since the generation order of a certain signal may change depending on the cycle, the generation order is obtained for each cycle, and the generation order is determined according to the magnitude of the value (average rank).

  Details of the generation order extraction process performed by the generation order extraction unit 13 will be described with reference to FIGS. FIG. 10 is a flowchart showing the flow of occurrence order extraction processing. FIG. 11 is a diagram for explaining the order of occurrence in each cycle. FIG. 12 is a diagram illustrating the relationship between the average value (weight) of the occurrence order for each cycle and the occurrence order to be calculated.

The generation order extraction unit 13 first obtains the generation order of each signal in each cycle. Therefore, 1 is set to the index i representing the cycle (S1001). Then, referring to the state change time information 102 (FIG. 8), the rising time of each signal in cycle i is extracted (S1002). Then, the generation order of each signal in cycle i is obtained and stored as the generation order 104 for each cycle (S1003). Taking FIG. 11 as an example, in cycle i = 1, the generation order of signal s ₁ is “3”, the generation order of signal s ₂ is “7”, the generation order of signal s ₃ is “1”, and so on. become.

  For all M cycles, it is determined whether the generation order extraction for each cycle has been completed (S1004). If there is a cycle that has not yet been processed (S1004-YES), the cycle index i is incremented (S1005). The next cycle is processed.

As described above, there may be variations in the order of generation of each signal in each cycle. In FIG. 11, the signal s ₁ whose generation order is “3” in cycle i = ₁ is different in cycle i = 2. The occurrence order is “2”.

When the extraction of the occurrence rank for each cycle is completed for all cycles (S1004-NO), the average of the occurrence rank for all cycles is calculated for each signal (S1006). This value is obtained by the following formula. Note that ord _j (i) indicates the generation order of the signal s _j in the cycle i.

Specifically, taking FIG. 11 as an example, the generation order of the signal s ₁ for each cycle is “3”, “2”, and “3”, so the average order is “2.67”. Similarly, since the generation rank of the signal s ₂ for each cycle is “7”, “7”, “6”, the average rank is “6.67”. The same applies to other signals.

Then, the occurrence rank extraction unit 13 sorts the average of the occurrence ranks for each cycle in ascending order, determines the occurrence rank of each signal, and stores it as the occurrence rank 105 (S1007). FIG. 12 is a diagram showing an example of the generation order 105, and it can be seen that the generation order (ORD _j ) of each signal is determined in the order of the average value (ord _j ) of the generation order for each cycle.

[Ranked correlation coefficient calculation unit]
Next, the process for obtaining the correlation between signals performed by the ranked correlation coefficient calculation unit 14 will be described. FIG. 13 is a flowchart showing the flow of processing performed by the ranked correlation coefficient calculation unit 14. FIG. 14 is a diagram for explaining the processing.

The ranked correlation coefficient calculation unit 14 rearranges the feature amount data 103 of each signal s _j based on the occurrence rank 105 extracted by the occurrence rank extraction unit 13 (S1301). That is, as represented by the following expression, the variable x _j (i) is a feature quantity f _j (i) labeled with the occurrence order ORD _j .

When the occurrence rank 105 is obtained as shown in FIG. 12, as shown in FIG. 14, x ₁ is f ₃ with the occurrence rank being “1”, and x ₂ is the occurrence rank “2”. the f _5. In this way, the feature quantity data f _j is rearranged in the order of occurrence to obtain x _j . Since the rearrangement of the signals is performed based on the generation order, when j <k, x _k is a signal that occurs after x _j (having a high probability).

Next, the ranked correlation coefficient calculation unit 14 calculates the correlation coefficient matrix 106 between x _j (i) (S1302). Each element ρ _jk of the correlation coefficient matrix 106 is calculated as follows.

An example of the calculated correlation coefficient matrix 106 is shown in FIG. Since the correlation coefficient matrix is a symmetric matrix, it is sufficient to obtain only the lower triangular (or upper triangular) matrix. Further, as will be described later, it is only necessary to obtain j ≧ 3 in the correlation coefficient matrix ρ _jk .

[Causal structure acquisition unit]
The causal structure acquisition unit 15 creates a causal structure between signals based on the correlation coefficient calculated by the ranked correlation coefficient calculation unit 14. FIG. 15 is a flowchart showing the flow of causal structure acquisition processing performed by the causal structure acquisition unit 15.

First, x ₁ → x ₂ is defined as a causal structure (S1501). Note that “x _j → x _k ” means that the signal x _k is generated due to the generation of the signal x _j . Signal which causes some signals are the signals generated before the signal, the signal which causes x ₂ always becomes x _1.

Next, 3 is set to the index j representing the signal (S1502). Then, with reference to the correlation coefficient matrix 106, k that maximizes ρ _jk (k <j) is obtained. In this process, x _k satisfying k <j, that is, a signal having the highest correlation with x _j is searched for among signals whose generation order is earlier than x _j . In this way, k is obtained, and x _k → x _j is stored in the causal relationship list 107.

  It is determined whether all the signals have been processed (S1505). If there is a signal that has not been processed (S1505-YES), j is incremented (S1506), and the signal that causes the next signal is obtained. .

Specifically, taking FIG. 16 as an example, first, x ₁ → x ₂ is stored. Then, with j = 3, the largest of the correlation coefficients ρ ₃₁ and ρ ₃₂ is obtained. Here, it is assumed that ρ ₃₁ is the largest. In this case, a causal relationship of x ₁ → x ₃ is obtained.

Next, assuming j = 4, the largest one of the correlation coefficients ρ ₄₁ , ρ ₄₂ , ρ ₄₃ is obtained. Here, it is assumed that ρ ₄₂ is the largest. In this case, a causal relationship of x ₂ → x ₄ is added.

Further, assuming that j = 5, the largest of the correlation coefficients ρ ₅₁ , ρ ₅₂ , ρ ₅₃ , ρ ₅₄ is obtained. Here, it is assumed that ρ ₅₂ is the largest. In this case, a causal relationship of x ₂ → x ₅ is added.

In this way, by the processing for all the signals, tree-like causal structure list which starts x ₁ is created. Since the causal structure is created as described above, only one signal may cause a certain signal, whereas one signal may cause a plurality of signals.

[Display section]
The display unit 17 displays the causal relationship between the signals created as described above. At this time, the contact name of the signal is acquired by referring to the design information, and the display is performed using the contact name. Design information here refers to knowledge of people such as experts, production line design drawings, information held in the PLC (PLC ladder, signal names, character strings with explanations, device control, etc. Parameter).

  FIG. 17 is an image of the output screen when the user confirms the causal structure. As shown in the figure, the causal relationship of the signals in the production line is analyzed by the process analysis apparatus 1 and the causal structure is displayed as a graph on the display unit 17 such as a display device.

  FIG. 18 is a flowchart showing the flow of display processing performed by the display unit 17. The display unit 17 acquires a tree structure from the causal relationship list 107 and acquires a contact name corresponding to the variable name from the design information 110 (S1901). Then, the causal structure in which the variable name is replaced with the signal contact name is displayed in a graph (S1902). When the variable name is replaced with the signal contact name, the variable name is associated with the signal contact name according to the following equation.

  In the displayed graph, as shown in FIG. 19, the variable name is replaced with a contact name of a signal such as “cylinder 1 output drive” or “cylinder 1 return sensor”, so that the relationship between the devices becomes easy to understand.

The display unit 17 may display the entire causal structure as shown in FIG. 20, when selecting any variable on the causal structure as shown in FIG. 21 to 23, and the variable and x ₁ A route formed by connecting the lines with a single line may be displayed. FIG. 21 shows a path obtained by tracing the signal causing the variable x ₃ (the contact name of the signal is “cylinder 1 output sensor”) to the variable x ₁ . Similarly, FIG. 22 shows a path obtained by tracing the signal causing the variable x ₄ (“cylinder 2 output sensor”) to the variable x ₁ . Similarly, FIG. 23 shows a path obtained by tracing the signal causing the variable x ₅ (“cylinder 1 return sensor”) to the variable x ₁ . 20 to 23, the variable name “x _j ” is displayed next to the contact name. However, this is for explanation, and it is not actually necessary to display this variable name.

<Operation / Effect of Embodiment>
According to the process analysis apparatus 1 according to the present embodiment, order information (prior relationship) is acquired from time-series information of signals actually exchanged between the manufacturing apparatus 2 and the PLC 3, and the order information and the feature amount are obtained. Based on the correlation, a causal relationship between the signals can be created. By looking at the relationship between the signals, the relationship between the devices can be easily understood, so that when a failure or a sign thereof is detected, it is very useful for identifying the true cause. In particular, if only the path that follows the cause of the signal in which an abnormality has occurred is displayed by the function of displaying only a part of the causal relationships as shown in FIGS. 21 to 23, the true cause location is narrowed down and displayed. . Maintenance personnel only need to examine the devices related to the displayed route, and maintenance work is facilitated.

  In an actual production line, the operation order may change due to deterioration of the apparatus or improved maintenance. Although it takes a lot of work to correct the design information by acquiring this every time, the process analysis apparatus 1 can automatically acquire the causal relationship between the apparatuses by monitoring the signal. It is possible to obtain the current state of the production line.

<Modification>
In the description of the present embodiment, the signals are rearranged according to the generation order, the correlation coefficient is calculated, and for each signal x _j (3 ≦ j ≦ N), the signal x _{k with} the highest correlation (1 ≦ k < Although j) is the cause of x _j , the specific method for creating the causal structure is not limited to this.

The causal structure can be created by any method as long as the signal having the highest correlation with the target signal is determined as the cause of the target signal among the signals whose generation order is earlier than that of the target signal. The causal structure may be created by. For example, the correlation coefficient may be obtained without rearranging the signal s _j , and a signal satisfying the above condition may be searched for as a cause of the target signal.

(Second Embodiment)
In the said 1st Embodiment, the causal structure between each signal exchanged between PLC3 and the manufacturing apparatus 2 is acquired, and it uses for the maintenance of the manufacturing apparatus 3 by providing to a user (maintenance worker). . In the present embodiment, the acquired causal structure is utilized to identify an abnormal part of the manufacturing apparatus 3 and notify the user.

  FIG. 24 is a diagram showing a functional configuration of the process analysis apparatus 1 according to the present embodiment. The process analysis apparatus 1 according to the present embodiment is broadly divided into a causal structure learning unit 1a and a causal structure utilization unit 1b.

  The causal structure learning unit 1a creates and learns a causal structure in advance based on a signal (learning signal data) from the PLC 3 when the manufacturing apparatus 2 is operating normally. In this case, it is only necessary for a human to determine whether or not the manufacturing apparatus 2 is in a normal state. For example, after completion of a trial run at the time of introduction of the apparatus, after completion of maintenance, signal data prepared separately for learning The causal structure may be acquired based on the above. The acquired causal structure is stored in the learning result storage unit 24. The causal structure utilization unit 1b compares the learning result with a signal (diagnosis data) during normal operation, and diagnoses the operation status of the production line.

  Since it is assumed that the learning signal data is data in a normal state, “learning data” or the like may be hereinafter expressed as “normal data” or the like. In addition, since it is assumed that the diagnosis of the operating status is performed during normal operation, “diagnosis data” may be expressed as “normal data”, “normal operation data”, etc. .

<Causal Structure Learning Department>
As shown in FIG. 24, the causal structure learning unit 1 a includes a signal measurement collection unit 11 and a feature amount calculation unit 12.
, An occurrence rank extraction unit 13, a ranked correlation coefficient calculation unit 14, a causal structure acquisition unit 15, and a time difference calculation unit 20. Here, except for the time difference calculation unit 20, since it is the same as that described in the first embodiment, a detailed description is omitted, but a causal relationship is created by each of these functional units.

  The time difference calculation unit 20 calculates a difference (time difference) in occurrence time between signals having a causal relationship. Hereinafter, processing performed by the time difference calculation unit 20 will be described with reference to FIGS. FIG. 25 is a flowchart showing a process flow of the time difference calculation unit 20. FIG. 26 is a diagram illustrating a relationship between difference data for which the time difference calculation unit 20 calculates a time difference.

[Time difference calculator]
First, the time difference calculation unit 20 sets an index i representing the number of cycles and an index j representing the number of signals to 1 (S2501). Then, the causal relationship list 107 is referenced to search for the signal X _j that causes the signal x _j (S2502). For example, as shown in FIG. 26, if the causal relationship _{"x 1} → _{x 3"} is stored in the causal relation list 107, it causes the signal _{X 3} of the signal _{x 3} is a signal _{x 1.} The signal x ₃ in this case is the result signal for the cause signal.

Next, the time difference calculation unit 20 uses the state change time information 102 acquired from the signal data for learning and the occurrence order 105 to determine the difference between the occurrence times of the signal x _j and the cause signal X _j in the cycle i (x _j (I) −X _j (i)) is calculated (S2503). Here, the signal x _i are those rearranged in chronological order according to the formula a signal contacts s _k (number 4), the correspondence between the signal x and the contact s, obtained from the generator rank list 105. As shown in FIG. 26, taking as an example the situation of _{"x 1} → _{x 3",} from the generation rank list 105, the signal _{x 1} corresponds to the signal contacts _{s 3,} the signal _{x 3} are contacts _{s 1} of the signal It can be seen that If the correspondence relationship between the signal x and the contact s is known, the time difference between the signals can be calculated by obtaining the difference in the generation time of each contact in the cycle i from the state change time information 102. In the example of FIG. 26, the time difference “x ₁ → x ₃ ” can be obtained as a difference in signal generation time between the contact s ₃ and the contact s ₁ (s ₁ −s ₃ ). Is 10 milliseconds, and when i = 2, the time difference can be obtained as 11 milliseconds.

It is determined whether or not the time difference of all cycles has been obtained for the signal x _j (S2504). If there is an unprocessed cycle, the next cycle is processed (S2505). If processing has been performed for all cycles, it is determined whether time differences have been obtained for all signals (S2506). If there is an unprocessed signal, the next signal is processed (S2507).

  In this manner, the time difference calculation unit 20 obtains the difference in generation time between signals that are causal for all signals and all cycles, and creates time difference data 111. An example of the created time difference data 111 is shown in FIG. This calculation result is stored in the learning result storage unit 24 together with the causal relationship list 107 as time difference data 111 for learning.

<Causal structure utilization department>
The causal structure utilization unit 1b diagnoses whether the manufacturing apparatus 2 is operating normally by acquiring a signal from the PLC 3 during normal operation or the like and comparing it with the learning data during normal operation. Here, explanation will be made on the assumption that the causal structure utilization unit 1b obtains a signal from the PLC 3 constantly or periodically and diagnoses the operation status, but the maintenance staff usually instructs the diagnosis without making a diagnosis. Diagnosis may be made only at times.

As shown in FIG. 24, the causal structure utilization unit 1 b includes a diagnostic data signal measurement collection unit 21, a diagnostic data time difference calculation unit 22, and an abnormal point estimation unit 23. In FIG. 24, each functional unit of the causal structure utilization unit 1b is shown as being different from the causal structure learning unit 1a. However, this is for convenience of explanation, and the same program as the causal structure learning unit 1a. Each functional unit may be realized by the above.

[Data signal measurement and collection unit for diagnosis]
The diagnostic data signal measurement collection unit 21 acquires signal information from the PLC 3 during normal operation, and generates diagnostic signal information 201. FIG. 27 is a flowchart showing the contents of the process performed by the diagnostic data signal measurement / collection unit 21. Since the content of this process is the same as the process (flowchart in FIG. 2) of the signal measurement collection unit 11 of the causal structure learning unit 1a, the description is omitted. The signal acquired for diagnosis may be acquired over a plurality of cycles, or only one cycle may be acquired. In this embodiment, a diagnostic signal is acquired over a plurality of cycles.

[Diagnostic data time difference calculator]
Next, the processing of the time difference calculation unit 22 for diagnostic data will be described with reference to FIG. The time difference calculation unit 22 extracts the state change time information of each signal from the signal information 201 for diagnosis and creates the state change time information 202 for diagnosis (steps S2801 to S2804). This process is the same as part of the process of the feature quantity calculation unit 12 of the causal structure learning unit 1a (steps S701 to S704 in the flowchart of FIG. 7). Then, the time difference between the causal signals is calculated from the created diagnostic state change time information 202, and diagnostic time difference data 203 is created (steps S2805 to S2811). This process is the same as the time difference calculation process (FIG. 25) of the time difference calculation unit 20.

[Abnormal part estimation part]
The abnormal part estimation unit 23 compares the time difference data 111 for learning (during normal operation) with the time difference data 203 for diagnosis (during normal operation) to obtain abnormal part candidates. Here, a portion where the time difference between the signals is large between normal operation and normal operation is estimated as a portion having an abnormality.

  Hereinafter, the process of the abnormal part estimation part 23 is demonstrated with reference to FIG. FIG. 29 is a flowchart showing the flow of an abnormal location estimation process performed by the abnormal location estimation unit 23. FIG. 30 (a) is an example of an abnormal candidate list 204 to be created, and FIG. 30 (b) is a causal relationship with time difference information. A schematic diagram of the list is shown.

  The abnormal part estimation unit 23 first calculates a production cycle from the state change time information 201 for diagnosis (S2901). This production cycle can be calculated by calculating the cycle time of a signal representing a machine cycle among signals acquired for diagnosis. The signal representing the machine cycle is a signal that is used as a reference signal in the feature amount calculation unit 12 of the causal structure learning unit 1a and that is remeasured and collected as diagnostic data. The cycle time is expressed as a time difference from the rising edge of this reference signal to the next rising edge.

  Next, the abnormal part estimation part 23 reads the threshold value of a production cycle (S2902). The production cycle threshold value is a predetermined set value and can be set by the user. An example of a production cycle threshold setting screen is shown in FIG. As shown in FIG. 32, on a threshold setting screen on a display device such as a display, the user sets a threshold value using an input device such as a keyboard or a mouse. In this example, an upper limit value of a production cycle that is determined to be normal is input. For example, when it is determined that the production line scheduled for 30 seconds is normal up to 32 seconds, 32 seconds is input on the setting screen of FIG. However, a normal production cycle (for example, 30 seconds) and an allowable upper limit value (for example, 2 seconds) may be input. Further, instead of setting the threshold value via the display device and the input device, the threshold value may be set by reading a setting file or the like.

The abnormal part estimation unit 23 reads the production cycle threshold value, and then performs step S2901.
It is determined whether the production cycle calculated in (1) exceeds a threshold value (S2903). If cycle over (tact delay) has not occurred (S2903-NO), it is determined that no abnormality has occurred, and the abnormal point estimation process is not performed. On the other hand, when the cycle over has occurred (S2903-YES), the abnormal part is estimated.

In step S2904, from the learning time difference data 111 and the diagnostic time difference data 203, an average of the time differences between the causal signals for the learning and the diagnostic is obtained, and the difference between these two average values is calculated. . In the time difference data as shown in FIG. 26, the time difference between causal signals (such as x ₁ → x ₃ ) is stored for each cycle. By taking this average, the average value of the time difference data can be obtained.

  In the flowchart of FIG. 29, the average value of the time difference data for the learning data is obtained for each abnormal part estimation process. However, since the average value for the learning data is a value that does not change, it is obtained and stored only once. You can keep it.

Assume that the time difference between the signals is as shown in FIG. In this case, the average time difference between the signals x ₁ and the signal x ₃ is 10 milliseconds in the learning data, is 150 ms in the diagnostic data, the difference is 140 ms. Similarly the average time difference between the signals x ₁ and the signal x ₂ is 30 ms in the learning data, at 35 milliseconds diagnosis data, the difference is 5 ms. In this way, for each of the causal signals, the difference between the time difference data average value learning time and the diagnosis time is calculated. The deviation of the calculated time difference data average value is stored in the abnormality candidate list 204.

  The data in the abnormality candidate list 204 is sorted in descending order of time difference (displacement) (S2905). That is, abnormality candidate ranks are added to the abnormality candidate list 204 in descending order of time difference as shown in FIG. Note that the sort processing here refers to processing for determining the order of time difference deviation, and does not necessarily involve data movement (replacement) on the memory.

<Display section>
The display unit 17 displays a causal relationship list created from the learning signal data. In the present embodiment, when there is an abnormality in the production line, the abnormal part is highlighted and displayed.

  The display process performed by the display unit 17 will be described with reference to the flowchart of FIG. First, the display unit 17 acquires a causal relationship (tree structure) between signals from the causal relationship list 107, refers to the design information 110, and acquires a contact name corresponding to each signal as a variable name (S3101). Then, the causal structure is displayed in a graph using the contact name of the signal.

  Next, the display unit 17 determines whether an abnormality candidate list has been created, that is, whether there is an abnormality in the production cycle (S3103). If the abnormality candidate list 204 has not been created, no further display is performed. Accordingly, the output screen when the abnormality candidate list 204 is not created is as shown in FIG. FIG. 34 shows an output screen and data used to create the output screen. When there is no abnormality, a tree structure in which the causal structure is replaced with design information (signal name) is displayed on the output screen.

  On the other hand, if an abnormality candidate list has been created, that is, if there is an abnormality in the production cycle (threshold is exceeded) (S3103—YES), the abnormality candidate list 204, normal (for learning) time difference data 111, with reference to the normal (diagnostic) time difference data 203, the abnormality candidate display number 205, and the abnormality candidate display threshold 206, an abnormal part is highlighted among the already displayed causal structures (S3104). .

  FIG. 35 shows an output screen on which the estimated abnormality location is highlighted. FIG. 35 shows an output screen and data used to create the output screen. In this example, two places are highlighted as abnormal places.

  Here, depending on the settings of the abnormality candidate display number 205 and the abnormality candidate display threshold 206, it is determined which of the abnormality candidates is highlighted. The abnormality candidate display number 205 and the abnormality candidate display threshold 206 are input through a setting screen as shown in FIG. Or you may comprise so that a setting file may be read. The abnormality candidate display number 205 is a setting value that determines how many places are displayed as abnormality candidates. The abnormality candidate display threshold 206 is a setting value that determines how large a time difference shift is to be treated as an abnormality.

  Only one of the abnormality candidate display number 205 and the abnormality candidate display threshold 206 may be set. When only the abnormality candidate display number 205 is set, the number of set values is highlighted in order from the top of the abnormality candidate list 204. When only the abnormality candidate display threshold 206 is set, a portion of the abnormality candidate list 204 where the time difference shift is equal to or larger than the set value is highlighted. When both of the abnormality candidate display number 205 and the abnormality candidate display threshold 206 are set, from the abnormality candidate list 204, the part having the larger deviation among the positions where the deviation of the time difference is equal to or larger than the set value is set. Will be highlighted. In the example of FIG. 33, only the number of abnormal candidate displays is input, and in this case, as shown in FIG. 35, two locations with a large time difference are highlighted.

In FIG. 36, there is a large time difference between the signal x ₂ (cylinder 2 output drive) and the signal x ₄ (cylinder 3 output drive) (x ₂ → x ₄ ) from the normal state, and this point is abnormal. It is estimated that. Here, the result signal is highlighted. However, various methods can be adopted as a method for highlighting abnormal portions. FIG. 35 shows an example of highlighting.

  FIG. 35A shows a method of highlighting the result signal portion as shown in FIG. FIG. 35B shows a method for highlighting the cause signal portion. FIG. 35C shows a method of highlighting the transition portion between the cause signal and the result signal. FIG. 36D shows a method for highlighting all of the above. The highlighting may be performed by any method as long as the user can easily recognize the abnormal part on the output screen.

<Operation and effect of the embodiment>
According to the process analysis apparatus 1 according to the present embodiment, a causal relationship is created from time-series information of signals exchanged between the manufacturing apparatus 2 and the PLC 3, and a difference in generation time between signals having a causal relationship is acquired. it can. During normal operation, the time difference between the causal signals can be monitored. When an abnormality occurs, it is possible to estimate a portion where the time difference between the signals is large as an abnormal portion and display the screen in a format that is easy for the user to recognize.

  As described above, in the process analysis apparatus 1 according to the present embodiment, when an abnormality occurs, the abnormal part is automatically identified and displayed to the user, so that prompt handling and repair are possible. If the tact delay occurs, the manufacturing efficiency is lowered accordingly. Loss can be minimized by being able to deal with it quickly.

It is a figure which shows the functional block of the process analysis apparatus which concerns on 1st Embodiment. It is a flowchart which shows the flow of the process which the signal measurement collection part 11 performs. It is a figure which shows the example of the signal collected by the signal measurement collection part. It is a figure which shows the example of the signal information 101 memorize | stored by the signal measurement collection part. It is a figure explaining the cycle of a signal. It is a figure explaining the feature-value of the signal which the feature-value calculation part 12 calculates. It is a flowchart which shows the flow of the process which the feature-value calculation part 12 performs. It is a figure which shows the example of the state time change information 102 which the feature-value calculation part 12 produces. It is a figure which shows the example of the feature-value data 103 which the feature-value calculation part 12 produces. It is a flowchart which shows the flow of the process which the generation | occurrence | production order extraction part 13 performs. It is a figure explaining the generation | occurrence | production order in each cycle which the generation | occurrence | production order extraction part 13 extracts. It is a figure which shows the example of the generation | occurrence | production order 105 which the generation | occurrence | production order extraction part 13 produces. It is a flowchart which shows the flow of the process which the correlation coefficient calculation part 14 with a ranking performs. It is a figure explaining the correlation coefficient calculation process which the correlation coefficient calculation part 14 with a ranking performs. It is a flowchart which shows the flow of the process which the causal structure acquisition part 15 performs. It is a figure explaining the process which the causal structure acquisition part 15 performs. FIG. 10 is a diagram illustrating an output screen displayed on the display unit 17. It is a flowchart which shows the flow of the process which the display part 17 performs. It is a figure explaining the process which the display part 17 performs. It is a figure which shows the example which displayed the whole causal structure by the graph display of the causal structure by the display part. It is a figure which shows the example which displayed only the one part path | route of the causal structure by the graph display of the causal structure by the display part 17. FIG. It is a figure which shows the example which displayed only the one part path | route of the causal structure by the graph display of the causal structure by the display part 17. FIG. It is a figure which shows the example which displayed only the one part path | route of the causal structure by the graph display of the causal structure by the display part 17. FIG. It is a figure which shows the functional block of the process analysis apparatus which concerns on 2nd Embodiment. It is a flowchart which shows the flow of the process which the time difference calculation part 20 performs. 3 is a diagram for explaining a data flow for creating time difference data 111 and a data structure of time difference data 111. FIG. It is a flowchart which shows the flow of the process which the data signal measurement collection part 21 for a diagnosis performs. It is a flowchart which shows the flow of the process which the diagnostic data time difference calculation part 22 performs. It is a flowchart which shows the flow of the process which the abnormal location estimation part 23 performs. It is a figure which shows the example of the abnormality candidate list 204 which the abnormal location estimation part 23 produces. It is a flowchart which shows the flow of the process which the display part 17 performs. It is a figure which shows the example of the setting screen which inputs the threshold value for abnormality determination of a production cycle. It is a figure which shows the example of the setting screen which inputs the display number of an abnormal location, and the threshold value considered as abnormality. It is a figure which shows the example in case there is no abnormality in a manufacturing apparatus and the abnormal location estimation is not performed by the graph display of the causal structure by the display part 17. FIG. It is a figure which shows the example in case a manufacturing apparatus has abnormality and the abnormal estimated part is highlighted by the graph display of the causal structure by the display part 17. FIG. It is a figure which shows the example of the method of highlighting an abnormality estimated location in the graph display of the causal structure by the display part.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Process analysis apparatus 2 Manufacturing apparatus 3 Programmable logic controller 11 Signal measurement collection part 12 Feature-value calculation part 13 Generation | occurrence | production order extraction part 14 Ranking correlation coefficient calculation part 15 Causal structure acquisition part 17 Display part 20 Time difference calculation part 21 Diagnosis data Signal measurement collection unit 22 Diagnostic data time difference calculation unit 23 Abnormal part estimation unit 24 Learning result storage unit 101 Signal information 102 State change time information 103 Feature data 104 Generation order 105 per cycle Generation order 106 Correlation coefficient 107 Causal relation list 110 Design information 111 Time difference data for learning 201 Signal information for diagnosis 202 State change time information for diagnosis 203 Time difference data for diagnosis 204 Abnormality candidate list

Claims (19)

A process analysis device that analyzes the relationship between a plurality of signals that are input and output substantially periodically between a manufacturing device and a control device ,
Signal acquisition means for acquiring a control signal transmitted from the control device and a response signal corresponding to the control signal for each period ;
A feature amount calculating means for calculating a feature amount for each period of each signal acquired by the signal acquiring means;
Generation order extraction means for extracting the generation order for each period of the signals acquired by the signal acquisition means;
A causal structure creation means for creating a causal structure between signals based on a correlation between feature quantities and a signal generation order;
A process analysis apparatus comprising: The causal structure creating means creates a causal structure for each signal by using a signal having the highest correlation with the signal among the signals whose generation order is earlier than the signal as a cause of the signal. Item 2. The process analysis apparatus according to Item 1. The process analysis apparatus according to claim 1, wherein the generation order of each signal is calculated based on an average of the generation order in each cycle. The process analysis apparatus according to claim 1, wherein the feature amount of each signal is calculated as a difference in rise time between the signal and a predetermined reference signal. The process analysis apparatus according to claim 1, further comprising display means for displaying the created causal structure. The process analysis apparatus according to claim 5, wherein the display unit acquires a name corresponding to the signal from the design information, and displays the causal structure using the name. A process analysis method for analyzing a relationship between a plurality of signals input and output substantially periodically between a manufacturing apparatus and a control apparatus ,
Information processing device
Obtaining a control signal transmitted from the control device and a response signal corresponding to the control signal for each period ;
Calculating a feature amount for each period of each acquired signal ;
Extracting the order of occurrence of the acquired signals for each period ;
Creating a causal structure between signals based on the correlation between feature quantities and the order of signal generation;
Process analysis method to execute. In the step of creating the causal structure, for each signal, a causal structure is created by using a signal having the highest correlation with the signal among the signals whose generation order is earlier than the signal as a cause of the signal. The process analysis method according to claim 7. The process analysis method according to claim 7 or 8, wherein the generation order of each signal is calculated based on an average of the generation order in each cycle. The process analysis method according to claim 7, wherein the feature amount of each signal is calculated as a difference in rise time between the signal and a predetermined reference signal. The process analysis method according to claim 7, further comprising a step of displaying the created causal structure. The process analysis method according to claim 11, wherein in the step of displaying the causal structure, a name corresponding to the signal is acquired from the design information, and the causal structure is displayed using the name. A process analysis program for analyzing a relationship between a plurality of signals that are input and output substantially periodically between a manufacturing apparatus and a control apparatus ,
In the information processing device,
Obtaining a control signal transmitted from the control device and a response signal corresponding to the control signal for each period ;
Calculating a feature amount for each period of each acquired signal ;
Extracting the order of occurrence of the acquired signals for each period ;
Creating a causal structure between signals based on the correlation between feature quantities and the order of signal generation;
Process analysis program to execute A storage means for storing a causal structure between signals obtained in advance and a time difference between signals having a causal relationship as learning results using signal data for learning;
An abnormality location estimating means for calculating a time difference between causal signals using diagnostic signal data and estimating an abnormal location in the diagnostic data by comparing with a learning result. The process analysis apparatus according to claim 1. The abnormal location estimating means estimates that a location where a difference between a time difference between causal signals in the diagnostic signal data and a time difference in the learning result is larger than a predetermined threshold is an abnormal location. The process analysis apparatus according to claim 14, wherein the process analysis apparatus is characterized in that: The process analysis apparatus according to claim 14, further comprising: a display unit that displays the causal structure between the signals while highlighting the estimated abnormal part. The information processing apparatus further includes:
Using the learning signal data, determining and storing a causal structure between signals and a time difference between signals having a causal relationship;
Using the diagnostic signal data, calculating a time difference between the causal signals, and comparing the learning result with estimating the abnormal portion in the diagnostic data;
The process analysis method according to claim 7, wherein the process analysis method is executed. In the step of estimating the abnormal location, a location where the difference between the time difference between the causal signals in the signal data for diagnosis and the time difference in the learning result is larger than a predetermined threshold is estimated as an abnormal location. The process analysis method according to claim 17. The process analysis method according to claim 17 or 18, wherein the information processing apparatus further executes a step of displaying a causal structure between signals while emphasizing an estimated abnormal part.

Images