JP2013041173A - Failure prediction system and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique, even when there is an abnormality, caused by data abnormality of one parameter, in data of another parameter, effectively specifying a causal parameter and using it for improving accuracy of failure prediction.SOLUTION: A correlation coefficient calculation part 5 generates a correlation data string for each parameter pair for every parameter group consisting of a plurality of parameters. A threshold comparison part 6 determines whether or not a correlation coefficient exceeding a threshold is included in the correlation data string. When a parameter pair including the correlation coefficient exceeding the threshold is detected, a similarity calculation part 7 calculates a distance between the correlation data strings of each parameter pair of the parameter group. A causal parameter specification part 8 specifies a parameter that is common in the two correlation data strings the distance of which is the shortest (or the distance is the shortest with a proviso). A failure prediction part 9 predicts an occurrence of the failure caused by data abnormality of the causal parameter.

Description

  The present invention relates to a failure prediction system and a program.

  Conventionally, for a continuously operated system such as a plant, a method of identifying an abnormality based on whether or not the data obtained by observation exceeds a threshold value, and using a cross-correlation coefficient between two parameters A method for identifying an abnormality has been devised.

  For example, in a process having a large number of process values that are causally related in a tree structure, the process value is sequentially traced along one path from the result side to the cause side, and each time value is measured for each two adjacent process values. The cross-correlation coefficient of the measured data and the geometric mean of multiple cross-correlation coefficients along the path are calculated, and the process value on the result side is calculated according to the magnitude of the geometric mean obtained for all possible paths. An invention of a method for analyzing a variation in a process, characterized by analyzing the degree of the influence of a cause-side process value on the variation of the process (see Patent Document 1), has been proposed.

  For example, a remote monitoring system for remotely monitoring a plurality of facilities each having a plurality of sensors, wherein sensor value acquisition means for acquiring measured sensor values detected by the plurality of sensors transmitted from the facilities as needed And a plurality of measured sensor values detected by the plurality of sensors based on the measured sensor values detected by the plurality of sensors and acquired by the sensor value acquisition means when the facility is operating normally. A prediction model construction unit that obtains a correlation between them, and stores the correlation as a prediction model; and when the sensor value acquisition unit acquires the actual sensor values of the plurality of sensors, the acquired actual sensor value and the Predictive sensor values of the plurality of sensors are obtained from the predictive model stored by the predictive model construction means, and based on the difference between the actually measured sensor value and the predictive sensor value, Invention of remote monitoring systems have been proposed which is characterized by having a failure sign detection means for detecting the disabled sign (see Patent Document 2).

Japanese Patent Application Laid-Open No. 08-328644 JP 2005-149137 A

  The present invention relates to a technique for predicting a failure based on time-series data of a plurality of parameters indicating an internal state of a monitored device, and when abnormalities occur in data of other parameters due to abnormal data of a certain parameter. Even so, it is an object of the present invention to propose a technique that can effectively identify a causal parameter and use the failure prediction to improve accuracy.

  The present invention according to claim 1 is set by a setting means for setting a plurality of sets of a pair of parameters that are causally related to a change in value for a plurality of parameters indicating the internal state of the monitored device, and the setting means. Specific means for identifying a parameter common to the two sets having the most similar transition of the correlation degree, based on the transition of the correlation degree of the time-series change of the value between the parameters constituting each pair, obtained for each pair; And an output means for outputting a notification that a failure previously associated with the parameter specified by the specifying means occurs in the monitored device.

  According to a second aspect of the present invention, in the first aspect of the present invention, the setting means sets a plurality of parameters indicating an internal state of the monitored device at least for each of charging voltage, toner concentration, development, and laser power. For each parameter group classified into the parameter group, for each parameter classified into the parameter group, in the monitored device in a normal state, setting a plurality of sets of a pair of parameters causally related to the change in the value, It is a characteristic failure prediction system.

  According to a third aspect of the present invention, in the present invention according to the first and second aspects, the specifying unit is configured to determine a correlation coefficient between the parameters constituting each set for a plurality of periods in which a period of a target range is slid. Calculate the correlation coefficients of the calculated periods in time series to form a correlation data string, and specify the two sets having the closest distance between the correlation data strings as the two sets having the most similar correlation degree transitions This is a failure prediction system characterized by that.

  According to a fourth aspect of the present invention, in the first aspect of the present invention according to the first to third aspects, the specifying means is configured such that when there is no parameter common to the two sets having the most similar transition of the correlation, the two For a pair, the degree of similarity between the transition of the correlation degree in one pair and the transition of the correlation degree in the other pair set for the parameters that make up the pair, and the transition of the correlation degree in the other pair and the parameters that make up the pair A failure prediction system characterized by examining the degree of similarity with the transition of the correlation degree in the other set set for, and identifying the parameters common to the two sets with the most similar transition of the correlation degree is there.

  According to a fifth aspect of the present invention, there is provided a setting function in which a plurality of sets of a pair of parameters that are causally related to a change in value are set in a computer for a plurality of parameters indicating the internal state of the monitored device, and the setting function Based on the transition of the correlation degree of the time-series change of the value between the parameters of each pair obtained for each pair set by the above, the parameter common to the two pairs having the most similar transition of the correlation degree is specified A program for realizing a specific function and an output function for outputting a notification that a failure previously associated with the parameter specified by the specific function occurs in the monitored device.

  According to a sixth aspect of the present invention, in the present invention according to the fifth aspect, the setting function includes at least a plurality of parameters indicating an internal state of the monitored device, each type of at least charging voltage, toner concentration, development, and laser power. For each parameter group classified into the parameter group, for each parameter classified into the parameter group, in the monitored device in a normal state, setting a plurality of sets of a pair of parameters causally related to the change in the value, It is a featured program.

  According to a seventh aspect of the present invention, in the present invention according to the fifth and sixth aspects, each of the specific functions is a plurality of periods obtained by sliding a period of a target range with a correlation coefficient between the parameters constituting each set. Calculate the correlation coefficients of the calculated periods in time series to form a correlation data string, and specify the two sets having the closest distance between the correlation data strings as the two sets having the most similar correlation degree transitions This is a program characterized by that.

  According to an eighth aspect of the present invention, in the present invention according to the fifth to seventh aspects, the specific function is configured such that when there is no parameter common to the two sets having the most similar transition of the correlation degree, For a pair, the degree of similarity between the transition of the correlation degree in one pair and the transition of the correlation degree in the other pair set for the parameters that make up the pair, and the transition of the correlation degree in the other pair and the parameters that make up the pair Is a program characterized by examining the degree of similarity with the transition of the correlation degree in the other set set for, and specifying a parameter common to the two sets with the most similar transition of the correlation degree.

  According to the first and fifth aspects of the present invention, even when data of another parameter is abnormal due to data abnormality of a certain parameter, the cause parameter is effectively identified. Failure prediction can be used to improve accuracy.

  According to the second and sixth aspects of the present invention, it is possible to reduce the amount of calculation related to specifying the causal parameter. In addition, the occurrence of an abnormality in the monitored device can be detected with high sensitivity.

  According to the third and seventh aspects of the present invention, a slight change in the original data can be detected sensitively.

  According to the present invention according to claims 4 and 8, even when there is no parameter common to the two sets having the most similar time-series changes in the correlation degree, the cause parameter can be specified. .

It is a figure which shows the structural example of the failure prediction system which concerns on one Embodiment of this invention. It is a figure which shows an example of a parameter pair. It is a figure explaining the process by a cause parameter specific | specification part. It is a figure explaining the process by a cause parameter specific | specification part. It is a figure which shows an example of the correspondence table which matched the cause parameter and the failure relevant to the abnormality. It is a figure which shows an example of the processing flow by the failure prediction system of this example. It is a figure which shows an example of the processing flow which concerns on the specific process of a cause parameter. It is a figure which shows an example of the relationship between each time series data of two parameters, and its correlation data sequence. It is a figure explaining the example which specifies a cause parameter from three parameters. It is a figure explaining the example which specifies a cause parameter from three parameters. It is a figure which illustrates the hardware constitutions of a server apparatus.

A failure prediction system according to an embodiment of the present invention will be described with reference to the drawings.
First, the background will be described prior to the description of the failure prediction system according to the present example.
For example, in a copying machine, a large amount of data indicating the internal state of the machine such as process control parameters is collected during operation for machine maintenance. These time series data are often system parameter data having a complicated feedback loop, and anomalies may occur in other parameter data due to anomalies in data of a certain parameter. For this reason, when a failure is predicted from the transition of the cross-correlation coefficient of the time-series data, if multiple parameters with anomalies in the time-series data are detected, it is necessary to identify which parameter was the cause. There is an obstacle to prediction.
Therefore, in the failure prediction system of this example, the cause parameter is specified by the following configuration to help improve the accuracy of failure prediction.

FIG. 1 illustrates the configuration of the failure prediction system of this example.
The failure prediction system of this example includes a plurality of copying machines that are targets of failure prediction and a server device that predicts that a failure will occur in the copying machine based on data collected from each copying machine via a communication network. A data collection unit 1, a parameter classification unit 2, a group-specific parameter pair extraction unit 3, a group-specific parameter pair storage unit 4, and a cross-correlation coefficient calculation. A unit 5, a threshold comparison unit 6, a similarity calculation unit 7, a cause parameter identification unit 8, and a failure prediction unit 9 are provided. In this example, each of the functional units 1 to 9 is provided in one server device, but may be provided in a distributed manner in a plurality of server devices.

  The data collection unit 1 collects and accumulates data of a plurality of parameters indicating the internal state from each copying machine. The collected data includes a device ID for identifying the collection source copier, a collection date and time, a parameter type, and a value (setting value or measurement value). Based on the collected data, parameter values are arranged in order of collection date and time for the same device ID and the same parameter type, thereby obtaining parameter time-series data.

The parameter classification unit 2 classifies (groups) a plurality of parameters into a parameter group (group) in which parameters having commonality are collected. The classification is performed based on parameter attributes such as the location (part) where the parameters are collected and the contents. For example, parameters related to each system such as a charging voltage system, a toner concentration system, a development system, and a laser power system are classified. Some parameters may be classified into two or more parameter groups.
As described above, by classifying a plurality of parameters in advance, it is possible to greatly reduce the amount of calculation for specifying the cause parameter described later.

  The group-by-group parameter pair extraction unit 3 sets the correlation of time series data in a normal state (a state in which a failure and its precursor are not detected) for each parameter group (by group) classified by the parameter classification unit 2. Is extracted, and information on the pair is stored in the group-specific parameter pair storage unit 4. That is, a pair consisting of a pair of parameters that are causally related to a change in value is extracted and set in advance. The parameter pair may be extracted based on time-series data collected from one copying machine (for example, a test copying machine), or a plurality of copying machines (for example, operating copying machines). You may perform based on the time series data collected from.

In this example, based on the time-series data of each parameter collected by the data collection unit 1, the correlation between the time-series data (hereinafter referred to as the correlation between the time-series changes of the two parameter values in the normal state) , A correlation coefficient) is calculated, and a parameter pair whose normal state correlation coefficient is greater than or equal to a threshold value (0.9 in this example) is extracted.
Here, the correlation coefficient is a statistical index indicating the correlation (degree of similarity) between two data, and when the two data are exactly the same (or the change is the same), The number of relations is 1, and on the other hand, when it changes in exactly the opposite direction, it becomes -1. If the two data are not related at all, it becomes zero.

ここで、ｘ￣，ｙ￣は、それぞれ、ｘ＝｛ｘ_ｉ｝，ｙ＝｛ｙ_ｉ｝の相加平均である。
これは、各データの平均からのずれを表すベクトル（ｘ−ｘ￣）＝（ｘ_１−ｘ￣，・・・，ｘ_ｎ−ｘ￣），（ｙ−ｙ￣）＝（ｙ_１−ｙ￣，・・・，ｙ_ｎ−ｙ￣）のなす角の余弦である。また、この式は、共分散をそれぞれの標準偏差で割ったものに等しい。 For example, the correlation coefficient between the data string (x ₁ , x ₂ ,..., X _n ) and the data string (y ₁ , y ₂ ,..., Y _n ) can be obtained by (Equation 1). .

Here, x￣ and y￣ are arithmetic averages of x = {x _i } and y = {y _i }, respectively.
This is a vector (x−x∥) = (x ₁ −x∥,..., X _n− x∥), (y−y∥) = (y ₁ −y) representing a deviation from the average of each data. Cosine of the angle formed by ￣,..., Y _n −y￣). This equation is also equal to the covariance divided by the respective standard deviation.

The correlation coefficient may be calculated at a time over the entire period of interest (for example, 6 months) (that is, the correlation coefficient is calculated for the data transition of the entire 6 months). In this period, the calculation period may be sequentially shifted by a certain range (for example, one week) (that is, a correlation coefficient may be calculated for the data transition of one week), It is only necessary to detect parameter pairs that continuously show strong correlation during the period of interest. Note that parameters that do not correlate strongly with any parameter in the normal state are not used.
Thus, by pairing parameters having a strong correlation in the normal state, there is an effect that the correlation coefficient changes with high sensitivity when an abnormality occurs.

FIG. 2 shows an example of parameter pairs.
FIG. 2A is an example of a parameter pair in the parameter group of the charging voltage system, and is configured by parameters such as PA_a, PA_b, PA_c,. FIG. 2B shows an example of a parameter pair in the toner density system parameter group, which is composed of parameters such as PB_a, PB_b, PB_c,. FIG. 2C shows an example of a parameter pair in the development system parameter group, which is composed of parameters such as PC_a, PC_b, PC_c,. Each parameter group is configured to include at least three types of parameters. In this example, each parameter is provided for each color element such as Y (Yellow), M (Magenta), C (Cyan), K (Key Plate), and a numerical value (1) added to the end of the parameter name. Color elements are distinguished by (4) to (4).

The cross-correlation coefficient calculation unit 5 uses the parameter group-specific (group-specific) parameter pair information stored in the group-specific parameter pair storage unit 4 to determine the latest period (in this example, The degree of correlation (in this example, the correlation coefficient) between the time series data of the parameter pair in the last 5 days) is calculated. Note that the period of the target range for calculating the correlation coefficient (calculation period) is related to the steepness of change in the correlation coefficient (the steeper the smaller the number of days), and the calculation period is not strictly determined. What is necessary is just to determine in relation to abnormality detection.
In the cross-correlation coefficient calculation unit 5 of this example, as new data is collected by the data collection unit 1 (for example, every day), the correlation coefficient is slid while the calculation period is slid for each parameter pair for each parameter group. Are sequentially calculated. As a result, a data string in which the correlation coefficients obtained for each parameter pair are arranged in chronological order is set as a correlation data string (data string indicating the transition of the correlation degree) related to the parameter pair. That is, according to the cross-correlation coefficient calculation unit 5 of this example, a correlation data string can be obtained for each parameter pair for the target copying machine.

Here, the correlation data string will be described with reference to FIG.
FIG. 8A illustrates a graph showing time-series data of each of the parameters A and B (time-series changes in the values of the parameters), and FIG. 8B illustrates the parameters A and B. The graph which shows the correlation data sequence (transition of a correlation coefficient) which concerns on a pair is illustrated.
According to FIG. 8A, the time series data of the parameter A and the time series data of the parameter B appear to change in the same way. However, as shown in FIG. 8B, in the correlation data string created based on these time-series data, a slight change difference of the original data appears as a large fluctuation. For example, the difference between the portions surrounded by the dotted line in FIG. 8A is slight, but in FIG. 8B, the difference is represented by a large variation. Thus, by using the correlation data string, it is possible to sensitively detect a slight change in the original data.

  The threshold comparison unit 6 constantly monitors the correlation data string for each parameter pair obtained by the cross-correlation coefficient calculation unit 5 and compares it with a predetermined threshold (0.4 in this example). Then, when a parameter pair that includes a correlation coefficient exceeding the threshold (or higher than the threshold) in the correlation data string is detected, it is determined as an abnormal state, and the parameter group (group) to which the parameter pair belongs is specified. To do. That is, the parameter group to which the cause parameter candidate that caused the data abnormality belongs is specified. When the cause parameter belongs to a plurality of parameter groups, a plurality of parameter groups can be specified.

For the parameter group (group) including the cause parameter specified by the threshold comparison unit 6, the similarity calculation unit 7 calculates the similarity between the correlation data sequences based on the correlation data sequences of all parameter pairs in the parameter group. Is calculated.
In this example, the Euclidean distance between the correlation data strings in a predetermined period (for example, 5 days) is used as the similarity between the correlation data strings. That is, by regarding the correlation data string as a vector, the distance between the correlation data strings (between the vectors) can be used as an index indicating the similarity of the waveform change between the correlation data strings. In this case, the smaller the distance between the correlation data strings, the more similar the waveform changes of both data strings. If the distance is zero, it means that both data strings are equal.

本例では、ｎ＝３０で計算して距離を算出している。 For example, a correlation data string AB (x ₁ , x ₂ ,..., X _n ) related to a pair of parameters A and B and a correlation data string CD (y ₁ , y ₂ ,. .., Y _n ) and the distance D _{AB, CD} can be obtained by (Equation 2).

In this example, the distance is calculated by calculating with n = 30.

  The cause parameter specifying unit 8 uses the similarity (in this example) between the correlation data strings of the parameter pairs calculated by the similarity calculation unit 7 for the parameter group (group) including the cause parameter specified by the threshold comparison unit 6. ), The cause parameter causing the data abnormality is specified (estimated) based on the distance).

The cause parameter specifying process by the cause parameter specifying unit 8 will be described with reference to FIGS.
For example, as illustrated in FIG. 3, the distances D _AB and EA between the correlation data sequence AB related to the pair of parameters A and B and the correlation data sequence EA related to the pair of parameters E and A are all correlation data sequences. The shortest distance between them. At this time, the parameter A is common to the parameter pairs associated with the distances D _{AB and EA} . In this case, it is presumed that parameter A causes other parameters. Thus, when there is a parameter common to the two correlation data strings having the shortest distance, it can be estimated that the parameter is a cause parameter of data abnormality (decrease in correlation coefficient caused by a failure or the like).

For example, as illustrated in FIG. 4, the distances D _AB and EF between the correlation data sequence AB related to the pair of parameters A and B and the correlation data sequence EF related to the pair of parameters E and F are all correlated. The shortest distance between data strings is assumed. At this time, each parameter pair related to the distances D _{AB and EA} has no common parameter. In this case, the cause parameter is estimated by reexamining the shortest distance conditionally according to the following procedure.
(Procedure 1) The distance between one correlation data string AB and another correlation data string related to one of the parameters A and B that is the object is obtained. That is, distances D _{AB, AC} , D _{AB, AE} , D _{AB, AF} are obtained for parameter A, and distances D _{AB, BC} , D _{AB, BE} , D _{AB, BF} are obtained for parameter B.
(Procedure 2) The distance between the other correlation data string EF and another correlation data string related to one of the parameters E and F that is the object is obtained. That is, the distances D _{EF, EA} , D _{EF, EB} , D _{EF, EC} are obtained for the parameter E, and the distances D _{EF, FA} , D _{EF, FB} , D _{EF, FC} are obtained for the parameter F.
(Procedure 3) A parameter common to the correlation data string of the shortest distance among the distances obtained in (Procedure 1) and (Procedure 2) is estimated as a cause parameter. For example, if the distances D _AB and BC between the correlation data string AB and the correlation data string BC are the shortest, the parameter B common to these two correlation data strings is estimated as the cause parameter.
Since distance calculation is originally performed for all combinations, it is not necessary to recalculate the distances in (Procedure 1) and (Procedure 2).

The failure predicting unit 9 specifies (estimates) a failure that is predicted to occur due to data abnormality of the cause parameter for the cause parameter specified by the cause parameter specifying unit 8. In this example, as illustrated in FIG. 5, a correspondence table in which a cause parameter and a failure related to the abnormality (and a content of a treatment to be performed) are associated is held in a memory, and the correspondence table is referred to. Thus, the failure (and the content of the treatment to be performed) that is predicted to occur in the target copying machine is estimated.
Information on the failure estimated by the failure prediction unit 9 is notified to a maintenance person in charge of maintenance of the target copier. As a result, maintenance work such as maintenance can be systematically performed before a failure actually occurs.

The processing by the failure prediction system of this example will be described with reference to the processing flow illustrated in FIG.
In the failure prediction system of this example, as preprocessing, the parameter classification unit 2 classifies (groups) parameters in units of subsystems (charging voltage system, toner density system, development system, etc.) (step S1), and sets group-specific parameters. The pair extraction unit 3 extracts a parameter pair having a strong normal state correlation coefficient for each parameter group (group) and stores the parameter pair in the group-specific parameter pair storage unit 4 (step S2). These pre-processes do not need to be re-executed unless there is a change in the configuration of the parameter group, addition, correction, or deletion of parameters.

After the above preprocessing is performed, the following processing is repeated for each target copying machine.
First, the cross-correlation coefficient calculation unit 5 generates a correlation data string between the time series data of the parameter pairs in the most recent period (step S3), and the threshold comparison unit 6 is calculated by the cross-correlation coefficient calculation unit 5. It is determined whether or not a correlation coefficient exceeding a threshold is included in the correlation data string (step S4), and when a parameter pair including a correlation coefficient exceeding the threshold is detected, an abnormal state is determined. A parameter group (group) including the parameter pair is specified (step S5). Thereafter, the similarity calculation unit 7 and the cause parameter specifying unit 8 perform the cause parameter specifying process for the specified parameter group (group) (step S6), and the failure predicting unit 9 stores the specified cause parameter data. A fault that is predicted to occur due to an abnormality is identified (estimated) (step S7).

The cause parameter specifying process will be further described with reference to the process flow of FIG.
First, it is determined whether or not the number of parameters belonging to the specified parameter group (group) is two (that is, whether or not there is one parameter pair) (step S11).
If it is determined in step S11 that the number of parameters is two, one of the parameters is identified as the cause parameter (step S12).
On the other hand, if it is determined in step S11 that the number of parameters is two or more, the distance (similarity) between the correlation data strings is calculated based on the correlation data strings of all parameter pairs in the parameter group. (Step S13), it is determined whether there is a parameter common to the two correlation data strings having the shortest distance (Step S14).
If it is determined in step S14 that there is a parameter common to the two correlation data strings having the shortest distance, the common parameter is specified as a cause parameter (step S16).
On the other hand, if it is determined in step S14 that there is no parameter common to the two correlation data strings having the shortest distance, other correlation data related to the two correlation data strings and the four parameters that are the targets thereof. The distance to the column is checked, and two correlation data strings that are the shortest distance among them are specified (step S15), and the common parameter is specified as the cause parameter (step S16). That is, the common parameter in the correlation data string that is the shortest distance with a condition is specified as the cause parameter.

Here, as an example, a case where a cause parameter is specified from a parameter group including three parameters A, B, and C will be described with reference to FIGS. 9 and 10.
First, as shown in FIG. 9A, the cross-correlation coefficient calculation unit 5 performs a correlation data sequence AB related to a pair of parameters A and B, a correlation data sequence BC related to a pair of parameters B and C, a parameter C, A correlation data string CA relating to the pair A is calculated.
As a result, the data transition of FIG. 10A is obtained for the correlation data string AB, the data transition of FIG. 10B is obtained for the correlation data string BC, and the data transition of FIG. 10C for the correlation data string CA. Is obtained.
In this case, the threshold value comparison unit 6 detects that the correlation data string AB and the correlation data string BC include a correlation coefficient exceeding the threshold value, and shifts to a process for specifying a cause parameter for the parameter group.

In specifying the cause parameter, first, as shown in FIG. 9B, the similarity calculation unit 7 determines the distance D _{AB, BC} between the correlation data string AB and the correlation data string _BC , the correlation data string BC, and the correlation data string. distance _{D BC} and _{CA, CA,} the distance _{D CA} between the correlation data sequence CA correlation data sequence _{AB, AB} and calculated respectively.
As a result, as shown in FIG. 9C, it is assumed that distances D _{AB, BC} = 0.085, distances D _{BC, CA} = 0.251, and distances D _{CA, AB} = 0.191 are obtained.
In this case, since the relationship between the distances is the distance D _{AB, BC} <D _{CA, AB} <D _{BC, CA} , the cause parameter specifying unit 8 uses the parameter B which is a parameter common to the shortest distances D _{AB, BC.} Is identified as the cause parameter.

As described above, in the failure prediction system of this example, for each parameter group composed of a plurality of parameters, the cross-correlation coefficient calculation unit 5 generates a correlation data string for each parameter pair, and the threshold comparison unit 6 It is determined whether or not a correlation coefficient exceeding a threshold value is included in the data string, and when a parameter pair including a correlation coefficient exceeding the threshold value is detected, the similarity calculation unit 7 determines the parameter group to which the parameter belongs. The distance between the correlation data strings of each parameter pair is calculated, and the cause parameter specifying unit 8 specifies a parameter common to the two correlation data strings having the shortest distance (or the shortest distance under certain conditions) as the cause parameter. The failure predicting unit 9 predicts the occurrence of a failure due to the abnormal data of the cause parameter.
According to such a configuration, it is possible to effectively identify the cause parameter, and therefore it is possible to accurately predict the occurrence of a failure due to the data abnormality of the parameter.

  Here, the failure prediction system of this example performs failure prediction for a copying machine, but may also perform failure prediction for other devices. For example, as with a copying machine, a recording material such as paper Other types such as a printer (document printing device), a facsimile device (document transfer device), or a multi-function device having a combination of the functions of these devices. Failure prediction may be performed for the image forming apparatus. Further, the present invention is not limited to application to such an image forming apparatus, but can be applied to other apparatuses controlled using a large number of parameters having a complicated feedback loop.

  In the failure prediction system of this example, the failure prediction is performed by a server device separate from the monitored device (in this example, the copying machine). However, each monitored device predicts the failure of itself. You may make it the structure to perform. In this case, functions such as a data collection unit 1, a cross-correlation coefficient calculation unit 5, a threshold comparison unit 6, a similarity calculation unit 7, a cause parameter identification unit 8, and a failure prediction unit 9 are provided on the monitored device side. Just do it.

FIG. 11 illustrates the hardware configuration of the server device in the failure prediction system of this example.
In this example, a main memory such as a CPU (Central Processing Unit) 21 that performs various arithmetic processing, a RAM (Random Access Memory) 22 that is a work area of the CPU 21, and a ROM (Read Only Memory) 23 that records a basic control program. Apparatus, auxiliary storage device such as HDD (Hard Disk Drive) 24 for storing programs and various data according to an embodiment of the present invention, display device for displaying and outputting various information, and operations used for input operations by the operator The computer of the server device uses hardware resources such as an input / output I / F 25 which is an interface with an input device such as a button or a touch panel, and a communication I / F 26 which is an interface for performing wired or wireless communication with other devices. Has.
The program according to the embodiment of the present invention is read from the auxiliary storage device 24 and the like, loaded into the RAM 22, and executed by the CPU 21, thereby realizing each functional unit described above on the computer of the server device. .

Note that the program according to the embodiment of the present invention is based on, for example, the server according to the present example in a format read from an external storage medium such as a CD-ROM storing the program or a format received via a communication network. Set in the computer of the device.
Moreover, it is not restricted to the aspect which implement | achieves each function part by software configuration like this example, You may make it implement | achieve each function part with a dedicated hardware module.

  1: data collection unit, 2: parameter classification unit, 3: group-specific parameter pair extraction unit, 4: group-specific parameter pair storage unit, 5: cross-correlation coefficient calculation unit, 6: threshold comparison unit, 7: similarity calculation Part, 8: cause parameter specifying part, 9: failure predicting part

Claims (8)

For a plurality of parameters indicating the internal state of the monitored device, setting means for setting a plurality of sets consisting of a pair of parameters that are causally related to a change in value;
Based on the transition of the correlation degree of the time series change of the value between the parameters constituting each pair, which is obtained for each pair set by the setting means, the parameter common to the two pairs with the most similar transition of the correlation degree Identifying means for identifying
Output means for outputting a notification that a failure associated in advance with the parameter specified by the specifying means occurs in the monitored device;
A failure prediction system comprising: The setting means includes, for each parameter group classified into at least a plurality of parameters indicating an internal state of the monitored apparatus, into at least a class of charging voltage, toner density, development, and laser power. In the monitored device in the normal state, a plurality of sets of a pair of parameters that are causally related to the change in the value are set.
The failure prediction system according to claim 1. The specifying unit calculates a correlation coefficient between the parameters constituting each set for each of a plurality of periods obtained by sliding a period of the target range, and arranges the calculated correlation coefficients for each period in time series order to generate a correlation data string. And the two sets having the closest distance in the correlation data string are specified as the two sets having the most similar transition of the correlation degree.
The failure prediction system according to claim 1 or claim 2, characterized by that. When the parameter common to the two sets having the most similar transition of the correlation degree does not exist, the specifying means is set for the two sets and the correlation degree transition in one set and the parameters forming the set. The degree of similarity with the transition of the correlation degree in the other group, and the degree of similarity between the transition of the correlation degree in the other group and the transition of the correlation degree in the other group set for the parameters constituting the group, Identify the parameters that are common to the two pairs with the most similar transitions in correlation,
The failure prediction system according to any one of claims 1 to 3, wherein the failure prediction system is characterized in that: On the computer,
For a plurality of parameters indicating the internal state of the monitored device, a setting function that sets a plurality of sets of a pair of parameters that are causally related to a change in value;
A parameter that is obtained for each set set by the setting function, and is a parameter that is common to the two sets having the most similar transition of the correlation degree, based on the transition of the correlation degree of the time-series change of the value between the parameters constituting each pair Specific function to identify,
An output function for outputting a notification that a failure previously associated with the parameter specified by the specific function occurs in the monitored device;
Program to realize. The setting function includes, for each parameter group classified into at least a plurality of parameters indicating an internal state of the monitored apparatus, into at least a class of charging voltage, toner density, development, and laser power. In the monitored device in the normal state, a plurality of sets of a pair of parameters that are causally related to the change in the value are set.
The program according to claim 5. The specific function calculates a correlation coefficient between the parameters constituting each set for each of a plurality of periods obtained by sliding a period of a target range, and arranges the calculated correlation coefficients for each period in time-series order to generate a correlation data string. And the two sets having the closest distance in the correlation data string are specified as the two sets having the most similar transition of the correlation degree.
The program according to claim 5 or 6, characterized by the above. The specific function is set with respect to the transition of the correlation degree in one set and the parameters constituting the set for the two sets when there is no parameter common to the two sets having the most similar transition of the correlation level. The degree of similarity with the transition of the correlation degree in the other group, and the degree of similarity between the transition of the correlation degree in the other group and the transition of the correlation degree in the other group set for the parameters constituting the group, Identify the parameters that are common to the two pairs with the most similar transitions in correlation,
The program according to any one of claims 5 to 7, wherein:

Images