JP7043320B2 - State analyzer and state analysis method - Google Patents

Description

The present invention relates to a state analyzer and a state analysis method.

Patent Document 1 discloses a technique for acquiring data during operation of a work machine and diagnosing an abnormality in the work machine using a self-organizing map. In such a technical field, there is a demand to increase the types of data to be acquired in order to improve the accuracy of machine abnormality diagnosis. The more types of data, the greater the dimensions of the vector or matrix that represents the data set. On the other hand, as the dimension of the vector space increases, a phenomenon called a spherical concentration phenomenon is known in which the values of each vector are concentrated and distributed at the edges of the vector space. That is, the larger the dimension of the vector space, the longer the distance between the vectors due to the spherical concentration phenomenon. A method of preventing the spherical concentration of a vector by compressing the dimensions of the vector and the matrix is known.

Japanese Unexamined Patent Publication No. 2006-53818

By the way, there is a desire to predict the deterioration state or the presence or absence of a failure of a machine from the history of machine use. For example, each of the data related to a plurality of past time points of the machine is plotted on a graph, and the deterioration state or the presence or absence of a failure of the machine is predicted based on the change in the position of the plot. At this time, when there are many types of data to be acquired, it is possible to consider compression of the dimensions of the vector space as described above. However, if the dimension of the vector consisting of vital data representing the state related to the operation of the machine is compressed, the time-series relationship of the vector is lost, and the time-series of the vector representing the state of one machine is graphed. Even with plotting, it is difficult to predict where the vector will appear in the future.

An object of the present invention is to provide a state analyzer and a state analysis method capable of compressing the dimension of data related to the history of machine use so as not to lose the time-series relevance.

According to the first aspect of the present invention, the state analyzer changes monotonically with time with the operation of the machine for each of a plurality of snapshots of the state of the plurality of machines at a plurality of time points. It is provided with a vector acquisition unit for acquiring a plurality of state vectors representing a quantity related to a plurality of items indicating the states of, and a vector compression unit for acquiring a plurality of compression vectors obtained by compressing the dimensions of the plurality of state vectors.

According to the above aspects, the state analyzer can compress the dimensions of the data relating to the history of machine use without losing time-series relevance.

It is a schematic diagram which shows the structure of the state analysis system which concerns on one Embodiment. It is a schematic block diagram which shows the structure of the target machine which concerns on 1st Embodiment. It is a schematic block diagram which shows the structure of the state analysis apparatus which concerns on 1st Embodiment. It is a flowchart which shows the learning method by the state analyzer which concerns on 1st Embodiment. It is a flowchart which shows the state analysis method by the state analysis apparatus which concerns on 1st Embodiment. It is a figure which shows the example of the screen output by the state analysis apparatus which concerns on 1st Embodiment. It is a graph which shows the compression vector group classified into one cluster by the state analyzer which concerns on 1st Embodiment.

Hereinafter, embodiments will be described in detail with reference to the drawings.
<First embodiment>
"overall structure"
FIG. 1 is a schematic diagram showing a configuration of a state analysis system according to an embodiment.
The state analysis system 1 includes a plurality of target machines 10 and a state analysis device 20.

The target machine 10 is a target of state analysis by the state analysis device 20. Examples of the target machine 10 include a work machine such as a hydraulic excavator. The target machine 10 is provided with a plurality of sensors, and information related to the measured values of each sensor is transmitted to the state analyzer 20.

The state analysis device 20 outputs a screen showing the state of the target machine 10 based on the information received from the target machine 10. By visually recognizing the screen output by the state analysis device 20, the user can recognize the current state of the state analysis device 20 and predict the future state.

<< Configuration of target machine >>
FIG. 2 is a schematic block diagram showing the configuration of the target machine according to the first embodiment.
The target machine 10 includes a machine main body 11, a plurality of sensors 12, a data aggregation device 13, and a communication device 14.
Each of the plurality of sensors 12 measures vital data that changes with the operation of the machine body 11. Examples of vital data measured by a plurality of sensors 12 include the operating time (SMR: Service Meter Reading) of the machine body 11, the fuel consumption of the machine body 11, and the open / closed state of the relief valve of the hydraulic device included in the machine body 11. Examples include the traveling speed of the machine body 11.
The data aggregation device 13 collects measurement data from a plurality of sensors 12 and generates information to be transmitted to the state analysis device 20.
The communication device 14 transmits the information generated by the data aggregation device 13 to the state analysis device 20.

The data aggregation device 13 is a computer including a processor 131, a main memory 132, a storage 133, and an interface 134. The storage 133 stores a data aggregation program. The processor 131 reads the data aggregation program from the storage 133, expands it in the main memory 132, and executes processing according to the data aggregation program. The data aggregation device 13 is connected to the plurality of sensors 12 and the communication device 14 via the interface 134.

Examples of the storage 133 include semiconductor memory, disk media, tape media, and the like. The storage 133 may be an internal medium directly connected to the common communication line of the data aggregation device 13, or an external medium connected to the data aggregation device 13 via the interface 134. The storage 133 is a tangible storage medium for non-temporary storage.

The processor 131 includes a measured value acquisition unit 1311, a cumulative calculation unit 1312, a state recording unit 1313, and a transmission unit 1314 by executing a data aggregation program. Further, the processor 131 secures a storage area of the state storage unit 1321 in the main memory 132 by executing the data aggregation program.

The measured value acquisition unit 1311 acquires measured values from each of the plurality of sensors 12.

The cumulative calculation unit 1312 calculates the cumulative value of the measured values acquired by the measured value acquisition unit 1311 at least for the measured values that do not change monotonically with time. Monotonic changes include monotonically increasing (monotonically non-decreasing) and monotonically decreasing (monotonically non-increasing).
Examples of measured values that do not change monotonically with time include fuel consumption, relief valve open / closed state, running state, running speed, and the like. For example, the cumulative calculation unit 1312 calculates the cumulative fuel consumption by cumulatively calculating the fuel consumption. Further, for example, the cumulative calculation unit 1312 calculates the cumulative relief time by cumulatively calculating the time during which the relief valve is open. Further, for example, the cumulative calculation unit 1312 calculates the cumulative running time by cumulatively calculating the running time (the time when the running speed is equal to or higher than the threshold value). Further, for example, the cumulative calculation unit 1312 calculates the cumulative mileage by cumulatively calculating the running speed.
An example of a measured value that changes monotonically with time is SMR. That is, when the sensor 12 measures the cumulative value, the cumulative calculation unit 1312 does not have to perform the cumulative calculation for the measured value.

The state recording unit 1313 stores the cumulative value of the measured value calculated by the cumulative calculation unit 1312 and the measured value that changes monotonically with respect to the time measured by the sensor 12 in the state storage unit 1321 in association with the time. Hereinafter, the state at one time of one target machine 10 is referred to as a snapshot. Further, the cumulative value of the measured values related to one snapshot and the measured value that changes monotonically with respect to time, that is, the value that changes monotonically with time, which represents the state of the machine body 11 related to one snapshot, is the state information. Called.

The transmission unit 1314 transmits the state information stored in the state storage unit 1321 to the state analysis device 20 via the communication device 14.

<< Configuration of state analyzer >>
FIG. 3 is a schematic block diagram showing the configuration of the state analyzer according to the first embodiment.
The state analysis device 20 is a computer including a processor 21, a main memory 22, a storage 23, an interface 24, and a communication device 25. The storage 23 stores the state analysis program. The processor 21 reads the state analysis program from the storage 23, expands it in the main memory 22, and executes processing according to the state analysis program. The processor 21 is connected to the communication device 25 via the interface 24.

Examples of the storage 23 include semiconductor memory, disk media, tape media, and the like. The storage 23 may be an internal medium directly connected to the common communication line of the state analysis device 20, or an external medium connected to the state analysis device 20 via the interface 24. The storage 23 is a tangible storage medium for non-temporary storage.

By executing the state analysis program, the processor 21 executes a state information acquisition unit 2101, a vector acquisition unit 2102, a model generation unit 2103, a vector compression unit 2104, a clustering unit 2105, a classification unit 2106, a component information acquisition unit 2107, and a heat map generation unit. It includes a 2108, a graph generation unit 2109, and an output unit 2110. Further, the processor 21 secures a storage area of the state storage unit 2201, the model storage unit 2202, the cluster storage unit 2203, and the heat map storage unit 2204 in the main memory 22 by executing the state analysis program.

The state information acquisition unit 2101 receives state information related to a plurality of times from each of the plurality of target machines 10. That is, the state information acquisition unit 2101 acquires the state information related to the plurality of snapshots. Further, the state information acquisition unit 2101 stores each state information in the state storage unit 2201 in association with the ID of the target machine 10 and the time. That is, in the state storage unit 2201, the combination of the ID and the time of the target machine 10 constitutes a primary key (composite primary key).

The vector acquisition unit 2102 reads a plurality of state information from the state storage unit 1321 and acquires a state vector representing the state information. The state vector is a vector having the value of each state information as an element. The number of dimensions of the state vector is equal to the number of items in the state information.

The model generation unit 2103 generates a compression model for two-dimensionally compressing the state vector from a plurality of state vectors related to a plurality of times of the plurality of target machines 10 acquired by the vector acquisition unit 2102. In the first embodiment, an example of compressing a state vector using a self-organizing map (SOM) will be described. Therefore, the model generation unit 2103 specifies the parameters of the SOM model using a plurality of state vectors. The model generation unit 2103 stores the generated compression model in the model storage unit 2202. In other embodiments, principal component analysis, multidimensional scale method, singular value decomposition, autoencoder, or other dimensional compression method may be used as the dimensional compression method.

The vector compression unit 2104 compresses the state vector based on the compression model and generates a compression vector. The vector compression unit 2104 can obtain a compression vector group representing the usage history of one target machine 10 by compressing a plurality of state vectors related to a plurality of times of one target machine 10. In other words, the compression vector group is a time series of compression vectors related to one target machine 10.

The clustering unit 2105 divides the set of the compression vector group into a plurality of clusters based on the plurality of compression vector groups related to the plurality of target machines 10. As the clustering method, a k-means method, a shortest distance method, a longest distance method, a group average method, Ward's method, or another clustering method can be used. The similarity between compressed vector groups can be obtained by, for example, dynamic programming. The clustering unit 2105 stores the compression vector group related to the representative points of each cluster in the cluster storage unit 2203.

The classification unit 2106 classifies the compression vector group related to one target machine 10 into one cluster based on the information stored in the cluster storage unit 2203. For example, the classification unit 2106 obtains the distances between the compression vector group related to one target machine 10 and the plurality of representative points, and classifies the compression vector group into the clusters related to the representative points having the shortest distance.

The parts information acquisition unit 2107 acquires parts information indicating the parts replacement date and time and the parts replacement amount of the plurality of target machines 10. The component information may be stored in, for example, an external database, or may be stored in advance in the storage 23. The component information is an example of information related to the failure of the target machine 10.

The heat map generation unit 2108 generates a heat map showing the relationship between the state of the target machine 10 and the replacement amount of parts based on a plurality of compression vectors related to a plurality of times of the plurality of target machines 10 and component information. .. The heat map is a diagram in which each region obtained by dividing a two-dimensional plane defined by an axis related to the dimension of the compression vector by a grid is colored with a color representing the average component replacement amount in the state belonging to the region. In another embodiment, the heat map generation unit 2108 may generate a heat map representing parameters related to other failures, such as a heat map showing the probability of occurrence of a component failure. Further, in another embodiment, the heat map generation unit 2108 may generate a heat map in which a region divided by a honeycomb grid is arranged, or may generate a heat map in which a region is arranged on a spherical surface. good. The heat map generation unit 2108 stores the generated heat map in the heat map storage unit 2204.

The graph generation unit 2109 generates a graph in which a plot group representing a compression vector group related to one target machine 10 is attached to a two-dimensional plane defined by an axis related to the dimension of the compression vector. As will be described later, in the graph, the positions of the plurality of plots change in chronological order from the positions of the corners of the two-dimensional plane to the positions of the diagonal corners with a substantially monotonous change.

The output unit 2110 outputs a cluster to which the usage history of one target machine 10 belongs. Further, the output unit 2110 outputs a diagram in which the heat map generated by the heat map generation unit 2108 and the graph generated by the graph generation unit 2109 are superimposed. The output unit 2110 may display these information on a display (see FIG. 1) connected via the interface 24, or may transmit the information to an external terminal device via the communication device 25. Alternatively, it may be stored in an external storage medium via the interface 24.

《Learning method》
The state analyzer 20 generates a compression model, information on representative points of a plurality of clusters, and a heat map in advance before executing the analysis process of the usage history of one target machine 10.
FIG. 4 is a flowchart showing a learning method by the state analyzer according to the first embodiment.

The state information acquisition unit 2101 of the state analysis device 20 receives a time series of state information representing the usage history of the target machine 10 from each of the plurality of target machines 10 (step S1). That is, the state information acquisition unit 2101 acquires state information related to a plurality of items associated with a plurality of times from the plurality of target machines 10. The state information acquisition unit 2101 stores the received state information in the state storage unit 2201 in association with the ID of the target machine 10 and the time (step S2).

Next, the vector acquisition unit 2102 acquires a plurality of state vectors related to the state information associated with each primary key from the model storage unit 2202 (step S3). That is, the vector acquisition unit 2102 generates a state vector from the model storage unit 2202 for each ID and the state information associated with each time.

Next, the model generation unit 2103 trains the compression model based on the plurality of state vectors acquired by the vector acquisition unit 2102 (step S4), and stores the trained compression model in the model storage unit 2202 (step S5). ). Specifically, the model generation unit 2103 trains the compression model by the following procedure.
The model generation unit 2103 arranges a plurality of classes evenly on a two-dimensional plane. Each class has a weight vector (with the same number of dimensions as the number of items in the state information) that is random or specified by preprocessing such as principal component analysis of the state vector. A set of classes is an example of a compression model. The model generation unit 2103 randomly selects one state vector from a plurality of state vectors, and selects a weight vector related to the class closest to the selected state vector among the plurality of classes and the class in the vicinity thereof. Get closer to the value of the state vector. By repeatedly executing this process, the compression model can be trained.

Next, the vector compression unit 2104 selects a plurality of target machines 10 one by one, and executes the processes of the following steps S7 to S8 (step S6).
The vector compression unit 2104 identifies a plurality of state vectors acquired by the vector acquisition unit 2102 that are associated with the ID of the target machine 10 selected in step S6 (step S7). That is, the vector compression unit 2104 specifies a plurality of state vectors related to a plurality of times of one target machine 10. The vector compression unit 2104 compresses each of the specified plurality of state vectors based on the compression model to obtain a compression vector group (time series of compression vectors) representing the usage history of one target machine 10 (step S8). ..

When the vector compression unit 2104 obtains a compression vector group related to each of the plurality of target machines 10, the clustering unit 2105 divides the plurality of compression vector groups into a plurality of clusters (step S9). The clustering unit 2105 identifies a compression vector group related to the representative points of each of the plurality of clusters, associates the cluster ID with the compressed vector group related to the representative points, and stores them in the cluster storage unit 2203 (step S10).

Further, the component information acquisition unit 2107 acquires component information indicating the component replacement date and time and the component replacement amount of the plurality of target machines 10 (step S11). The heat map generation unit 2108 generates an initial value of the heat map by arranging each class of the compression model learned in step S4 on a two-dimensional plane (step S12). The value of the class on the heat map is the average value of the parts replacement amount.

Next, the heat map generation unit 2108 selects a combination of the component replacement date and time and the component replacement amount included in the component information one by one, and executes the processes of steps S14 to S15 below (step S13).
Based on the compression model, the heat map generation unit 2108 identifies a class on the heat map to which the compression vector group related to the component replacement date and time related to the selected combination belongs (step S14). The heat map generation unit 2108 updates the value of the specified class based on the component replacement amount related to the selected combination (step S15).

When the heat map generation unit 2108 sets a value in the heat map class based on the component information, each class is colored in a color corresponding to the average value of the component replacement amount related to the class (step S16). For example, the heat map generation unit 2108 may lower the brightness of the class in which the average value of the parts replacement amount is low and increase the brightness of the class in which the average value of the parts replacement amount is high. Further, for example, the heat map generation unit 2108 may bring the hue of the class having a low average value of the parts replacement amount closer to blue and the hue of the class having a high average value of the parts replacement amount closer to red. The heat map generation unit 2108 stores the generated heat map in the heat map storage unit 2204 (step S17).

By the above procedure, the state analyzer 20 can store the compression model, the information of the representative points of the plurality of clusters, and the heat map in the main memory 22.

<< State analysis method >>
When the above preparation is completed, the state analysis device 20 can analyze the state of any target machine 10.
FIG. 5 is a flowchart showing a state analysis method by the state analysis device according to the first embodiment.

The state information acquisition unit 2101 of the state analysis device 20 receives a time series of state information from one target machine 10 (step S51). That is, the state information acquisition unit 2101 acquires state information related to a plurality of items associated with a plurality of times from one target machine 10. Next, the vector acquisition unit 2102 acquires a plurality of state vectors based on the received state information (step S52). That is, the vector acquisition unit 2102 generates a state vector for each state information associated with each time of one target machine 10.

Next, the vector compression unit 2104 compresses each of the generated state vectors based on the compression model stored in the model storage unit 2202 to obtain a compression vector group representing the usage history of one target machine 10. (Step S53). Next, the classification unit 2106 classifies the target machine 10 into one cluster based on the distance between the representative points of each cluster stored in the cluster storage unit 2203 and the obtained compression vector group (step S54).

Next, the graph generation unit 2109 attaches a plot group representing the compression vector group obtained in step S53 to the two-dimensional plane defined by the axis related to the dimension of the compression vector, and draws each plot in a straight line in chronological order. Generate a connected graph (step S55). Further, the graph generation unit 2109 generates a graph representing the representative points of the clusters classified in step S54 (step S56).
The output unit 2110 reads the heat map from the heat map storage unit 2204, and generates a composite graph in which the heat map, the generated graph related to the target machine 10, and the graph representing the cluster center are superimposed (step S57). The output unit 2110 outputs a composite graph (step S58).

FIG. 6 is a diagram showing an example of a screen output by the state analyzer according to the first embodiment.
As shown in FIG. 6, among the composite graphs G, the graph G1 related to the target machine 10 extends substantially monotonously from the upper right to the lower left in chronological order. That is, the user identifies the lower left point of the graph G1 related to the target machine 10 and confirms the color of the heat map H related to the point, whereby the replacement amount of the parts in the current state of the target machine 10 is calculated. The average value can be recognized. Further, since the graph G2 representing the cluster center is drawn together with the graph G1 related to the target machine 10, the user recognizes the extending direction of the graph G2 representing the cluster center, so that the graph G1 related to the target machine 10 grows. The direction can be predicted and the future state of the target machine 10 can be predicted. For example, in the example shown in FIG. 6, the graph G2 representing the center of the cluster is expected to extend to the left, and the graph G1 related to the target machine 10 is also expected to extend to the left. As a result, the user can recognize that the target machine 10 may reach a state where the average value of the parts replacement amount appears high in the future, and the operation method of the target machine 10 may be changed or in advance. You can take measures such as performing maintenance.

《Action / Effect》
As described above, according to the first embodiment, the state analyzer 20 acquires a state vector at each of a plurality of time points of the target machine 10 and compresses the dimensions of the plurality of state vectors to compress the plurality of compression vectors. To get. The element of the state vector represents a quantity related to a plurality of items indicating a state that changes monotonically with time with the operation of the target machine 10. As a result, the state analyzer 20 can compress the dimension of the data related to the history of use of the target machine 10 so as not to lose the time-series relationship.

Here, by using a time series of a state vector having quantities related to a plurality of items indicating a state that changes monotonically with time with the operation of the target machine 10, the time series is related even after the dimension reduction. Explain why you can keep your sex.
When the first element (for example, SMR) having a state vector increases monotonically with time, the value of the first element becomes smaller as the target machine 10 is newer and larger as the target machine 10 becomes older. That is, the first element has a relation of monotonically increasing in the time series of the state vector related to one target machine 10.
On the other hand, when the second element (for example, the remaining life) of the state vector decreases monotonically with time, the value of the second element becomes larger as the target machine 10 is newer and smaller as the target machine 10 becomes older. That is, the second element has a relation of monotonically decreasing in the time series of the state vector related to one target machine 10.

Here, a dimension compression algorithm such as SOM is configured to compress dimensions so as to maintain information as much as possible. Therefore, when each element of the state vector has a relation of monotonically changing with respect to the operating time of the target machine 10 (especially when all the elements of the state vector have a relation of monotonically changing with respect to the operating time). ,) Dimensional compression is done so that the relationship is maintained. For example, in SOM, the compression vector related to the relatively new target machine 10 and the compression vector related to the relatively old target machine 10 are dimensionally compressed so as to be located diagonally to each other in the state space. As a result, the compression vector group obtained by compressing the time series of the state vectors for one target machine 10 changes substantially monotonously in the time series.
The state analyzer 20 according to the first embodiment compresses the dimension of the state vector by SOM. According to SOM, since the dimensional compression is performed while preserving the phase relationship between the vectors, the continuity of the time series in the compressed vector group can be well maintained. In particular, when the element of the state vector contains a non-linear value, it is preferable to use a non-linear dimensional compression method such as SOM rather than a linear dimensional compression method such as principal component analysis.
Further, in other embodiments, it is not always necessary that all the elements of the state vector have a relationship of monotonically changing with respect to the operating time. For example, in other embodiments, the state vector may be a vector dominated by elements having a relationship of monotonically changing with respect to the operating time.

In the first embodiment, the state analyzer 20 compresses the state vector into two dimensions for the output of the graph and the heat map, but in other embodiments, the state vector has three or more dimensions. It may be compressed to a number.

Further, according to the first embodiment, the state vector includes the cumulative value of vital data related to the operation of the target machine 10 as an element. By calculating the cumulative value of vital data, the data aggregation device 13 of the target machine 10 can easily obtain an amount indicating a state of monotonously changing with time as the target machine 10 operates. In another embodiment, the state vector may not necessarily include the cumulative value of vital data as an element.

Further, according to the first embodiment, the state analyzer 20 outputs a graph in which plot groups representing a plurality of compression vectors related to a plurality of time points of the target machine 10 are attached on a two-dimensional plane. As a result, the user can easily recognize the usage history and the current state of the target machine 10. The state analyzer 20 according to the first embodiment attaches a plot group on a two-dimensional plane, but the present invention is not limited to this. In another embodiment, the state analyzer 20 may output a three-dimensional graph with a group of plots on the three-dimensional space.

Further, according to the first embodiment, the state analyzer 20 classifies the states of the target machine 10 based on the compression vector group of the plurality of target machines 10. Thereby, the user can understand the tendency of the past usage of the target machine 10 and the tendency of the future state change by recognizing the classification result of one target machine 10. The state analyzer 20 according to another embodiment does not necessarily have to classify the state of the target machine 10 based on the compression vector group.
FIG. 7 is a graph showing a compression vector group classified into one cluster by the state analyzer according to the first embodiment. As shown in FIG. 7, it can be seen that the compression vector groups of the target machines 10 classified into one cluster all draw similar trajectories. From this, it can be seen that the state analyzer 20 can appropriately classify the usage of the target machine 10 by clustering the compression vector group.

Further, according to the first embodiment, the state analyzer 20 generates a heat map in which a plurality of regions colored with a color representing the average value of the parts replacement amount of the target machine 10 are arranged, and one target. The plot group related to the machine 10 and the heat map are superimposed and output. As a result, the user can recognize the risk of current or future parts replacement of the target machine 10. The state analyzer 20 according to another embodiment does not necessarily have to output a heat map.

<Other embodiments>
Although one embodiment has been described in detail with reference to the drawings, the specific configuration is not limited to the above-mentioned one, and various design changes and the like can be made.

In the above-described embodiment, the data aggregation device 13 of the target machine 10 calculates the cumulative value of the measured values of the sensor 12 and transmits it to the state analysis device 20, but the present invention is not limited to this. For example, in another embodiment, when all the measured values of the sensor 12 change monotonically with time, the data aggregation device 13 may transmit the measured values of the sensor 12 to the state analysis device 20 as they are. good. In another embodiment, the data aggregation device 13 may transmit the measured value of the sensor 12 to the state analysis device 20 as it is, and the state analysis device 20 may calculate the cumulative value of the measured value. Further, even if not all the measured values of the sensor 12 change monotonically with time, the data aggregation device 13 transmits the measured values of the sensor 12 to the state analysis device 20 as they are, and the state analysis device 20 May calculate the cumulative value. Alternatively, the data aggregation device 13 calculates the cumulative value of the measured value of the sensor 12 for one day and sends it to the state analysis device 20, and the state analysis device 20 accumulates the cumulative value for one day for one month. Both the data aggregation device 13 and the state analysis device 20 may share the calculation of the cumulative value.

In the above-described embodiment, the state analyzer 20 performs both learning of the compression model and dimensional compression based on the trained compression model, but is not limited to this. For example, in another embodiment, the compression model is trained by the server device, the trained compression model is stored in a device (for example, a data aggregation device 13) included in the target machine 10, and the dimension of the state vector is stored in the target machine 10. At least one of compression and display of graphs and heatmaps may be made. That is, the state analysis device 20 may be provided in the target machine 10.

1 ... State analysis system 10 ... Target machine 11 ... Machine body 12 ... Sensor 13 ... Data aggregation device 14 ... Communication device 131 ... Processor 132 ... Main memory 133 ... Storage 134 ... Interface 1311 ... Measurement value acquisition unit 1312 ... Cumulative calculation unit 1313 ... Status recording unit 1314 ... Transmission unit 1321 ... State storage unit 20 ... State analysis device 21 ... Processor 22 ... Main memory 23 ... Storage 24 ... Interface 25 ... Communication device 2101 ... Status information acquisition unit 2102 ... Vector acquisition unit 2103 ... Model generation Unit 2104 ... Vector compression unit 2105 ... Clustering unit 2106 ... Classification unit 2107 ... Part information acquisition unit 2108 ... Heat map generation unit 2109 ... Graph generation unit 2110 ... Output unit 2201 ... State storage unit 2202 ... Model storage unit 2203 ... Cluster storage unit 2204 ... Heat map storage unit

Claims (7)

A vector acquisition unit that acquires a plurality of state vectors representing quantities related to a plurality of items indicating the state of the machine, which changes monotonically with time with the operation of the machine at each of a plurality of time points.
A state analyzer including a vector compression unit for obtaining a plurality of compression vectors obtained by compressing the dimensions of the plurality of state vectors. The state analyzer according to claim 1, wherein the plurality of state vectors are vectors including cumulative values of data relating to a plurality of items generated by the operation of the machine as elements. The dimension of the compression vector is two-dimensional or three-dimensional,
A graph generator that generates a graph including a plurality of axes related to the dimensions of the compression vector and a plot group representing a plurality of compression vectors related to a plurality of time points of one machine.
The state analyzer according to claim 1 or 2, further comprising an output unit that outputs the graph. A model generation unit for generating a compression model for compressing the state vector based on a plurality of state vectors related to a plurality of time points of the plurality of machines is provided.
The state analyzer according to any one of claims 1 to 3, wherein the vector compression unit compresses a plurality of state vectors related to a plurality of time points of one machine based on the compression model. The state analyzer according to claim 4, further comprising a classification unit for classifying the states of the machine based on the plurality of compression vectors. The dimension of the compression vector is two-dimensional,
Generates a heat map in which a plurality of regions representing the state of the machine and a plurality of regions colored with colors representing parameters related to the failure in the state are arranged on a plane defined by an axis related to the dimension. Heat map generator and
The state analyzer according to claim 4 or 5, further comprising a plot group representing a plurality of compression vectors representing a plurality of compression vectors related to a plurality of time points of one machine and an output unit for outputting the heat map. A step of acquiring a plurality of state vectors representing quantities related to a plurality of items indicating the state of the machine, which changes monotonically with time with the operation of the machine at each of a plurality of time points of the plurality of machines.
A state analysis method comprising the steps of obtaining a plurality of compressed vectors obtained by compressing the dimensions of the plurality of state vectors.

Images